From 5cbd3f29e354b221bffa639efeed491ea6036d5d Mon Sep 17 00:00:00 2001 From: wehub-resource-sync Date: Mon, 13 Jul 2026 12:41:19 +0800 Subject: [PATCH] chore: import upstream snapshot with attribution --- .agents/skills/add-function-body/SKILL.md | 120 + .agents/skills/add-op/SKILL.md | 99 + .../add-op/references/node-test-pattern.md | 39 + .../references/reference-impl-pattern.md | 56 + .agents/skills/add-shape-inference/SKILL.md | 140 + .agents/skills/onnxtxt/SKILL.md | 121 + .clang-format | 94 + .clang-tidy | 86 + .claude/instructions/add-function-body.md | 13 + .claude/instructions/add-op.md | 12 + .claude/instructions/add-shape-inference.md | 14 + .editorconfig | 7 + .git-blame-ignore-revs | 8 + .gitattributes | 7 + .github/ISSUE_TEMPLATE/bug.md | 42 + .github/ISSUE_TEMPLATE/config.yml | 14 + .github/ISSUE_TEMPLATE/feature_request.yml | 71 + .github/ISSUE_TEMPLATE/operator.md | 24 + .github/ISSUE_TEMPLATE/question.md | 24 + .github/copilot-instructions.md | 17 + .github/dependabot.yml | 40 + .github/install_test.yml | 44 + .../instructions/onnx-defs.instructions.md | 28 + .github/pull_request_template.md | 5 + .github/release.yml | 43 + .github/workflows/check_pr_label.yml | 43 + .github/workflows/check_urls.yml | 47 + .github/workflows/clang-tidy.yml | 40 + .github/workflows/codeql.yml | 88 + .github/workflows/copilot-setup-steps.yml | 47 + .github/workflows/create_release.yml | 437 + .github/workflows/dco_merge_group.yml | 17 + .github/workflows/dependency-review.yml | 22 + .github/workflows/fuzz.yml | 129 + .github/workflows/lint.yml | 117 + .github/workflows/main.yml | 280 + .github/workflows/pages.yml | 55 + .github/workflows/pixi_build.yml | 124 + .../workflows/preview_source_dist_test.yml | 43 + .github/workflows/release_linux_cibw.yml | 101 + .github/workflows/release_macos_cibw.yml | 97 + .github/workflows/release_pyodide_cibw.yml | 83 + .github/workflows/release_sdist.yml | 66 + .github/workflows/release_windows_cibw.yml | 108 + .github/workflows/scorecard.yml | 48 + .github/workflows/stale.yml | 38 + .github/workflows/win_no_exception_ci.yml | 81 + .gitignore | 118 + .gitmodules | 0 .lintrunner.toml | 221 + .lycheeignore | 15 + .pre-commit-config.yaml | 102 + CLAUDE.md | 81 + CMakeLists.txt | 681 + CODEOWNERS | 18 + CODE_OF_CONDUCT.md | 1 + CONTRIBUTING.md | 271 + INSTALL.md | 261 + LICENSE | 202 + LICENSES/Apache-2.0.txt | 73 + LICENSES/BSD-2-Clause.txt | 9 + NOTICE | 59 + README.md | 141 + README.wehub.md | 7 + RELEASE-MANAGEMENT.md | 129 + REUSE.toml | 236 + ROADMAP.md | 141 + SECURITY.md | 73 + VERSION_NUMBER | 1 + cmake/ONNXConfig.cmake.in | 12 + cmake/ONNXConfigVersion.cmake.in | 11 + cmake/Utils.cmake | 208 + cmake/external/FindSanitizer.cmake | 166 + cmake/external/googletest.cmake | 14 + cmake/summary.cmake | 90 + cmake/unittest.cmake | 39 + codecov.yml | 8 + community/logo_request.md | 16 + community/readme.md | 168 + community/repo_guidelines.md | 40 + community/sc-election-guidelines.md | 71 + community/sigs.md | 19 + community/working-groups.md | 29 + docs/AddFunctionBody.md | 263 + docs/AddNewOp.md | 190 + docs/AssuranceCase.md | 111 + docs/Broadcasting.md | 79 + docs/CIPipelines.md | 59 + docs/Changelog-ml.md | 1349 + docs/Changelog.md | 33668 ++++++++++++ docs/DefineDifferentiability.md | 126 + docs/DimensionDenotation.md | 56 + docs/ExternalData.md | 101 + docs/ExternalDataSecurity.md | 163 + docs/IR.md | 628 + docs/ImplementingAnOnnxBackend.md | 127 + docs/ManagingExperimentalOps.md | 42 + docs/MetadataProps.md | 35 + docs/ONNXTypes.md | 105 + docs/OnnxBackendTest.md | 24 + docs/OnnxReleases.md | 182 + docs/OpConventions.md | 17 + docs/Operators-ml.md | 1214 + docs/Operators.md | 44256 ++++++++++++++++ docs/Overview.md | 23 + docs/PythonAPIOverview.md | 503 + docs/ReleaseVerification.md | 39 + docs/Relicensing.md | 15 + docs/ShapeInference.md | 197 + docs/Syntax.md | 97 + docs/TestCoverage-ml.md | 350 + docs/TestCoverage.md | 32001 +++++++++++ docs/TypeDenotation.md | 62 + docs/VersionConverter.md | 55 + docs/Versioning.md | 205 + docs/docsgen/Makefile | 24 + docs/docsgen/make.bat | 39 + docs/docsgen/source/_static/ONNX-Logo.svg | 83 + docs/docsgen/source/_static/css/custom.css | 36 + .../source/_static/diff2html-ui-slim.min.js | 1 + docs/docsgen/source/_static/diff2html.min.css | 1 + .../source/_static/onnx-horizontal-color.png | Bin 0 -> 70339 bytes .../source/_static/onnx-horizontal-white.png | Bin 0 -> 46498 bytes docs/docsgen/source/_static/readme.txt | 16 + docs/docsgen/source/api/backend.md | 35 + docs/docsgen/source/api/checker.md | 17 + docs/docsgen/source/api/classes.md | 294 + docs/docsgen/source/api/compose.md | 44 + docs/docsgen/source/api/defs.md | 62 + .../source/api/external_data_helper.md | 61 + docs/docsgen/source/api/helper.md | 163 + docs/docsgen/source/api/index.md | 71 + docs/docsgen/source/api/inliner.md | 13 + docs/docsgen/source/api/model_container.md | 20 + docs/docsgen/source/api/numpy_helper.md | 63 + docs/docsgen/source/api/parser.md | 25 + docs/docsgen/source/api/printer.md | 7 + docs/docsgen/source/api/reference.md | 45 + docs/docsgen/source/api/serialization.md | 119 + docs/docsgen/source/api/shape_inference.md | 25 + docs/docsgen/source/api/tools.md | 14 + docs/docsgen/source/api/utils.md | 14 + docs/docsgen/source/api/version_converter.md | 7 + docs/docsgen/source/conf.py | 107 + docs/docsgen/source/expect_onnxruntime.md | 75 + docs/docsgen/source/index.md | 19 + docs/docsgen/source/intro/concepts.md | 364 + docs/docsgen/source/intro/converters.md | 317 + docs/docsgen/source/intro/images/dot_att.png | Bin 0 -> 35875 bytes docs/docsgen/source/intro/images/dot_if.png | Bin 0 -> 94683 bytes .../docsgen/source/intro/images/dot_if_py.png | Bin 0 -> 77079 bytes .../source/intro/images/dot_linreg.png | Bin 0 -> 25490 bytes .../source/intro/images/dot_linreg2.png | Bin 0 -> 27739 bytes docs/docsgen/source/intro/images/dot_scan.png | Bin 0 -> 121455 bytes .../source/intro/images/dot_scan_py.png | Bin 0 -> 54026 bytes docs/docsgen/source/intro/images/iff.png | Bin 0 -> 42935 bytes docs/docsgen/source/intro/images/linreg1.png | Bin 0 -> 7832 bytes docs/docsgen/source/intro/images/linreg2.png | Bin 0 -> 27462 bytes docs/docsgen/source/intro/images/scanop.png | Bin 0 -> 38876 bytes docs/docsgen/source/intro/index.md | 15 + docs/docsgen/source/intro/python.md | 1601 + docs/docsgen/source/onnx-favicon.png | Bin 0 -> 7259 bytes docs/docsgen/source/onnx_sphinx.py | 905 + docs/docsgen/source/repo-docs/.gitignore | 3 + docs/docsgen/source/repo-docs/index.md | 16 + .../source/technical/InPlaceKVCache.png | Bin 0 -> 86992 bytes docs/docsgen/source/technical/float4.md | 121 + docs/docsgen/source/technical/float8.md | 232 + docs/docsgen/source/technical/index.md | 21 + docs/docsgen/source/technical/int2.md | 43 + docs/docsgen/source/technical/int4.md | 58 + docs/docsgen/source/technical/kv_cache.md | 25 + docs/images/onnx_hub_arch.svg | 1 + docs/onnx-horizontal-color.png | Bin 0 -> 70339 bytes docs/proposals/0000-template.md | 101 + .../0001-ArchiveFileFormatProposal.md | 47 + docs/proposals/0002-NLP-in-ONNX-proposal.md | 77 + docs/proposals/0003-ONNXIFIproposal.md | 127 + docs/proposals/0004-FunctionsProposal.md | 33 + .../0005-SymbolicShapeInfProposal.md | 178 + .../proposals/0006-ONNXMultiDeviceProposal.md | 191 + docs/proposals/0007-ShardingFormalism.md | 269 + docs/proposals/README.md | 34 + .../images/composing_broadcast_axes.png | Bin 0 -> 6903 bytes onnx/CMakeLists.txt | 21 + onnx/__init__.py | 470 + onnx/_mapping.py | 119 + onnx/backend/__init__.py | 3 + onnx/backend/base.py | 148 + onnx/backend/sample/__init__.py | 3 + onnx/backend/sample/ops/__init__.py | 30 + onnx/backend/sample/ops/abs.py | 8 + onnx/backend/test/__init__.py | 8 + onnx/backend/test/case/__init__.py | 14 + onnx/backend/test/case/base.py | 47 + onnx/backend/test/case/model/__init__.py | 81 + onnx/backend/test/case/model/expand.py | 91 + onnx/backend/test/case/model/gradient.py | 110 + onnx/backend/test/case/model/sequence.py | 456 + onnx/backend/test/case/model/shrink.py | 42 + onnx/backend/test/case/model/sign.py | 36 + onnx/backend/test/case/model/single_relu.py | 36 + .../test/case/model/stringnormalizer.py | 205 + onnx/backend/test/case/node/__init__.py | 474 + .../test/case/node/_image_decoder_data.py | 6353 +++ onnx/backend/test/case/node/abs.py | 24 + onnx/backend/test/case/node/acos.py | 28 + onnx/backend/test/case/node/acosh.py | 28 + onnx/backend/test/case/node/adagrad.py | 116 + onnx/backend/test/case/node/adam.py | 135 + onnx/backend/test/case/node/add.py | 60 + onnx/backend/test/case/node/affinegrid.py | 208 + .../test/case/node/ai_onnx_ml/__init__.py | 0 .../ai_onnx_ml/array_feature_extractor.py | 31 + .../test/case/node/ai_onnx_ml/binarizer.py | 28 + .../case/node/ai_onnx_ml/label_encoder.py | 101 + .../case/node/ai_onnx_ml/tree_ensemble.py | 123 + onnx/backend/test/case/node/and_.py | 76 + onnx/backend/test/case/node/argmax.py | 256 + onnx/backend/test/case/node/argmin.py | 256 + onnx/backend/test/case/node/asin.py | 28 + onnx/backend/test/case/node/asinh.py | 28 + onnx/backend/test/case/node/atan.py | 28 + onnx/backend/test/case/node/atanh.py | 28 + onnx/backend/test/case/node/attention.py | 2699 + onnx/backend/test/case/node/averagepool.py | 695 + onnx/backend/test/case/node/batchnorm.py | 137 + onnx/backend/test/case/node/bernoulli.py | 59 + onnx/backend/test/case/node/bitcast.py | 147 + onnx/backend/test/case/node/bitshift.py | 100 + onnx/backend/test/case/node/bitwiseand.py | 55 + onnx/backend/test/case/node/bitwisenot.py | 36 + onnx/backend/test/case/node/bitwiseor.py | 52 + onnx/backend/test/case/node/bitwisexor.py | 55 + onnx/backend/test/case/node/blackmanwindow.py | 56 + onnx/backend/test/case/node/cast.py | 398 + onnx/backend/test/case/node/castlike.py | 324 + .../test/case/node/causal_conv_with_state.py | 335 + onnx/backend/test/case/node/ceil.py | 28 + onnx/backend/test/case/node/celu.py | 97 + .../backend/test/case/node/center_crop_pad.py | 130 + onnx/backend/test/case/node/clip.py | 144 + onnx/backend/test/case/node/col2im.py | 349 + onnx/backend/test/case/node/compress.py | 99 + onnx/backend/test/case/node/concat.py | 56 + onnx/backend/test/case/node/constant.py | 29 + .../backend/test/case/node/constantofshape.py | 64 + onnx/backend/test/case/node/conv.py | 257 + onnx/backend/test/case/node/convinteger.py | 103 + onnx/backend/test/case/node/convtranspose.py | 532 + onnx/backend/test/case/node/cos.py | 28 + onnx/backend/test/case/node/cosh.py | 28 + onnx/backend/test/case/node/cumprod.py | 129 + onnx/backend/test/case/node/cumsum.py | 112 + onnx/backend/test/case/node/deformconv.py | 161 + onnx/backend/test/case/node/depthtospace.py | 98 + .../test/case/node/dequantizelinear.py | 373 + onnx/backend/test/case/node/det.py | 38 + onnx/backend/test/case/node/dft.py | 152 + onnx/backend/test/case/node/div.py | 78 + onnx/backend/test/case/node/dropout.py | 209 + .../test/case/node/dynamicquantizelinear.py | 70 + onnx/backend/test/case/node/einsum.py | 92 + onnx/backend/test/case/node/elu.py | 37 + onnx/backend/test/case/node/equal.py | 92 + onnx/backend/test/case/node/erf.py | 26 + onnx/backend/test/case/node/exp.py | 28 + onnx/backend/test/case/node/expand.py | 66 + onnx/backend/test/case/node/eyelike.py | 60 + onnx/backend/test/case/node/flatten.py | 65 + onnx/backend/test/case/node/flexattention.py | 583 + onnx/backend/test/case/node/floor.py | 28 + onnx/backend/test/case/node/gather.py | 91 + onnx/backend/test/case/node/gatherelements.py | 92 + onnx/backend/test/case/node/gathernd.py | 121 + onnx/backend/test/case/node/gelu.py | 52 + onnx/backend/test/case/node/gemm.py | 157 + .../test/case/node/globalaveragepool.py | 44 + onnx/backend/test/case/node/globalmaxpool.py | 44 + onnx/backend/test/case/node/greater.py | 68 + onnx/backend/test/case/node/greater_equal.py | 68 + onnx/backend/test/case/node/gridsample.py | 642 + .../test/case/node/groupnormalization.py | 78 + onnx/backend/test/case/node/gru.py | 264 + onnx/backend/test/case/node/hammingwindow.py | 48 + onnx/backend/test/case/node/hannwindow.py | 45 + onnx/backend/test/case/node/hardmax.py | 92 + onnx/backend/test/case/node/hardsigmoid.py | 39 + onnx/backend/test/case/node/hardswish.py | 30 + onnx/backend/test/case/node/identity.py | 93 + onnx/backend/test/case/node/if_.py | 210 + onnx/backend/test/case/node/image_decoder.py | 249 + onnx/backend/test/case/node/instancenorm.py | 58 + onnx/backend/test/case/node/isinf.py | 56 + onnx/backend/test/case/node/isnan.py | 36 + .../test/case/node/layernormalization.py | 178 + onnx/backend/test/case/node/leakyrelu.py | 39 + onnx/backend/test/case/node/less.py | 68 + onnx/backend/test/case/node/less_equal.py | 68 + .../test/case/node/linear_attention.py | 541 + onnx/backend/test/case/node/log.py | 28 + onnx/backend/test/case/node/logsoftmax.py | 92 + onnx/backend/test/case/node/loop.py | 459 + .../backend/test/case/node/lpnormalization.py | 80 + onnx/backend/test/case/node/lppool.py | 298 + onnx/backend/test/case/node/lrn.py | 70 + onnx/backend/test/case/node/lstm.py | 278 + onnx/backend/test/case/node/matmul.py | 62 + onnx/backend/test/case/node/matmulinteger.py | 60 + onnx/backend/test/case/node/max.py | 66 + onnx/backend/test/case/node/maxpool.py | 725 + onnx/backend/test/case/node/maxunpool.py | 67 + onnx/backend/test/case/node/mean.py | 47 + .../case/node/meanvariancenormalization.py | 49 + .../backend/test/case/node/melweightmatrix.py | 90 + onnx/backend/test/case/node/min.py | 66 + onnx/backend/test/case/node/mish.py | 23 + onnx/backend/test/case/node/mod.py | 177 + onnx/backend/test/case/node/momentum.py | 162 + onnx/backend/test/case/node/mul.py | 73 + onnx/backend/test/case/node/neg.py | 28 + .../case/node/negativeloglikelihoodloss.py | 585 + .../test/case/node/nonmaxsuppression.py | 475 + onnx/backend/test/case/node/nonzero.py | 26 + onnx/backend/test/case/node/not_.py | 32 + onnx/backend/test/case/node/onehot.py | 159 + .../test/case/node/optionalgetelement.py | 80 + .../test/case/node/optionalhaselement.py | 93 + onnx/backend/test/case/node/or_.py | 76 + onnx/backend/test/case/node/pad.py | 129 + onnx/backend/test/case/node/pow.py | 106 + onnx/backend/test/case/node/prelu.py | 40 + onnx/backend/test/case/node/qlinearconv.py | 82 + onnx/backend/test/case/node/qlinearmatmul.py | 149 + onnx/backend/test/case/node/quantizelinear.py | 499 + onnx/backend/test/case/node/rangeop.py | 102 + onnx/backend/test/case/node/reciprocal.py | 28 + onnx/backend/test/case/node/reduce_log_sum.py | 104 + .../test/case/node/reduce_log_sum_exp.py | 193 + onnx/backend/test/case/node/reducel1.py | 189 + onnx/backend/test/case/node/reducel2.py | 207 + onnx/backend/test/case/node/reducemax.py | 231 + onnx/backend/test/case/node/reducemean.py | 174 + onnx/backend/test/case/node/reducemin.py | 234 + onnx/backend/test/case/node/reduceprod.py | 187 + onnx/backend/test/case/node/reducesum.py | 247 + .../backend/test/case/node/reducesumsquare.py | 194 + .../test/case/node/regex_full_match.py | 68 + onnx/backend/test/case/node/relu.py | 24 + onnx/backend/test/case/node/reshape.py | 82 + onnx/backend/test/case/node/resize.py | 1714 + .../backend/test/case/node/reversesequence.py | 86 + .../test/case/node/rmsnormalization.py | 126 + onnx/backend/test/case/node/rnn.py | 220 + onnx/backend/test/case/node/roialign.py | 446 + .../backend/test/case/node/rotaryembedding.py | 231 + onnx/backend/test/case/node/round.py | 62 + onnx/backend/test/case/node/scan.py | 155 + onnx/backend/test/case/node/scatter.py | 109 + .../backend/test/case/node/scatterelements.py | 266 + onnx/backend/test/case/node/scatternd.py | 271 + onnx/backend/test/case/node/selu.py | 49 + onnx/backend/test/case/node/sequence_map.py | 304 + onnx/backend/test/case/node/sequenceinsert.py | 75 + onnx/backend/test/case/node/shape.py | 64 + onnx/backend/test/case/node/shrink.py | 37 + onnx/backend/test/case/node/sigmoid.py | 30 + onnx/backend/test/case/node/sign.py | 24 + onnx/backend/test/case/node/sin.py | 28 + onnx/backend/test/case/node/sinh.py | 28 + onnx/backend/test/case/node/size.py | 35 + onnx/backend/test/case/node/slice.py | 178 + onnx/backend/test/case/node/softmax.py | 91 + .../test/case/node/softmaxcrossentropy.py | 1158 + onnx/backend/test/case/node/softplus.py | 30 + onnx/backend/test/case/node/softsign.py | 28 + onnx/backend/test/case/node/spacetodepth.py | 66 + onnx/backend/test/case/node/split.py | 378 + .../backend/test/case/node/splittosequence.py | 80 + onnx/backend/test/case/node/sqrt.py | 28 + onnx/backend/test/case/node/squeeze.py | 37 + onnx/backend/test/case/node/stft.py | 70 + onnx/backend/test/case/node/string_concat.py | 69 + onnx/backend/test/case/node/string_split.py | 151 + .../test/case/node/stringnormalizer.py | 148 + onnx/backend/test/case/node/sub.py | 73 + onnx/backend/test/case/node/sum.py | 47 + onnx/backend/test/case/node/swish.py | 36 + onnx/backend/test/case/node/tan.py | 28 + onnx/backend/test/case/node/tanh.py | 28 + onnx/backend/test/case/node/tensorscatter.py | 174 + .../backend/test/case/node/tfidfvectorizer.py | 264 + .../backend/test/case/node/thresholdedrelu.py | 41 + onnx/backend/test/case/node/tile.py | 38 + onnx/backend/test/case/node/topk.py | 262 + onnx/backend/test/case/node/transpose.py | 47 + onnx/backend/test/case/node/trilu.py | 453 + onnx/backend/test/case/node/unique.py | 229 + onnx/backend/test/case/node/unsqueeze.py | 91 + onnx/backend/test/case/node/upsample.py | 58 + onnx/backend/test/case/node/where.py | 42 + onnx/backend/test/case/node/xor.py | 76 + onnx/backend/test/case/test_case.py | 29 + onnx/backend/test/case/utils.py | 45 + onnx/backend/test/cmd_tools.py | 190 + onnx/backend/test/data/light/README.md | 16 + .../test/data/light/light_bvlc_alexnet.onnx | Bin 0 -> 3968 bytes .../data/light/light_bvlc_alexnet_output_0.pb | 1 + .../test/data/light/light_densenet121.onnx | Bin 0 -> 214344 bytes .../data/light/light_densenet121_output_0.pb | 1 + .../test/data/light/light_inception_v1.onnx | Bin 0 -> 36869 bytes .../data/light/light_inception_v1_output_0.pb | 1 + .../test/data/light/light_inception_v2.onnx | Bin 0 -> 159024 bytes .../data/light/light_inception_v2_output_0.pb | 1 + .../test/data/light/light_resnet50.onnx | Bin 0 -> 79770 bytes .../data/light/light_resnet50_output_0.pb | 1 + .../test/data/light/light_shufflenet.onnx | Bin 0 -> 67666 bytes .../data/light/light_shufflenet_output_0.pb | 1 + .../test/data/light/light_squeezenet.onnx | Bin 0 -> 15618 bytes .../data/light/light_squeezenet_output_0.pb | 1 + onnx/backend/test/data/light/light_vgg19.onnx | Bin 0 -> 9311 bytes .../test/data/light/light_vgg19_output_0.pb | 1 + .../test/data/light/light_zfnet512.onnx | Bin 0 -> 4506 bytes .../data/light/light_zfnet512_output_0.pb | 1 + .../test/data/node/test_abs/model.onnx | Bin 0 -> 97 bytes .../node/test_abs/test_data_set_0/input_0.pb | 1 + .../node/test_abs/test_data_set_0/output_0.pb | 1 + .../test/data/node/test_acos/model.onnx | Bin 0 -> 99 bytes .../node/test_acos/test_data_set_0/input_0.pb | 1 + .../test_acos/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_acos_example/model.onnx | Bin 0 -> 91 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | 3 + .../test/data/node/test_acosh/model.onnx | Bin 0 -> 101 bytes .../test_acosh/test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_acosh/test_data_set_0/output_0.pb | 1 + .../data/node/test_acosh_example/model.onnx | Bin 0 -> 93 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_adagrad/model.onnx | Bin 0 -> 320 bytes .../test_adagrad/test_data_set_0/input_0.pb | 1 + .../test_adagrad/test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_adagrad/test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_adagrad/test_data_set_0/input_3.pb | Bin 0 -> 13 bytes .../test_adagrad/test_data_set_0/input_4.pb | Bin 0 -> 13 bytes .../test_adagrad/test_data_set_0/output_0.pb | 1 + .../test_adagrad/test_data_set_0/output_1.pb | 1 + .../node/test_adagrad_multiple/model.onnx | Bin 0 -> 465 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 18 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 18 bytes .../test_data_set_0/input_6.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_7.pb | Bin 0 -> 18 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../test_data_set_0/output_3.pb | 1 + .../test/data/node/test_adam/model.onnx | Bin 0 -> 371 bytes .../node/test_adam/test_data_set_0/input_0.pb | 1 + .../node/test_adam/test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../node/test_adam/test_data_set_0/input_2.pb | 1 + .../node/test_adam/test_data_set_0/input_3.pb | Bin 0 -> 17 bytes .../node/test_adam/test_data_set_0/input_4.pb | 1 + .../node/test_adam/test_data_set_0/input_5.pb | 1 + .../test_adam/test_data_set_0/output_0.pb | 1 + .../test_adam/test_data_set_0/output_1.pb | 1 + .../test_adam/test_data_set_0/output_2.pb | 1 + .../data/node/test_adam_multiple/model.onnx | Bin 0 -> 572 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 18 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 18 bytes .../test_data_set_0/input_6.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_7.pb | Bin 0 -> 18 bytes .../test_data_set_0/input_8.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_9.pb | Bin 0 -> 18 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../test_data_set_0/output_3.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_4.pb | 1 + .../test_data_set_0/output_5.pb | 1 + .../test/data/node/test_add/model.onnx | Bin 0 -> 129 bytes .../node/test_add/test_data_set_0/input_0.pb | 1 + .../node/test_add/test_data_set_0/input_1.pb | 1 + .../node/test_add/test_data_set_0/output_0.pb | Bin 0 -> 256 bytes .../test/data/node/test_add_bcast/model.onnx | Bin 0 -> 127 bytes .../test_add_bcast/test_data_set_0/input_0.pb | 1 + .../test_add_bcast/test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 256 bytes .../test/data/node/test_add_int16/model.onnx | Bin 0 -> 135 bytes .../test_add_int16/test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_add_int16/test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 135 bytes .../test/data/node/test_add_int8/model.onnx | Bin 0 -> 134 bytes .../test_add_int8/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_add_int8/test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_add_int8/test_data_set_0/output_0.pb | 1 + .../test/data/node/test_add_uint16/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 135 bytes .../test/data/node/test_add_uint32/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 256 bytes .../test/data/node/test_add_uint64/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test/data/node/test_add_uint8/model.onnx | Bin 0 -> 135 bytes .../test_add_uint8/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_add_uint8/test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | 3 + .../data/node/test_affine_grid_2d/model.onnx | Bin 0 -> 182 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 44 bytes .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 196 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 44 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 27753 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 44 bytes .../test_data_set_0/output_0.pb | 1 + .../test_affine_grid_2d_expanded/model.onnx | Bin 0 -> 22975 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 44 bytes .../test_data_set_0/output_0.pb | 2 + .../data/node/test_affine_grid_3d/model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 113 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 52 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2901 bytes .../model.onnx | Bin 0 -> 200 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 113 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 52 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2901 bytes .../model.onnx | Bin 0 -> 27757 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 113 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 52 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2901 bytes .../test_affine_grid_3d_expanded/model.onnx | Bin 0 -> 22979 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 113 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 52 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2901 bytes .../model.onnx | 19 + .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 25 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../node/test_ai_onnx_ml_binarizer/model.onnx | Bin 0 -> 164 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 221 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../model.onnx | Bin 0 -> 210 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../model.onnx | Bin 0 -> 288 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../model.onnx | Bin 0 -> 281 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../model.onnx | Bin 0 -> 656 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 107 bytes .../model.onnx | Bin 0 -> 598 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../test/data/node/test_and2d/model.onnx | Bin 0 -> 119 bytes .../test_and2d/test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_and2d/test_data_set_0/input_1.pb | Bin 0 -> 23 bytes .../test_and2d/test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test/data/node/test_and3d/model.onnx | Bin 0 -> 131 bytes .../test_and3d/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_and3d/test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_and3d/test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../test/data/node/test_and4d/model.onnx | Bin 0 -> 143 bytes .../test_and4d/test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../test_and4d/test_data_set_0/input_1.pb | Bin 0 -> 376 bytes .../test_and4d/test_data_set_0/output_0.pb | Bin 0 -> 378 bytes .../data/node/test_and_bcast3v1d/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../data/node/test_and_bcast3v2d/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 31 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../data/node/test_and_bcast4v2d/model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 41 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 378 bytes .../data/node/test_and_bcast4v3d/model.onnx | Bin 0 -> 147 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 378 bytes .../data/node/test_and_bcast4v4d/model.onnx | Bin 0 -> 151 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 105 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 378 bytes .../model.onnx | Bin 0 -> 149 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 194 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 157 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 114 bytes .../model.onnx | Bin 0 -> 201 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 114 bytes .../test_argmax_keepdims_example/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 203 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../test_argmax_keepdims_random/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../model.onnx | Bin 0 -> 210 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../model.onnx | Bin 0 -> 182 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 226 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../model.onnx | Bin 0 -> 233 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../model.onnx | Bin 0 -> 158 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../test_argmax_no_keepdims_random/model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 80 bytes .../model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 80 bytes .../model.onnx | Bin 0 -> 149 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 194 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 157 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 114 bytes .../model.onnx | Bin 0 -> 201 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 114 bytes .../test_argmin_keepdims_example/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 203 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../test_argmin_keepdims_random/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../model.onnx | Bin 0 -> 210 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../model.onnx | Bin 0 -> 182 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 226 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../model.onnx | Bin 0 -> 233 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../model.onnx | Bin 0 -> 158 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../test_argmin_no_keepdims_random/model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 80 bytes .../model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 80 bytes .../test/data/node/test_asin/model.onnx | Bin 0 -> 99 bytes .../node/test_asin/test_data_set_0/input_0.pb | 1 + .../test_asin/test_data_set_0/output_0.pb | 1 + .../data/node/test_asin_example/model.onnx | Bin 0 -> 91 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_asinh/model.onnx | Bin 0 -> 101 bytes .../test_asinh/test_data_set_0/input_0.pb | 1 + .../test_asinh/test_data_set_0/output_0.pb | 1 + .../data/node/test_asinh_example/model.onnx | Bin 0 -> 93 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_atan/model.onnx | Bin 0 -> 99 bytes .../node/test_atan/test_data_set_0/input_0.pb | 1 + .../test_atan/test_data_set_0/output_0.pb | 1 + .../data/node/test_atan_example/model.onnx | Bin 0 -> 91 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_atanh/model.onnx | Bin 0 -> 101 bytes .../test_atanh/test_data_set_0/input_0.pb | 1 + .../test_atanh/test_data_set_0/output_0.pb | 2 + .../data/node/test_atanh_example/model.onnx | Bin 0 -> 93 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 265 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 144 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 144 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 144 bytes .../model.onnx | Bin 0 -> 17451 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 144 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 144 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 144 bytes .../model.onnx | Bin 0 -> 361 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 144 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 62 bytes .../model.onnx | Bin 0 -> 17630 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 144 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 62 bytes .../model.onnx | Bin 0 -> 361 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 144 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 62 bytes .../model.onnx | Bin 0 -> 19373 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 144 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 62 bytes .../model.onnx | Bin 0 -> 388 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 19551 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 46 bytes .../data/node/test_attention_3d/model.onnx | Bin 0 -> 210 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_attention_3d_attn_mask/model.onnx | Bin 0 -> 260 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../model.onnx | Bin 0 -> 14836 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../node/test_attention_3d_causal/model.onnx | Bin 0 -> 235 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../model.onnx | Bin 0 -> 16872 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../model.onnx | Bin 0 -> 227 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../model.onnx | Bin 0 -> 277 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../model.onnx | Bin 0 -> 18019 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../model.onnx | Bin 0 -> 252 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../model.onnx | Bin 0 -> 20700 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../model.onnx | Bin 0 -> 16022 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../model.onnx | Bin 0 -> 251 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../model.onnx | Bin 0 -> 17317 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../model.onnx | Bin 0 -> 254 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../model.onnx | Bin 0 -> 18518 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../model.onnx | Bin 0 -> 487 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2905 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 4348 bytes .../model.onnx | Bin 0 -> 19614 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1454 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2905 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 974 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 4348 bytes .../test_attention_3d_expanded/model.onnx | Bin 0 -> 12855 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../node/test_attention_3d_gqa/model.onnx | Bin 0 -> 214 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../model.onnx | Bin 0 -> 264 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../model.onnx | Bin 0 -> 15584 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../test_attention_3d_gqa_causal/model.onnx | Bin 0 -> 239 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../model.onnx | Bin 0 -> 17772 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../test_attention_3d_gqa_expanded/model.onnx | Bin 0 -> 13602 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../test_attention_3d_gqa_scaled/model.onnx | Bin 0 -> 238 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../model.onnx | Bin 0 -> 14895 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../test_attention_3d_gqa_softcap/model.onnx | Bin 0 -> 241 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../model.onnx | Bin 0 -> 15953 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../model.onnx | Bin 0 -> 480 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../model.onnx | Bin 0 -> 18287 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2318 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../node/test_attention_3d_scaled/model.onnx | Bin 0 -> 234 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | 7 + .../model.onnx | Bin 0 -> 14151 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | 7 + .../node/test_attention_3d_softcap/model.onnx | Bin 0 -> 237 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../model.onnx | Bin 0 -> 15165 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../model.onnx | Bin 0 -> 233 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 17140 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 476 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../model.onnx | Bin 0 -> 17531 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../model.onnx | Bin 0 -> 549 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 584 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 20553 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 19590 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 606 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 22294 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 587 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 21120 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1166 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 782 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../data/node/test_attention_4d/model.onnx | Bin 0 -> 185 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_attention_4d_attn_mask/model.onnx | Bin 0 -> 235 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_attention_4d_attn_mask_3d/model.onnx | Bin 0 -> 246 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 216 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 271 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 216 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 17315 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 216 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 12988 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 216 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_attention_4d_attn_mask_4d/model.onnx | Bin 0 -> 246 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 600 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 271 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 600 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 17315 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 600 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 12988 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 600 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 240 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 251 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 13931 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 13441 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 12506 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../node/test_attention_4d_causal/model.onnx | Bin 0 -> 210 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 14628 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 341 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_2.pb | 3 + .../test_data_set_0/input_3.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | 4 + .../model.onnx | Bin 0 -> 26162 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_2.pb | 3 + .../test_data_set_0/input_3.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | 4 + .../model.onnx | Bin 0 -> 287 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 48 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 400 bytes .../model.onnx | Bin 0 -> 22605 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 48 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 400 bytes .../model.onnx | Bin 0 -> 291 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 32 bytes .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 23550 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 32 bytes .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 306 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/input_3.pb | Bin 0 -> 32 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 27105 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/input_3.pb | Bin 0 -> 32 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 438 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 599 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 601 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 1370 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 1372 bytes .../model.onnx | Bin 0 -> 19737 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 599 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 601 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 1370 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 1372 bytes .../model.onnx | Bin 0 -> 315 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 408 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 19186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 408 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 252 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 15195 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 227 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 17964 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 13489 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 226 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 14580 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 229 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 15752 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../model.onnx | Bin 0 -> 462 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2905 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 4348 bytes .../model.onnx | Bin 0 -> 16615 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2905 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 4348 bytes .../model.onnx | Bin 0 -> 477 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 600 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2905 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 4348 bytes .../model.onnx | Bin 0 -> 17746 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 600 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2905 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 4348 bytes .../model.onnx | Bin 0 -> 477 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1752 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2905 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 4348 bytes .../model.onnx | Bin 0 -> 17746 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1456 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1752 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2905 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 976 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 4348 bytes .../test_attention_4d_expanded/model.onnx | Bin 0 -> 10815 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../node/test_attention_4d_fp16/model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 400 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 11603 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 400 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../node/test_attention_4d_gqa/model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../model.onnx | Bin 0 -> 239 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../model.onnx | Bin 0 -> 13138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../test_attention_4d_gqa_causal/model.onnx | Bin 0 -> 214 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../model.onnx | Bin 0 -> 15412 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../model.onnx | Bin 0 -> 284 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 21897 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 289 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 144 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/input_3.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | 3 + .../model.onnx | Bin 0 -> 23078 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 144 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/input_3.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | 3 + .../test_attention_4d_gqa_expanded/model.onnx | Bin 0 -> 11446 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../test_attention_4d_gqa_scaled/model.onnx | Bin 0 -> 213 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../model.onnx | Bin 0 -> 12536 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../test_attention_4d_gqa_softcap/model.onnx | Bin 0 -> 216 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../model.onnx | Bin 0 -> 13565 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../model.onnx | Bin 0 -> 455 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../model.onnx | Bin 0 -> 15492 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2320 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../model.onnx | Bin 0 -> 460 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_1.pb | 3 + .../test_data_set_0/input_2.pb | Bin 0 -> 592 bytes .../test_data_set_0/input_3.pb | 2 + .../test_data_set_0/input_4.pb | 3 + .../test_data_set_0/input_5.pb | Bin 0 -> 1177 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_1.pb | 4 + .../test_data_set_0/output_2.pb | Bin 0 -> 1756 bytes .../model.onnx | Bin 0 -> 16292 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_1.pb | 3 + .../test_data_set_0/input_2.pb | Bin 0 -> 592 bytes .../test_data_set_0/input_3.pb | 2 + .../test_data_set_0/input_4.pb | 3 + .../test_data_set_0/input_5.pb | Bin 0 -> 1177 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_1.pb | 4 + .../test_data_set_0/output_2.pb | Bin 0 -> 1756 bytes .../node/test_attention_4d_scaled/model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 11908 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../node/test_attention_4d_softcap/model.onnx | Bin 0 -> 212 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 12893 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../model.onnx | Bin 0 -> 264 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 208 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 208 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 15039 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 208 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 208 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 271 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 208 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 208 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 16223 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 208 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 208 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 451 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../model.onnx | Bin 0 -> 14852 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../model.onnx | Bin 0 -> 523 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 559 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 575 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 600 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 600 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 600 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 24336 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 600 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 18741 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 600 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 575 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1752 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 600 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1752 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 24336 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1752 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 18741 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1752 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 17437 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 16621 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 308 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 2327 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 2329 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 3482 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 3484 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 1759 bytes .../model.onnx | Bin 0 -> 266 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 607 bytes .../model.onnx | Bin 0 -> 341 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 607 bytes .../model.onnx | Bin 0 -> 14251 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 607 bytes .../model.onnx | Bin 0 -> 13324 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 607 bytes .../model.onnx | Bin 0 -> 363 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 607 bytes .../model.onnx | Bin 0 -> 15710 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 607 bytes .../model.onnx | Bin 0 -> 344 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 607 bytes .../model.onnx | Bin 0 -> 14724 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1168 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 784 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 607 bytes .../model.onnx | Bin 0 -> 271 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 144 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 144 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 144 bytes .../model.onnx | Bin 0 -> 20605 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 144 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 144 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 144 bytes .../test_averagepool_1d_default/model.onnx | Bin 0 -> 145 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_averagepool_2d_ceil/model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 260 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_averagepool_2d_default/model.onnx | Bin 0 -> 156 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 11548 bytes .../test_averagepool_2d_dilations/model.onnx | Bin 0 -> 214 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../node/test_averagepool_2d_pads/model.onnx | Bin 0 -> 172 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 9424 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 10816 bytes .../model.onnx | Bin 0 -> 216 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 9424 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 10816 bytes .../model.onnx | Bin 0 -> 184 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 115 bytes .../model.onnx | Bin 0 -> 228 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 115 bytes .../model.onnx | Bin 0 -> 216 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../test_averagepool_2d_same_lower/model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 12304 bytes .../test_averagepool_2d_same_upper/model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 12304 bytes .../test_averagepool_2d_strides/model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1216 bytes .../test_averagepool_3d_default/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 393235 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 357511 bytes .../model.onnx | Bin 0 -> 303 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 131091 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2066 bytes .../model.onnx | Bin 0 -> 302 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 131091 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2934 bytes .../model.onnx | Bin 0 -> 303 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 131091 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2066 bytes .../model.onnx | Bin 0 -> 302 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 131091 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2934 bytes .../model.onnx | Bin 0 -> 234 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 274 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../test_basic_conv_with_padding/model.onnx | Bin 0 -> 201 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 115 bytes .../model.onnx | Bin 0 -> 204 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 308 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 546 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 92 bytes .../model.onnx | Bin 0 -> 323 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 165 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../node/test_batchnorm_epsilon/model.onnx | Bin 0 -> 250 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 364 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 2 + .../node/test_batchnorm_example/model.onnx | Bin 0 -> 231 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 345 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 2 + .../test/data/node/test_bernoulli/model.onnx | Bin 0 -> 93 bytes .../test_bernoulli/test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 89 bytes .../node/test_bernoulli_double/model.onnx | Bin 0 -> 114 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 89 bytes .../test_bernoulli_double_expanded/model.onnx | Bin 0 -> 446 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 89 bytes .../node/test_bernoulli_expanded/model.onnx | Bin 0 -> 409 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 89 bytes .../data/node/test_bernoulli_seed/model.onnx | Bin 0 -> 114 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../test_bernoulli_seed_expanded/model.onnx | Bin 0 -> 452 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../model.onnx | Bin 0 -> 128 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../test_bitcast_bool_to_uint8/model.onnx | Bin 0 -> 114 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 13 bytes .../test_bitcast_float32_to_int32/model.onnx | Bin 0 -> 117 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_bitcast_float64_to_int64/model.onnx | Bin 0 -> 117 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../test_bitcast_int32_to_float32/model.onnx | Bin 0 -> 117 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_bitcast_int64_to_float64/model.onnx | Bin 0 -> 117 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../test_bitcast_int8_to_uint8/model.onnx | Bin 0 -> 114 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 13 bytes .../model.onnx | Bin 0 -> 116 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 11 bytes .../test_bitcast_uint16_to_int16/model.onnx | Bin 0 -> 116 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test_bitcast_uint32_to_int32/model.onnx | Bin 0 -> 116 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../node/test_bitshift_left_uint16/model.onnx | Bin 0 -> 145 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../node/test_bitshift_left_uint32/model.onnx | Bin 0 -> 145 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../node/test_bitshift_left_uint64/model.onnx | Bin 0 -> 145 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../node/test_bitshift_left_uint8/model.onnx | Bin 0 -> 144 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_bitshift_right_uint16/model.onnx | Bin 0 -> 147 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test_bitshift_right_uint32/model.onnx | Bin 0 -> 147 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_bitshift_right_uint64/model.onnx | Bin 0 -> 147 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../node/test_bitshift_right_uint8/model.onnx | Bin 0 -> 146 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 12 bytes .../node/test_bitwise_and_i16_3d/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 142 bytes .../node/test_bitwise_and_i32_2d/model.onnx | Bin 0 -> 154 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 167 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 49 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 503 bytes .../model.onnx | Bin 0 -> 182 bytes .../test_data_set_0/input_0.pb | 5 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 385 bytes .../data/node/test_bitwise_not_2d/model.onnx | Bin 0 -> 127 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 69 bytes .../data/node/test_bitwise_not_3d/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 143 bytes .../data/node/test_bitwise_not_4d/model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | 5 + .../test_data_set_0/output_0.pb | 1 + .../node/test_bitwise_or_i16_4d/model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 376 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 384 bytes .../node/test_bitwise_or_i32_2d/model.onnx | Bin 0 -> 149 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 163 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 49 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 502 bytes .../test_bitwise_or_ui8_bcast_4v3d/model.onnx | Bin 0 -> 178 bytes .../test_data_set_0/input_0.pb | 5 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../node/test_bitwise_xor_i16_3d/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 142 bytes .../node/test_bitwise_xor_i32_2d/model.onnx | Bin 0 -> 154 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 68 bytes .../model.onnx | Bin 0 -> 167 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 49 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 503 bytes .../model.onnx | Bin 0 -> 182 bytes .../test_data_set_0/input_0.pb | 5 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 385 bytes .../data/node/test_blackmanwindow/model.onnx | Bin 0 -> 99 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../test_blackmanwindow_expanded/model.onnx | Bin 0 -> 4244 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../test_blackmanwindow_symmetric/model.onnx | Bin 0 -> 126 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../model.onnx | Bin 0 -> 4814 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../test_cast_BFLOAT16_to_FLOAT/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../node/test_cast_DOUBLE_to_FLOAT/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 111 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../test_cast_DOUBLE_to_FLOAT16/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 111 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40 bytes .../test_cast_FLOAT16_to_DOUBLE/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 112 bytes .../test_cast_FLOAT16_to_FLOAT/model.onnx | Bin 0 -> 137 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 144 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 146 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 146 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../node/test_cast_FLOAT16_to_INT2/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_cast_FLOAT16_to_INT4/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/output_0.pb | 1 + .../test_cast_FLOAT16_to_UINT2/model.onnx | Bin 0 -> 137 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../test_cast_FLOAT16_to_UINT4/model.onnx | Bin 0 -> 137 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/output_0.pb | 1 + .../test_cast_FLOAT4E2M1_to_FLOAT/model.onnx | Bin 0 -> 140 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 144 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 146 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 144 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 144 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 146 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../test_cast_FLOAT8E5M2_to_FLOAT/model.onnx | Bin 0 -> 140 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../test_cast_FLOAT_to_BFLOAT16/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40 bytes .../node/test_cast_FLOAT_to_DOUBLE/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 112 bytes .../test_cast_FLOAT_to_FLOAT16/model.onnx | Bin 0 -> 137 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40 bytes .../test_cast_FLOAT_to_FLOAT4E2M1/model.onnx | Bin 0 -> 140 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 144 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../test_cast_FLOAT_to_FLOAT8E5M2/model.onnx | Bin 0 -> 140 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 144 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../node/test_cast_FLOAT_to_INT2/model.onnx | Bin 0 -> 134 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_cast_FLOAT_to_INT4/model.onnx | Bin 0 -> 134 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_cast_FLOAT_to_UINT2/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_cast_FLOAT_to_UINT4/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_cast_INT2_to_FLOAT/model.onnx | Bin 0 -> 134 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 44 bytes .../node/test_cast_INT2_to_FLOAT16/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../node/test_cast_INT2_to_INT8/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../node/test_cast_INT4_to_FLOAT/model.onnx | Bin 0 -> 134 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 116 bytes .../node/test_cast_INT4_to_FLOAT16/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../node/test_cast_INT4_to_INT8/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../node/test_cast_UINT2_to_FLOAT/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 44 bytes .../test_cast_UINT2_to_FLOAT16/model.onnx | Bin 0 -> 137 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../node/test_cast_UINT2_to_UINT8/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../node/test_cast_UINT4_to_FLOAT/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 116 bytes .../test_cast_UINT4_to_FLOAT16/model.onnx | Bin 0 -> 137 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../node/test_cast_UINT4_to_UINT8/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../model.onnx | Bin 0 -> 145 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 48 bytes .../model.onnx | Bin 0 -> 147 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 184 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 47 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 172 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 172 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 162 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../test_castlike_DOUBLE_to_FLOAT/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 111 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../model.onnx | Bin 0 -> 162 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 111 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40 bytes .../model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 111 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 111 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../model.onnx | Bin 0 -> 162 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 112 bytes .../model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 112 bytes .../test_castlike_FLOAT16_to_FLOAT/model.onnx | Bin 0 -> 161 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 195 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 168 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 199 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 199 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 195 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../test_castlike_FLOAT16_to_INT2/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../test_castlike_FLOAT16_to_INT4/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../test_castlike_FLOAT16_to_UINT2/model.onnx | Bin 0 -> 161 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../test_castlike_FLOAT16_to_UINT4/model.onnx | Bin 0 -> 161 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 164 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 195 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 193 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 168 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 199 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 168 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 195 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 168 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 199 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 164 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 195 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 46 bytes .../model.onnx | Bin 0 -> 193 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 162 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40 bytes .../model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40 bytes .../test_castlike_FLOAT_to_DOUBLE/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 112 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 112 bytes .../test_castlike_FLOAT_to_FLOAT16/model.onnx | Bin 0 -> 161 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40 bytes .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40 bytes .../model.onnx | Bin 0 -> 164 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 193 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 168 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 195 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 164 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 168 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 193 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../test_castlike_FLOAT_to_INT2/model.onnx | Bin 0 -> 158 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../test_castlike_FLOAT_to_INT4/model.onnx | Bin 0 -> 158 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../test_castlike_FLOAT_to_UINT2/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../test_castlike_FLOAT_to_UINT4/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | 1 + .../test_castlike_INT2_to_FLOAT/model.onnx | Bin 0 -> 158 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 44 bytes .../test_castlike_INT2_to_FLOAT16/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 44 bytes .../test_castlike_INT2_to_INT8/model.onnx | Bin 0 -> 157 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../test_castlike_INT4_to_FLOAT/model.onnx | Bin 0 -> 158 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 116 bytes .../test_castlike_INT4_to_FLOAT16/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 116 bytes .../test_castlike_INT4_to_INT8/model.onnx | Bin 0 -> 157 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../test_castlike_UINT2_to_FLOAT/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 44 bytes .../test_castlike_UINT2_to_FLOAT16/model.onnx | Bin 0 -> 161 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 44 bytes .../test_castlike_UINT2_to_UINT8/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../test_castlike_UINT4_to_FLOAT/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 116 bytes .../test_castlike_UINT4_to_FLOAT16/model.onnx | Bin 0 -> 161 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 116 bytes .../test_castlike_UINT4_to_UINT8/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 199 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 211 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 195 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 199 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 211 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 207 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 195 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 207 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 193 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 205 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 258 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 3450 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 247 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 3106 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 325 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 2228 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 246 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 146 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | Bin 0 -> 73 bytes .../model.onnx | Bin 0 -> 3075 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 146 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | Bin 0 -> 73 bytes .../model.onnx | Bin 0 -> 257 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 25 bytes .../model.onnx | Bin 0 -> 3419 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 25 bytes .../model.onnx | Bin 0 -> 267 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 154 bytes .../model.onnx | Bin 0 -> 3731 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 154 bytes .../model.onnx | Bin 0 -> 269 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 3793 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 274 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 146 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | Bin 0 -> 73 bytes .../model.onnx | Bin 0 -> 3991 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 146 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | Bin 0 -> 73 bytes .../model.onnx | Bin 0 -> 333 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 3229 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 277 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 4069 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 277 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 338 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 2490 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 3259 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 305 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 2282 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | 1 + .../test/data/node/test_ceil/model.onnx | Bin 0 -> 99 bytes .../node/test_ceil/test_data_set_0/input_0.pb | 1 + .../test_ceil/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_ceil_example/model.onnx | Bin 0 -> 91 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../test/data/node/test_celu/model.onnx | Bin 0 -> 124 bytes .../node/test_celu/test_data_set_0/input_0.pb | 2 + .../test_celu/test_data_set_0/output_0.pb | 2 + .../data/node/test_celu_bfloat16/model.onnx | Bin 0 -> 109 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 19 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../test_celu_bfloat16_expanded/model.onnx | Bin 0 -> 658 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 19 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../data/node/test_celu_expanded/model.onnx | Bin 0 -> 591 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../data/node/test_celu_float16/model.onnx | Bin 0 -> 108 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 19 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../test_celu_float16_expanded/model.onnx | Bin 0 -> 647 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 19 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../node/test_center_crop_pad_crop/model.onnx | Bin 0 -> 153 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2414 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 854 bytes .../model.onnx | Bin 0 -> 161 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1934 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1214 bytes .../model.onnx | Bin 0 -> 2658 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1934 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1214 bytes .../model.onnx | Bin 0 -> 177 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1934 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1094 bytes .../model.onnx | Bin 0 -> 3485 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1934 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1094 bytes .../model.onnx | Bin 0 -> 177 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1934 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1094 bytes .../model.onnx | Bin 0 -> 3485 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1934 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1094 bytes .../model.onnx | Bin 0 -> 2410 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2414 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 854 bytes .../model.onnx | Bin 0 -> 204 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1934 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1094 bytes .../model.onnx | Bin 0 -> 3864 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1934 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1094 bytes .../node/test_center_crop_pad_pad/model.onnx | Bin 0 -> 151 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 854 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2414 bytes .../model.onnx | Bin 0 -> 2379 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 854 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2414 bytes .../test/data/node/test_clip/model.onnx | Bin 0 -> 139 bytes .../node/test_clip/test_data_set_0/input_0.pb | 1 + .../node/test_clip/test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../node/test_clip/test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_clip/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test_clip_default_inbounds/model.onnx | Bin 0 -> 104 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 115 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 109 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 12 bytes .../model.onnx | Bin 0 -> 120 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 12 bytes .../test_clip_default_int8_max/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../model.onnx | Bin 0 -> 311 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../test_clip_default_int8_min/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../model.onnx | Bin 0 -> 309 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../node/test_clip_default_max/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test_clip_default_max_expanded/model.onnx | Bin 0 -> 296 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_clip_default_min/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test_clip_default_min_expanded/model.onnx | Bin 0 -> 294 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_clip_example/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_clip_example_expanded/model.onnx | Bin 0 -> 578 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../data/node/test_clip_expanded/model.onnx | Bin 0 -> 530 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_clip_inbounds/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_clip_inbounds_expanded/model.onnx | Bin 0 -> 587 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_clip_min_greater_than_max/model.onnx | Bin 0 -> 144 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 685 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../data/node/test_clip_outbounds/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_clip_outbounds_expanded/model.onnx | Bin 0 -> 596 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../node/test_clip_splitbounds/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_clip_splitbounds_expanded/model.onnx | Bin 0 -> 613 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_col2im/model.onnx | Bin 0 -> 206 bytes .../test_col2im/test_data_set_0/input_0.pb | Bin 0 -> 117 bytes .../test_col2im/test_data_set_0/input_1.pb | Bin 0 -> 35 bytes .../test_col2im/test_data_set_0/input_2.pb | Bin 0 -> 35 bytes .../test_col2im/test_data_set_0/output_0.pb | Bin 0 -> 120 bytes .../test/data/node/test_col2im_5d/model.onnx | Bin 0 -> 213 bytes .../test_col2im_5d/test_data_set_0/input_0.pb | Bin 0 -> 498 bytes .../test_col2im_5d/test_data_set_0/input_1.pb | Bin 0 -> 43 bytes .../test_col2im_5d/test_data_set_0/input_2.pb | Bin 0 -> 43 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 503 bytes .../node/test_col2im_dilations/model.onnx | Bin 0 -> 236 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 97 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 35 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 165 bytes .../data/node/test_col2im_pads/model.onnx | Bin 0 -> 230 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 318 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 35 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 120 bytes .../data/node/test_col2im_strides/model.onnx | Bin 0 -> 232 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 162 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 35 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 120 bytes .../test/data/node/test_compress_0/model.onnx | Bin 0 -> 169 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../test/data/node/test_compress_1/model.onnx | Bin 0 -> 169 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 19 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 28 bytes .../test_compress_default_axis/model.onnx | Bin 0 -> 163 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 22 bytes .../test_compress_negative_axis/model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 19 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 28 bytes .../node/test_concat_1d_axis_0/model.onnx | Bin 0 -> 161 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 22 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../test_concat_1d_axis_negative_1/model.onnx | Bin 0 -> 179 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 22 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../node/test_concat_2d_axis_0/model.onnx | Bin 0 -> 173 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 32 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 32 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 48 bytes .../node/test_concat_2d_axis_1/model.onnx | Bin 0 -> 173 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 32 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 32 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 48 bytes .../test_concat_2d_axis_negative_1/model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 32 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 32 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 48 bytes .../test_concat_2d_axis_negative_2/model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 32 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 32 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 48 bytes .../node/test_concat_3d_axis_0/model.onnx | Bin 0 -> 185 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 50 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../node/test_concat_3d_axis_1/model.onnx | Bin 0 -> 185 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 50 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../node/test_concat_3d_axis_2/model.onnx | Bin 0 -> 185 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 50 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../test_concat_3d_axis_negative_1/model.onnx | Bin 0 -> 203 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 50 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../test_concat_3d_axis_negative_2/model.onnx | Bin 0 -> 203 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 50 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../test_concat_3d_axis_negative_3/model.onnx | Bin 0 -> 203 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 50 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../test/data/node/test_constant/model.onnx | Bin 0 -> 224 bytes .../test_constant/test_data_set_0/output_0.pb | 1 + .../data/node/test_constant_pad/model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 76 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_constant_pad_axes/model.onnx | Bin 0 -> 217 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 44 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 28 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 226 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 44 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 156 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 109 bytes .../model.onnx | Bin 0 -> 148 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 9 bytes .../test_constantofshape_int_zeros/model.onnx | Bin 0 -> 147 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 25 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 252 bytes .../test_conv_with_autopad_same/model.onnx | Bin 0 -> 226 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 236 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 156 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 224 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 156 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 39 bytes .../test_conv_with_strides_padding/model.onnx | Bin 0 -> 221 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 156 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 63 bytes .../test_convinteger_with_padding/model.onnx | Bin 0 -> 268 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 144 bytes .../model.onnx | Bin 0 -> 208 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../data/node/test_convtranspose/model.onnx | Bin 0 -> 158 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 87 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 216 bytes .../node/test_convtranspose_1d/model.onnx | Bin 0 -> 148 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 53 bytes .../node/test_convtranspose_3d/model.onnx | Bin 0 -> 173 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 258 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 234 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1698 bytes .../model.onnx | Bin 0 -> 216 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 87 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 304 bytes .../test_convtranspose_dilations/model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 31 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 115 bytes .../test_convtranspose_group_2/model.onnx | Bin 0 -> 180 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 87 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 87 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 216 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 232 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 87 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 616 bytes .../model.onnx | Bin 0 -> 266 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 87 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 656 bytes .../model.onnx | Bin 0 -> 212 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 87 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 656 bytes .../node/test_convtranspose_pad/model.onnx | Bin 0 -> 205 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 87 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 656 bytes .../node/test_convtranspose_pads/model.onnx | Bin 0 -> 200 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 87 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 184 bytes .../test/data/node/test_cos/model.onnx | Bin 0 -> 97 bytes .../node/test_cos/test_data_set_0/input_0.pb | 1 + .../node/test_cos/test_data_set_0/output_0.pb | 3 + .../data/node/test_cos_example/model.onnx | Bin 0 -> 89 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_cosh/model.onnx | Bin 0 -> 99 bytes .../node/test_cosh/test_data_set_0/input_0.pb | 1 + .../test_cosh/test_data_set_0/output_0.pb | 1 + .../data/node/test_cosh_example/model.onnx | Bin 0 -> 91 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_cumprod_1d/model.onnx | Bin 0 -> 114 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../node/test_cumprod_1d_exclusive/model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../model.onnx | Bin 0 -> 148 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../node/test_cumprod_1d_reverse/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../model.onnx | Bin 0 -> 167 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../node/test_cumprod_2d_axis_0/model.onnx | Bin 0 -> 129 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../node/test_cumprod_2d_axis_1/model.onnx | Bin 0 -> 129 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../node/test_cumprod_2d_int32/model.onnx | Bin 0 -> 128 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 35 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../test_cumprod_2d_negative_axis/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../test/data/node/test_cumsum_1d/model.onnx | Bin 0 -> 112 bytes .../test_cumsum_1d/test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_cumsum_1d/test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../node/test_cumsum_1d_exclusive/model.onnx | Bin 0 -> 140 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../test_cumsum_1d_int32_exclusive/model.onnx | Bin 0 -> 146 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../node/test_cumsum_1d_reverse/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../node/test_cumsum_2d_axis_0/model.onnx | Bin 0 -> 127 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../node/test_cumsum_2d_axis_1/model.onnx | Bin 0 -> 127 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../data/node/test_cumsum_2d_int32/model.onnx | Bin 0 -> 126 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 35 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../test_cumsum_2d_negative_axis/model.onnx | Bin 0 -> 134 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../model.onnx | Bin 0 -> 310 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 149 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 286 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 87 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 47 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 277 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 175 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 208 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 208 bytes .../node/test_depthtospace_example/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 208 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 208 bytes .../node/test_dequantizelinear/model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test_dequantizelinear_axis/model.onnx | Bin 0 -> 211 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 27 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 87 bytes .../test_dequantizelinear_blocked/model.onnx | Bin 0 -> 270 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 69 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 111 bytes .../test_dequantizelinear_e4m3fn/model.onnx | Bin 0 -> 156 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../model.onnx | Bin 0 -> 164 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../model.onnx | Bin 0 -> 205 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 19 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../test_dequantizelinear_e5m2/model.onnx | Bin 0 -> 154 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../test_dequantizelinear_int16/model.onnx | Bin 0 -> 180 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test_dequantizelinear_int2/model.onnx | Bin 0 -> 196 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test_dequantizelinear_int4/model.onnx | Bin 0 -> 196 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../test_dequantizelinear_uint16/model.onnx | Bin 0 -> 181 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test_dequantizelinear_uint2/model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test_dequantizelinear_uint4/model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../test/data/node/test_det_2d/model.onnx | Bin 0 -> 84 bytes .../test_det_2d/test_data_set_0/input_0.pb | Bin 0 -> 27 bytes .../test_det_2d/test_data_set_0/output_0.pb | Bin 0 -> 11 bytes .../test/data/node/test_det_nd/model.onnx | Bin 0 -> 92 bytes .../test_det_nd/test_data_set_0/input_0.pb | Bin 0 -> 61 bytes .../test_det_nd/test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_dft/model.onnx | Bin 0 -> 129 bytes .../node/test_dft/test_data_set_0/input_0.pb | Bin 0 -> 416 bytes .../node/test_dft/test_data_set_0/input_1.pb | Bin 0 -> 18 bytes .../node/test_dft/test_data_set_0/output_0.pb | Bin 0 -> 816 bytes .../test/data/node/test_dft_axis/model.onnx | Bin 0 -> 134 bytes .../test_dft_axis/test_data_set_0/input_0.pb | Bin 0 -> 416 bytes .../test_dft_axis/test_data_set_0/input_1.pb | Bin 0 -> 18 bytes .../test_dft_axis/test_data_set_0/output_0.pb | Bin 0 -> 816 bytes .../node/test_dft_axis_opset19/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 416 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 816 bytes .../data/node/test_dft_inverse/model.onnx | Bin 0 -> 154 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 816 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 18 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 816 bytes .../node/test_dft_inverse_opset19/model.onnx | Bin 0 -> 150 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 816 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 816 bytes .../test/data/node/test_dft_irfft/model.onnx | Bin 0 -> 169 bytes .../test_dft_irfft/test_data_set_0/input_0.pb | Bin 0 -> 496 bytes .../test_dft_irfft/test_data_set_0/input_1.pb | Bin 0 -> 18 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 416 bytes .../node/test_dft_irfft_opset19/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 416 bytes .../data/node/test_dft_opset19/model.onnx | Bin 0 -> 126 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 416 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 816 bytes .../test/data/node/test_dft_rfft/model.onnx | Bin 0 -> 151 bytes .../test_dft_rfft/test_data_set_0/input_0.pb | Bin 0 -> 416 bytes .../test_dft_rfft/test_data_set_0/input_1.pb | Bin 0 -> 18 bytes .../test_dft_rfft/test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../node/test_dft_rfft_opset19/model.onnx | Bin 0 -> 148 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 416 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test/data/node/test_div/model.onnx | Bin 0 -> 125 bytes .../node/test_div/test_data_set_0/input_0.pb | 1 + .../node/test_div/test_data_set_0/input_1.pb | 1 + .../node/test_div/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_div_bcast/model.onnx | Bin 0 -> 123 bytes .../test_div_bcast/test_data_set_0/input_0.pb | 1 + .../test_div_bcast/test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_div_example/model.onnx | Bin 0 -> 109 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../test/data/node/test_div_int16/model.onnx | Bin 0 -> 131 bytes .../test_div_int16/test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_div_int16/test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../data/node/test_div_int32_trunc/model.onnx | Bin 0 -> 113 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 25 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test/data/node/test_div_int8/model.onnx | Bin 0 -> 130 bytes .../test_div_int8/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_div_int8/test_data_set_0/input_1.pb | 4 + .../test_div_int8/test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../test/data/node/test_div_uint16/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test/data/node/test_div_uint32/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_div_uint64/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 494 bytes .../test/data/node/test_div_uint8/model.onnx | Bin 0 -> 131 bytes .../test_div_uint8/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_div_uint8/test_data_set_0/input_1.pb | 7 + .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../data/node/test_dropout_default/model.onnx | Bin 0 -> 126 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_dropout_default_mask/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 182 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../node/test_dropout_default_old/model.onnx | Bin 0 -> 101 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_dropout_default_ratio/model.onnx | Bin 0 -> 148 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_dropout_random_old/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_dynamicquantizelinear/model.onnx | Bin 0 -> 184 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 3145 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 3600 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 205 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | Bin 0 -> 19 bytes .../model.onnx | Bin 0 -> 3608 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | Bin 0 -> 19 bytes .../test/data/node/test_edge_pad/model.onnx | Bin 0 -> 154 bytes .../test_edge_pad/test_data_set_0/input_0.pb | Bin 0 -> 256 bytes .../test_edge_pad/test_data_set_0/input_1.pb | Bin 0 -> 76 bytes .../test_edge_pad/test_data_set_0/output_0.pb | Bin 0 -> 520 bytes .../test_einsum_batch_diagonal/model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 614 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_einsum_batch_matmul/model.onnx | Bin 0 -> 177 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | 3 + .../test_data_set_0/output_0.pb | Bin 0 -> 334 bytes .../node/test_einsum_inner_prod/model.onnx | Bin 0 -> 134 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_einsum_scalar/model.onnx | Bin 0 -> 105 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test/data/node/test_einsum_sum/model.onnx | Bin 0 -> 117 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_einsum_transpose/model.onnx | Bin 0 -> 128 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_elu/model.onnx | Bin 0 -> 114 bytes .../node/test_elu/test_data_set_0/input_0.pb | 1 + .../node/test_elu/test_data_set_0/output_0.pb | 1 + .../data/node/test_elu_default/model.onnx | Bin 0 -> 105 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1317 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_elu_example/model.onnx | Bin 0 -> 106 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 1301 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../node/test_elu_expanded_ver18/model.onnx | Bin 0 -> 1147 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_equal/model.onnx | Bin 0 -> 129 bytes .../test_equal/test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_equal/test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_equal/test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../data/node/test_equal_bcast/model.onnx | Bin 0 -> 127 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../data/node/test_equal_int16/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../test/data/node/test_equal_int8/model.onnx | Bin 0 -> 134 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../data/node/test_equal_string/model.onnx | Bin 0 -> 112 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 11 bytes .../test_equal_string_broadcast/model.onnx | Bin 0 -> 122 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 11 bytes .../data/node/test_equal_uint16/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../data/node/test_equal_uint32/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../data/node/test_equal_uint64/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../data/node/test_equal_uint8/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../test/data/node/test_erf/model.onnx | Bin 0 -> 105 bytes .../node/test_erf/test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../node/test_erf/test_data_set_0/output_0.pb | Bin 0 -> 12304 bytes .../test/data/node/test_exp/model.onnx | Bin 0 -> 97 bytes .../node/test_exp/test_data_set_0/input_0.pb | 1 + .../node/test_exp/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_exp_example/model.onnx | Bin 0 -> 89 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../node/test_expand_dim_changed/model.onnx | Bin 0 -> 168 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 41 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 165 bytes .../node/test_expand_dim_unchanged/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 66 bytes .../model.onnx | Bin 0 -> 148 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 91 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 91 bytes .../node/test_eyelike_with_dtype/model.onnx | Bin 0 -> 122 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 107 bytes .../test_eyelike_without_dtype/model.onnx | Bin 0 -> 111 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../data/node/test_flatten_axis0/model.onnx | Bin 0 -> 124 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_flatten_axis1/model.onnx | Bin 0 -> 124 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_flatten_axis2/model.onnx | Bin 0 -> 124 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_flatten_axis3/model.onnx | Bin 0 -> 124 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_flatten_default_axis/model.onnx | Bin 0 -> 118 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_flatten_negative_axis1/model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_flatten_negative_axis2/model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_flatten_negative_axis3/model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_flatten_negative_axis4/model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_flexattention/model.onnx | Bin 0 -> 228 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 3088 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 3088 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 4112 bytes .../test_flexattention_causal_mask/model.onnx | Bin 0 -> 855 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 5 + .../model.onnx | Bin 0 -> 10822 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 5 + .../model.onnx | Bin 0 -> 244 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 3088 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 6160 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 8208 bytes .../model.onnx | Bin 0 -> 7963 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 3088 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 6160 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 8208 bytes .../node/test_flexattention_double/model.onnx | Bin 0 -> 235 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 8208 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 6160 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 6160 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 8208 bytes .../model.onnx | Bin 0 -> 7170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 8208 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 6160 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 6160 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 8208 bytes .../model.onnx | Bin 0 -> 6554 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 3088 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 3088 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 4112 bytes .../node/test_flexattention_fp16/model.onnx | Bin 0 -> 233 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2064 bytes .../test_data_set_0/input_1.pb | 4 + .../test_data_set_0/input_2.pb | Bin 0 -> 1552 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2064 bytes .../model.onnx | Bin 0 -> 7281 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2064 bytes .../test_data_set_0/input_1.pb | 4 + .../test_data_set_0/input_2.pb | Bin 0 -> 1552 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2064 bytes .../node/test_flexattention_gqa/model.onnx | Bin 0 -> 232 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1552 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1552 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 4112 bytes .../model.onnx | Bin 0 -> 6906 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1552 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1552 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 4112 bytes .../test_flexattention_prob_mod/model.onnx | Bin 0 -> 400 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 7536 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 879 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 11772 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../node/test_flexattention_scaled/model.onnx | Bin 0 -> 252 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 3088 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 3088 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 4112 bytes .../model.onnx | Bin 0 -> 6061 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 3088 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 3088 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 4112 bytes .../test_flexattention_score_mod/model.onnx | Bin 0 -> 404 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 7625 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_flexattention_soft_cap/model.onnx | Bin 0 -> 461 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 7945 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 272 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../test/data/node/test_floor/model.onnx | Bin 0 -> 101 bytes .../test_floor/test_data_set_0/input_0.pb | 1 + .../test_floor/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_floor_example/model.onnx | Bin 0 -> 93 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_gather_0/model.onnx | Bin 0 -> 165 bytes .../test_gather_0/test_data_set_0/input_0.pb | 1 + .../test_gather_0/test_data_set_0/input_1.pb | Bin 0 -> 39 bytes .../test_gather_0/test_data_set_0/output_0.pb | 1 + .../test/data/node/test_gather_1/model.onnx | Bin 0 -> 165 bytes .../test_gather_1/test_data_set_0/input_0.pb | 1 + .../test_gather_1/test_data_set_0/input_1.pb | Bin 0 -> 39 bytes .../test_gather_1/test_data_set_0/output_0.pb | 1 + .../node/test_gather_2d_indices/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_gather_elements_0/model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 49 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 27 bytes .../node/test_gather_elements_1/model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 65 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../model.onnx | Bin 0 -> 185 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 65 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../test_gather_negative_indices/model.onnx | Bin 0 -> 156 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 52 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 39 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_gathernd_example_float32/model.onnx | Bin 0 -> 180 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 48 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 34 bytes .../test_gathernd_example_int32/model.onnx | Bin 0 -> 162 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 49 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 22 bytes .../model.onnx | Bin 0 -> 200 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 48 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 32 bytes .../data/node/test_gelu_default_1/model.onnx | Bin 0 -> 93 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_gelu_default_1_expanded/model.onnx | Bin 0 -> 1429 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../data/node/test_gelu_default_2/model.onnx | Bin 0 -> 109 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 4 + .../test_gelu_default_2_expanded/model.onnx | Bin 0 -> 1445 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 4 + .../data/node/test_gelu_tanh_1/model.onnx | Bin 0 -> 114 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../node/test_gelu_tanh_1_expanded/model.onnx | Bin 0 -> 2239 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../data/node/test_gelu_tanh_2/model.onnx | Bin 0 -> 130 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_gelu_tanh_2_expanded/model.onnx | Bin 0 -> 2255 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_gemm_all_attributes/model.onnx | Bin 0 -> 218 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_gemm_alpha/model.onnx | Bin 0 -> 163 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_gemm_beta/model.onnx | Bin 0 -> 161 bytes .../test_gemm_beta/test_data_set_0/input_0.pb | 1 + .../test_gemm_beta/test_data_set_0/input_1.pb | 1 + .../test_gemm_beta/test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_gemm_default_matrix_bias/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../node/test_gemm_default_no_bias/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../test_gemm_default_scalar_bias/model.onnx | Bin 0 -> 151 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 168 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_gemm_default_vector_bias/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../test_gemm_default_zero_bias/model.onnx | Bin 0 -> 158 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | 1 + .../data/node/test_gemm_transposeA/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../data/node/test_gemm_transposeB/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_globalaveragepool/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 145 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../data/node/test_globalmaxpool/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_globalmaxpool_precomputed/model.onnx | Bin 0 -> 137 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 51 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../test/data/node/test_greater/model.onnx | Bin 0 -> 145 bytes .../test_greater/test_data_set_0/input_0.pb | 1 + .../test_greater/test_data_set_0/input_1.pb | 1 + .../test_greater/test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../data/node/test_greater_bcast/model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../data/node/test_greater_equal/model.onnx | Bin 0 -> 171 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../node/test_greater_equal_bcast/model.onnx | Bin 0 -> 169 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../model.onnx | Bin 0 -> 447 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../test_greater_equal_expanded/model.onnx | Bin 0 -> 425 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../node/test_greater_equal_int16/model.onnx | Bin 0 -> 177 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../model.onnx | Bin 0 -> 455 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../node/test_greater_equal_int8/model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../model.onnx | Bin 0 -> 450 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../node/test_greater_equal_uint16/model.onnx | Bin 0 -> 178 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../model.onnx | Bin 0 -> 460 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../node/test_greater_equal_uint32/model.onnx | Bin 0 -> 178 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../model.onnx | Bin 0 -> 460 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../node/test_greater_equal_uint64/model.onnx | Bin 0 -> 178 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../model.onnx | Bin 0 -> 460 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../node/test_greater_equal_uint8/model.onnx | Bin 0 -> 177 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../model.onnx | Bin 0 -> 455 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 85 bytes .../data/node/test_greater_int16/model.onnx | Bin 0 -> 151 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../data/node/test_greater_int8/model.onnx | Bin 0 -> 150 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../data/node/test_greater_uint16/model.onnx | Bin 0 -> 153 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../data/node/test_greater_uint32/model.onnx | Bin 0 -> 153 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../data/node/test_greater_uint64/model.onnx | Bin 0 -> 153 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../data/node/test_greater_uint8/model.onnx | Bin 0 -> 151 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../test/data/node/test_gridsample/model.onnx | Bin 0 -> 225 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 307 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 160 bytes .../model.onnx | Bin 0 -> 217 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../node/test_gridsample_bicubic/model.onnx | Bin 0 -> 184 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 235 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 235 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../node/test_gridsample_bilinear/model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 237 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 237 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../test_gridsample_border_padding/model.onnx | Bin 0 -> 200 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../node/test_gridsample_nearest/model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 237 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 237 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 208 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 247 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 213 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 81 bytes .../model.onnx | Bin 0 -> 247 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 213 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 81 bytes .../model.onnx | Bin 0 -> 247 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 213 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 81 bytes .../model.onnx | Bin 0 -> 247 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 213 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 81 bytes .../test_gridsample_zeros_padding/model.onnx | Bin 0 -> 198 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 82 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 237 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 208 bytes .../model.onnx | Bin 0 -> 7018 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 208 bytes .../model.onnx | Bin 0 -> 218 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 7018 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../data/node/test_gru_batchwise/model.onnx | Bin 0 -> 235 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 7 + .../test_data_set_0/output_1.pb | 7 + .../data/node/test_gru_defaults/model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_gru_seq_length/model.onnx | Bin 0 -> 215 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 85 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../test_gru_with_initial_bias/model.onnx | Bin 0 -> 222 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 83 bytes .../test_data_set_0/output_0.pb | 1 + .../data/node/test_hammingwindow/model.onnx | Bin 0 -> 97 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../test_hammingwindow_expanded/model.onnx | Bin 0 -> 4131 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../test_hammingwindow_symmetric/model.onnx | Bin 0 -> 124 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 4701 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | 2 + .../test/data/node/test_hannwindow/model.onnx | Bin 0 -> 91 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../node/test_hannwindow_expanded/model.onnx | Bin 0 -> 3792 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../node/test_hannwindow_symmetric/model.onnx | Bin 0 -> 118 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../model.onnx | Bin 0 -> 4362 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../data/node/test_hardmax_axis_0/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_hardmax_axis_1/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_hardmax_axis_2/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_hardmax_default_axis/model.onnx | Bin 0 -> 118 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_hardmax_example/model.onnx | Bin 0 -> 105 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../test_hardmax_negative_axis/model.onnx | Bin 0 -> 141 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_hardmax_one_hot/model.onnx | Bin 0 -> 105 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 27 bytes .../data/node/test_hardsigmoid/model.onnx | Bin 0 -> 146 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_hardsigmoid_default/model.onnx | Bin 0 -> 121 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1832 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_hardsigmoid_example/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 1816 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 1643 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_hardswish/model.onnx | Bin 0 -> 109 bytes .../test_hardswish/test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_hardswish_expanded/model.onnx | Bin 0 -> 266 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_identity/model.onnx | Bin 0 -> 115 bytes .../test_identity/test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_identity/test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../data/node/test_identity_opt/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 42 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../node/test_identity_sequence/model.onnx | Bin 0 -> 96 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 65 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 65 bytes .../backend/test/data/node/test_if/model.onnx | Bin 0 -> 321 bytes .../node/test_if/test_data_set_0/input_0.pb | 1 + .../node/test_if/test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../test/data/node/test_if_opt/model.onnx | Bin 0 -> 419 bytes .../test_if_opt/test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_if_opt/test_data_set_0/output_0.pb | Bin 0 -> 44 bytes .../test/data/node/test_if_seq/model.onnx | Bin 0 -> 389 bytes .../test_if_seq/test_data_set_0/input_0.pb | 1 + .../test_if_seq/test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 3140 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3091 bytes .../model.onnx | Bin 0 -> 168 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1901 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3091 bytes .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1072 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3091 bytes .../model.onnx | Bin 0 -> 178 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1072 bytes .../test_data_set_0/output_0.pb | 10 + .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1072 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3091 bytes .../model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 326 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3091 bytes .../model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 3099 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3091 bytes .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 3226 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3091 bytes .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 566 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3091 bytes .../node/test_instancenorm_epsilon/model.onnx | Bin 0 -> 206 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../node/test_instancenorm_example/model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 39 bytes .../test/data/node/test_isinf/model.onnx | Bin 0 -> 85 bytes .../test_isinf/test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_isinf/test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../data/node/test_isinf_float16/model.onnx | Bin 0 -> 93 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../data/node/test_isinf_negative/model.onnx | Bin 0 -> 118 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../data/node/test_isinf_positive/model.onnx | Bin 0 -> 118 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test/data/node/test_isnan/model.onnx | Bin 0 -> 85 bytes .../test_isnan/test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_isnan/test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../data/node/test_isnan_float16/model.onnx | Bin 0 -> 93 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test_l1normalization_axis_0/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | 1 + .../test_l1normalization_axis_1/model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | 1 + .../test_l1normalization_axis_last/model.onnx | Bin 0 -> 164 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 61 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 61 bytes .../test_l2normalization_axis_0/model.onnx | Bin 0 -> 151 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 61 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 61 bytes .../test_l2normalization_axis_1/model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 261 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6048 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6300 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 253 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6040 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6292 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 272 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6633 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6915 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 280 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6641 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6923 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 312 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6600 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6876 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 304 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6592 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6868 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 296 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6584 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6860 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 315 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 7171 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 7478 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 323 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 71 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 7179 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 71 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 7486 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 71 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 331 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 7187 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 7494 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 309 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 496 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6096 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 496 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6348 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 496 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 301 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6088 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6340 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 293 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6080 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6332 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 285 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 114 bytes .../test_data_set_0/output_2.pb | 2 + .../model.onnx | Bin 0 -> 6072 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 114 bytes .../test_data_set_0/output_2.pb | 2 + .../model.onnx | Bin 0 -> 6324 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 114 bytes .../test_data_set_0/output_2.pb | 2 + .../model.onnx | Bin 0 -> 304 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 114 bytes .../test_data_set_0/output_2.pb | 2 + .../model.onnx | Bin 0 -> 6665 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 114 bytes .../test_data_set_0/output_2.pb | 2 + .../model.onnx | Bin 0 -> 6947 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 114 bytes .../test_data_set_0/output_2.pb | 2 + .../model.onnx | Bin 0 -> 312 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6673 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6955 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 320 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6681 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6963 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 328 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 496 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6689 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 496 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 6971 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 496 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../model.onnx | Bin 0 -> 276 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 114 bytes .../test_data_set_0/output_2.pb | 2 + .../model.onnx | Bin 0 -> 6273 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 114 bytes .../test_data_set_0/output_2.pb | 2 + .../model.onnx | Bin 0 -> 6537 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 114 bytes .../test_data_set_0/output_2.pb | 2 + .../test/data/node/test_leakyrelu/model.onnx | Bin 0 -> 126 bytes .../test_leakyrelu/test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_leakyrelu_default/model.onnx | Bin 0 -> 117 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 996 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_leakyrelu_example/model.onnx | Bin 0 -> 118 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 980 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../node/test_leakyrelu_expanded/model.onnx | Bin 0 -> 887 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_less/model.onnx | Bin 0 -> 133 bytes .../node/test_less/test_data_set_0/input_0.pb | 1 + .../node/test_less/test_data_set_0/input_1.pb | 1 + .../test_less/test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../test/data/node/test_less_bcast/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../test/data/node/test_less_equal/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../node/test_less_equal_bcast/model.onnx | Bin 0 -> 157 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../test_less_equal_bcast_expanded/model.onnx | Bin 0 -> 411 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../node/test_less_equal_expanded/model.onnx | Bin 0 -> 388 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../node/test_less_equal_int16/model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../test_less_equal_int16_expanded/model.onnx | Bin 0 -> 419 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../data/node/test_less_equal_int8/model.onnx | Bin 0 -> 164 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../test_less_equal_int8_expanded/model.onnx | Bin 0 -> 414 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../node/test_less_equal_uint16/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../model.onnx | Bin 0 -> 424 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../node/test_less_equal_uint32/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../model.onnx | Bin 0 -> 424 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../node/test_less_equal_uint64/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../model.onnx | Bin 0 -> 424 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../node/test_less_equal_uint8/model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../test_less_equal_uint8_expanded/model.onnx | Bin 0 -> 419 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 82 bytes .../test/data/node/test_less_int16/model.onnx | Bin 0 -> 139 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../test/data/node/test_less_int8/model.onnx | Bin 0 -> 138 bytes .../test_less_int8/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_less_int8/test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../data/node/test_less_uint16/model.onnx | Bin 0 -> 140 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../data/node/test_less_uint32/model.onnx | Bin 0 -> 140 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../data/node/test_less_uint64/model.onnx | Bin 0 -> 140 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../test/data/node/test_less_uint8/model.onnx | Bin 0 -> 139 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 76 bytes .../model.onnx | Bin 0 -> 438 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 274 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 2073 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 12824 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 274 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 2073 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../test_linear_attention_delta/model.onnx | Bin 0 -> 376 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 11036 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 411 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 145 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 13126 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 145 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../test_linear_attention_fp16/model.onnx | Bin 0 -> 383 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 528 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 530 bytes .../test_data_set_0/input_3.pb | 2 + .../test_data_set_0/input_4.pb | 3 + .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 1052 bytes .../model.onnx | Bin 0 -> 13970 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 528 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 530 bytes .../test_data_set_0/input_3.pb | 2 + .../test_data_set_0/input_4.pb | 3 + .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 1052 bytes .../test_linear_attention_gated/model.onnx | Bin 0 -> 376 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1042 bytes .../test_data_set_0/output_0.pb | 6 + .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 390 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 145 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 402 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 48 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 15068 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 48 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 13262 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 145 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 394 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2066 bytes .../test_data_set_0/input_1.pb | 8 + .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 2067 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 15862 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2066 bytes .../test_data_set_0/input_1.pb | 8 + .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 2067 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 394 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2066 bytes .../test_data_set_0/input_1.pb | 3 + .../test_data_set_0/input_2.pb | Bin 0 -> 274 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 274 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 2067 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 540 bytes .../model.onnx | Bin 0 -> 15862 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2066 bytes .../test_data_set_0/input_1.pb | 3 + .../test_data_set_0/input_2.pb | Bin 0 -> 274 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 274 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 2067 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 540 bytes .../model.onnx | Bin 0 -> 10208 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 1042 bytes .../test_data_set_0/output_0.pb | 6 + .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 391 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 12059 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../test_linear_attention_linear/model.onnx | Bin 0 -> 339 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 8993 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 350 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 274 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 10170 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 274 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 275 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 449 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 2073 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 145 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 14415 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 2073 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 145 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 444 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 2073 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 145 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../model.onnx | Bin 0 -> 13688 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 1040 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 2073 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 1042 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 145 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1043 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 2076 bytes .../test/data/node/test_log/model.onnx | Bin 0 -> 97 bytes .../node/test_log/test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../node/test_log/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_log_example/model.onnx | Bin 0 -> 89 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../node/test_logsoftmax_axis_0/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1058 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1110 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_logsoftmax_axis_1/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1058 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1110 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_logsoftmax_axis_2/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1058 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1110 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test_logsoftmax_default_axis/model.onnx | Bin 0 -> 124 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1161 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1210 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_logsoftmax_example_1/model.onnx | Bin 0 -> 113 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1111 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1157 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | 1 + .../test_logsoftmax_large_number/model.onnx | Bin 0 -> 116 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1153 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/output_0.pb | 1 + .../test_logsoftmax_negative_axis/model.onnx | Bin 0 -> 147 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1175 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1225 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_loop11/model.onnx | Bin 0 -> 726 bytes .../test_loop11/test_data_set_0/input_0.pb | Bin 0 -> 24 bytes .../test_loop11/test_data_set_0/input_1.pb | 1 + .../test_loop11/test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_loop11/test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../test_loop11/test_data_set_0/output_1.pb | Bin 0 -> 38 bytes .../test/data/node/test_loop13_seq/model.onnx | Bin 0 -> 735 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 111 bytes .../data/node/test_loop16_seq_none/model.onnx | Bin 0 -> 1149 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 25 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 121 bytes .../test_lpnormalization_default/model.onnx | Bin 0 -> 129 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 61 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 61 bytes .../node/test_lppool_1d_default/model.onnx | Bin 0 -> 162 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../node/test_lppool_2d_default/model.onnx | Bin 0 -> 156 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 11548 bytes .../node/test_lppool_2d_dilations/model.onnx | Bin 0 -> 196 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/output_0.pb | 1 + .../data/node/test_lppool_2d_pads/model.onnx | Bin 0 -> 172 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 9424 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 10816 bytes .../node/test_lppool_2d_same_lower/model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 12304 bytes .../node/test_lppool_2d_same_upper/model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 12304 bytes .../node/test_lppool_2d_strides/model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1216 bytes .../node/test_lppool_3d_default/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 393235 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 357511 bytes .../test/data/node/test_lrn/model.onnx | Bin 0 -> 168 bytes .../node/test_lrn/test_data_set_0/input_0.pb | Bin 0 -> 2516 bytes .../node/test_lrn/test_data_set_0/output_0.pb | Bin 0 -> 2516 bytes .../data/node/test_lrn_default/model.onnx | Bin 0 -> 126 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2516 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2516 bytes .../data/node/test_lstm_batchwise/model.onnx | Bin 0 -> 237 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../data/node/test_lstm_defaults/model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_lstm_with_initial_bias/model.onnx | Bin 0 -> 224 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 140 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_lstm_with_peepholes/model.onnx | Bin 0 -> 377 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 107 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_6.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_7.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_matmul_1d_1d/model.onnx | Bin 0 -> 109 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_matmul_1d_3d/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_matmul_2d/model.onnx | Bin 0 -> 122 bytes .../test_matmul_2d/test_data_set_0/input_0.pb | 1 + .../test_matmul_2d/test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_matmul_3d/model.onnx | Bin 0 -> 134 bytes .../test_matmul_3d/test_data_set_0/input_0.pb | 1 + .../test_matmul_3d/test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_matmul_4d/model.onnx | Bin 0 -> 146 bytes .../test_matmul_4d/test_data_set_0/input_0.pb | 1 + .../test_matmul_4d/test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_matmul_4d_1d/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 111 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_matmul_bcast/model.onnx | Bin 0 -> 149 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 4 + .../data/node/test_matmulinteger/model.onnx | Bin 0 -> 218 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | 1 + .../data/node/test_max_example/model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_max_float16/model.onnx | Bin 0 -> 139 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 20 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 20 bytes .../data/node/test_max_float32/model.onnx | Bin 0 -> 139 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_max_float64/model.onnx | Bin 0 -> 139 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 38 bytes .../test/data/node/test_max_int16/model.onnx | Bin 0 -> 137 bytes .../test_max_int16/test_data_set_0/input_0.pb | Bin 0 -> 20 bytes .../test_max_int16/test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 20 bytes .../test/data/node/test_max_int32/model.onnx | Bin 0 -> 137 bytes .../test_max_int32/test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_max_int32/test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../test/data/node/test_max_int64/model.onnx | Bin 0 -> 137 bytes .../test_max_int64/test_data_set_0/input_0.pb | Bin 0 -> 38 bytes .../test_max_int64/test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 38 bytes .../test/data/node/test_max_int8/model.onnx | Bin 0 -> 136 bytes .../test_max_int8/test_data_set_0/input_0.pb | 1 + .../test_max_int8/test_data_set_0/input_1.pb | 1 + .../test_max_int8/test_data_set_0/output_0.pb | 1 + .../data/node/test_max_one_input/model.onnx | Bin 0 -> 111 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_max_two_inputs/model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../test/data/node/test_max_uint16/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 20 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 20 bytes .../test/data/node/test_max_uint32/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../test/data/node/test_max_uint64/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 38 bytes .../test/data/node/test_max_uint8/model.onnx | Bin 0 -> 137 bytes .../test_max_uint8/test_data_set_0/input_0.pb | 1 + .../test_max_uint8/test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_maxpool_1d_default/model.onnx | Bin 0 -> 137 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_maxpool_2d_ceil/model.onnx | Bin 0 -> 181 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 207 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../node/test_maxpool_2d_default/model.onnx | Bin 0 -> 147 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 11548 bytes .../node/test_maxpool_2d_dilations/model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../data/node/test_maxpool_2d_pads/model.onnx | Bin 0 -> 164 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 9424 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 10816 bytes .../model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 115 bytes .../model.onnx | Bin 0 -> 208 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 178 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../test_maxpool_2d_same_lower/model.onnx | Bin 0 -> 178 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 12304 bytes .../test_maxpool_2d_same_upper/model.onnx | Bin 0 -> 178 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 12304 bytes .../node/test_maxpool_2d_strides/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12304 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1216 bytes .../node/test_maxpool_2d_uint8/model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../node/test_maxpool_3d_default/model.onnx | Bin 0 -> 158 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 393235 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 357511 bytes .../node/test_maxpool_3d_dilations/model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 274 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../model.onnx | Bin 0 -> 215 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 274 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../model.onnx | Bin 0 -> 239 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 131091 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2934 bytes .../model.onnx | Bin 0 -> 220 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 216 bytes .../model.onnx | Bin 0 -> 244 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 115 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 47 bytes .../model.onnx | Bin 0 -> 262 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 32 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 48 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 52 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 115 bytes .../model.onnx | Bin 0 -> 223 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 32 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 48 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../data/node/test_mean_example/model.onnx | Bin 0 -> 172 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_mean_one_input/model.onnx | Bin 0 -> 113 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_mean_two_inputs/model.onnx | Bin 0 -> 144 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_melweightmatrix/model.onnx | Bin 0 -> 300 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 22 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 305 bytes .../data/node/test_min_example/model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_min_float16/model.onnx | Bin 0 -> 139 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 20 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 20 bytes .../data/node/test_min_float32/model.onnx | Bin 0 -> 139 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_min_float64/model.onnx | Bin 0 -> 139 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 38 bytes .../test/data/node/test_min_int16/model.onnx | Bin 0 -> 137 bytes .../test_min_int16/test_data_set_0/input_0.pb | Bin 0 -> 20 bytes .../test_min_int16/test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 20 bytes .../test/data/node/test_min_int32/model.onnx | Bin 0 -> 137 bytes .../test_min_int32/test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_min_int32/test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../test/data/node/test_min_int64/model.onnx | Bin 0 -> 137 bytes .../test_min_int64/test_data_set_0/input_0.pb | Bin 0 -> 38 bytes .../test_min_int64/test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 38 bytes .../test/data/node/test_min_int8/model.onnx | Bin 0 -> 136 bytes .../test_min_int8/test_data_set_0/input_0.pb | 1 + .../test_min_int8/test_data_set_0/input_1.pb | 1 + .../test_min_int8/test_data_set_0/output_0.pb | 1 + .../data/node/test_min_one_input/model.onnx | Bin 0 -> 111 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_min_two_inputs/model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../test/data/node/test_min_uint16/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 20 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 20 bytes .../test/data/node/test_min_uint32/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../test/data/node/test_min_uint64/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 38 bytes .../test/data/node/test_min_uint8/model.onnx | Bin 0 -> 137 bytes .../test_min_uint8/test_data_set_0/input_0.pb | 1 + .../test_min_uint8/test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_mish/model.onnx | Bin 0 -> 85 bytes .../node/test_mish/test_data_set_0/input_0.pb | Bin 0 -> 40012 bytes .../test_mish/test_data_set_0/output_0.pb | Bin 0 -> 40012 bytes .../data/node/test_mish_expanded/model.onnx | Bin 0 -> 309 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 40012 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40012 bytes .../data/node/test_mod_broadcast/model.onnx | Bin 0 -> 127 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../data/node/test_mod_int64_fmod/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 57 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 57 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 57 bytes .../test_mod_mixed_sign_float16/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_mod_mixed_sign_float32/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../test_mod_mixed_sign_float64/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 57 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 57 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 57 bytes .../node/test_mod_mixed_sign_int16/model.onnx | Bin 0 -> 118 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../node/test_mod_mixed_sign_int32/model.onnx | Bin 0 -> 118 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../node/test_mod_mixed_sign_int64/model.onnx | Bin 0 -> 118 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 57 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 57 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 57 bytes .../node/test_mod_mixed_sign_int8/model.onnx | Bin 0 -> 117 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test/data/node/test_mod_uint16/model.onnx | Bin 0 -> 108 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test/data/node/test_mod_uint32/model.onnx | Bin 0 -> 108 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_mod_uint64/model.onnx | Bin 0 -> 108 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../test/data/node/test_mod_uint8/model.onnx | Bin 0 -> 107 bytes .../test_mod_uint8/test_data_set_0/input_0.pb | 1 + .../test_mod_uint8/test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 12 bytes .../test/data/node/test_momentum/model.onnx | Bin 0 -> 333 bytes .../test_momentum/test_data_set_0/input_0.pb | 1 + .../test_momentum/test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_momentum/test_data_set_0/input_2.pb | 1 + .../test_momentum/test_data_set_0/input_3.pb | Bin 0 -> 17 bytes .../test_momentum/test_data_set_0/input_4.pb | 1 + .../test_momentum/test_data_set_0/output_0.pb | 1 + .../test_momentum/test_data_set_0/output_1.pb | 1 + .../node/test_momentum_multiple/model.onnx | Bin 0 -> 478 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 18 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_5.pb | Bin 0 -> 18 bytes .../test_data_set_0/input_6.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_7.pb | Bin 0 -> 18 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_data_set_0/output_2.pb | 1 + .../test_data_set_0/output_3.pb | 1 + .../test/data/node/test_mul/model.onnx | Bin 0 -> 125 bytes .../node/test_mul/test_data_set_0/input_0.pb | 1 + .../node/test_mul/test_data_set_0/input_1.pb | 1 + .../node/test_mul/test_data_set_0/output_0.pb | 3 + .../test/data/node/test_mul_bcast/model.onnx | Bin 0 -> 123 bytes .../test_mul_bcast/test_data_set_0/input_0.pb | 1 + .../test_mul_bcast/test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../data/node/test_mul_example/model.onnx | Bin 0 -> 109 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_mul_int16/model.onnx | Bin 0 -> 131 bytes .../test_mul_int16/test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_mul_int16/test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test/data/node/test_mul_int8/model.onnx | Bin 0 -> 130 bytes .../test_mul_int8/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_mul_int8/test_data_set_0/input_1.pb | 7 + .../test_mul_int8/test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../test/data/node/test_mul_uint16/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test/data/node/test_mul_uint32/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_mul_uint64/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 494 bytes .../test/data/node/test_mul_uint8/model.onnx | Bin 0 -> 131 bytes .../test_mul_uint8/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_mul_uint8/test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 73 bytes .../test/data/node/test_mvn/model.onnx | Bin 0 -> 127 bytes .../node/test_mvn/test_data_set_0/input_0.pb | 2 + .../node/test_mvn/test_data_set_0/output_0.pb | 1 + .../data/node/test_mvn_expanded/model.onnx | Bin 0 -> 1714 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../node/test_mvn_expanded_ver18/model.onnx | Bin 0 -> 1901 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_neg/model.onnx | Bin 0 -> 97 bytes .../node/test_neg/test_data_set_0/input_0.pb | 1 + .../node/test_neg/test_data_set_0/output_0.pb | 1 + .../data/node/test_neg_example/model.onnx | Bin 0 -> 89 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../node/test_nesterov_momentum/model.onnx | Bin 0 -> 342 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test/data/node/test_nllloss_NC/model.onnx | Bin 0 -> 181 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_nllloss_NC_expanded/model.onnx | Bin 0 -> 1487 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | 1 + .../data/node/test_nllloss_NCd1/model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 64 bytes .../test_data_set_0/output_0.pb | 1 + .../test_nllloss_NCd1_expanded/model.onnx | Bin 0 -> 1702 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 64 bytes .../test_data_set_0/output_0.pb | 1 + .../data/node/test_nllloss_NCd1_ii/model.onnx | Bin 0 -> 211 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 64 bytes .../test_data_set_0/output_0.pb | 1 + .../test_nllloss_NCd1_ii_expanded/model.onnx | Bin 0 -> 4799 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 64 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 271 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 161 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 6057 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 161 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_nllloss_NCd1_weight/model.onnx | Bin 0 -> 224 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 2696 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_nllloss_NCd1_weight_ii/model.onnx | Bin 0 -> 248 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 5288 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_nllloss_NCd1d2/model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 449 bytes .../test_nllloss_NCd1d2_expanded/model.onnx | Bin 0 -> 1579 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 449 bytes .../model.onnx | Bin 0 -> 246 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 6244 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 212 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 2033 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 210 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 2013 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 251 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 449 bytes .../model.onnx | Bin 0 -> 2211 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 449 bytes .../model.onnx | Bin 0 -> 254 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 3324 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 252 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 2730 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 276 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 5854 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2180 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 883 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 280 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 10822 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 4341 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2179 bytes .../model.onnx | Bin 0 -> 5510 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 10822 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 4341 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2179 bytes .../model.onnx | Bin 0 -> 252 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 5445 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 269 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129627 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51866 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 3092 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129627 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51866 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 266 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129627 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51866 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25944 bytes .../model.onnx | Bin 0 -> 1974 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129627 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51866 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25944 bytes .../model.onnx | Bin 0 -> 410 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 113 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 98 bytes .../model.onnx | Bin 0 -> 380 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 113 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 98 bytes .../model.onnx | Bin 0 -> 376 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 178 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 50 bytes .../model.onnx | Bin 0 -> 383 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_3.pb | 1 + .../test_data_set_0/input_4.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 74 bytes .../model.onnx | Bin 0 -> 378 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 113 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 74 bytes .../model.onnx | Bin 0 -> 371 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 50 bytes .../model.onnx | Bin 0 -> 376 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 113 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 98 bytes .../model.onnx | Bin 0 -> 387 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 113 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 74 bytes .../model.onnx | Bin 0 -> 372 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 210 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 66 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 122 bytes .../model.onnx | Bin 0 -> 372 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 113 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 66 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 42 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 122 bytes .../data/node/test_nonzero_example/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../test/data/node/test_not_2d/model.onnx | Bin 0 -> 96 bytes .../test_not_2d/test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_not_2d/test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test/data/node/test_not_3d/model.onnx | Bin 0 -> 104 bytes .../test_not_3d/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_not_3d/test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../test/data/node/test_not_4d/model.onnx | Bin 0 -> 112 bytes .../test_not_4d/test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../test_not_4d/test_data_set_0/output_0.pb | Bin 0 -> 378 bytes .../test_onehot_negative_indices/model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 131 bytes .../model.onnx | Bin 0 -> 192 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 71 bytes .../node/test_onehot_with_axis/model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 174 bytes .../test_onehot_with_negative_axis/model.onnx | Bin 0 -> 207 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 174 bytes .../node/test_onehot_without_axis/model.onnx | Bin 0 -> 171 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 156 bytes .../model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 46 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 34 bytes .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 42 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 161 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 42 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 34 bytes .../model.onnx | Bin 0 -> 157 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 13 bytes .../model.onnx | Bin 0 -> 134 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 13 bytes .../model.onnx | Bin 0 -> 129 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 13 bytes .../model.onnx | Bin 0 -> 127 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 13 bytes .../model.onnx | Bin 0 -> 167 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 18 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 13 bytes .../model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 42 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 163 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 42 bytes .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_or2d/model.onnx | Bin 0 -> 115 bytes .../node/test_or2d/test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../node/test_or2d/test_data_set_0/input_1.pb | Bin 0 -> 23 bytes .../test_or2d/test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../test/data/node/test_or3d/model.onnx | Bin 0 -> 127 bytes .../node/test_or3d/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../node/test_or3d/test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_or3d/test_data_set_0/output_0.pb | Bin 0 -> 74 bytes .../test/data/node/test_or4d/model.onnx | Bin 0 -> 139 bytes .../node/test_or4d/test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../node/test_or4d/test_data_set_0/input_1.pb | Bin 0 -> 376 bytes .../test_or4d/test_data_set_0/output_0.pb | Bin 0 -> 377 bytes .../data/node/test_or_bcast3v1d/model.onnx | Bin 0 -> 127 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 74 bytes .../data/node/test_or_bcast3v2d/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 31 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 74 bytes .../data/node/test_or_bcast4v2d/model.onnx | Bin 0 -> 139 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 41 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 377 bytes .../data/node/test_or_bcast4v3d/model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 377 bytes .../data/node/test_or_bcast4v4d/model.onnx | Bin 0 -> 147 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 105 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 377 bytes .../test/data/node/test_pow/model.onnx | Bin 0 -> 125 bytes .../node/test_pow/test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../node/test_pow/test_data_set_0/input_1.pb | 1 + .../node/test_pow/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_pow_bcast_array/model.onnx | Bin 0 -> 121 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 35 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../node/test_pow_bcast_scalar/model.onnx | Bin 0 -> 110 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../data/node/test_pow_example/model.onnx | Bin 0 -> 109 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_pow_types_float32_int32/model.onnx | Bin 0 -> 121 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_pow_types_float32_int64/model.onnx | Bin 0 -> 121 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_pow_types_float32_uint32/model.onnx | Bin 0 -> 122 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_pow_types_float32_uint64/model.onnx | Bin 0 -> 122 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_pow_types_int32_float32/model.onnx | Bin 0 -> 121 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_pow_types_int32_int32/model.onnx | Bin 0 -> 119 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_pow_types_int64_float32/model.onnx | Bin 0 -> 121 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../test_pow_types_int64_int64/model.onnx | Bin 0 -> 119 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../data/node/test_prelu_broadcast/model.onnx | Bin 0 -> 139 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_prelu_broadcast_expanded/model.onnx | Bin 0 -> 676 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_prelu_example/model.onnx | Bin 0 -> 145 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_prelu_example_expanded/model.onnx | Bin 0 -> 665 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../data/node/test_qlinearconv/model.onnx | Bin 0 -> 360 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/input_6.pb | 1 + .../test_data_set_0/input_7.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 64 bytes .../model.onnx | Bin 0 -> 372 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/input_4.pb | 2 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/input_6.pb | 2 + .../test_data_set_0/input_7.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 372 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/input_6.pb | 1 + .../test_data_set_0/input_7.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 373 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 19 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/input_4.pb | 2 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/input_6.pb | 2 + .../test_data_set_0/input_7.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 373 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 19 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 23 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/input_6.pb | 1 + .../test_data_set_0/input_7.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 384 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_4.pb | 2 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/input_6.pb | 2 + .../test_data_set_0/input_7.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 384 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/input_6.pb | 1 + .../test_data_set_0/input_7.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 385 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_4.pb | 2 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/input_6.pb | 2 + .../test_data_set_0/input_7.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 385 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/input_5.pb | 1 + .../test_data_set_0/input_6.pb | 1 + .../test_data_set_0/input_7.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_quantizelinear/model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../node/test_quantizelinear_axis/model.onnx | Bin 0 -> 207 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 87 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 27 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 245 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 41 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 28 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 219 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 41 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../test_quantizelinear_e4m3fn/model.onnx | Bin 0 -> 181 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 14 bytes .../node/test_quantizelinear_e5m2/model.onnx | Bin 0 -> 179 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 14 bytes .../test_quantizelinear_float4e2m1/model.onnx | Bin 0 -> 210 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 27 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_quantizelinear_int16/model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../node/test_quantizelinear_int2/model.onnx | Bin 0 -> 204 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 27 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_quantizelinear_int4/model.onnx | Bin 0 -> 204 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 27 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_quantizelinear_uint16/model.onnx | Bin 0 -> 177 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 57 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../node/test_quantizelinear_uint2/model.onnx | Bin 0 -> 205 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 27 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_quantizelinear_uint4/model.onnx | Bin 0 -> 205 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 27 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 18 bytes .../model.onnx | Bin 0 -> 177 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 3091 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 18 bytes .../model.onnx | Bin 0 -> 3054 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 13 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 18 bytes .../model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 22 bytes .../model.onnx | Bin 0 -> 2647 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 22 bytes .../model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 22 bytes .../model.onnx | Bin 0 -> 2647 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 22 bytes .../test/data/node/test_reciprocal/model.onnx | Bin 0 -> 111 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_reciprocal_example/model.onnx | Bin 0 -> 103 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../model.onnx | Bin 0 -> 200 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 415 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 199 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 412 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 393 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 390 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_reduce_l1_empty_set/model.onnx | Bin 0 -> 180 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 355 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 379 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 376 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 421 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 201 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 418 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 200 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 980 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 199 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 971 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 922 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 913 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_reduce_l2_empty_set/model.onnx | Bin 0 -> 180 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 800 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 872 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 863 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 998 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 201 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 989 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_reduce_log_sum_asc_axes/model.onnx | Bin 0 -> 180 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 28 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 373 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 28 bytes .../test_data_set_0/output_0.pb | 1 + .../test_reduce_log_sum_default/model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 378 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../test_reduce_log_sum_desc_axes/model.onnx | Bin 0 -> 181 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 28 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 376 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 28 bytes .../test_data_set_0/output_0.pb | 1 + .../test_reduce_log_sum_empty_set/model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 384 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 216 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 27 bytes .../model.onnx | Bin 0 -> 1121 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 27 bytes .../model.onnx | Bin 0 -> 215 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 1112 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 206 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 65 bytes .../model.onnx | Bin 0 -> 1062 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 65 bytes .../model.onnx | Bin 0 -> 205 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 1053 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 196 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 940 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 203 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 67 bytes .../model.onnx | Bin 0 -> 1003 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 67 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 994 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 217 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 67 bytes .../model.onnx | Bin 0 -> 1130 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 67 bytes .../model.onnx | Bin 0 -> 216 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 1121 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 112 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 396 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_reduce_max_bool_inputs/model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 22 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 175 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 175 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 192 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_reduce_max_empty_set/model.onnx | Bin 0 -> 182 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 203 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 204 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 203 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 194 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 193 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 205 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 204 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_reduce_min_bool_inputs/model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 22 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 175 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 192 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_reduce_min_empty_set/model.onnx | Bin 0 -> 182 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 203 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 178 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 177 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 194 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 193 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../test_reduce_prod_empty_set/model.onnx | Bin 0 -> 184 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 205 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 204 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 201 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 192 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 191 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 223 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 231 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 67 bytes .../node/test_reduce_sum_empty_set/model.onnx | Bin 0 -> 182 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 204 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 203 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 215 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 465 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 214 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 462 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 205 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 443 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 41 bytes .../model.onnx | Bin 0 -> 204 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 440 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 195 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 405 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 426 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 201 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 423 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 216 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 468 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../model.onnx | Bin 0 -> 215 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 465 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_reflect_pad/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 256 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 76 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 520 bytes .../test_regex_full_match_basic/model.onnx | Bin 0 -> 148 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 12 bytes .../model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test_regex_full_match_empty/model.onnx | Bin 0 -> 180 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 9 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 11 bytes .../test/data/node/test_relu/model.onnx | Bin 0 -> 99 bytes .../node/test_relu/test_data_set_0/input_0.pb | 1 + .../test_relu/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_relu_expanded_ver18/model.onnx | Bin 0 -> 335 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 192 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 20 bytes .../test_reshape_extended_dims/model.onnx | Bin 0 -> 172 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_reshape_negative_dim/model.onnx | Bin 0 -> 167 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 181 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | 1 + .../data/node/test_reshape_one_dim/model.onnx | Bin 0 -> 154 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_reshape_reduced_dims/model.onnx | Bin 0 -> 163 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 173 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 180 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_reshape_zero_dim/model.onnx | Bin 0 -> 167 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 258 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 252 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 214 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 47 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 254 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 47 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 216 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 268 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 47 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 185 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 213 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 215 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 263 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 27 bytes .../model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 47 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 27 bytes .../model.onnx | Bin 0 -> 258 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 47 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 23 bytes .../model.onnx | Bin 0 -> 260 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 47 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 39 bytes .../test_resize_tf_crop_and_resize/model.onnx | Bin 0 -> 261 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 43 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 285 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 285 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 313 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 43 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 51 bytes .../model.onnx | Bin 0 -> 184 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 256 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 250 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 244 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 186 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../model.onnx | Bin 0 -> 252 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 79 bytes .../model.onnx | Bin 0 -> 266 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 95 bytes .../model.onnx | Bin 0 -> 188 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 111 bytes .../model.onnx | Bin 0 -> 212 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 111 bytes .../model.onnx | Bin 0 -> 212 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 111 bytes .../model.onnx | Bin 0 -> 183 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 376 bytes .../model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 240 bytes .../model.onnx | Bin 0 -> 211 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 240 bytes .../model.onnx | Bin 0 -> 211 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 240 bytes .../model.onnx | Bin 0 -> 277 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 285 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 256 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 212 bytes .../model.onnx | Bin 0 -> 258 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../model.onnx | Bin 0 -> 304 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 272 bytes .../test_reversesequence_batch/model.onnx | Bin 0 -> 201 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 53 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../node/test_reversesequence_time/model.onnx | Bin 0 -> 200 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 75 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 53 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../model.onnx | Bin 0 -> 163 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 2869 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 2865 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 178 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 3282 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 182 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 3286 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 3145 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 198 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 3141 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 194 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 3137 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 213 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../model.onnx | Bin 0 -> 3564 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../model.onnx | Bin 0 -> 217 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 3 + .../model.onnx | Bin 0 -> 3568 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 3 + .../model.onnx | Bin 0 -> 221 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 3572 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 2893 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 496 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 183 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 3 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 2889 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 3 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 179 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 2885 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 175 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 2881 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 194 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 3298 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 198 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 3302 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | 4 + .../model.onnx | Bin 0 -> 3306 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | 4 + .../model.onnx | Bin 0 -> 206 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 3310 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../model.onnx | Bin 0 -> 3094 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 496 bytes .../data/node/test_rnn_seq_length/model.onnx | Bin 0 -> 215 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 85 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../test_roialign_aligned_false/model.onnx | Bin 0 -> 352 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 416 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 62 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | 3 + .../test_roialign_aligned_true/model.onnx | Bin 0 -> 344 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 416 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 62 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | 2 + .../node/test_roialign_mode_max/model.onnx | Bin 0 -> 363 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 416 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 62 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | 2 + .../node/test_rotary_embedding/model.onnx | Bin 0 -> 275 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 70 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../test_rotary_embedding_3d_input/model.onnx | Bin 0 -> 294 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 786 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 70 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 787 bytes .../model.onnx | Bin 0 -> 7049 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 786 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 70 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 787 bytes .../test_rotary_embedding_expanded/model.onnx | Bin 0 -> 5952 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 70 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 307 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 70 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 9026 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 820 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 70 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 253 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 7104 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 285 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 10581 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 293 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 7858 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 352 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 420 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 70 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 10558 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 420 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 70 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 789 bytes .../model.onnx | Bin 0 -> 320 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 420 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 70 bytes .../test_data_set_0/output_0.pb | 4 + .../model.onnx | Bin 0 -> 7090 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 788 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 420 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 70 bytes .../test_data_set_0/output_0.pb | 4 + .../test/data/node/test_round/model.onnx | Bin 0 -> 85 bytes .../test_round/test_data_set_0/input_0.pb | Bin 0 -> 69 bytes .../test_round/test_data_set_0/output_0.pb | Bin 0 -> 69 bytes .../node/test_scan9_multi_state/model.onnx | Bin 0 -> 533 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 27 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 28 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 22 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 35 bytes .../data/node/test_scan9_scalar/model.onnx | Bin 0 -> 327 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 29 bytes .../test/data/node/test_scan9_sum/model.onnx | Bin 0 -> 356 bytes .../test_scan9_sum/test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_scan9_sum/test_data_set_0/input_1.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 35 bytes .../test/data/node/test_scan_sum/model.onnx | Bin 0 -> 373 bytes .../test_scan_sum/test_data_set_0/input_0.pb | Bin 0 -> 25 bytes .../test_scan_sum/test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_scan_sum/test_data_set_0/output_0.pb | Bin 0 -> 19 bytes .../test_scan_sum/test_data_set_0/output_1.pb | Bin 0 -> 37 bytes .../model.onnx | Bin 0 -> 216 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 34 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 250 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 34 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 228 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 34 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 246 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 34 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 246 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 34 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 246 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 34 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../model.onnx | Bin 0 -> 206 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 41 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../node/test_scatter_with_axis/model.onnx | Bin 0 -> 199 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 34 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 31 bytes .../node/test_scatter_without_axis/model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 65 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 41 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 47 bytes .../test/data/node/test_scatternd/model.onnx | Bin 0 -> 192 bytes .../test_scatternd/test_data_set_0/input_0.pb | Bin 0 -> 273 bytes .../test_scatternd/test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_scatternd/test_data_set_0/input_2.pb | Bin 0 -> 148 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 270 bytes .../data/node/test_scatternd_add/model.onnx | Bin 0 -> 217 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 273 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 148 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 270 bytes .../data/node/test_scatternd_max/model.onnx | Bin 0 -> 217 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 273 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 148 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 270 bytes .../model.onnx | Bin 0 -> 222 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 27 bytes .../data/node/test_scatternd_min/model.onnx | Bin 0 -> 217 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 273 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 148 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 270 bytes .../model.onnx | Bin 0 -> 222 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 30 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 27 bytes .../node/test_scatternd_multiply/model.onnx | Bin 0 -> 222 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 273 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 148 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 270 bytes .../model.onnx | Bin 0 -> 231 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 156 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1672 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 156 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 282 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 156 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 381 bytes .../model.onnx | Bin 0 -> 1959 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 156 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 381 bytes .../model.onnx | Bin 0 -> 250 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 10818 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 4336 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2176 bytes .../model.onnx | Bin 0 -> 1789 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 10818 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 4336 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2176 bytes .../model.onnx | Bin 0 -> 309 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 10818 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 4336 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2176 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 10825 bytes .../model.onnx | Bin 0 -> 2091 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 10818 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 4336 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2176 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 10825 bytes .../model.onnx | Bin 0 -> 212 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1639 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 259 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 1921 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 2 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 229 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129623 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51861 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1614 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129623 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51861 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 296 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129623 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51861 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 129630 bytes .../model.onnx | Bin 0 -> 1913 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129623 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51861 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 129630 bytes .../model.onnx | Bin 0 -> 236 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129623 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51861 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25941 bytes .../model.onnx | Bin 0 -> 1663 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129623 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51861 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25941 bytes .../model.onnx | Bin 0 -> 303 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129623 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51861 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25941 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 129630 bytes .../model.onnx | Bin 0 -> 1965 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 129623 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 51861 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25941 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 129630 bytes .../test/data/node/test_sce_mean/model.onnx | Bin 0 -> 148 bytes .../test_sce_mean/test_data_set_0/input_0.pb | 1 + .../test_sce_mean/test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_sce_mean/test_data_set_0/output_0.pb | 1 + .../data/node/test_sce_mean_3d/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_sce_mean_3d_expanded/model.onnx | Bin 0 -> 1305 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_sce_mean_3d_log_prob/model.onnx | Bin 0 -> 211 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 140 bytes .../model.onnx | Bin 0 -> 1572 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 140 bytes .../node/test_sce_mean_expanded/model.onnx | Bin 0 -> 1252 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_sce_mean_log_prob/model.onnx | Bin 0 -> 196 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 1512 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_sce_mean_no_weight_ii/model.onnx | Bin 0 -> 183 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../test_sce_mean_no_weight_ii_3d/model.onnx | Bin 0 -> 194 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1522 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 245 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 140 bytes .../model.onnx | Bin 0 -> 1802 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 140 bytes .../test_sce_mean_no_weight_ii_4d/model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 856 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 350 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1530 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 856 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 350 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 257 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 856 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 350 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 863 bytes .../model.onnx | Bin 0 -> 1814 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 856 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 350 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 863 bytes .../model.onnx | Bin 0 -> 1469 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 230 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 1742 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../data/node/test_sce_mean_weight/model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_sce_mean_weight_expanded/model.onnx | Bin 0 -> 1378 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_sce_mean_weight_ii/model.onnx | Bin 0 -> 200 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_sce_mean_weight_ii_3d/model.onnx | Bin 0 -> 211 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1497 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 262 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 140 bytes .../model.onnx | Bin 0 -> 1774 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 140 bytes .../test_sce_mean_weight_ii_4d/model.onnx | Bin 0 -> 219 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 856 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 350 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1505 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 856 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 350 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 274 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 856 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 350 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 863 bytes .../model.onnx | Bin 0 -> 1786 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 856 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 350 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | Bin 0 -> 863 bytes .../model.onnx | Bin 0 -> 1444 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 247 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 1714 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_sce_mean_weight_log_prob/model.onnx | Bin 0 -> 223 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 1644 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test/data/node/test_sce_none/model.onnx | Bin 0 -> 153 bytes .../test_sce_none/test_data_set_0/input_0.pb | 1 + .../test_sce_none/test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_sce_none/test_data_set_0/output_0.pb | 1 + .../node/test_sce_none_expanded/model.onnx | Bin 0 -> 1256 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_sce_none_log_prob/model.onnx | Bin 0 -> 200 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 1516 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../node/test_sce_none_weights/model.onnx | Bin 0 -> 181 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_sce_none_weights_expanded/model.onnx | Bin 0 -> 1397 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_sce_none_weights_log_prob/model.onnx | Bin 0 -> 228 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 1664 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test/data/node/test_sce_sum/model.onnx | Bin 0 -> 146 bytes .../test_sce_sum/test_data_set_0/input_0.pb | 1 + .../test_sce_sum/test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_sce_sum/test_data_set_0/output_0.pb | 1 + .../node/test_sce_sum_expanded/model.onnx | Bin 0 -> 1236 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_sce_sum_log_prob/model.onnx | Bin 0 -> 194 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test_sce_sum_log_prob_expanded/model.onnx | Bin 0 -> 1494 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../test/data/node/test_selu/model.onnx | Bin 0 -> 133 bytes .../node/test_selu/test_data_set_0/input_0.pb | 1 + .../test_selu/test_data_set_0/output_0.pb | 2 + .../data/node/test_selu_default/model.onnx | Bin 0 -> 107 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 1680 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_selu_example/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 1664 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../node/test_selu_expanded_ver18/model.onnx | Bin 0 -> 1464 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_sequence_insert_at_back/model.onnx | Bin 0 -> 181 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 108 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 147 bytes .../test_sequence_insert_at_front/model.onnx | Bin 0 -> 216 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 108 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 147 bytes .../model.onnx | Bin 0 -> 272 bytes .../test_data_set_0/input_0.pb | 5 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | 6 + .../model.onnx | Bin 0 -> 2583 bytes .../test_data_set_0/input_0.pb | 5 + .../test_data_set_0/input_1.pb | 2 + .../test_data_set_0/output_0.pb | 6 + .../model.onnx | Bin 0 -> 268 bytes .../test_data_set_0/input_0.pb | 3 + .../test_data_set_0/input_1.pb | 4 + .../test_data_set_0/output_0.pb | 3 + .../model.onnx | Bin 0 -> 2502 bytes .../test_data_set_0/input_0.pb | 3 + .../test_data_set_0/input_1.pb | 4 + .../test_data_set_0/output_0.pb | 3 + .../model.onnx | Bin 0 -> 248 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17452 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 106 bytes .../model.onnx | Bin 0 -> 2232 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17452 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 106 bytes .../model.onnx | Bin 0 -> 220 bytes .../test_data_set_0/input_0.pb | 5 + .../test_data_set_0/output_0.pb | 5 + .../model.onnx | Bin 0 -> 350 bytes .../test_data_set_0/input_0.pb | 3 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 3 + .../test_data_set_0/output_1.pb | 2 + .../model.onnx | Bin 0 -> 3664 bytes .../test_data_set_0/input_0.pb | 3 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 3 + .../test_data_set_0/output_1.pb | 2 + .../model.onnx | Bin 0 -> 2307 bytes .../test_data_set_0/input_0.pb | 5 + .../test_data_set_0/output_0.pb | 5 + .../model.onnx | Bin 0 -> 346 bytes .../test_data_set_0/input_0.pb | 3 + .../test_data_set_0/input_1.pb | 3 + .../test_data_set_0/output_0.pb | 3 + .../test_data_set_0/output_1.pb | 3 + .../model.onnx | Bin 0 -> 3524 bytes .../test_data_set_0/input_0.pb | 3 + .../test_data_set_0/input_1.pb | 3 + .../test_data_set_0/output_0.pb | 3 + .../test_data_set_0/output_1.pb | 3 + .../test/data/node/test_shape/model.onnx | Bin 0 -> 93 bytes .../test_shape/test_data_set_0/input_0.pb | 1 + .../test_shape/test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../data/node/test_shape_clip_end/model.onnx | Bin 0 -> 114 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../node/test_shape_clip_start/model.onnx | Bin 0 -> 127 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 33 bytes .../data/node/test_shape_end_1/model.onnx | Bin 0 -> 111 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../node/test_shape_end_negative_1/model.onnx | Bin 0 -> 129 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../data/node/test_shape_example/model.onnx | Bin 0 -> 97 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../data/node/test_shape_start_1/model.onnx | Bin 0 -> 115 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../node/test_shape_start_1_end_2/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../model.onnx | Bin 0 -> 151 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 9 bytes .../test_shape_start_negative_1/model.onnx | Bin 0 -> 133 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../data/node/test_shrink_hard/model.onnx | Bin 0 -> 109 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../model.onnx | Bin 0 -> 1761 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../data/node/test_shrink_soft/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../model.onnx | Bin 0 -> 1761 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../test/data/node/test_sigmoid/model.onnx | Bin 0 -> 105 bytes .../test_sigmoid/test_data_set_0/input_0.pb | 1 + .../test_sigmoid/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_sigmoid_example/model.onnx | Bin 0 -> 97 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_sign/model.onnx | Bin 0 -> 83 bytes .../node/test_sign/test_data_set_0/input_0.pb | Bin 0 -> 53 bytes .../test_sign/test_data_set_0/output_0.pb | Bin 0 -> 53 bytes .../node/test_simple_rnn_batchwise/model.onnx | Bin 0 -> 242 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 77 bytes .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../node/test_simple_rnn_defaults/model.onnx | Bin 0 -> 196 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 229 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 49 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/input_3.pb | Bin 0 -> 51 bytes .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_sin/model.onnx | Bin 0 -> 97 bytes .../node/test_sin/test_data_set_0/input_0.pb | 1 + .../node/test_sin/test_data_set_0/output_0.pb | 1 + .../data/node/test_sin_example/model.onnx | Bin 0 -> 89 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_sinh/model.onnx | Bin 0 -> 99 bytes .../node/test_sinh/test_data_set_0/input_0.pb | 1 + .../test_sinh/test_data_set_0/output_0.pb | 1 + .../data/node/test_sinh_example/model.onnx | Bin 0 -> 91 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_size/model.onnx | Bin 0 -> 87 bytes .../node/test_size/test_data_set_0/input_0.pb | 1 + .../test_size/test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../data/node/test_size_example/model.onnx | Bin 0 -> 91 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 15 bytes .../test/data/node/test_slice/model.onnx | Bin 0 -> 212 bytes .../test_slice/test_data_set_0/input_0.pb | Bin 0 -> 4014 bytes .../test_slice/test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_slice/test_data_set_0/input_2.pb | Bin 0 -> 28 bytes .../test_slice/test_data_set_0/input_3.pb | Bin 0 -> 28 bytes .../test_slice/test_data_set_0/input_4.pb | Bin 0 -> 29 bytes .../test_slice/test_data_set_0/output_0.pb | 3 + .../node/test_slice_default_axes/model.onnx | Bin 0 -> 171 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4014 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 36 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 814 bytes .../node/test_slice_default_steps/model.onnx | Bin 0 -> 198 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4014 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 36 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 36 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 814 bytes .../test_slice_end_out_of_bounds/model.onnx | Bin 0 -> 230 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4014 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 22 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 20 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 20 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3614 bytes .../test/data/node/test_slice_neg/model.onnx | Bin 0 -> 216 bytes .../test_slice_neg/test_data_set_0/input_0.pb | Bin 0 -> 4014 bytes .../test_slice_neg/test_data_set_0/input_1.pb | Bin 0 -> 22 bytes .../test_slice_neg/test_data_set_0/input_2.pb | 1 + .../test_slice_neg/test_data_set_0/input_3.pb | Bin 0 -> 20 bytes .../test_slice_neg/test_data_set_0/input_4.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3614 bytes .../data/node/test_slice_neg_steps/model.onnx | Bin 0 -> 222 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4014 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 36 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 36 bytes .../test_data_set_0/input_4.pb | 1 + .../test_data_set_0/output_0.pb | 3 + .../node/test_slice_negative_axes/model.onnx | Bin 0 -> 198 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4014 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 36 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 36 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 814 bytes .../test_slice_start_out_of_bounds/model.onnx | Bin 0 -> 232 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 4014 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 22 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 20 bytes .../test_data_set_0/input_3.pb | Bin 0 -> 20 bytes .../test_data_set_0/input_4.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 13 bytes .../data/node/test_softmax_axis_0/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_softmax_axis_0_expanded/model.onnx | Bin 0 -> 859 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 905 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../data/node/test_softmax_axis_1/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_softmax_axis_1_expanded/model.onnx | Bin 0 -> 859 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 905 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../data/node/test_softmax_axis_2/model.onnx | Bin 0 -> 125 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test_softmax_axis_2_expanded/model.onnx | Bin 0 -> 859 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 905 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_softmax_default_axis/model.onnx | Bin 0 -> 118 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 951 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 994 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_softmax_example/model.onnx | Bin 0 -> 105 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | 1 + .../test_softmax_example_expanded/model.onnx | Bin 0 -> 881 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 919 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_softmax_large_number/model.onnx | Bin 0 -> 110 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 943 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 986 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 43 bytes .../test_data_set_0/output_0.pb | 1 + .../test_softmax_negative_axis/model.onnx | Bin 0 -> 141 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 963 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 1007 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_softplus/model.onnx | Bin 0 -> 107 bytes .../test_softplus/test_data_set_0/input_0.pb | 1 + .../test_softplus/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_softplus_example/model.onnx | Bin 0 -> 99 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 653 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | 1 + .../test_softplus_expanded_ver18/model.onnx | Bin 0 -> 596 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_softsign/model.onnx | Bin 0 -> 107 bytes .../test_softsign/test_data_set_0/input_0.pb | 1 + .../test_softsign/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../node/test_softsign_example/model.onnx | Bin 0 -> 99 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 661 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test_softsign_expanded_ver18/model.onnx | Bin 0 -> 605 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_spacetodepth/model.onnx | Bin 0 -> 141 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 592 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 592 bytes .../node/test_spacetodepth_example/model.onnx | Bin 0 -> 149 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 111 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 111 bytes .../model.onnx | Bin 0 -> 254 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 41 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 20 bytes .../model.onnx | Bin 0 -> 249 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 79 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 42 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 42 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 34 bytes .../model.onnx | Bin 0 -> 212 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 24 bytes .../model.onnx | Bin 0 -> 232 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 24 bytes .../node/test_split_equal_parts_2d/model.onnx | Bin 0 -> 202 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 42 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 42 bytes .../model.onnx | Bin 0 -> 190 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 42 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 42 bytes .../model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 24 bytes .../model.onnx | Bin 0 -> 229 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 24 bytes .../node/test_split_to_sequence_1/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 86 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 19 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 109 bytes .../node/test_split_to_sequence_2/model.onnx | Bin 0 -> 163 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 86 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 99 bytes .../model.onnx | Bin 0 -> 161 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 86 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 127 bytes .../model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 221 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 34 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 50 bytes .../model.onnx | Bin 0 -> 221 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 63 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 34 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 50 bytes .../model.onnx | Bin 0 -> 206 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 206 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 32 bytes .../model.onnx | Bin 0 -> 229 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 16 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 16 bytes .../model.onnx | Bin 0 -> 229 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 13 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 16 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 16 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 16 bytes .../test/data/node/test_sqrt/model.onnx | Bin 0 -> 99 bytes .../node/test_sqrt/test_data_set_0/input_0.pb | 1 + .../test_sqrt/test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_sqrt_example/model.onnx | Bin 0 -> 91 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_squeeze/model.onnx | Bin 0 -> 135 bytes .../test_squeeze/test_data_set_0/input_0.pb | 1 + .../test_squeeze/test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_squeeze/test_data_set_0/output_0.pb | 1 + .../test_squeeze_negative_axes/model.onnx | Bin 0 -> 149 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_stft/model.onnx | Bin 0 -> 199 bytes .../node/test_stft/test_data_set_0/input_0.pb | Bin 0 -> 532 bytes .../node/test_stft/test_data_set_0/input_1.pb | Bin 0 -> 24 bytes .../node/test_stft/test_data_set_0/input_2.pb | Bin 0 -> 26 bytes .../test_stft/test_data_set_0/output_0.pb | Bin 0 -> 1101 bytes .../node/test_stft_with_window/model.onnx | Bin 0 -> 201 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 532 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 24 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 78 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1101 bytes .../data/node/test_string_concat/model.onnx | Bin 0 -> 130 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 14 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_string_concat_utf8/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_string_split_basic/model.onnx | Bin 0 -> 176 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 193 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 50 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 192 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | Bin 0 -> 38 bytes .../test_string_split_empty_tensor/model.onnx | Bin 0 -> 162 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 7 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 18 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 14 bytes .../test_string_split_maxsplit/model.onnx | Bin 0 -> 189 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 83 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 48 bytes .../test_string_split_no_delimiter/model.onnx | Bin 0 -> 164 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_data_set_0/output_1.pb | Bin 0 -> 38 bytes .../model.onnx | Bin 0 -> 222 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 197 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 222 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 214 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 9 bytes .../model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 156 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_sub/model.onnx | Bin 0 -> 125 bytes .../node/test_sub/test_data_set_0/input_0.pb | 1 + .../node/test_sub/test_data_set_0/input_1.pb | 1 + .../node/test_sub/test_data_set_0/output_0.pb | 1 + .../test/data/node/test_sub_bcast/model.onnx | Bin 0 -> 123 bytes .../test_sub_bcast/test_data_set_0/input_0.pb | 1 + .../test_sub_bcast/test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../data/node/test_sub_example/model.onnx | Bin 0 -> 109 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_sub_int16/model.onnx | Bin 0 -> 131 bytes .../test_sub_int16/test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_sub_int16/test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test/data/node/test_sub_int8/model.onnx | Bin 0 -> 130 bytes .../test_sub_int8/test_data_set_0/input_0.pb | 1 + .../test_sub_int8/test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_sub_int8/test_data_set_0/output_0.pb | 8 + .../test/data/node/test_sub_uint16/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../test/data/node/test_sub_uint32/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_sub_uint64/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 494 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 494 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 494 bytes .../test/data/node/test_sub_uint8/model.onnx | Bin 0 -> 131 bytes .../test_sub_uint8/test_data_set_0/input_0.pb | 1 + .../test_sub_uint8/test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_data_set_0/output_0.pb | 5 + .../data/node/test_sum_example/model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_sum_one_input/model.onnx | Bin 0 -> 111 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../data/node/test_sum_two_inputs/model.onnx | Bin 0 -> 142 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 26 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 26 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 26 bytes .../test/data/node/test_swish/model.onnx | Bin 0 -> 102 bytes .../test_swish/test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_swish/test_data_set_0/output_0.pb | 1 + .../data/node/test_swish_expanded/model.onnx | Bin 0 -> 548 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | 1 + .../test/data/node/test_tan/model.onnx | Bin 0 -> 97 bytes .../node/test_tan/test_data_set_0/input_0.pb | 1 + .../node/test_tan/test_data_set_0/output_0.pb | 1 + .../data/node/test_tan_example/model.onnx | Bin 0 -> 89 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../test/data/node/test_tanh/model.onnx | Bin 0 -> 99 bytes .../node/test_tanh/test_data_set_0/input_0.pb | 1 + .../test_tanh/test_data_set_0/output_0.pb | 1 + .../data/node/test_tanh_example/model.onnx | Bin 0 -> 91 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 21 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../data/node/test_tensorscatter/model.onnx | Bin 0 -> 273 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 185 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 60 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 188 bytes .../node/test_tensorscatter_3d/model.onnx | Bin 0 -> 245 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 263 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 138 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 266 bytes .../test_tensorscatter_circular/model.onnx | Bin 0 -> 284 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 185 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 100 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 188 bytes .../model.onnx | Bin 0 -> 323 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 67 bytes .../model.onnx | Bin 0 -> 323 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 67 bytes .../model.onnx | Bin 0 -> 325 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 67 bytes .../model.onnx | Bin 0 -> 310 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 57 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 37 bytes .../model.onnx | Bin 0 -> 298 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 57 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 21 bytes .../model.onnx | Bin 0 -> 309 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 57 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 37 bytes .../model.onnx | Bin 0 -> 311 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 57 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 37 bytes .../data/node/test_thresholdedrelu/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test_thresholdedrelu_default/model.onnx | Bin 0 -> 129 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 993 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test_thresholdedrelu_example/model.onnx | Bin 0 -> 130 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../model.onnx | Bin 0 -> 977 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../model.onnx | Bin 0 -> 905 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test/data/node/test_tile/model.onnx | Bin 0 -> 127 bytes .../node/test_tile/test_data_set_0/input_0.pb | 1 + .../node/test_tile/test_data_set_0/input_1.pb | Bin 0 -> 41 bytes .../test_tile/test_data_set_0/output_0.pb | 2 + .../node/test_tile_precomputed/model.onnx | Bin 0 -> 123 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 27 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 25 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../test/data/node/test_top_k/model.onnx | Bin 0 -> 172 bytes .../test_top_k/test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_top_k/test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_top_k/test_data_set_0/output_0.pb | Bin 0 -> 52 bytes .../test_top_k/test_data_set_0/output_1.pb | Bin 0 -> 89 bytes .../node/test_top_k_negative_axis/model.onnx | Bin 0 -> 195 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 52 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 89 bytes .../node/test_top_k_same_values/model.onnx | Bin 0 -> 172 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 41 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 39 bytes .../node/test_top_k_same_values_2d/model.onnx | Bin 0 -> 187 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 107 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 88 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 89 bytes .../test_top_k_same_values_largest/model.onnx | Bin 0 -> 180 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 41 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 39 bytes .../data/node/test_top_k_smallest/model.onnx | Bin 0 -> 212 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 59 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 52 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 89 bytes .../data/node/test_top_k_uint64/model.onnx | Bin 0 -> 179 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 107 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 88 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 89 bytes .../node/test_training_dropout/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 11 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test_training_dropout_default/model.onnx | Bin 0 -> 168 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 11 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../model.onnx | Bin 0 -> 201 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 11 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 73 bytes .../test_training_dropout_mask/model.onnx | Bin 0 -> 193 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 11 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 254 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 73 bytes .../model.onnx | Bin 0 -> 171 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 11 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 204 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 11 bytes .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_data_set_0/output_1.pb | 1 + .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | Bin 0 -> 170 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../node/test_transpose_default/model.onnx | Bin 0 -> 141 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test/data/node/test_tril/model.onnx | Bin 0 -> 106 bytes .../node/test_tril/test_data_set_0/input_0.pb | Bin 0 -> 172 bytes .../test_tril/test_data_set_0/output_0.pb | Bin 0 -> 172 bytes .../test/data/node/test_tril_neg/model.onnx | Bin 0 -> 126 bytes .../test_tril_neg/test_data_set_0/input_0.pb | Bin 0 -> 172 bytes .../test_tril_neg/test_data_set_0/input_1.pb | 1 + .../test_tril_neg/test_data_set_0/output_0.pb | Bin 0 -> 172 bytes .../node/test_tril_one_row_neg/model.onnx | Bin 0 -> 126 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../data/node/test_tril_out_neg/model.onnx | Bin 0 -> 130 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 172 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 172 bytes .../data/node/test_tril_out_pos/model.onnx | Bin 0 -> 130 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 172 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 172 bytes .../test/data/node/test_tril_pos/model.onnx | Bin 0 -> 126 bytes .../test_tril_pos/test_data_set_0/input_0.pb | Bin 0 -> 172 bytes .../test_tril_pos/test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_tril_pos/test_data_set_0/output_0.pb | Bin 0 -> 172 bytes .../data/node/test_tril_square/model.onnx | Bin 0 -> 121 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 158 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 158 bytes .../data/node/test_tril_square_neg/model.onnx | Bin 0 -> 141 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 158 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 158 bytes .../test/data/node/test_tril_zero/model.onnx | Bin 0 -> 135 bytes .../test_tril_zero/test_data_set_0/input_0.pb | Bin 0 -> 13 bytes .../test_tril_zero/test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 13 bytes .../test/data/node/test_triu/model.onnx | Bin 0 -> 92 bytes .../node/test_triu/test_data_set_0/input_0.pb | Bin 0 -> 172 bytes .../test_triu/test_data_set_0/output_0.pb | Bin 0 -> 172 bytes .../test/data/node/test_triu_neg/model.onnx | Bin 0 -> 112 bytes .../test_triu_neg/test_data_set_0/input_0.pb | Bin 0 -> 172 bytes .../test_triu_neg/test_data_set_0/input_1.pb | 1 + .../test_triu_neg/test_data_set_0/output_0.pb | Bin 0 -> 172 bytes .../data/node/test_triu_one_row/model.onnx | Bin 0 -> 124 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 133 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 133 bytes .../node/test_triu_out_neg_out/model.onnx | Bin 0 -> 120 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 172 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 172 bytes .../data/node/test_triu_out_pos/model.onnx | Bin 0 -> 116 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 172 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 172 bytes .../test/data/node/test_triu_pos/model.onnx | Bin 0 -> 112 bytes .../test_triu_pos/test_data_set_0/input_0.pb | Bin 0 -> 172 bytes .../test_triu_pos/test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_triu_pos/test_data_set_0/output_0.pb | Bin 0 -> 172 bytes .../data/node/test_triu_square/model.onnx | Bin 0 -> 107 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 158 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 158 bytes .../data/node/test_triu_square_neg/model.onnx | Bin 0 -> 127 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 158 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 158 bytes .../test/data/node/test_triu_zero/model.onnx | Bin 0 -> 113 bytes .../test_triu_zero/test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_triu_zero/test_data_set_0/input_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 11 bytes .../data/node/test_unique_length_1/model.onnx | Bin 0 -> 222 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 23 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 31 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 22 bytes .../model.onnx | Bin 0 -> 237 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 47 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 71 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 46 bytes .../test_unique_sorted_with_axis/model.onnx | Bin 0 -> 251 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 47 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 31 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 47 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 262 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 77 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 61 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 39 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 55 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 38 bytes .../model.onnx | Bin 0 -> 269 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 47 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 35 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 31 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 47 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 30 bytes .../model.onnx | Bin 0 -> 218 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 33 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 47 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 71 bytes .../test_data_set_0/output_3.pb | Bin 0 -> 46 bytes .../node/test_unsqueeze_axis_0/model.onnx | Bin 0 -> 146 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_unsqueeze_axis_1/model.onnx | Bin 0 -> 146 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_unsqueeze_axis_2/model.onnx | Bin 0 -> 146 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | 1 + .../test_unsqueeze_negative_axes/model.onnx | Bin 0 -> 162 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../node/test_unsqueeze_three_axes/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 36 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_unsqueeze_two_axes/model.onnx | Bin 0 -> 153 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 28 bytes .../test_data_set_0/output_0.pb | 1 + .../test_unsqueeze_unsorted_axes/model.onnx | Bin 0 -> 162 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | Bin 0 -> 36 bytes .../test_data_set_0/output_0.pb | 1 + .../node/test_upsample_nearest/model.onnx | Bin 0 -> 174 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 31 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 111 bytes .../data/node/test_where_example/model.onnx | Bin 0 -> 166 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 27 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 27 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 27 bytes .../node/test_where_long_example/model.onnx | Bin 0 -> 171 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 43 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 43 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 43 bytes .../test/data/node/test_wrap_pad/model.onnx | Bin 0 -> 154 bytes .../test_wrap_pad/test_data_set_0/input_0.pb | Bin 0 -> 256 bytes .../test_wrap_pad/test_data_set_0/input_1.pb | Bin 0 -> 76 bytes .../test_wrap_pad/test_data_set_0/output_0.pb | Bin 0 -> 520 bytes .../test/data/node/test_xor2d/model.onnx | Bin 0 -> 119 bytes .../test_xor2d/test_data_set_0/input_0.pb | Bin 0 -> 23 bytes .../test_xor2d/test_data_set_0/input_1.pb | Bin 0 -> 23 bytes .../test_xor2d/test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test/data/node/test_xor3d/model.onnx | Bin 0 -> 131 bytes .../test_xor3d/test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_xor3d/test_data_set_0/input_1.pb | Bin 0 -> 73 bytes .../test_xor3d/test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../test/data/node/test_xor4d/model.onnx | Bin 0 -> 143 bytes .../test_xor4d/test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../test_xor4d/test_data_set_0/input_1.pb | Bin 0 -> 376 bytes .../test_xor4d/test_data_set_0/output_0.pb | Bin 0 -> 378 bytes .../data/node/test_xor_bcast3v1d/model.onnx | Bin 0 -> 131 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../data/node/test_xor_bcast3v2d/model.onnx | Bin 0 -> 135 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 73 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 31 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 75 bytes .../data/node/test_xor_bcast4v2d/model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 41 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 378 bytes .../data/node/test_xor_bcast4v3d/model.onnx | Bin 0 -> 147 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 376 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 133 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 378 bytes .../data/node/test_xor_bcast4v4d/model.onnx | Bin 0 -> 151 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 39 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 105 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 378 bytes .../test_AvgPool1d/model.onnx | Bin 0 -> 234 bytes .../test_AvgPool1d/test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_AvgPool1d_stride/model.onnx | Bin 0 -> 234 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 155 bytes .../test_data_set_0/output_0.pb | 1 + .../test_AvgPool2d/model.onnx | Bin 0 -> 180 bytes .../test_AvgPool2d/test_data_set_0/input_0.pb | Bin 0 -> 877 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 229 bytes .../test_AvgPool2d_stride/model.onnx | Bin 0 -> 180 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 877 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 229 bytes .../test_AvgPool3d/model.onnx | Bin 0 -> 196 bytes .../test_AvgPool3d/test_data_set_0/input_0.pb | Bin 0 -> 1551 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 207 bytes .../test_AvgPool3d_stride/model.onnx | Bin 0 -> 196 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 3015 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 196 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1551 bytes .../test_data_set_0/output_0.pb | 3 + .../test_BatchNorm1d_3d_input_eval/model.onnx | Bin 0 -> 378 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 251 bytes .../test_data_set_0/output_0.pb | 1 + .../test_BatchNorm2d_eval/model.onnx | Bin 0 -> 354 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 877 bytes .../test_data_set_0/output_0.pb | 3 + .../test_BatchNorm2d_momentum_eval/model.onnx | Bin 0 -> 354 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 877 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 877 bytes .../test_BatchNorm3d_eval/model.onnx | Bin 0 -> 362 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1551 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1551 bytes .../test_BatchNorm3d_momentum_eval/model.onnx | Bin 0 -> 362 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1551 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1551 bytes .../test_ConstantPad2d/model.onnx | Bin 0 -> 177 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 397 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1861 bytes .../pytorch-converted/test_Conv1d/model.onnx | Bin 0 -> 525 bytes .../test_Conv1d/test_data_set_0/input_0.pb | Bin 0 -> 331 bytes .../test_Conv1d/test_data_set_0/output_0.pb | Bin 0 -> 331 bytes .../test_Conv1d_dilated/model.onnx | Bin 0 -> 525 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 251 bytes .../test_Conv1d_groups/model.onnx | Bin 0 -> 433 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 203 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 203 bytes .../test_Conv1d_pad1/model.onnx | Bin 0 -> 525 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 331 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 411 bytes .../test_Conv1d_pad1size1/model.onnx | 31 + .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_Conv1d_pad2/model.onnx | Bin 0 -> 685 bytes .../test_data_set_0/input_0.pb | 4 + .../test_data_set_0/output_0.pb | Bin 0 -> 411 bytes .../test_Conv1d_pad2size1/model.onnx | Bin 0 -> 601 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_Conv1d_stride/model.onnx | Bin 0 -> 525 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 331 bytes .../test_data_set_0/output_0.pb | 1 + .../pytorch-converted/test_Conv2d/model.onnx | Bin 0 -> 593 bytes .../test_Conv2d/test_data_set_0/input_0.pb | Bin 0 -> 853 bytes .../test_Conv2d/test_data_set_0/output_0.pb | Bin 0 -> 653 bytes .../test_Conv2d_depthwise/model.onnx | Bin 0 -> 449 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1165 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 525 bytes .../test_Conv2d_depthwise_padded/model.onnx | Bin 0 -> 449 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1165 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1165 bytes .../test_Conv2d_depthwise_strided/model.onnx | Bin 0 -> 449 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1165 bytes .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 609 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1165 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1037 bytes .../test_Conv2d_dilated/model.onnx | Bin 0 -> 513 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1549 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 157 bytes .../test_Conv2d_groups/model.onnx | Bin 0 -> 601 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 973 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 781 bytes .../test_Conv2d_groups_thnn/model.onnx | Bin 0 -> 601 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 973 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 781 bytes .../test_Conv2d_no_bias/model.onnx | Bin 0 -> 546 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | Bin 0 -> 525 bytes .../test_Conv2d_padding/model.onnx | Bin 0 -> 737 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 877 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 301 bytes .../test_Conv2d_strided/model.onnx | Bin 0 -> 737 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 877 bytes .../test_data_set_0/output_0.pb | 3 + .../pytorch-converted/test_Conv3d/model.onnx | Bin 0 -> 1481 bytes .../test_Conv3d/test_data_set_0/input_0.pb | Bin 0 -> 1455 bytes .../test_Conv3d/test_data_set_0/output_0.pb | 1 + .../test_Conv3d_dilated/model.onnx | Bin 0 -> 713 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 3015 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 879 bytes .../test_Conv3d_dilated_strided/model.onnx | Bin 0 -> 713 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 3015 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 271 bytes .../test_Conv3d_groups/model.onnx | Bin 0 -> 1633 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2575 bytes .../test_data_set_0/output_0.pb | 1 + .../test_Conv3d_no_bias/model.onnx | Bin 0 -> 1434 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1455 bytes .../test_data_set_0/output_0.pb | 2 + .../test_Conv3d_stride/model.onnx | Bin 0 -> 713 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 3015 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 271 bytes .../test_Conv3d_stride_padding/model.onnx | Bin 0 -> 713 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 3015 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 879 bytes .../test_ConvTranspose2d/model.onnx | Bin 0 -> 772 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 517 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3853 bytes .../test_ConvTranspose2d_no_bias/model.onnx | Bin 0 -> 725 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 517 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3853 bytes .../pytorch-converted/test_ELU/model.onnx | Bin 0 -> 120 bytes .../test_ELU/test_data_set_0/input_0.pb | 1 + .../test_ELU/test_data_set_0/output_0.pb | 1 + .../test_Embedding/model.onnx | 17 + .../test_Embedding/test_data_set_0/input_0.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | 1 + .../test_Embedding_sparse/model.onnx | 17 + .../test_data_set_0/input_0.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | 1 + .../pytorch-converted/test_GLU/model.onnx | 17 + .../test_GLU/test_data_set_0/input_0.pb | Bin 0 -> 128 bytes .../test_GLU/test_data_set_0/output_0.pb | Bin 0 -> 68 bytes .../pytorch-converted/test_GLU_dim/model.onnx | 19 + .../test_GLU_dim/test_data_set_0/input_0.pb | Bin 0 -> 851 bytes .../test_GLU_dim/test_data_set_0/output_0.pb | Bin 0 -> 431 bytes .../test_LeakyReLU/model.onnx | 15 + .../test_LeakyReLU/test_data_set_0/input_0.pb | Bin 0 -> 130 bytes .../test_data_set_0/output_0.pb | 1 + .../test_LeakyReLU_with_negval/model.onnx | Bin 0 -> 126 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../pytorch-converted/test_Linear/model.onnx | Bin 0 -> 585 bytes .../test_Linear/test_data_set_0/input_0.pb | 3 + .../test_Linear/test_data_set_0/output_0.pb | Bin 0 -> 137 bytes .../test_Linear_no_bias/model.onnx | Bin 0 -> 492 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 169 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 137 bytes .../test_LogSoftmax/model.onnx | 15 + .../test_data_set_0/input_0.pb | Bin 0 -> 809 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 809 bytes .../test_MaxPool1d/model.onnx | Bin 0 -> 160 bytes .../test_MaxPool1d/test_data_set_0/input_0.pb | Bin 0 -> 331 bytes .../test_data_set_0/output_0.pb | 3 + .../test_MaxPool1d_stride/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | 4 + .../test_data_set_0/output_0.pb | 2 + .../model.onnx | 19 + .../test_data_set_0/input_0.pb | Bin 0 -> 880014 bytes .../test_data_set_0/output_0.pb | 154 + .../test_MaxPool2d/model.onnx | 18 + .../test_MaxPool2d/test_data_set_0/input_0.pb | Bin 0 -> 601 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 205 bytes .../model.onnx | 25 + .../test_data_set_0/input_0.pb | Bin 0 -> 4000017 bytes .../test_data_set_0/output_0.pb | 1 + .../test_MaxPool3d/model.onnx | Bin 0 -> 192 bytes .../test_MaxPool3d/test_data_set_0/input_0.pb | Bin 0 -> 3015 bytes .../test_data_set_0/output_0.pb | 1 + .../test_MaxPool3d_stride/model.onnx | Bin 0 -> 192 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 3015 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 207 bytes .../test_MaxPool3d_stride_padding/model.onnx | 20 + .../test_data_set_0/input_0.pb | Bin 0 -> 3015 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 663 bytes .../test_PReLU_1d/model.onnx | Bin 0 -> 140 bytes .../test_PReLU_1d/test_data_set_0/input_0.pb | Bin 0 -> 106 bytes .../test_PReLU_1d/test_data_set_0/output_0.pb | Bin 0 -> 106 bytes .../test_PReLU_1d_multiparam/model.onnx | Bin 0 -> 148 bytes .../test_data_set_0/input_0.pb | 2 + .../test_data_set_0/output_0.pb | 2 + .../test_PReLU_2d/model.onnx | Bin 0 -> 148 bytes .../test_PReLU_2d/test_data_set_0/input_0.pb | 3 + .../test_PReLU_2d/test_data_set_0/output_0.pb | 3 + .../test_PReLU_2d_multiparam/model.onnx | Bin 0 -> 157 bytes .../test_data_set_0/input_0.pb | 3 + .../test_data_set_0/output_0.pb | 3 + .../test_PReLU_3d/model.onnx | Bin 0 -> 157 bytes .../test_PReLU_3d/test_data_set_0/input_0.pb | Bin 0 -> 2895 bytes .../test_PReLU_3d/test_data_set_0/output_0.pb | Bin 0 -> 2895 bytes .../test_PReLU_3d_multiparam/model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2895 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 2895 bytes .../test_PixelShuffle/model.onnx | Bin 0 -> 331 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 589 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 589 bytes .../test_PoissonNLLLLoss_no_reduce/model.onnx | Bin 0 -> 569 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 409 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 409 bytes .../pytorch-converted/test_ReLU/model.onnx | 15 + .../test_ReLU/test_data_set_0/input_0.pb | Bin 0 -> 493 bytes .../test_ReLU/test_data_set_0/output_0.pb | Bin 0 -> 493 bytes .../test_ReflectionPad2d/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 1549 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 3973 bytes .../test_ReplicationPad2d/model.onnx | Bin 0 -> 156 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 397 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1861 bytes .../pytorch-converted/test_SELU/model.onnx | 13 + .../test_SELU/test_data_set_0/input_0.pb | Bin 0 -> 130 bytes .../test_SELU/test_data_set_0/output_0.pb | 1 + .../pytorch-converted/test_Sigmoid/model.onnx | 15 + .../test_Sigmoid/test_data_set_0/input_0.pb | Bin 0 -> 493 bytes .../test_Sigmoid/test_data_set_0/output_0.pb | 4 + .../pytorch-converted/test_Softmax/model.onnx | 14 + .../test_Softmax/test_data_set_0/input_0.pb | Bin 0 -> 809 bytes .../test_Softmax/test_data_set_0/output_0.pb | Bin 0 -> 809 bytes .../pytorch-converted/test_Softmin/model.onnx | 16 + .../test_Softmin/test_data_set_0/input_0.pb | Bin 0 -> 809 bytes .../test_Softmin/test_data_set_0/output_0.pb | Bin 0 -> 809 bytes .../test_Softplus/model.onnx | 13 + .../test_Softplus/test_data_set_0/input_0.pb | Bin 0 -> 809 bytes .../test_Softplus/test_data_set_0/output_0.pb | Bin 0 -> 809 bytes .../test_Softsign/model.onnx | Bin 0 -> 191 bytes .../test_Softsign/test_data_set_0/input_0.pb | Bin 0 -> 130 bytes .../test_Softsign/test_data_set_0/output_0.pb | 2 + .../pytorch-converted/test_Tanh/model.onnx | 15 + .../test_Tanh/test_data_set_0/input_0.pb | Bin 0 -> 493 bytes .../test_Tanh/test_data_set_0/output_0.pb | Bin 0 -> 493 bytes .../test_ZeroPad2d/model.onnx | Bin 0 -> 177 bytes .../test_ZeroPad2d/test_data_set_0/input_0.pb | Bin 0 -> 397 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1861 bytes .../test_log_softmax_dim3/model.onnx | 17 + .../test_data_set_0/input_0.pb | Bin 0 -> 493 bytes .../test_data_set_0/output_0.pb | 3 + .../test_log_softmax_lastdim/model.onnx | 13 + .../test_data_set_0/input_0.pb | Bin 0 -> 1034 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1034 bytes .../test_softmax_functional_dim3/model.onnx | 16 + .../test_data_set_0/input_0.pb | 4 + .../test_data_set_0/output_0.pb | Bin 0 -> 493 bytes .../test_softmax_lastdim/model.onnx | 12 + .../test_data_set_0/input_0.pb | Bin 0 -> 1034 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 1034 bytes .../test_operator_add_broadcast/model.onnx | 18 + .../test_data_set_0/input_0.pb | Bin 0 -> 56 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 56 bytes .../model.onnx | Bin 0 -> 151 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 56 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 56 bytes .../model.onnx | 18 + .../test_data_set_0/input_0.pb | Bin 0 -> 56 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 30 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 56 bytes .../model.onnx | Bin 0 -> 151 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 56 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 32 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 56 bytes .../test_operator_addconstant/model.onnx | Bin 0 -> 158 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 56 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 56 bytes .../test_operator_addmm/model.onnx | Bin 0 -> 245 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 32 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/input_2.pb | 1 + .../test_data_set_0/output_0.pb | 2 + .../test_operator_basic/model.onnx | 25 + .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_operator_chunk/model.onnx | Bin 0 -> 138 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 18 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 14 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 10 bytes .../test_operator_clip/model.onnx | Bin 0 -> 126 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 56 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 56 bytes .../test_operator_concat2/model.onnx | 17 + .../test_data_set_0/input_0.pb | Bin 0 -> 32 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 56 bytes .../test_operator_conv/model.onnx | Bin 0 -> 7746 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 2560015 bytes .../test_data_set_0/output_0.pb | 1 + .../test_operator_convtranspose/model.onnx | Bin 0 -> 617 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 493 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 4333 bytes .../test_operator_exp/model.onnx | 11 + .../test_data_set_0/input_0.pb | Bin 0 -> 56 bytes .../test_data_set_0/output_0.pb | 3 + .../test_operator_flatten/model.onnx | 14 + .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_operator_index/model.onnx | Bin 0 -> 165 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 12 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 10 bytes .../test_operator_max/model.onnx | 16 + .../test_data_set_0/input_0.pb | Bin 0 -> 56 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_operator_maxpool/model.onnx | Bin 0 -> 160 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 64012 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 30732 bytes .../test_operator_min/model.onnx | 16 + .../test_data_set_0/input_0.pb | Bin 0 -> 56 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 56 bytes .../test_operator_mm/model.onnx | Bin 0 -> 214 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 32 bytes .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_operator_non_float_params/model.onnx | Bin 0 -> 181 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 40 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 40 bytes .../test_operator_pad/model.onnx | Bin 0 -> 159 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 44 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 120 bytes .../test_operator_params/model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 24 bytes .../test_data_set_0/output_0.pb | 1 + .../test_operator_permute2/model.onnx | Bin 0 -> 157 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 20 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 20 bytes .../test_operator_pow/model.onnx | 22 + .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/input_1.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 108 bytes .../test_operator_reduced_mean/model.onnx | Bin 0 -> 144 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | 18 + .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_operator_reduced_sum/model.onnx | Bin 0 -> 143 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 3 + .../model.onnx | 17 + .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 3 + .../test_operator_repeat/model.onnx | Bin 0 -> 183 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 262 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_operator_selu/model.onnx | 15 + .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../test_operator_sqrt/model.onnx | 11 + .../test_data_set_0/input_0.pb | Bin 0 -> 56 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 56 bytes .../model.onnx | Bin 0 -> 291 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 81934 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 81934 bytes .../model.onnx | 33 + .../test_data_set_0/input_0.pb | Bin 0 -> 10 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 10 bytes .../test_operator_view/model.onnx | 12 + .../test_data_set_0/input_0.pb | Bin 0 -> 10 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 12 bytes .../data/real/test_bvlc_alexnet/data.json | 1 + .../test/data/real/test_densenet121/data.json | 1 + .../data/real/test_inception_v1/data.json | 1 + .../data/real/test_inception_v2/data.json | 1 + .../test/data/real/test_resnet50/data.json | 1 + .../test/data/real/test_shufflenet/data.json | 1 + .../test/data/real/test_squeezenet/data.json | 1 + .../test/data/real/test_vgg19/data.json | 1 + .../test/data/real/test_zfnet512/data.json | 1 + .../test_expand_shape_model1/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 25 bytes .../test_expand_shape_model2/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 29 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 49 bytes .../test_expand_shape_model3/model.onnx | Bin 0 -> 132 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 121 bytes .../test_expand_shape_model4/model.onnx | Bin 0 -> 136 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 25 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 45 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 340 bytes .../simple/test_gradient_of_add/model.onnx | Bin 0 -> 264 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 15 bytes .../test_gradient_of_add_and_mul/model.onnx | Bin 0 -> 297 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 11 bytes .../test_data_set_0/output_1.pb | Bin 0 -> 15 bytes .../test_data_set_0/output_2.pb | Bin 0 -> 15 bytes .../simple/test_sequence_model1/model.onnx | Bin 0 -> 371 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 61 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 158 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 63 bytes .../simple/test_sequence_model2/model.onnx | Bin 0 -> 322 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 109 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 111 bytes .../simple/test_sequence_model3/model.onnx | Bin 0 -> 437 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 109 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 111 bytes .../simple/test_sequence_model4/model.onnx | Bin 0 -> 219 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 109 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 304 bytes .../simple/test_sequence_model5/model.onnx | Bin 0 -> 251 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 109 bytes .../test_data_set_0/input_2.pb | Bin 0 -> 109 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 306 bytes .../simple/test_sequence_model6/model.onnx | Bin 0 -> 156 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 109 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../simple/test_sequence_model7/model.onnx | Bin 0 -> 209 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../simple/test_sequence_model8/model.onnx | Bin 0 -> 157 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 9 bytes .../test_data_set_0/input_1.pb | Bin 0 -> 38 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 17 bytes .../test/data/simple/test_shrink/model.onnx | Bin 0 -> 115 bytes .../test_shrink/test_data_set_0/input_0.pb | Bin 0 -> 29 bytes .../test_shrink/test_data_set_0/output_0.pb | Bin 0 -> 29 bytes .../data/simple/test_sign_model/model.onnx | Bin 0 -> 90 bytes .../test_data_set_0/input_0.pb | Bin 0 -> 37 bytes .../test_data_set_0/output_0.pb | Bin 0 -> 37 bytes .../simple/test_single_relu_model/model.onnx | Bin 0 -> 98 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 185 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 153 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 185 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 185 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | Bin 0 -> 9 bytes .../model.onnx | Bin 0 -> 167 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + .../model.onnx | Bin 0 -> 128 bytes .../test_data_set_0/input_0.pb | 1 + .../test_data_set_0/output_0.pb | 1 + onnx/backend/test/loader/__init__.py | 66 + onnx/backend/test/report/__init__.py | 50 + onnx/backend/test/report/base.py | 8 + onnx/backend/test/report/coverage.py | 266 + onnx/backend/test/runner/__init__.py | 529 + onnx/backend/test/runner/item.py | 21 + onnx/backend/test/stat_coverage.py | 299 + onnx/bin/__init__.py | 3 + onnx/bin/checker.py | 27 + onnx/checker.cc | 1550 + onnx/checker.h | 204 + onnx/checker.py | 171 + onnx/common/CMakeLists.txt | 34 + onnx/common/array_ref.h | 201 + onnx/common/assertions.cc | 24 + onnx/common/assertions.h | 62 + onnx/common/common.h | 54 + onnx/common/constants.h | 46 + onnx/common/file_utils.h | 38 + onnx/common/graph_node_list.h | 165 + onnx/common/interned_strings.cc | 80 + onnx/common/interned_strings.h | 247 + onnx/common/ir.h | 1424 + onnx/common/ir_pb_converter.cc | 771 + onnx/common/ir_pb_converter.h | 47 + onnx/common/model_helpers.cc | 37 + onnx/common/model_helpers.h | 26 + onnx/common/path.h | 88 + onnx/common/platform_helpers.h | 17 + onnx/common/proto_util.h | 41 + onnx/common/safe_math.h | 69 + onnx/common/scoped_resource.h | 86 + onnx/common/status.cc | 89 + onnx/common/status.h | 99 + onnx/common/tensor.h | 236 + onnx/common/version.h.in | 20 + onnx/common/visitor.h | 118 + onnx/compose.py | 756 + onnx/cpp2py_export.cc | 814 + onnx/defs/CMakeLists.txt | 54 + onnx/defs/__init__.py | 158 + onnx/defs/attr_proto_util.cc | 73 + onnx/defs/attr_proto_util.h | 44 + onnx/defs/controlflow/CMakeLists.txt | 9 + onnx/defs/controlflow/defs.cc | 194 + onnx/defs/controlflow/old.cc | 2708 + onnx/defs/controlflow/utils.cc | 377 + onnx/defs/controlflow/utils.h | 27 + onnx/defs/data_propagators.h | 88 + onnx/defs/data_type_utils.cc | 405 + onnx/defs/data_type_utils.h | 72 + onnx/defs/doc_strings.cc | 1395 + onnx/defs/doc_strings.h | 86 + onnx/defs/function.cc | 226 + onnx/defs/function.h | 247 + onnx/defs/gen_doc.py | 440 + onnx/defs/gen_shape_inference_information.py | 31 + onnx/defs/generator/CMakeLists.txt | 9 + onnx/defs/generator/defs.cc | 510 + onnx/defs/generator/old.cc | 1034 + onnx/defs/generator/utils.cc | 111 + onnx/defs/generator/utils.h | 34 + onnx/defs/image/CMakeLists.txt | 6 + onnx/defs/image/defs.cc | 68 + onnx/defs/logical/CMakeLists.txt | 7 + onnx/defs/logical/defs.cc | 349 + onnx/defs/logical/old.cc | 265 + onnx/defs/math/CMakeLists.txt | 9 + onnx/defs/math/defs.cc | 3101 ++ onnx/defs/math/old.cc | 3642 ++ onnx/defs/math/utils.cc | 289 + onnx/defs/math/utils.h | 60 + onnx/defs/nn/CMakeLists.txt | 9 + onnx/defs/nn/defs.cc | 4788 ++ onnx/defs/nn/old.cc | 4633 ++ onnx/defs/nn/utils.cc | 259 + onnx/defs/nn/utils.h | 32 + onnx/defs/object_detection/CMakeLists.txt | 7 + onnx/defs/object_detection/defs.cc | 169 + onnx/defs/object_detection/old.cc | 252 + onnx/defs/operator_sets.h | 1575 + onnx/defs/operator_sets_ml.h | 109 + onnx/defs/operator_sets_preview.h | 39 + onnx/defs/operator_sets_training.h | 24 + onnx/defs/optional/CMakeLists.txt | 7 + onnx/defs/optional/defs.cc | 154 + onnx/defs/optional/old.cc | 83 + onnx/defs/parser.cc | 1056 + onnx/defs/parser.h | 510 + onnx/defs/preview/CMakeLists.txt | 6 + onnx/defs/preview/defs.cc | 511 + onnx/defs/printer.cc | 533 + onnx/defs/printer.h | 50 + onnx/defs/quantization/CMakeLists.txt | 7 + onnx/defs/quantization/defs.cc | 320 + onnx/defs/quantization/old.cc | 921 + onnx/defs/reduction/CMakeLists.txt | 9 + onnx/defs/reduction/defs.cc | 183 + onnx/defs/reduction/old.cc | 455 + onnx/defs/reduction/utils.cc | 172 + onnx/defs/reduction/utils.h | 42 + onnx/defs/rnn/CMakeLists.txt | 7 + onnx/defs/rnn/defs.cc | 373 + onnx/defs/rnn/old.cc | 1490 + onnx/defs/schema.cc | 1705 + onnx/defs/schema.h | 1433 + onnx/defs/sequence/CMakeLists.txt | 9 + onnx/defs/sequence/defs.cc | 630 + onnx/defs/sequence/old.cc | 18 + onnx/defs/sequence/utils.cc | 194 + onnx/defs/sequence/utils.h | 27 + onnx/defs/shape_inference.cc | 610 + onnx/defs/shape_inference.h | 1018 + onnx/defs/tensor/CMakeLists.txt | 9 + onnx/defs/tensor/defs.cc | 3891 ++ onnx/defs/tensor/old.cc | 7803 +++ onnx/defs/tensor/utils.cc | 503 + onnx/defs/tensor/utils.h | 60 + onnx/defs/tensor_proto_util.cc | 155 + onnx/defs/tensor_proto_util.h | 21 + onnx/defs/tensor_util.cc | 63 + onnx/defs/tensor_util.h | 16 + onnx/defs/text/CMakeLists.txt | 6 + onnx/defs/text/defs.cc | 203 + onnx/defs/traditionalml/CMakeLists.txt | 8 + onnx/defs/traditionalml/defs.cc | 1271 + onnx/defs/traditionalml/old.cc | 658 + onnx/defs/traditionalml/utils.h | 35 + onnx/defs/training/CMakeLists.txt | 6 + onnx/defs/training/defs.cc | 621 + onnx/defs/type_builders.h | 132 + onnx/external_data_helper.py | 414 + onnx/frontend/__init__.py | 3 + onnx/fuzz/README.md | 113 + onnx/fuzz/fuzz_checker.py | 27 + onnx/fuzz/fuzz_compose.py | 147 + onnx/fuzz/fuzz_model_loader.py | 31 + onnx/fuzz/fuzz_parser.py | 28 + onnx/fuzz/fuzz_shape_inference.py | 189 + onnx/fuzz/fuzz_version_converter.py | 57 + onnx/fuzz/make_seed_corpus.py | 550 + onnx/gen_proto.py | 272 + onnx/helper.py | 1381 + onnx/inliner.py | 60 + onnx/inliner/CMakeLists.txt | 7 + onnx/inliner/inliner.cc | 806 + onnx/inliner/inliner.h | 148 + onnx/model_container.py | 346 + onnx/numpy_helper.py | 662 + onnx/onnx-data.in.proto | 150 + onnx/onnx-data.proto | 155 + onnx/onnx-data.proto3 | 155 + onnx/onnx-ml.proto | 1016 + onnx/onnx-ml.proto3 | 1016 + onnx/onnx-operators-ml.proto | 136 + onnx/onnx-operators-ml.proto3 | 136 + onnx/onnx-operators.in.proto | 131 + onnx/onnx-operators.proto | 136 + onnx/onnx-operators.proto3 | 136 + onnx/onnx-operators_pb.h | 12 + onnx/onnx.in.proto | 1013 + onnx/onnx.proto | 1000 + onnx/onnx.proto3 | 1000 + onnx/onnx_cpp2py_export/__init__.pyi | 5 + onnx/onnx_cpp2py_export/checker.pyi | 48 + onnx/onnx_cpp2py_export/defs.pyi | 214 + onnx/onnx_cpp2py_export/inliner.pyi | 26 + onnx/onnx_cpp2py_export/parser.pyi | 27 + onnx/onnx_cpp2py_export/printer.pyi | 4 + onnx/onnx_cpp2py_export/shape_inference.pyi | 45 + onnx/onnx_cpp2py_export/version_converter.pyi | 4 + onnx/onnx_pb.h | 56 + onnx/parser.py | 73 + onnx/printer.py | 21 + onnx/proto_utils.h | 75 + onnx/py.typed | 0 onnx/py_utils.h | 22 + onnx/reference/README.md | 43 + onnx/reference/__init__.py | 8 + onnx/reference/op_run.py | 682 + onnx/reference/ops/__init__.py | 8 + onnx/reference/ops/_helpers.py | 67 + onnx/reference/ops/_op.py | 182 + onnx/reference/ops/_op_common_indices.py | 37 + onnx/reference/ops/_op_common_pool.py | 301 + onnx/reference/ops/_op_common_random.py | 55 + onnx/reference/ops/_op_common_window.py | 26 + onnx/reference/ops/_op_list.py | 627 + onnx/reference/ops/_quant_utils.py | 56 + onnx/reference/ops/aionnx_preview/__init__.py | 8 + onnx/reference/ops/aionnx_preview/_op_list.py | 85 + .../ops/aionnx_preview/op_flex_attention.py | 160 + .../ops/aionnx_preview_training/__init__.py | 9 + .../ops/aionnx_preview_training/_op_list.py | 89 + .../_op_run_training.py | 10 + .../ops/aionnx_preview_training/op_adagrad.py | 64 + .../ops/aionnx_preview_training/op_adam.py | 105 + .../aionnx_preview_training/op_momentum.py | 82 + onnx/reference/ops/aionnxml/__init__.py | 6 + .../ops/aionnxml/_common_classifier.py | 79 + onnx/reference/ops/aionnxml/_op_list.py | 121 + .../ops/aionnxml/_op_run_aionnxml.py | 8 + .../aionnxml/op_array_feature_extractor.py | 47 + onnx/reference/ops/aionnxml/op_binarizer.py | 15 + .../ops/aionnxml/op_dict_vectorizer.py | 47 + .../ops/aionnxml/op_feature_vectorizer.py | 31 + onnx/reference/ops/aionnxml/op_imputer.py | 47 + .../ops/aionnxml/op_label_encoder.py | 50 + .../ops/aionnxml/op_linear_classifier.py | 98 + .../ops/aionnxml/op_linear_regressor.py | 26 + onnx/reference/ops/aionnxml/op_normalizer.py | 41 + .../ops/aionnxml/op_one_hot_encoder.py | 53 + onnx/reference/ops/aionnxml/op_scaler.py | 12 + .../ops/aionnxml/op_svm_classifier.py | 333 + onnx/reference/ops/aionnxml/op_svm_helper.py | 97 + .../ops/aionnxml/op_svm_regressor.py | 43 + .../ops/aionnxml/op_tree_ensemble.py | 268 + .../aionnxml/op_tree_ensemble_classifier.py | 132 + .../ops/aionnxml/op_tree_ensemble_helper.py | 103 + .../aionnxml/op_tree_ensemble_regressor.py | 109 + onnx/reference/ops/experimental/__init__.py | 8 + onnx/reference/ops/experimental/_op_list.py | 72 + .../ops/experimental/_op_run_experimental.py | 10 + onnx/reference/ops/experimental/op_im2col.py | 35 + onnx/reference/ops/op_abs.py | 13 + onnx/reference/ops/op_acos.py | 13 + onnx/reference/ops/op_acosh.py | 13 + onnx/reference/ops/op_add.py | 13 + onnx/reference/ops/op_affine_grid.py | 90 + onnx/reference/ops/op_and.py | 13 + onnx/reference/ops/op_argmax.py | 42 + onnx/reference/ops/op_argmin.py | 42 + onnx/reference/ops/op_asin.py | 13 + onnx/reference/ops/op_asinh.py | 13 + onnx/reference/ops/op_atan.py | 13 + onnx/reference/ops/op_atanh.py | 13 + onnx/reference/ops/op_attention.py | 336 + onnx/reference/ops/op_attribute_has_value.py | 33 + onnx/reference/ops/op_average_pool.py | 107 + onnx/reference/ops/op_batch_normalization.py | 101 + onnx/reference/ops/op_bernoulli.py | 17 + onnx/reference/ops/op_bitcast.py | 29 + onnx/reference/ops/op_bitshift.py | 17 + onnx/reference/ops/op_bitwise_and.py | 13 + onnx/reference/ops/op_bitwise_not.py | 13 + onnx/reference/ops/op_bitwise_or.py | 13 + onnx/reference/ops/op_bitwise_xor.py | 13 + onnx/reference/ops/op_blackman_window.py | 30 + onnx/reference/ops/op_cast.py | 54 + onnx/reference/ops/op_cast_like.py | 27 + .../ops/op_causal_conv_with_state.py | 71 + onnx/reference/ops/op_ceil.py | 13 + onnx/reference/ops/op_celu.py | 19 + onnx/reference/ops/op_center_crop_pad.py | 50 + onnx/reference/ops/op_clip.py | 25 + onnx/reference/ops/op_col2im.py | 213 + onnx/reference/ops/op_compress.py | 13 + onnx/reference/ops/op_concat.py | 22 + onnx/reference/ops/op_concat_from_sequence.py | 31 + onnx/reference/ops/op_constant.py | 119 + onnx/reference/ops/op_constant_of_shape.py | 29 + onnx/reference/ops/op_conv.py | 316 + onnx/reference/ops/op_conv_integer.py | 52 + onnx/reference/ops/op_conv_transpose.py | 139 + onnx/reference/ops/op_cos.py | 13 + onnx/reference/ops/op_cosh.py | 13 + onnx/reference/ops/op_cum_prod.py | 31 + onnx/reference/ops/op_cum_sum.py | 31 + onnx/reference/ops/op_deform_conv.py | 181 + onnx/reference/ops/op_depth_to_space.py | 46 + onnx/reference/ops/op_dequantize_linear.py | 91 + onnx/reference/ops/op_det.py | 13 + onnx/reference/ops/op_dft.py | 146 + onnx/reference/ops/op_div.py | 30 + onnx/reference/ops/op_dropout.py | 66 + .../ops/op_dynamic_quantize_linear.py | 32 + onnx/reference/ops/op_einsum.py | 21 + onnx/reference/ops/op_elu.py | 14 + onnx/reference/ops/op_equal.py | 13 + onnx/reference/ops/op_erf.py | 19 + onnx/reference/ops/op_exp.py | 13 + onnx/reference/ops/op_expand.py | 18 + onnx/reference/ops/op_eyelike.py | 31 + onnx/reference/ops/op_flatten.py | 16 + onnx/reference/ops/op_floor.py | 13 + onnx/reference/ops/op_gather.py | 23 + onnx/reference/ops/op_gather_elements.py | 47 + onnx/reference/ops/op_gathernd.py | 58 + onnx/reference/ops/op_gemm.py | 77 + onnx/reference/ops/op_global_average_pool.py | 21 + onnx/reference/ops/op_global_max_pool.py | 22 + onnx/reference/ops/op_greater.py | 13 + onnx/reference/ops/op_greater_or_equal.py | 13 + onnx/reference/ops/op_grid_sample.py | 365 + onnx/reference/ops/op_gru.py | 132 + onnx/reference/ops/op_hamming_window.py | 23 + onnx/reference/ops/op_hann_window.py | 20 + onnx/reference/ops/op_hard_sigmoid.py | 16 + onnx/reference/ops/op_hardmax.py | 24 + onnx/reference/ops/op_identity.py | 13 + onnx/reference/ops/op_if.py | 72 + onnx/reference/ops/op_image_decoder.py | 33 + .../ops/op_instance_normalization.py | 21 + onnx/reference/ops/op_isinf.py | 20 + onnx/reference/ops/op_isnan.py | 13 + onnx/reference/ops/op_layer_normalization.py | 73 + onnx/reference/ops/op_leaky_relu.py | 23 + onnx/reference/ops/op_less.py | 13 + onnx/reference/ops/op_less_or_equal.py | 13 + onnx/reference/ops/op_linear_attention.py | 181 + onnx/reference/ops/op_log.py | 13 + onnx/reference/ops/op_log_softmax.py | 15 + onnx/reference/ops/op_loop.py | 86 + onnx/reference/ops/op_lp_normalization.py | 19 + onnx/reference/ops/op_lp_pool.py | 43 + onnx/reference/ops/op_lrn.py | 28 + onnx/reference/ops/op_lstm.py | 155 + onnx/reference/ops/op_matmul.py | 25 + onnx/reference/ops/op_matmul_integer.py | 19 + onnx/reference/ops/op_max.py | 25 + onnx/reference/ops/op_max_pool.py | 413 + onnx/reference/ops/op_max_unpool.py | 58 + onnx/reference/ops/op_mean.py | 14 + onnx/reference/ops/op_mel_weight_matrix.py | 61 + onnx/reference/ops/op_min.py | 25 + onnx/reference/ops/op_mod.py | 18 + onnx/reference/ops/op_mul.py | 13 + onnx/reference/ops/op_neg.py | 13 + .../ops/op_negative_log_likelihood_loss.py | 84 + onnx/reference/ops/op_non_max_suppression.py | 271 + onnx/reference/ops/op_non_zero.py | 15 + onnx/reference/ops/op_not.py | 13 + onnx/reference/ops/op_one_hot.py | 32 + onnx/reference/ops/op_optional.py | 18 + onnx/reference/ops/op_optional_get_element.py | 13 + onnx/reference/ops/op_optional_has_element.py | 13 + onnx/reference/ops/op_or.py | 13 + onnx/reference/ops/op_pad.py | 65 + onnx/reference/ops/op_pool_common.py | 345 + onnx/reference/ops/op_pow.py | 17 + onnx/reference/ops/op_prelu.py | 34 + onnx/reference/ops/op_qlinear_conv.py | 78 + onnx/reference/ops/op_qlinear_matmul.py | 29 + onnx/reference/ops/op_quantize_linear.py | 191 + onnx/reference/ops/op_random_normal.py | 16 + onnx/reference/ops/op_random_normal_like.py | 19 + onnx/reference/ops/op_random_uniform.py | 16 + onnx/reference/ops/op_random_uniform_like.py | 19 + onnx/reference/ops/op_range.py | 39 + onnx/reference/ops/op_reciprocal.py | 14 + onnx/reference/ops/op_reduce_l1.py | 34 + onnx/reference/ops/op_reduce_l2.py | 34 + onnx/reference/ops/op_reduce_log_sum.py | 34 + onnx/reference/ops/op_reduce_log_sum_exp.py | 43 + onnx/reference/ops/op_reduce_max.py | 46 + onnx/reference/ops/op_reduce_mean.py | 35 + onnx/reference/ops/op_reduce_min.py | 50 + onnx/reference/ops/op_reduce_prod.py | 30 + onnx/reference/ops/op_reduce_sum.py | 35 + onnx/reference/ops/op_reduce_sum_square.py | 30 + onnx/reference/ops/op_regex_full_match.py | 36 + onnx/reference/ops/op_relu.py | 13 + onnx/reference/ops/op_reshape.py | 38 + onnx/reference/ops/op_resize.py | 665 + onnx/reference/ops/op_reverse_sequence.py | 20 + onnx/reference/ops/op_rms_normalization.py | 49 + onnx/reference/ops/op_rnn.py | 155 + onnx/reference/ops/op_roi_align.py | 296 + onnx/reference/ops/op_rotary_embedding.py | 119 + onnx/reference/ops/op_round.py | 13 + onnx/reference/ops/op_scan.py | 181 + onnx/reference/ops/op_scatter_elements.py | 119 + onnx/reference/ops/op_scatternd.py | 34 + onnx/reference/ops/op_selu.py | 15 + onnx/reference/ops/op_sequence_at.py | 11 + onnx/reference/ops/op_sequence_construct.py | 11 + onnx/reference/ops/op_sequence_empty.py | 11 + onnx/reference/ops/op_sequence_erase.py | 17 + onnx/reference/ops/op_sequence_insert.py | 52 + onnx/reference/ops/op_sequence_length.py | 17 + onnx/reference/ops/op_sequence_map.py | 33 + onnx/reference/ops/op_shape.py | 35 + onnx/reference/ops/op_shrink.py | 19 + onnx/reference/ops/op_sigmoid.py | 27 + onnx/reference/ops/op_sign.py | 13 + onnx/reference/ops/op_sin.py | 13 + onnx/reference/ops/op_sinh.py | 13 + onnx/reference/ops/op_size.py | 13 + onnx/reference/ops/op_slice.py | 75 + onnx/reference/ops/op_softmax.py | 19 + .../ops/op_softmax_cross_entropy_loss.py | 94 + onnx/reference/ops/op_softplus.py | 16 + onnx/reference/ops/op_softsign.py | 16 + onnx/reference/ops/op_space_to_depth.py | 33 + onnx/reference/ops/op_split.py | 51 + onnx/reference/ops/op_split_to_sequence.py | 54 + onnx/reference/ops/op_sqrt.py | 17 + onnx/reference/ops/op_squeeze.py | 45 + onnx/reference/ops/op_stft.py | 167 + onnx/reference/ops/op_string_concat.py | 23 + onnx/reference/ops/op_string_normalizer.py | 149 + onnx/reference/ops/op_string_split.py | 46 + onnx/reference/ops/op_sub.py | 13 + onnx/reference/ops/op_sum.py | 11 + onnx/reference/ops/op_swish.py | 14 + onnx/reference/ops/op_tan.py | 13 + onnx/reference/ops/op_tanh.py | 13 + onnx/reference/ops/op_tensor_scatter.py | 50 + onnx/reference/ops/op_tfidf_vectorizer.py | 380 + onnx/reference/ops/op_thresholded_relu.py | 14 + onnx/reference/ops/op_tile.py | 13 + onnx/reference/ops/op_topk.py | 111 + onnx/reference/ops/op_transpose.py | 20 + onnx/reference/ops/op_trilu.py | 16 + onnx/reference/ops/op_unique.py | 51 + onnx/reference/ops/op_unsqueeze.py | 52 + onnx/reference/ops/op_upsample.py | 20 + onnx/reference/ops/op_where.py | 21 + onnx/reference/ops/op_xor.py | 13 + onnx/reference/ops_optimized/__init__.py | 10 + .../ops_optimized/op_conv_optimized.py | 189 + onnx/reference/reference_evaluator.py | 617 + onnx/serialization.py | 212 + onnx/shape_inference.py | 181 + onnx/shape_inference/CMakeLists.txt | 8 + onnx/shape_inference/attribute_binder.h | 69 + onnx/shape_inference/implementation.cc | 1169 + onnx/shape_inference/implementation.h | 538 + onnx/string_utils.h | 43 + onnx/test/basic_test.py | 293 + onnx/test/checker_test.py | 1538 + onnx/test/cmake/CMakeLists.txt | 9 + onnx/test/cmake/main.cc | 29 + onnx/test/compose_test.py | 1122 + onnx/test/cpp/checker_test.cc | 186 + onnx/test/cpp/data_propagation_test.cc | 412 + onnx/test/cpp/function_context_test.cc | 281 + onnx/test/cpp/function_get_test.cc | 47 + onnx/test/cpp/function_verify_test.cc | 568 + onnx/test/cpp/inliner_test.cc | 502 + onnx/test/cpp/ir_test.cc | 72 + onnx/test/cpp/op_reg_test.cc | 24 + onnx/test/cpp/parser_test.cc | 858 + onnx/test/cpp/schema_registration_test.cc | 269 + onnx/test/cpp/shape_inference_test.cc | 660 + onnx/test/cpp/tensor_test.cc | 47 + onnx/test/cpp/test_main.cc | 13 + onnx/test/cpp/type_builders_test.cc | 53 + onnx/test/cpp/utf8_conversion_test.cc | 27 + onnx/test/data_propagation_test.py | 245 + onnx/test/function_inference_test.py | 119 + onnx/test/function_test.py | 221 + onnx/test/helper_test.py | 1172 + onnx/test/inference_function_test.py | 298 + onnx/test/inliner_test.py | 140 + onnx/test/model_container_refeval_test.py | 136 + onnx/test/model_container_test.py | 133 + onnx/test/model_inference_test.py | 272 + onnx/test/node_shape_inference_test.py | 24 + onnx/test/numpy_helper_test.py | 324 + onnx/test/parser_test.py | 410 + onnx/test/printer_test.py | 68 + onnx/test/reference_evaluator_ml_test.py | 2353 + onnx/test/reference_evaluator_model_test.py | 124 + onnx/test/reference_evaluator_test.py | 6512 +++ onnx/test/schema_test.py | 563 + onnx/test/serialization_test.py | 98 + onnx/test/shape_inference_test.py | 13184 +++++ onnx/test/symbolic_shape_test.py | 193 + onnx/test/test_backend_reference.py | 184 + onnx/test/test_backend_test.py | 120 + onnx/test/test_external_data.py | 1289 + onnx/test/tools_test.py | 321 + onnx/test/training_tool_test.py | 95 + onnx/test/utils_test.py | 79 + .../automatic_conversion_test_base.py | 160 + .../automatic_downgrade_test.py | 130 + .../automatic_upgrade_test.py | 2110 + onnx/test/version_converter_test.py | 3027 ++ onnx/test/version_utils.py | 20 + onnx/tools/__init__.py | 3 + onnx/tools/replace_constants.py | 421 + onnx/tools/update_model_dims.py | 98 + onnx/utils.py | 324 + onnx/version.py | 24 + onnx/version_converter.py | 43 + onnx/version_converter/BaseConverter.h | 90 + onnx/version_converter/CMakeLists.txt | 12 + .../adapters/Attention_24_23.h | 45 + .../version_converter/adapters/CMakeLists.txt | 50 + onnx/version_converter/adapters/adapter.h | 69 + .../adapters/axes_attribute_to_input.h | 48 + .../adapters/axes_input_to_attribute.h | 62 + .../adapters/axis_attribute_to_input.h | 72 + .../adapters/axis_input_to_attribute.h | 90 + .../adapters/batch_normalization_13_14.h | 32 + .../broadcast_backward_compatibility.h | 61 + .../broadcast_forward_compatibility.h | 86 + onnx/version_converter/adapters/cast_9_8.h | 33 + onnx/version_converter/adapters/clip_10_11.h | 56 + onnx/version_converter/adapters/compatible.h | 28 + .../adapters/dropout_11_12.h | 47 + .../adapters/extend_supported_types.h | 104 + onnx/version_converter/adapters/gemm_6_7.h | 57 + onnx/version_converter/adapters/gemm_7_6.h | 66 + .../adapters/gridsample_19_20.h | 36 + .../adapters/group_normalization_20_21.h | 127 + onnx/version_converter/adapters/maxpool_8_7.h | 34 + .../adapters/no_previous_version.h | 30 + onnx/version_converter/adapters/pad_10_11.h | 56 + onnx/version_converter/adapters/q_dq_21_20.h | 87 + onnx/version_converter/adapters/range_27_26.h | 61 + .../adapters/remove_consumed_inputs.h | 30 + onnx/version_converter/adapters/reshape_4_5.h | 49 + onnx/version_converter/adapters/reshape_5_4.h | 65 + .../version_converter/adapters/resize_10_11.h | 53 + onnx/version_converter/adapters/scan_8_9.h | 65 + onnx/version_converter/adapters/scan_9_8.h | 93 + .../adapters/scatter_10_11.h | 46 + .../adapters/scatter_16_15.h | 66 + .../adapters/scatter_18_17.h | 33 + onnx/version_converter/adapters/slice_9_10.h | 54 + .../adapters/softmax_12_13.h | 86 + .../adapters/softmax_13_12.h | 66 + onnx/version_converter/adapters/split_12_13.h | 45 + onnx/version_converter/adapters/split_13_12.h | 61 + onnx/version_converter/adapters/split_17_18.h | 36 + onnx/version_converter/adapters/sum_8_7.h | 39 + onnx/version_converter/adapters/topk_9_10.h | 43 + .../version_converter/adapters/transformers.h | 100 + .../adapters/type_restriction.h | 63 + .../version_converter/adapters/upsample_6_7.h | 48 + .../version_converter/adapters/upsample_8_9.h | 48 + .../adapters/upsample_9_10.h | 44 + .../version_converter/adapters/upsample_9_8.h | 79 + onnx/version_converter/convert.cc | 161 + onnx/version_converter/convert.h | 998 + onnx/version_converter/helper.cc | 83 + onnx/version_converter/helper.h | 50 + pixi.lock | 16948 ++++++ pixi.toml | 180 + pyproject.toml | 282 + renovate.json5 | 105 + requirements-dev.txt | 13 + requirements-lintrunner.txt | 11 + requirements-release_test.txt | 13 + sbom.cdx.json | 75 + tools/check_namespace.py | 30 + tools/gen_coverage_report.py | 95 + tools/pixi-shim.py | 32 + tools/pretty_clang_tidy.py | 35 + tools/update_doc.bat | 19 + tools/update_doc.sh | 35 + typos.toml | 60 + 9424 files changed, 332934 insertions(+) create mode 100644 .agents/skills/add-function-body/SKILL.md create mode 100644 .agents/skills/add-op/SKILL.md create mode 100644 .agents/skills/add-op/references/node-test-pattern.md create mode 100644 .agents/skills/add-op/references/reference-impl-pattern.md create mode 100644 .agents/skills/add-shape-inference/SKILL.md create mode 100644 .agents/skills/onnxtxt/SKILL.md create mode 100644 .clang-format create mode 100644 .clang-tidy create mode 100644 .claude/instructions/add-function-body.md create mode 100644 .claude/instructions/add-op.md create mode 100644 .claude/instructions/add-shape-inference.md create mode 100644 .editorconfig create mode 100644 .git-blame-ignore-revs create mode 100644 .gitattributes create mode 100644 .github/ISSUE_TEMPLATE/bug.md create mode 100644 .github/ISSUE_TEMPLATE/config.yml create mode 100644 .github/ISSUE_TEMPLATE/feature_request.yml create mode 100644 .github/ISSUE_TEMPLATE/operator.md create mode 100644 .github/ISSUE_TEMPLATE/question.md create mode 100644 .github/copilot-instructions.md create mode 100644 .github/dependabot.yml create mode 100644 .github/install_test.yml create mode 100644 .github/instructions/onnx-defs.instructions.md create mode 100644 .github/pull_request_template.md create mode 100644 .github/release.yml create mode 100644 .github/workflows/check_pr_label.yml create mode 100644 .github/workflows/check_urls.yml create mode 100644 .github/workflows/clang-tidy.yml create mode 100644 .github/workflows/codeql.yml create mode 100644 .github/workflows/copilot-setup-steps.yml create mode 100644 .github/workflows/create_release.yml create mode 100644 .github/workflows/dco_merge_group.yml create mode 100644 .github/workflows/dependency-review.yml create mode 100644 .github/workflows/fuzz.yml create mode 100644 .github/workflows/lint.yml create mode 100644 .github/workflows/main.yml create mode 100644 .github/workflows/pages.yml create mode 100644 .github/workflows/pixi_build.yml create mode 100644 .github/workflows/preview_source_dist_test.yml create mode 100644 .github/workflows/release_linux_cibw.yml create mode 100644 .github/workflows/release_macos_cibw.yml create mode 100644 .github/workflows/release_pyodide_cibw.yml create mode 100644 .github/workflows/release_sdist.yml create mode 100644 .github/workflows/release_windows_cibw.yml create mode 100644 .github/workflows/scorecard.yml create mode 100644 .github/workflows/stale.yml create mode 100644 .github/workflows/win_no_exception_ci.yml create mode 100644 .gitignore create mode 100644 .gitmodules create mode 100644 .lintrunner.toml create mode 100644 .lycheeignore create mode 100644 .pre-commit-config.yaml create mode 100644 CLAUDE.md create mode 100644 CMakeLists.txt create mode 100644 CODEOWNERS create mode 100644 CODE_OF_CONDUCT.md create mode 100644 CONTRIBUTING.md create mode 100644 INSTALL.md create mode 100644 LICENSE create mode 100644 LICENSES/Apache-2.0.txt create mode 100644 LICENSES/BSD-2-Clause.txt create mode 100644 NOTICE create mode 100644 README.md create mode 100644 README.wehub.md create mode 100644 RELEASE-MANAGEMENT.md create mode 100644 REUSE.toml create mode 100644 ROADMAP.md create mode 100644 SECURITY.md create mode 100644 VERSION_NUMBER create mode 100644 cmake/ONNXConfig.cmake.in create mode 100644 cmake/ONNXConfigVersion.cmake.in create mode 100644 cmake/Utils.cmake create mode 100644 cmake/external/FindSanitizer.cmake create mode 100644 cmake/external/googletest.cmake create mode 100644 cmake/summary.cmake create mode 100644 cmake/unittest.cmake create mode 100644 codecov.yml create mode 100644 community/logo_request.md create mode 100644 community/readme.md create mode 100644 community/repo_guidelines.md create mode 100644 community/sc-election-guidelines.md create mode 100644 community/sigs.md create mode 100644 community/working-groups.md create mode 100644 docs/AddFunctionBody.md create mode 100644 docs/AddNewOp.md create mode 100644 docs/AssuranceCase.md create mode 100644 docs/Broadcasting.md create mode 100644 docs/CIPipelines.md create mode 100644 docs/Changelog-ml.md create mode 100644 docs/Changelog.md create mode 100644 docs/DefineDifferentiability.md create mode 100644 docs/DimensionDenotation.md create mode 100644 docs/ExternalData.md create mode 100644 docs/ExternalDataSecurity.md create mode 100644 docs/IR.md create mode 100644 docs/ImplementingAnOnnxBackend.md create mode 100644 docs/ManagingExperimentalOps.md create mode 100644 docs/MetadataProps.md create mode 100644 docs/ONNXTypes.md create mode 100644 docs/OnnxBackendTest.md create mode 100644 docs/OnnxReleases.md create mode 100644 docs/OpConventions.md create mode 100644 docs/Operators-ml.md create mode 100644 docs/Operators.md create mode 100644 docs/Overview.md create mode 100644 docs/PythonAPIOverview.md create mode 100644 docs/ReleaseVerification.md create mode 100644 docs/Relicensing.md create mode 100644 docs/ShapeInference.md create mode 100644 docs/Syntax.md create mode 100644 docs/TestCoverage-ml.md create mode 100644 docs/TestCoverage.md create mode 100644 docs/TypeDenotation.md create mode 100644 docs/VersionConverter.md create mode 100644 docs/Versioning.md create mode 100644 docs/docsgen/Makefile create mode 100644 docs/docsgen/make.bat create mode 100644 docs/docsgen/source/_static/ONNX-Logo.svg create mode 100644 docs/docsgen/source/_static/css/custom.css create mode 100644 docs/docsgen/source/_static/diff2html-ui-slim.min.js create mode 100644 docs/docsgen/source/_static/diff2html.min.css create mode 100644 docs/docsgen/source/_static/onnx-horizontal-color.png create mode 100644 docs/docsgen/source/_static/onnx-horizontal-white.png create mode 100644 docs/docsgen/source/_static/readme.txt create mode 100644 docs/docsgen/source/api/backend.md create mode 100644 docs/docsgen/source/api/checker.md create mode 100644 docs/docsgen/source/api/classes.md create mode 100644 docs/docsgen/source/api/compose.md create mode 100644 docs/docsgen/source/api/defs.md create mode 100644 docs/docsgen/source/api/external_data_helper.md create mode 100644 docs/docsgen/source/api/helper.md create mode 100644 docs/docsgen/source/api/index.md create mode 100644 docs/docsgen/source/api/inliner.md create mode 100644 docs/docsgen/source/api/model_container.md create mode 100644 docs/docsgen/source/api/numpy_helper.md create mode 100644 docs/docsgen/source/api/parser.md create mode 100644 docs/docsgen/source/api/printer.md create mode 100644 docs/docsgen/source/api/reference.md create mode 100644 docs/docsgen/source/api/serialization.md create mode 100644 docs/docsgen/source/api/shape_inference.md create mode 100644 docs/docsgen/source/api/tools.md create mode 100644 docs/docsgen/source/api/utils.md create mode 100644 docs/docsgen/source/api/version_converter.md create mode 100644 docs/docsgen/source/conf.py create mode 100644 docs/docsgen/source/expect_onnxruntime.md create mode 100644 docs/docsgen/source/index.md create mode 100644 docs/docsgen/source/intro/concepts.md create mode 100644 docs/docsgen/source/intro/converters.md create mode 100644 docs/docsgen/source/intro/images/dot_att.png create mode 100644 docs/docsgen/source/intro/images/dot_if.png create mode 100644 docs/docsgen/source/intro/images/dot_if_py.png create mode 100644 docs/docsgen/source/intro/images/dot_linreg.png create mode 100644 docs/docsgen/source/intro/images/dot_linreg2.png create mode 100644 docs/docsgen/source/intro/images/dot_scan.png create mode 100644 docs/docsgen/source/intro/images/dot_scan_py.png create mode 100644 docs/docsgen/source/intro/images/iff.png create mode 100644 docs/docsgen/source/intro/images/linreg1.png create mode 100644 docs/docsgen/source/intro/images/linreg2.png create mode 100644 docs/docsgen/source/intro/images/scanop.png create mode 100644 docs/docsgen/source/intro/index.md create mode 100644 docs/docsgen/source/intro/python.md create mode 100644 docs/docsgen/source/onnx-favicon.png create mode 100644 docs/docsgen/source/onnx_sphinx.py create mode 100644 docs/docsgen/source/repo-docs/.gitignore create mode 100644 docs/docsgen/source/repo-docs/index.md create mode 100644 docs/docsgen/source/technical/InPlaceKVCache.png create mode 100644 docs/docsgen/source/technical/float4.md create mode 100644 docs/docsgen/source/technical/float8.md create mode 100644 docs/docsgen/source/technical/index.md create mode 100644 docs/docsgen/source/technical/int2.md create mode 100644 docs/docsgen/source/technical/int4.md create mode 100644 docs/docsgen/source/technical/kv_cache.md create mode 100644 docs/images/onnx_hub_arch.svg create mode 100644 docs/onnx-horizontal-color.png create mode 100644 docs/proposals/0000-template.md create mode 100644 docs/proposals/0001-ArchiveFileFormatProposal.md create mode 100644 docs/proposals/0002-NLP-in-ONNX-proposal.md create mode 100644 docs/proposals/0003-ONNXIFIproposal.md create mode 100644 docs/proposals/0004-FunctionsProposal.md create mode 100644 docs/proposals/0005-SymbolicShapeInfProposal.md create mode 100644 docs/proposals/0006-ONNXMultiDeviceProposal.md create mode 100644 docs/proposals/0007-ShardingFormalism.md create mode 100644 docs/proposals/README.md create mode 100644 docs/proposals/images/composing_broadcast_axes.png create mode 100644 onnx/CMakeLists.txt create mode 100644 onnx/__init__.py create mode 100644 onnx/_mapping.py create mode 100644 onnx/backend/__init__.py create mode 100644 onnx/backend/base.py create mode 100644 onnx/backend/sample/__init__.py create mode 100644 onnx/backend/sample/ops/__init__.py create mode 100644 onnx/backend/sample/ops/abs.py create mode 100644 onnx/backend/test/__init__.py create mode 100644 onnx/backend/test/case/__init__.py create mode 100644 onnx/backend/test/case/base.py create mode 100644 onnx/backend/test/case/model/__init__.py create mode 100644 onnx/backend/test/case/model/expand.py create mode 100644 onnx/backend/test/case/model/gradient.py create mode 100644 onnx/backend/test/case/model/sequence.py create mode 100644 onnx/backend/test/case/model/shrink.py create mode 100644 onnx/backend/test/case/model/sign.py create mode 100644 onnx/backend/test/case/model/single_relu.py create mode 100644 onnx/backend/test/case/model/stringnormalizer.py create mode 100644 onnx/backend/test/case/node/__init__.py create mode 100644 onnx/backend/test/case/node/_image_decoder_data.py create mode 100644 onnx/backend/test/case/node/abs.py create mode 100644 onnx/backend/test/case/node/acos.py create mode 100644 onnx/backend/test/case/node/acosh.py create mode 100644 onnx/backend/test/case/node/adagrad.py create mode 100644 onnx/backend/test/case/node/adam.py create mode 100644 onnx/backend/test/case/node/add.py create mode 100644 onnx/backend/test/case/node/affinegrid.py create mode 100644 onnx/backend/test/case/node/ai_onnx_ml/__init__.py create mode 100644 onnx/backend/test/case/node/ai_onnx_ml/array_feature_extractor.py create mode 100644 onnx/backend/test/case/node/ai_onnx_ml/binarizer.py create mode 100644 onnx/backend/test/case/node/ai_onnx_ml/label_encoder.py create mode 100644 onnx/backend/test/case/node/ai_onnx_ml/tree_ensemble.py create mode 100644 onnx/backend/test/case/node/and_.py create mode 100644 onnx/backend/test/case/node/argmax.py create mode 100644 onnx/backend/test/case/node/argmin.py create mode 100644 onnx/backend/test/case/node/asin.py create mode 100644 onnx/backend/test/case/node/asinh.py create mode 100644 onnx/backend/test/case/node/atan.py create mode 100644 onnx/backend/test/case/node/atanh.py create mode 100644 onnx/backend/test/case/node/attention.py create mode 100644 onnx/backend/test/case/node/averagepool.py create mode 100644 onnx/backend/test/case/node/batchnorm.py create mode 100755 onnx/backend/test/case/node/bernoulli.py create mode 100644 onnx/backend/test/case/node/bitcast.py create mode 100644 onnx/backend/test/case/node/bitshift.py create mode 100644 onnx/backend/test/case/node/bitwiseand.py create mode 100644 onnx/backend/test/case/node/bitwisenot.py create mode 100644 onnx/backend/test/case/node/bitwiseor.py create mode 100644 onnx/backend/test/case/node/bitwisexor.py create mode 100644 onnx/backend/test/case/node/blackmanwindow.py create mode 100644 onnx/backend/test/case/node/cast.py create mode 100644 onnx/backend/test/case/node/castlike.py create mode 100644 onnx/backend/test/case/node/causal_conv_with_state.py create mode 100644 onnx/backend/test/case/node/ceil.py create mode 100644 onnx/backend/test/case/node/celu.py create mode 100644 onnx/backend/test/case/node/center_crop_pad.py create mode 100644 onnx/backend/test/case/node/clip.py create mode 100644 onnx/backend/test/case/node/col2im.py create mode 100644 onnx/backend/test/case/node/compress.py create mode 100644 onnx/backend/test/case/node/concat.py create mode 100644 onnx/backend/test/case/node/constant.py create mode 100644 onnx/backend/test/case/node/constantofshape.py create mode 100644 onnx/backend/test/case/node/conv.py create mode 100644 onnx/backend/test/case/node/convinteger.py create mode 100644 onnx/backend/test/case/node/convtranspose.py create mode 100644 onnx/backend/test/case/node/cos.py create mode 100644 onnx/backend/test/case/node/cosh.py create mode 100644 onnx/backend/test/case/node/cumprod.py create mode 100644 onnx/backend/test/case/node/cumsum.py create mode 100644 onnx/backend/test/case/node/deformconv.py create mode 100755 onnx/backend/test/case/node/depthtospace.py create mode 100644 onnx/backend/test/case/node/dequantizelinear.py create mode 100644 onnx/backend/test/case/node/det.py create mode 100644 onnx/backend/test/case/node/dft.py create mode 100644 onnx/backend/test/case/node/div.py create mode 100644 onnx/backend/test/case/node/dropout.py create mode 100644 onnx/backend/test/case/node/dynamicquantizelinear.py create mode 100644 onnx/backend/test/case/node/einsum.py create mode 100644 onnx/backend/test/case/node/elu.py create mode 100644 onnx/backend/test/case/node/equal.py create mode 100644 onnx/backend/test/case/node/erf.py create mode 100644 onnx/backend/test/case/node/exp.py create mode 100644 onnx/backend/test/case/node/expand.py create mode 100644 onnx/backend/test/case/node/eyelike.py create mode 100644 onnx/backend/test/case/node/flatten.py create mode 100644 onnx/backend/test/case/node/flexattention.py create mode 100644 onnx/backend/test/case/node/floor.py create mode 100644 onnx/backend/test/case/node/gather.py create mode 100644 onnx/backend/test/case/node/gatherelements.py create mode 100644 onnx/backend/test/case/node/gathernd.py create mode 100644 onnx/backend/test/case/node/gelu.py create mode 100644 onnx/backend/test/case/node/gemm.py create mode 100644 onnx/backend/test/case/node/globalaveragepool.py create mode 100644 onnx/backend/test/case/node/globalmaxpool.py create mode 100644 onnx/backend/test/case/node/greater.py create mode 100644 onnx/backend/test/case/node/greater_equal.py create mode 100644 onnx/backend/test/case/node/gridsample.py create mode 100644 onnx/backend/test/case/node/groupnormalization.py create mode 100644 onnx/backend/test/case/node/gru.py create mode 100644 onnx/backend/test/case/node/hammingwindow.py create mode 100644 onnx/backend/test/case/node/hannwindow.py create mode 100644 onnx/backend/test/case/node/hardmax.py create mode 100644 onnx/backend/test/case/node/hardsigmoid.py create mode 100644 onnx/backend/test/case/node/hardswish.py create mode 100644 onnx/backend/test/case/node/identity.py create mode 100644 onnx/backend/test/case/node/if_.py create mode 100644 onnx/backend/test/case/node/image_decoder.py create mode 100644 onnx/backend/test/case/node/instancenorm.py create mode 100644 onnx/backend/test/case/node/isinf.py create mode 100644 onnx/backend/test/case/node/isnan.py create mode 100644 onnx/backend/test/case/node/layernormalization.py create mode 100644 onnx/backend/test/case/node/leakyrelu.py create mode 100644 onnx/backend/test/case/node/less.py create mode 100644 onnx/backend/test/case/node/less_equal.py create mode 100644 onnx/backend/test/case/node/linear_attention.py create mode 100644 onnx/backend/test/case/node/log.py create mode 100644 onnx/backend/test/case/node/logsoftmax.py create mode 100644 onnx/backend/test/case/node/loop.py create mode 100644 onnx/backend/test/case/node/lpnormalization.py create mode 100644 onnx/backend/test/case/node/lppool.py create mode 100644 onnx/backend/test/case/node/lrn.py create mode 100644 onnx/backend/test/case/node/lstm.py create mode 100644 onnx/backend/test/case/node/matmul.py create mode 100644 onnx/backend/test/case/node/matmulinteger.py create mode 100644 onnx/backend/test/case/node/max.py create mode 100644 onnx/backend/test/case/node/maxpool.py create mode 100644 onnx/backend/test/case/node/maxunpool.py create mode 100644 onnx/backend/test/case/node/mean.py create mode 100644 onnx/backend/test/case/node/meanvariancenormalization.py create mode 100644 onnx/backend/test/case/node/melweightmatrix.py create mode 100644 onnx/backend/test/case/node/min.py create mode 100644 onnx/backend/test/case/node/mish.py create mode 100644 onnx/backend/test/case/node/mod.py create mode 100644 onnx/backend/test/case/node/momentum.py create mode 100644 onnx/backend/test/case/node/mul.py create mode 100644 onnx/backend/test/case/node/neg.py create mode 100644 onnx/backend/test/case/node/negativeloglikelihoodloss.py create mode 100644 onnx/backend/test/case/node/nonmaxsuppression.py create mode 100644 onnx/backend/test/case/node/nonzero.py create mode 100644 onnx/backend/test/case/node/not_.py create mode 100644 onnx/backend/test/case/node/onehot.py create mode 100644 onnx/backend/test/case/node/optionalgetelement.py create mode 100644 onnx/backend/test/case/node/optionalhaselement.py create mode 100644 onnx/backend/test/case/node/or_.py create mode 100644 onnx/backend/test/case/node/pad.py create mode 100644 onnx/backend/test/case/node/pow.py create mode 100644 onnx/backend/test/case/node/prelu.py create mode 100644 onnx/backend/test/case/node/qlinearconv.py create mode 100644 onnx/backend/test/case/node/qlinearmatmul.py create mode 100644 onnx/backend/test/case/node/quantizelinear.py create mode 100644 onnx/backend/test/case/node/rangeop.py create mode 100644 onnx/backend/test/case/node/reciprocal.py create mode 100644 onnx/backend/test/case/node/reduce_log_sum.py create mode 100644 onnx/backend/test/case/node/reduce_log_sum_exp.py create mode 100644 onnx/backend/test/case/node/reducel1.py create mode 100644 onnx/backend/test/case/node/reducel2.py create mode 100644 onnx/backend/test/case/node/reducemax.py create mode 100644 onnx/backend/test/case/node/reducemean.py create mode 100644 onnx/backend/test/case/node/reducemin.py create mode 100644 onnx/backend/test/case/node/reduceprod.py create mode 100644 onnx/backend/test/case/node/reducesum.py create mode 100644 onnx/backend/test/case/node/reducesumsquare.py create mode 100644 onnx/backend/test/case/node/regex_full_match.py create mode 100644 onnx/backend/test/case/node/relu.py create mode 100644 onnx/backend/test/case/node/reshape.py create mode 100644 onnx/backend/test/case/node/resize.py create mode 100644 onnx/backend/test/case/node/reversesequence.py create mode 100644 onnx/backend/test/case/node/rmsnormalization.py create mode 100644 onnx/backend/test/case/node/rnn.py create mode 100644 onnx/backend/test/case/node/roialign.py create mode 100644 onnx/backend/test/case/node/rotaryembedding.py create mode 100644 onnx/backend/test/case/node/round.py create mode 100644 onnx/backend/test/case/node/scan.py create mode 100644 onnx/backend/test/case/node/scatter.py create mode 100644 onnx/backend/test/case/node/scatterelements.py create mode 100644 onnx/backend/test/case/node/scatternd.py create mode 100644 onnx/backend/test/case/node/selu.py create mode 100755 onnx/backend/test/case/node/sequence_map.py create mode 100644 onnx/backend/test/case/node/sequenceinsert.py create mode 100644 onnx/backend/test/case/node/shape.py create mode 100644 onnx/backend/test/case/node/shrink.py create mode 100644 onnx/backend/test/case/node/sigmoid.py create mode 100644 onnx/backend/test/case/node/sign.py create mode 100644 onnx/backend/test/case/node/sin.py create mode 100644 onnx/backend/test/case/node/sinh.py create mode 100644 onnx/backend/test/case/node/size.py create mode 100644 onnx/backend/test/case/node/slice.py create mode 100644 onnx/backend/test/case/node/softmax.py create mode 100644 onnx/backend/test/case/node/softmaxcrossentropy.py create mode 100644 onnx/backend/test/case/node/softplus.py create mode 100644 onnx/backend/test/case/node/softsign.py create mode 100644 onnx/backend/test/case/node/spacetodepth.py create mode 100644 onnx/backend/test/case/node/split.py create mode 100644 onnx/backend/test/case/node/splittosequence.py create mode 100644 onnx/backend/test/case/node/sqrt.py create mode 100644 onnx/backend/test/case/node/squeeze.py create mode 100644 onnx/backend/test/case/node/stft.py create mode 100644 onnx/backend/test/case/node/string_concat.py create mode 100644 onnx/backend/test/case/node/string_split.py create mode 100644 onnx/backend/test/case/node/stringnormalizer.py create mode 100644 onnx/backend/test/case/node/sub.py create mode 100644 onnx/backend/test/case/node/sum.py create mode 100644 onnx/backend/test/case/node/swish.py create mode 100644 onnx/backend/test/case/node/tan.py create mode 100644 onnx/backend/test/case/node/tanh.py create mode 100644 onnx/backend/test/case/node/tensorscatter.py create mode 100644 onnx/backend/test/case/node/tfidfvectorizer.py create mode 100644 onnx/backend/test/case/node/thresholdedrelu.py create mode 100644 onnx/backend/test/case/node/tile.py create mode 100644 onnx/backend/test/case/node/topk.py create mode 100644 onnx/backend/test/case/node/transpose.py create mode 100644 onnx/backend/test/case/node/trilu.py create mode 100644 onnx/backend/test/case/node/unique.py create mode 100644 onnx/backend/test/case/node/unsqueeze.py create mode 100644 onnx/backend/test/case/node/upsample.py create mode 100644 onnx/backend/test/case/node/where.py create mode 100644 onnx/backend/test/case/node/xor.py create mode 100644 onnx/backend/test/case/test_case.py create mode 100644 onnx/backend/test/case/utils.py create mode 100644 onnx/backend/test/cmd_tools.py create mode 100644 onnx/backend/test/data/light/README.md create mode 100644 onnx/backend/test/data/light/light_bvlc_alexnet.onnx create mode 100644 onnx/backend/test/data/light/light_bvlc_alexnet_output_0.pb create mode 100644 onnx/backend/test/data/light/light_densenet121.onnx create mode 100644 onnx/backend/test/data/light/light_densenet121_output_0.pb create mode 100644 onnx/backend/test/data/light/light_inception_v1.onnx create mode 100644 onnx/backend/test/data/light/light_inception_v1_output_0.pb create mode 100644 onnx/backend/test/data/light/light_inception_v2.onnx create mode 100644 onnx/backend/test/data/light/light_inception_v2_output_0.pb create mode 100644 onnx/backend/test/data/light/light_resnet50.onnx create mode 100644 onnx/backend/test/data/light/light_resnet50_output_0.pb create mode 100644 onnx/backend/test/data/light/light_shufflenet.onnx create mode 100644 onnx/backend/test/data/light/light_shufflenet_output_0.pb create mode 100644 onnx/backend/test/data/light/light_squeezenet.onnx create mode 100644 onnx/backend/test/data/light/light_squeezenet_output_0.pb create mode 100644 onnx/backend/test/data/light/light_vgg19.onnx create mode 100644 onnx/backend/test/data/light/light_vgg19_output_0.pb create mode 100644 onnx/backend/test/data/light/light_zfnet512.onnx create mode 100644 onnx/backend/test/data/light/light_zfnet512_output_0.pb create mode 100644 onnx/backend/test/data/node/test_abs/model.onnx create mode 100644 onnx/backend/test/data/node/test_abs/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_abs/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_acos/model.onnx create mode 100644 onnx/backend/test/data/node/test_acos/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_acos/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_acos_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_acos_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_acos_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_acosh/model.onnx create mode 100644 onnx/backend/test/data/node/test_acosh/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_acosh/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_acosh_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_acosh_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_acosh_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_adagrad/model.onnx create mode 100644 onnx/backend/test/data/node/test_adagrad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_adagrad/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_adagrad/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_adagrad/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_adagrad/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_adagrad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_adagrad/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/model.onnx create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_adagrad_multiple/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_adam/model.onnx create mode 100644 onnx/backend/test/data/node/test_adam/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_adam/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_adam/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_adam/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_adam/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_adam/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_adam/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_adam/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_adam/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/model.onnx create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/input_8.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/input_9.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/output_4.pb create mode 100644 onnx/backend/test/data/node/test_adam_multiple/test_data_set_0/output_5.pb create mode 100644 onnx/backend/test/data/node/test_add/model.onnx create mode 100644 onnx/backend/test/data/node/test_add/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_add/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_add/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_add_bcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_add_bcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_add_bcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_add_bcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_add_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_add_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_add_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_add_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_add_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_add_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_add_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_add_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_add_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_add_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_add_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_add_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_add_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_add_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_add_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_add_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_add_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_add_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_add_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_add_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_add_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_add_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_add_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_add_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_align_corners/model.onnx create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_align_corners/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_align_corners/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_align_corners/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_align_corners_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_align_corners_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_align_corners_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_align_corners_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_2d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_align_corners/model.onnx create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_align_corners/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_align_corners/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_align_corners/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_align_corners_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_align_corners_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_align_corners_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_align_corners_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_affine_grid_3d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_array_feature_extractor/model.onnx create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_array_feature_extractor/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_array_feature_extractor/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_array_feature_extractor/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_binarizer/model.onnx create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_binarizer/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_binarizer/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_string_int/model.onnx create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_string_int/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_string_int/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_string_int_no_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_string_int_no_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_string_int_no_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_tensor_mapping/model.onnx create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_tensor_mapping/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_tensor_mapping/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_tensor_value_only_mapping/model.onnx create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_tensor_value_only_mapping/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_label_encoder_tensor_value_only_mapping/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_tree_ensemble_set_membership/model.onnx create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_tree_ensemble_set_membership/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_tree_ensemble_set_membership/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_tree_ensemble_single_tree/model.onnx create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_tree_ensemble_single_tree/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_ai_onnx_ml_tree_ensemble_single_tree/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_and2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_and2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_and2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_and2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_and3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_and3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_and3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_and3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_and4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_and4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_and4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_and4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast3v1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_and_bcast3v1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast3v1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast3v1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast3v2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_and_bcast3v2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast3v2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast3v2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast4v2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_and_bcast4v2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast4v2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast4v2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast4v3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_and_bcast4v3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast4v3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast4v3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast4v4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_and_bcast4v4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast4v4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_and_bcast4v4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_example_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_example_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_example_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_random_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_random_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_default_axis_random_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_example_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_example_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_example_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_random_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_random_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_keepdims_random_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_example_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_example_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_example_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_random_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_random_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_negative_axis_keepdims_random_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_example_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_example_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_example_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_random_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_random_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmax_no_keepdims_random_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_example_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_example_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_example_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_random_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_random_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_default_axis_random_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_example_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_example_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_example_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_random_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_random_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_keepdims_random_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_example_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_example_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_example_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_random_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_random_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_negative_axis_keepdims_random_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_example_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_example_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_example_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_random_select_last_index/model.onnx create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_random_select_last_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_argmin_no_keepdims_random_select_last_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_asin/model.onnx create mode 100644 onnx/backend/test/data/node/test_asin/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_asin/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_asin_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_asin_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_asin_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_asinh/model.onnx create mode 100644 onnx/backend/test/data/node/test_asinh/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_asinh/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_asinh_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_asinh_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_asinh_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_atan/model.onnx create mode 100644 onnx/backend/test/data/node/test_atan/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_atan/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_atan_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_atan_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_atan_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_atanh/model.onnx create mode 100644 onnx/backend/test/data/node/test_atanh/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_atanh/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_atanh_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_atanh_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_atanh_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_boolmask_fullymasked_row_nan_robustness_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_23_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_fullymasked_qk_matmul_output_mode3_zero_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_24_qk_matmul_output_mode3_softmax_precision_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_attn_mask_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_causal/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_causal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_causal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_causal/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_causal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_causal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_causal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_causal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_causal_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_causal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_attn_mask_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_causal/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_causal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_causal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_causal/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_causal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_causal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_causal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_causal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_causal_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_causal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_scaled/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_scaled/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_scaled/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_scaled/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_scaled/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_scaled_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_scaled_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_scaled_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_scaled_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_scaled_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_softcap/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_softcap/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_softcap/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_softcap/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_softcap/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_softcap_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_softcap_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_softcap_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_softcap_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_sizes_softcap_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_diff_heads_with_past_and_present_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_attn_mask_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_causal/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_causal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_causal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_causal/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_causal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_causal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_causal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_causal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_causal_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_causal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_scaled/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_scaled/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_scaled/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_scaled/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_scaled/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_scaled_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_scaled_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_scaled_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_scaled_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_scaled_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_softcap/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_softcap/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_softcap/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_softcap/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_softcap/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_softcap_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_softcap_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_softcap_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_softcap_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_softcap_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_gqa_with_past_and_present_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_scaled/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_scaled/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_scaled/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_scaled/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_scaled/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_scaled_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_scaled_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_scaled_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_scaled_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_scaled_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_softcap/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_softcap/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_softcap/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_softcap/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_softcap/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_softcap_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_softcap_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_softcap_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_softcap_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_softcap_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_transpose_verification/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_transpose_verification/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_transpose_verification/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_transpose_verification/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_transpose_verification/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_transpose_verification_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_transpose_verification_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_transpose_verification_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_transpose_verification_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_transpose_verification_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_expanded/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softcap_expanded/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_3d_with_past_and_present_qk_matmul_softmax_expanded/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_causal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_3d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_causal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_4d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_4d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_bool_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_attn_mask_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_attn_mask_composition_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_batch_prefill_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_continued_prefill_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_nonpad_negative_offset_structural_empty_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_causal_with_past_and_present_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_mask4d_padded_kv_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_attn_mask_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_causal/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_causal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_causal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_causal/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_causal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_causal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_causal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_causal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_causal_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_causal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_scaled/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_scaled/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_scaled/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_scaled/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_scaled/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_scaled_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_scaled_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_scaled_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_scaled_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_scaled_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_softcap/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_softcap/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_softcap/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_softcap/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_softcap/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_softcap_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_softcap_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_softcap_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_softcap_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_sizes_softcap_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask3d_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_diff_heads_with_past_and_present_mask4d_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_fp16/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_fp16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_fp16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_fp16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_fp16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_fp16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_fp16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_fp16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_fp16_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_fp16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_attn_mask_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_causal_nonpad_decode_fp16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_scaled/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_scaled/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_scaled/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_scaled/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_scaled/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_scaled_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_scaled_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_scaled_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_scaled_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_scaled_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_softcap/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_softcap/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_softcap/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_softcap/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_softcap/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_softcap_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_softcap_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_softcap_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_softcap_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_softcap_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_gqa_with_past_and_present_fp16_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_scaled/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_scaled/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_scaled/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_scaled/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_scaled/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_scaled_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_scaled_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_scaled_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_scaled_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_scaled_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_softcap_neginf_mask_poison_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal_expanded/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_expanded/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal_expanded/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_expanded/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_bias_expanded/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_past_and_present_qk_matmul_expanded/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_bias_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softcap_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_4d_with_qk_matmul_softmax_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_attention_causal_boolmask_nan_robustness_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_1d_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_1d_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_1d_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_ceil/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_ceil/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_ceil/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_ceil_last_window_starts_on_pad/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_ceil_last_window_starts_on_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_ceil_last_window_starts_on_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_dilations/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_dilations/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_dilations/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_pads/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_pads/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_pads/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_pads_count_include_pad/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_pads_count_include_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_pads_count_include_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_pads/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_pads/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_pads/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_pads_count_include_pad/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_pads_count_include_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_pads_count_include_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_same_upper/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_same_upper/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_same_upper/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_strides/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_strides/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_precomputed_strides/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_same_lower/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_same_lower/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_same_lower/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_same_upper/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_same_upper/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_same_upper/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_strides/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_strides/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_2d_strides/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_0_ceil_mode_is_False/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_0_ceil_mode_is_False/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_0_ceil_mode_is_False/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_0_ceil_mode_is_True/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_0_ceil_mode_is_True/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_0_ceil_mode_is_True/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_1_ceil_mode_is_False/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_1_ceil_mode_is_False/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_1_ceil_mode_is_False/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_1_ceil_mode_is_True/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_1_ceil_mode_is_True/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_large_count_include_pad_is_1_ceil_mode_is_True/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_small/model.onnx create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_small/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_averagepool_3d_dilations_small/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_basic_conv_with_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_basic_conv_with_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_basic_conv_with_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_basic_conv_with_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_basic_conv_without_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_basic_conv_without_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_basic_conv_without_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_basic_conv_without_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_basic_deform_conv_with_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_basic_deform_conv_with_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_basic_deform_conv_with_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_basic_deform_conv_with_padding/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_basic_deform_conv_with_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_basic_deform_conv_without_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_basic_deform_conv_without_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_basic_deform_conv_without_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_basic_deform_conv_without_padding/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_basic_deform_conv_without_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon_training_mode/model.onnx create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon_training_mode/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon_training_mode/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon_training_mode/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon_training_mode/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon_training_mode/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon_training_mode/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon_training_mode/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_epsilon_training_mode/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_batchnorm_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example_training_mode/model.onnx create mode 100644 onnx/backend/test/data/node/test_batchnorm_example_training_mode/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example_training_mode/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example_training_mode/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example_training_mode/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example_training_mode/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example_training_mode/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example_training_mode/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_batchnorm_example_training_mode/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli/model.onnx create mode 100644 onnx/backend/test/data/node/test_bernoulli/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli_double/model.onnx create mode 100644 onnx/backend/test/data/node/test_bernoulli_double/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli_double/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli_double_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_bernoulli_double_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli_double_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_bernoulli_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli_seed/model.onnx create mode 100644 onnx/backend/test/data/node/test_bernoulli_seed/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli_seed/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli_seed_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_bernoulli_seed_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bernoulli_seed_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_2d_float32_to_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitcast_2d_float32_to_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_2d_float32_to_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_bool_to_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitcast_bool_to_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_bool_to_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_float32_to_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitcast_float32_to_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_float32_to_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_float64_to_int64/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitcast_float64_to_int64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_float64_to_int64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_int32_to_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitcast_int32_to_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_int32_to_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_int64_to_float64/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitcast_int64_to_float64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_int64_to_float64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_int8_to_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitcast_int8_to_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_int8_to_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_scalar_float32_to_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitcast_scalar_float32_to_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_scalar_float32_to_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_uint16_to_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitcast_uint16_to_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_uint16_to_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_uint32_to_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitcast_uint32_to_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitcast_uint32_to_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_left_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitshift_right_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_i16_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_and_i16_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_i16_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_i16_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_i32_2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_and_i32_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_i32_2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_i32_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_ui64_bcast_3v1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_and_ui64_bcast_3v1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_ui64_bcast_3v1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_ui64_bcast_3v1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_ui8_bcast_4v3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_and_ui8_bcast_4v3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_ui8_bcast_4v3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_and_ui8_bcast_4v3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_not_2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_not_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_not_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_not_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_not_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_not_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_not_4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_not_4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_not_4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_i16_4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_or_i16_4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_i16_4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_i16_4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_i32_2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_or_i32_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_i32_2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_i32_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_ui64_bcast_3v1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_or_ui64_bcast_3v1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_ui64_bcast_3v1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_ui64_bcast_3v1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_ui8_bcast_4v3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_or_ui8_bcast_4v3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_ui8_bcast_4v3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_or_ui8_bcast_4v3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_i16_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_i16_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_i16_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_i16_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_i32_2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_i32_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_i32_2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_i32_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_ui64_bcast_3v1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_ui64_bcast_3v1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_ui64_bcast_3v1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_ui64_bcast_3v1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_ui8_bcast_4v3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_ui8_bcast_4v3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_ui8_bcast_4v3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_bitwise_xor_ui8_bcast_4v3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_blackmanwindow/model.onnx create mode 100644 onnx/backend/test/data/node/test_blackmanwindow/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_blackmanwindow/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_blackmanwindow_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_blackmanwindow_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_blackmanwindow_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_blackmanwindow_symmetric/model.onnx create mode 100644 onnx/backend/test/data/node/test_blackmanwindow_symmetric/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_blackmanwindow_symmetric/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_blackmanwindow_symmetric_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_blackmanwindow_symmetric_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_blackmanwindow_symmetric_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_BFLOAT16_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_BFLOAT16_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_BFLOAT16_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_DOUBLE_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_DOUBLE_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_DOUBLE_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_DOUBLE_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_DOUBLE_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_DOUBLE_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_DOUBLE/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_DOUBLE/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_DOUBLE/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT4E2M1/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT4E2M1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT4E2M1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E4M3FN/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E4M3FN/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E4M3FN/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E4M3FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E4M3FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E4M3FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E5M2/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E5M2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E5M2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E5M2FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E5M2FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_FLOAT8E5M2FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_INT2/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_INT2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_INT2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_INT4/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_INT4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_INT4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_UINT2/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_UINT2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_UINT2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_UINT4/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_UINT4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT16_to_UINT4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT4E2M1_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT4E2M1_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT4E2M1_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT4E2M1_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT4E2M1_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT4E2M1_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FNUZ_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FNUZ_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FNUZ_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FNUZ_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FNUZ_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FNUZ_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FN_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FN_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FN_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FN_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FN_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E4M3FN_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2FNUZ_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2FNUZ_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2FNUZ_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2FNUZ_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2FNUZ_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2FNUZ_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT8E5M2_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_BFLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_BFLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_BFLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_DOUBLE/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_DOUBLE/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_DOUBLE/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT4E2M1/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT4E2M1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT4E2M1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E4M3FN/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E4M3FN/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E4M3FN/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E4M3FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E4M3FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E4M3FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E5M2/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E5M2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E5M2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E5M2FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E5M2FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_FLOAT8E5M2FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_INT2/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_INT2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_INT2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_INT4/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_INT4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_INT4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_UINT2/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_UINT2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_UINT2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_UINT4/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_UINT4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_FLOAT_to_UINT4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT2_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_INT2_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT2_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT2_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_INT2_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT2_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT2_to_INT8/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_INT2_to_INT8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT2_to_INT8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT4_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_INT4_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT4_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT4_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_INT4_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT4_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT4_to_INT8/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_INT4_to_INT8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_INT4_to_INT8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT2_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_UINT2_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT2_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT2_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_UINT2_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT2_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT2_to_UINT8/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_UINT2_to_UINT8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT2_to_UINT8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT4_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_UINT4_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT4_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT4_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_UINT4_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT4_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT4_to_UINT8/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_UINT4_to_UINT8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_UINT4_to_UINT8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT16_to_FLOAT8E8M0/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT16_to_FLOAT8E8M0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT16_to_FLOAT8E8M0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT8E8M0_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT8E8M0_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT8E8M0_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT8E8M0_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT8E8M0_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT8E8M0_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT_to_FLOAT8E8M0/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT_to_FLOAT8E8M0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_e8m0_FLOAT_to_FLOAT8E8M0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E4M3FN/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E4M3FN/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E4M3FN/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E5M2/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E5M2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E5M2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E4M3FN/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E4M3FN/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E4M3FN/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E5M2/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E5M2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E5M2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cast_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_BFLOAT16_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_BFLOAT16_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_BFLOAT16_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_BFLOAT16_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_BFLOAT16_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_BFLOAT16_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_BFLOAT16_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_BFLOAT16_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_DOUBLE_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_DOUBLE/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_DOUBLE/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_DOUBLE/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_DOUBLE/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_DOUBLE_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_DOUBLE_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_DOUBLE_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_DOUBLE_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT4E2M1/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT4E2M1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT4E2M1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT4E2M1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT4E2M1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT4E2M1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT4E2M1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT4E2M1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FN/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FN/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FN/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FN/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FNUZ/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FNUZ_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FN_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FN_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FN_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E4M3FN_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2FNUZ/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2FNUZ_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT8E5M2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT2/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT4/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT4_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT4_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT4_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_INT4_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT2/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT4/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT4_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT4_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT4_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT16_to_UINT4_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT4E2M1_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FNUZ_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E4M3FN_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2FNUZ_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT8E5M2_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_BFLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_BFLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_BFLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_BFLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_BFLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_BFLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_BFLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_BFLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_DOUBLE/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_DOUBLE/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_DOUBLE/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_DOUBLE/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_DOUBLE_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_DOUBLE_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_DOUBLE_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_DOUBLE_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT4E2M1/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT4E2M1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT4E2M1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT4E2M1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT4E2M1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT4E2M1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT4E2M1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT4E2M1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FN/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FN/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FN/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FN/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FNUZ/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FNUZ_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FN_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FN_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FN_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E4M3FN_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2FNUZ/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2FNUZ_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_FLOAT8E5M2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT2/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT4/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT4_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT4_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT4_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_INT4_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT2/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT4/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT4_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT4_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT4_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_FLOAT_to_UINT4_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_INT8/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_INT8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_INT8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_INT8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_INT8_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_INT8_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_INT8_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT2_to_INT8_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_INT8/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_INT8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_INT8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_INT8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_INT8_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_INT8_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_INT8_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_INT4_to_INT8_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_UINT8/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_UINT8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_UINT8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_UINT8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_UINT8_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_UINT8_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_UINT8_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT2_to_UINT8_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT16/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_FLOAT_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_UINT8/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_UINT8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_UINT8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_UINT8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_UINT8_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_UINT8_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_UINT8_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_UINT4_to_UINT8_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FN/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FN/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FN/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FN/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FN_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FN_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FN_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E4M3FN_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT16_to_FLOAT8E5M2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FN/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FN/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FN/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FN/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FN_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FN_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FN_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E4M3FN_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_castlike_no_saturate_FLOAT_to_FLOAT8E5M2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_b1_c1_degenerate/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_b1_c1_degenerate/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_b1_c1_degenerate/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_b1_c1_degenerate/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_b1_c1_degenerate/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_b1_c1_degenerate_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_b1_c1_degenerate_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_b1_c1_degenerate_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_b1_c1_degenerate_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_b1_c1_degenerate_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_basic/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_basic/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_basic/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_basic/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_basic/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_basic_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_basic_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_basic_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_basic_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_basic_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_decode_step_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_fp16/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_fp16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_fp16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_fp16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_fp16/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_fp16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_fp16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_fp16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_fp16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_fp16_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_kernel_size_one/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_kernel_size_one/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_kernel_size_one/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_kernel_size_one/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_kernel_size_one/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_kernel_size_one_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_kernel_size_one_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_kernel_size_one_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_kernel_size_one_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_kernel_size_one_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_short_input_no_past_state/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_short_input_no_past_state/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_short_input_no_past_state/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_short_input_no_past_state/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_short_input_no_past_state/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_short_input_no_past_state_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_short_input_no_past_state_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_short_input_no_past_state_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_short_input_no_past_state_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_short_input_no_past_state_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_fp16/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_fp16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_fp16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_fp16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_fp16/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_fp16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_fp16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_fp16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_fp16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_fp16_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_silu_with_past_state_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_swish_alias/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_swish_alias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_swish_alias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_swish_alias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_swish_alias/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_swish_alias_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_swish_alias_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_swish_alias_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_swish_alias_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_swish_alias_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_and_past_state_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_bias_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_causal_conv_with_state_with_past_state_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_ceil/model.onnx create mode 100644 onnx/backend/test/data/node/test_ceil/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_ceil/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_ceil_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_ceil_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_ceil_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_celu/model.onnx create mode 100644 onnx/backend/test/data/node/test_celu/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_celu/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_celu_bfloat16/model.onnx create mode 100644 onnx/backend/test/data/node/test_celu_bfloat16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_celu_bfloat16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_celu_bfloat16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_celu_bfloat16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_celu_bfloat16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_celu_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_celu_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_celu_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_celu_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_celu_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_celu_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_celu_float16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_celu_float16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_celu_float16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_and_pad/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_and_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_and_pad/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_and_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_and_pad_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_and_pad_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_and_pad_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_and_pad_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_chw/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_chw/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_chw/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_chw/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_chw_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_chw_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_chw_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_chw_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_hwc/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_hwc/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_hwc/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_hwc/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_hwc_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_hwc_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_hwc_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_axes_hwc_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_negative_axes_hwc/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_negative_axes_hwc/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_negative_axes_hwc/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_negative_axes_hwc/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_negative_axes_hwc_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_negative_axes_hwc_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_negative_axes_hwc_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_crop_negative_axes_hwc_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_pad/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_pad/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_pad_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_pad_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_pad_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_center_crop_pad_pad_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_inbounds/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_inbounds/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_inbounds/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_inbounds_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_inbounds_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_inbounds_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_inbounds/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_inbounds/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_inbounds/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_inbounds_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_inbounds_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_inbounds_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_max/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_max/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_max/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_max/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_max_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_max_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_max_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_max_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_min/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_min/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_min/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_min/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_min_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_min_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_min_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_int8_min_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_max/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_max/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_max/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_max/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_max_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_max_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_max_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_max_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_min/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_min/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_min/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_min/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_min_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_default_min_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_min_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_default_min_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_example/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_example_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_inbounds/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_inbounds/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_inbounds/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_inbounds/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_inbounds/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_inbounds_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_inbounds_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_inbounds_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_inbounds_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_inbounds_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_min_greater_than_max/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_min_greater_than_max/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_min_greater_than_max/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_min_greater_than_max/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_min_greater_than_max/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_min_greater_than_max_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_min_greater_than_max_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_min_greater_than_max_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_min_greater_than_max_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_min_greater_than_max_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_outbounds/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_outbounds/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_outbounds/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_outbounds/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_outbounds/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_outbounds_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_outbounds_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_outbounds_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_outbounds_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_outbounds_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_splitbounds/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_splitbounds/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_splitbounds/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_splitbounds/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_splitbounds/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_splitbounds_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_clip_splitbounds_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_clip_splitbounds_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_clip_splitbounds_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_clip_splitbounds_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_col2im/model.onnx create mode 100644 onnx/backend/test/data/node/test_col2im/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_col2im/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_col2im/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_col2im/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_col2im_5d/model.onnx create mode 100644 onnx/backend/test/data/node/test_col2im_5d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_col2im_5d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_col2im_5d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_col2im_5d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_col2im_dilations/model.onnx create mode 100644 onnx/backend/test/data/node/test_col2im_dilations/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_col2im_dilations/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_col2im_dilations/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_col2im_dilations/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_col2im_pads/model.onnx create mode 100644 onnx/backend/test/data/node/test_col2im_pads/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_col2im_pads/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_col2im_pads/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_col2im_pads/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_col2im_strides/model.onnx create mode 100644 onnx/backend/test/data/node/test_col2im_strides/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_col2im_strides/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_col2im_strides/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_col2im_strides/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_compress_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_compress_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_compress_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_compress_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_compress_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_compress_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_compress_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_compress_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_compress_default_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_compress_default_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_compress_default_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_compress_default_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_compress_negative_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_compress_negative_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_compress_negative_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_compress_negative_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_1d_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_1d_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_1d_axis_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_1d_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_1d_axis_negative_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_1d_axis_negative_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_1d_axis_negative_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_1d_axis_negative_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_negative_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_negative_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_negative_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_negative_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_negative_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_negative_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_negative_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_2d_axis_negative_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_3/model.onnx create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_concat_3d_axis_negative_3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_constant/model.onnx create mode 100644 onnx/backend/test/data/node/test_constant/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad/model.onnx create mode 100644 onnx/backend/test/data/node/test_constant_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_constant_pad_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad_axes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad_axes/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad_negative_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_constant_pad_negative_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad_negative_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad_negative_axes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad_negative_axes/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_constant_pad_negative_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_constantofshape_float_ones/model.onnx create mode 100644 onnx/backend/test/data/node/test_constantofshape_float_ones/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_constantofshape_float_ones/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_constantofshape_int_shape_zero/model.onnx create mode 100644 onnx/backend/test/data/node/test_constantofshape_int_shape_zero/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_constantofshape_int_shape_zero/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_constantofshape_int_zeros/model.onnx create mode 100644 onnx/backend/test/data/node/test_constantofshape_int_zeros/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_constantofshape_int_zeros/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_autopad_same/model.onnx create mode 100644 onnx/backend/test/data/node/test_conv_with_autopad_same/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_autopad_same/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_autopad_same/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_and_asymmetric_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_and_asymmetric_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_and_asymmetric_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_and_asymmetric_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_no_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_no_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_no_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_no_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_conv_with_strides_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convinteger_with_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_convinteger_with_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convinteger_with_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convinteger_with_padding/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_convinteger_with_padding/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_convinteger_with_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convinteger_without_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_convinteger_without_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convinteger_without_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convinteger_without_padding/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_convinteger_without_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose_1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_autopad_same/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose_autopad_same/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_autopad_same/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_autopad_same/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_dilations/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose_dilations/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_dilations/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_dilations/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_group_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose_group_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_group_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_group_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_group_2_image_3/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose_group_2_image_3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_group_2_image_3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_group_2_image_3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_kernel_shape/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose_kernel_shape/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_kernel_shape/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_kernel_shape/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_output_shape/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose_output_shape/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_output_shape/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_output_shape/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_pad/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_pad/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_pads/model.onnx create mode 100644 onnx/backend/test/data/node/test_convtranspose_pads/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_pads/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_convtranspose_pads/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cos/model.onnx create mode 100644 onnx/backend/test/data/node/test_cos/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cos/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cos_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_cos_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cos_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cosh/model.onnx create mode 100644 onnx/backend/test/data/node/test_cosh/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cosh/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cosh_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_cosh_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cosh_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumprod_1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_exclusive/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_exclusive/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_exclusive/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_exclusive/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_int32_exclusive/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_int32_exclusive/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_int32_exclusive/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_int32_exclusive/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_reverse/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_reverse/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_reverse/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_reverse/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_reverse_exclusive/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_reverse_exclusive/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_reverse_exclusive/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_1d_reverse_exclusive/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_axis_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_axis_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_axis_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_axis_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_axis_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_int32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_negative_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_negative_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_negative_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumprod_2d_negative_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumsum_1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_exclusive/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_exclusive/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_exclusive/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_exclusive/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_int32_exclusive/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_int32_exclusive/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_int32_exclusive/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_int32_exclusive/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_reverse/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_reverse/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_reverse/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_reverse/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_reverse_exclusive/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_reverse_exclusive/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_reverse_exclusive/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_1d_reverse_exclusive/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_axis_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_axis_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_axis_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_axis_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_axis_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_int32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_negative_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_negative_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_negative_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_cumsum_2d_negative_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_mask_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_mask_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_mask_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_mask_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_mask_bias/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_mask_bias/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_mask_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_multiple_offset_groups/model.onnx create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_multiple_offset_groups/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_multiple_offset_groups/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_multiple_offset_groups/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_deform_conv_with_multiple_offset_groups/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_depthtospace_crd_mode_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_depthtospace_crd_mode_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_depthtospace_crd_mode_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_depthtospace_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_depthtospace_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_depthtospace_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_axis/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_blocked/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_blocked/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_blocked/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_blocked/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_blocked/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn_float16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn_zero_point/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn_zero_point/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn_zero_point/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn_zero_point/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e4m3fn_zero_point/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e5m2/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e5m2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e5m2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_e5m2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_float4e2m1/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_float4e2m1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_float4e2m1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_float4e2m1/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_float4e2m1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int2/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int2/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int4/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int4/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_int4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint2/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint2/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint4/model.onnx create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint4/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_dequantizelinear_uint4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_det_2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_det_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_det_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_det_nd/model.onnx create mode 100644 onnx/backend/test/data/node/test_det_nd/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_det_nd/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dft/model.onnx create mode 100644 onnx/backend/test/data/node/test_dft/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dft/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dft/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_dft_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dft_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_axis_opset19/model.onnx create mode 100644 onnx/backend/test/data/node/test_dft_axis_opset19/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_axis_opset19/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_inverse/model.onnx create mode 100644 onnx/backend/test/data/node/test_dft_inverse/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_inverse/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dft_inverse/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_inverse_opset19/model.onnx create mode 100644 onnx/backend/test/data/node/test_dft_inverse_opset19/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_inverse_opset19/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_irfft/model.onnx create mode 100644 onnx/backend/test/data/node/test_dft_irfft/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_irfft/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dft_irfft/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_irfft_opset19/model.onnx create mode 100644 onnx/backend/test/data/node/test_dft_irfft_opset19/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_irfft_opset19/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_opset19/model.onnx create mode 100644 onnx/backend/test/data/node/test_dft_opset19/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_opset19/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_rfft/model.onnx create mode 100644 onnx/backend/test/data/node/test_dft_rfft/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_rfft/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dft_rfft/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_rfft_opset19/model.onnx create mode 100644 onnx/backend/test/data/node/test_dft_rfft_opset19/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dft_rfft_opset19/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_div/model.onnx create mode 100644 onnx/backend/test/data/node/test_div/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_div/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_div/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_div_bcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_div_bcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_div_bcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_div_bcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_div_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_div_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_div_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_div_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_div_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_div_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_div_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_div_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_div_int32_trunc/model.onnx create mode 100644 onnx/backend/test/data/node/test_div_int32_trunc/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_div_int32_trunc/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_div_int32_trunc/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_div_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_div_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_div_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_div_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_div_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_div_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_div_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_div_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_div_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_div_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_div_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_div_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_div_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_div_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_div_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_div_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_div_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_div_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_div_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_div_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_dropout_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_dropout_default_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_mask/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_mask_ratio/model.onnx create mode 100644 onnx/backend/test/data/node/test_dropout_default_mask_ratio/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_mask_ratio/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_mask_ratio/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_mask_ratio/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_old/model.onnx create mode 100644 onnx/backend/test/data/node/test_dropout_default_old/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_old/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_ratio/model.onnx create mode 100644 onnx/backend/test/data/node/test_dropout_default_ratio/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_ratio/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_dropout_default_ratio/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_random_old/model.onnx create mode 100644 onnx/backend/test/data/node/test_dropout_random_old/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dropout_random_old/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear/model.onnx create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_max_adjusted/model.onnx create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_max_adjusted/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_max_adjusted/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_max_adjusted/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_max_adjusted/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_max_adjusted_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_max_adjusted_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_max_adjusted_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_max_adjusted_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_max_adjusted_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_min_adjusted/model.onnx create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_min_adjusted/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_min_adjusted/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_min_adjusted/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_min_adjusted/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_min_adjusted_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_min_adjusted_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_min_adjusted_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_min_adjusted_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_dynamicquantizelinear_min_adjusted_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_edge_pad/model.onnx create mode 100644 onnx/backend/test/data/node/test_edge_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_edge_pad/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_edge_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_batch_diagonal/model.onnx create mode 100644 onnx/backend/test/data/node/test_einsum_batch_diagonal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_batch_diagonal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_batch_matmul/model.onnx create mode 100644 onnx/backend/test/data/node/test_einsum_batch_matmul/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_batch_matmul/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_einsum_batch_matmul/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_inner_prod/model.onnx create mode 100644 onnx/backend/test/data/node/test_einsum_inner_prod/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_inner_prod/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_einsum_inner_prod/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_scalar/model.onnx create mode 100644 onnx/backend/test/data/node/test_einsum_scalar/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_scalar/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_sum/model.onnx create mode 100644 onnx/backend/test/data/node/test_einsum_sum/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_sum/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_transpose/model.onnx create mode 100644 onnx/backend/test/data/node/test_einsum_transpose/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_einsum_transpose/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_elu/model.onnx create mode 100644 onnx/backend/test/data/node/test_elu/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_elu/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_elu_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_elu_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_elu_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_elu_default_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_elu_default_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_elu_default_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_elu_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_elu_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_elu_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_elu_example_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_elu_example_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_elu_example_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_elu_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_elu_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_elu_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_equal/model.onnx create mode 100644 onnx/backend/test/data/node/test_equal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_equal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_equal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_bcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_equal_bcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_bcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_equal_bcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_equal_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_equal_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_equal_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_equal_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_string/model.onnx create mode 100644 onnx/backend/test/data/node/test_equal_string/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_string/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_equal_string/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_string_broadcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_equal_string_broadcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_string_broadcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_equal_string_broadcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_equal_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_equal_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_equal_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_equal_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_equal_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_erf/model.onnx create mode 100644 onnx/backend/test/data/node/test_erf/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_erf/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_exp/model.onnx create mode 100644 onnx/backend/test/data/node/test_exp/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_exp/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_exp_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_exp_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_exp_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_expand_dim_changed/model.onnx create mode 100644 onnx/backend/test/data/node/test_expand_dim_changed/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_expand_dim_changed/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_expand_dim_changed/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_expand_dim_unchanged/model.onnx create mode 100644 onnx/backend/test/data/node/test_expand_dim_unchanged/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_expand_dim_unchanged/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_expand_dim_unchanged/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_eyelike_populate_off_main_diagonal/model.onnx create mode 100644 onnx/backend/test/data/node/test_eyelike_populate_off_main_diagonal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_eyelike_populate_off_main_diagonal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_eyelike_with_dtype/model.onnx create mode 100644 onnx/backend/test/data/node/test_eyelike_with_dtype/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_eyelike_with_dtype/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_eyelike_without_dtype/model.onnx create mode 100644 onnx/backend/test/data/node/test_eyelike_without_dtype/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_eyelike_without_dtype/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_axis0/model.onnx create mode 100644 onnx/backend/test/data/node/test_flatten_axis0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_axis0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_axis1/model.onnx create mode 100644 onnx/backend/test/data/node/test_flatten_axis1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_axis1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_axis2/model.onnx create mode 100644 onnx/backend/test/data/node/test_flatten_axis2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_axis2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_axis3/model.onnx create mode 100644 onnx/backend/test/data/node/test_flatten_axis3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_axis3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_default_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_flatten_default_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_default_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis1/model.onnx create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis2/model.onnx create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis3/model.onnx create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis4/model.onnx create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flatten_negative_axis4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_causal_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_causal_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_causal_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_causal_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_causal_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_causal_mask_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_causal_mask_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_causal_mask_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_causal_mask_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_causal_mask_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_diff_head_sizes/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_diff_head_sizes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_diff_head_sizes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_diff_head_sizes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_diff_head_sizes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_diff_head_sizes_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_diff_head_sizes_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_diff_head_sizes_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_diff_head_sizes_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_diff_head_sizes_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_double/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_double/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_double/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_double/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_double/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_double_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_double_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_double_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_double_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_double_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_fp16/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_fp16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_fp16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_fp16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_fp16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_fp16_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_fp16_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_fp16_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_fp16_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_fp16_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_gqa/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_gqa/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_gqa/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_gqa/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_gqa/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_gqa_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_gqa_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_gqa_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_gqa_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_gqa_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_prob_mod/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_prob_mod/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_prob_mod/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_prob_mod/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_prob_mod/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_prob_mod_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_prob_mod_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_prob_mod_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_prob_mod_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_prob_mod_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_relative_positional/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_relative_positional/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_relative_positional/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_relative_positional/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_relative_positional/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_relative_positional_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_relative_positional_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_relative_positional_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_relative_positional_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_relative_positional_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_scaled/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_scaled/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_scaled/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_scaled/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_scaled/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_scaled_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_scaled_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_scaled_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_scaled_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_scaled_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_score_mod/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_score_mod/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_score_mod/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_score_mod/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_score_mod/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_score_mod_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_score_mod_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_score_mod_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_score_mod_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_score_mod_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_soft_cap/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_soft_cap/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_soft_cap/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_soft_cap/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_soft_cap/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_soft_cap_expanded_ver26/model.onnx create mode 100644 onnx/backend/test/data/node/test_flexattention_soft_cap_expanded_ver26/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_soft_cap_expanded_ver26/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_soft_cap_expanded_ver26/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_flexattention_soft_cap_expanded_ver26/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_floor/model.onnx create mode 100644 onnx/backend/test/data/node/test_floor/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_floor/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_floor_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_floor_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_floor_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_gather_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gather_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gather_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gather_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_2d_indices/model.onnx create mode 100644 onnx/backend/test/data/node/test_gather_2d_indices/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_2d_indices/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gather_2d_indices/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_elements_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_gather_elements_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_elements_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gather_elements_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_elements_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gather_elements_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_elements_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gather_elements_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_elements_negative_indices/model.onnx create mode 100644 onnx/backend/test/data/node/test_gather_elements_negative_indices/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_elements_negative_indices/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gather_elements_negative_indices/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_negative_indices/model.onnx create mode 100644 onnx/backend/test/data/node/test_gather_negative_indices/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gather_negative_indices/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gather_negative_indices/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gathernd_example_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_gathernd_example_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gathernd_example_float32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gathernd_example_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gathernd_example_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_gathernd_example_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gathernd_example_int32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gathernd_example_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gathernd_example_int32_batch_dim1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gathernd_example_int32_batch_dim1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gathernd_example_int32_batch_dim1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gathernd_example_int32_batch_dim1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_default_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gelu_default_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_default_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_default_1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_gelu_default_1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_default_1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_default_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_gelu_default_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_default_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_default_2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_gelu_default_2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_default_2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gelu_tanh_2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_all_attributes/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_all_attributes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_all_attributes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_all_attributes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gemm_all_attributes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_alpha/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_alpha/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_alpha/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_alpha/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gemm_alpha/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_beta/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_beta/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_beta/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_beta/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gemm_beta/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_matrix_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_default_matrix_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_matrix_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_matrix_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_matrix_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_no_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_default_no_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_no_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_no_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_scalar_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_default_scalar_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_scalar_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_scalar_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_scalar_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_single_elem_vector_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_default_single_elem_vector_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_single_elem_vector_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_single_elem_vector_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_single_elem_vector_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_vector_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_default_vector_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_vector_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_vector_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_vector_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_zero_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_default_zero_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_zero_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_zero_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gemm_default_zero_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_transposeA/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_transposeA/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_transposeA/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_transposeA/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gemm_transposeA/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_transposeB/model.onnx create mode 100644 onnx/backend/test/data/node/test_gemm_transposeB/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gemm_transposeB/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gemm_transposeB/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gemm_transposeB/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_globalaveragepool/model.onnx create mode 100644 onnx/backend/test/data/node/test_globalaveragepool/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_globalaveragepool/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_globalaveragepool_precomputed/model.onnx create mode 100644 onnx/backend/test/data/node/test_globalaveragepool_precomputed/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_globalaveragepool_precomputed/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_globalmaxpool/model.onnx create mode 100644 onnx/backend/test/data/node/test_globalmaxpool/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_globalmaxpool/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_globalmaxpool_precomputed/model.onnx create mode 100644 onnx/backend/test/data/node/test_globalmaxpool_precomputed/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_globalmaxpool_precomputed/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_bcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_bcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_bcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_bcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_bcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_bcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_bcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_bcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_bcast_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_bcast_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_bcast_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_bcast_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_int16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int8_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_int8_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int8_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_int8_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint32_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint32_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint32_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint32_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint64_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint64_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint64_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint64_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint8_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint8_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint8_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_equal_uint8_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_greater_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_greater_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_aligncorners_true/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_aligncorners_true/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_aligncorners_true/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_aligncorners_true/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic_align_corners_0_additional_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic_align_corners_0_additional_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic_align_corners_0_additional_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic_align_corners_0_additional_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic_align_corners_1_additional_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic_align_corners_1_additional_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic_align_corners_1_additional_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bicubic_align_corners_1_additional_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear_align_corners_0_additional_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear_align_corners_0_additional_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear_align_corners_0_additional_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear_align_corners_0_additional_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear_align_corners_1_additional_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear_align_corners_1_additional_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear_align_corners_1_additional_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_bilinear_align_corners_1_additional_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_border_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_border_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_border_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_border_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest_align_corners_0_additional_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest_align_corners_0_additional_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest_align_corners_0_additional_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest_align_corners_0_additional_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest_align_corners_1_additional_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest_align_corners_1_additional_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest_align_corners_1_additional_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_nearest_align_corners_1_additional_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_reflection_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_reflection_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_reflection_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_reflection_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_bilinear_align_corners_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_bilinear_align_corners_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_bilinear_align_corners_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_bilinear_align_corners_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_bilinear_align_corners_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_bilinear_align_corners_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_bilinear_align_corners_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_bilinear_align_corners_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_nearest_align_corners_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_nearest_align_corners_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_nearest_align_corners_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_nearest_align_corners_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_nearest_align_corners_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_nearest_align_corners_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_nearest_align_corners_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_volumetric_nearest_align_corners_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_zeros_padding/model.onnx create mode 100644 onnx/backend/test/data/node/test_gridsample_zeros_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_zeros_padding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gridsample_zeros_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_group_normalization_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_epsilon/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_group_normalization_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_epsilon_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_group_normalization_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_example/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_group_normalization_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_example_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_group_normalization_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gru_batchwise/model.onnx create mode 100644 onnx/backend/test/data/node/test_gru_batchwise/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gru_batchwise/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gru_batchwise/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gru_batchwise/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gru_batchwise/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_gru_defaults/model.onnx create mode 100644 onnx/backend/test/data/node/test_gru_defaults/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gru_defaults/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gru_defaults/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gru_defaults/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gru_seq_length/model.onnx create mode 100644 onnx/backend/test/data/node/test_gru_seq_length/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gru_seq_length/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gru_seq_length/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gru_seq_length/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_gru_seq_length/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_gru_with_initial_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_gru_with_initial_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_gru_with_initial_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_gru_with_initial_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_gru_with_initial_bias/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_gru_with_initial_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hammingwindow/model.onnx create mode 100644 onnx/backend/test/data/node/test_hammingwindow/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hammingwindow/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hammingwindow_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_hammingwindow_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hammingwindow_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hammingwindow_symmetric/model.onnx create mode 100644 onnx/backend/test/data/node/test_hammingwindow_symmetric/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hammingwindow_symmetric/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hammingwindow_symmetric_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_hammingwindow_symmetric_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hammingwindow_symmetric_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hannwindow/model.onnx create mode 100644 onnx/backend/test/data/node/test_hannwindow/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hannwindow/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hannwindow_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_hannwindow_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hannwindow_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hannwindow_symmetric/model.onnx create mode 100644 onnx/backend/test/data/node/test_hannwindow_symmetric/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hannwindow_symmetric/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hannwindow_symmetric_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_hannwindow_symmetric_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hannwindow_symmetric_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardmax_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_axis_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardmax_axis_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_axis_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_axis_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardmax_axis_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_axis_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_default_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardmax_default_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_default_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardmax_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_negative_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardmax_negative_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_negative_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_one_hot/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardmax_one_hot/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardmax_one_hot/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardsigmoid/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_default_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_default_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_default_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_example_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_example_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_example_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardsigmoid_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardswish/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardswish/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardswish/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_hardswish_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_hardswish_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_hardswish_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_identity/model.onnx create mode 100644 onnx/backend/test/data/node/test_identity/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_identity/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_identity_opt/model.onnx create mode 100644 onnx/backend/test/data/node/test_identity_opt/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_identity_opt/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_identity_sequence/model.onnx create mode 100644 onnx/backend/test/data/node/test_identity_sequence/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_identity_sequence/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_if/model.onnx create mode 100644 onnx/backend/test/data/node/test_if/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_if/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_if_opt/model.onnx create mode 100644 onnx/backend/test/data/node/test_if_opt/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_if_opt/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_if_seq/model.onnx create mode 100644 onnx/backend/test/data/node/test_if_seq/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_if_seq/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_bmp_rgb/model.onnx create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_bmp_rgb/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_bmp_rgb/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg2k_rgb/model.onnx create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg2k_rgb/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg2k_rgb/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg_bgr/model.onnx create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg_bgr/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg_bgr/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg_grayscale/model.onnx create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg_grayscale/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg_grayscale/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg_rgb/model.onnx create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg_rgb/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_jpeg_rgb/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_png_rgb/model.onnx create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_png_rgb/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_png_rgb/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_pnm_rgb/model.onnx create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_pnm_rgb/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_pnm_rgb/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_tiff_rgb/model.onnx create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_tiff_rgb/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_tiff_rgb/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_webp_rgb/model.onnx create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_webp_rgb/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_image_decoder_decode_webp_rgb/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_instancenorm_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_instancenorm_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_instancenorm_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_instancenorm_epsilon/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_instancenorm_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_instancenorm_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_instancenorm_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_instancenorm_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_instancenorm_example/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_instancenorm_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_isinf/model.onnx create mode 100644 onnx/backend/test/data/node/test_isinf/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_isinf/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_isinf_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_isinf_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_isinf_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_isinf_negative/model.onnx create mode 100644 onnx/backend/test/data/node/test_isinf_negative/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_isinf_negative/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_isinf_positive/model.onnx create mode 100644 onnx/backend/test/data/node/test_isinf_positive/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_isinf_positive/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_isnan/model.onnx create mode 100644 onnx/backend/test/data/node/test_isnan/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_isnan/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_isnan_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_isnan_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_isnan_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_l1normalization_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_l1normalization_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_l1normalization_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_l1normalization_axis_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_l1normalization_axis_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_l1normalization_axis_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_l1normalization_axis_last/model.onnx create mode 100644 onnx/backend/test/data/node/test_l1normalization_axis_last/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_l1normalization_axis_last/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_l2normalization_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_l2normalization_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_l2normalization_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_l2normalization_axis_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_l2normalization_axis_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_l2normalization_axis_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis0_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis1_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_1_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_2d_axis_negative_2_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis0_epsilon_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis1_epsilon_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis2_epsilon_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_1_epsilon_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_2_epsilon_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_3d_axis_negative_3_epsilon_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis0_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis1_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis2_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis3_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_1_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_2_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_3_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_4d_axis_negative_4_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded_ver18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded_ver18/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded_ver18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_layer_normalization_default_axis_expanded_ver18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu/model.onnx create mode 100644 onnx/backend/test/data/node/test_leakyrelu/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_leakyrelu_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu_default_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_leakyrelu_default_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu_default_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_leakyrelu_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_leakyrelu_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_leakyrelu_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_leakyrelu_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less/model.onnx create mode 100644 onnx/backend/test/data/node/test_less/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_bcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_bcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_bcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_bcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_bcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_bcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_bcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_bcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_bcast_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_bcast_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_bcast_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_bcast_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_int16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int8_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_int8_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int8_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_int8_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_uint16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint32_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_uint32_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint32_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint32_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint64_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_uint64_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint64_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint64_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint8_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_equal_uint8_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint8_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_equal_uint8_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_less_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_less_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_less_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_less_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_decode_step_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_delta_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_explicit_scale_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_fp16_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_beta_scalar_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_gqa_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_delta_mqa_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_gated_per_head_decay_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_linear_t1_no_past_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_no_past_explicit_zeros_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past_expanded/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past_expanded/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_linear_attention_prefill_with_past_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_log/model.onnx create mode 100644 onnx/backend/test/data/node/test_log/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_log/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_log_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_log_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_log_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_0_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_0_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_0_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_0_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_0_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_0_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_1_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_1_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_1_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_2_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_2_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_axis_2_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_default_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_default_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_default_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_default_axis_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_default_axis_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_default_axis_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_default_axis_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_default_axis_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_default_axis_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_example_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_example_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_example_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_example_1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_example_1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_example_1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_example_1_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_example_1_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_example_1_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_large_number/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_large_number/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_large_number/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_large_number_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_large_number_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_large_number_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_large_number_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_large_number_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_large_number_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_negative_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_negative_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_negative_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_negative_axis_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_negative_axis_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_negative_axis_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_negative_axis_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_logsoftmax_negative_axis_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_logsoftmax_negative_axis_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_loop11/model.onnx create mode 100644 onnx/backend/test/data/node/test_loop11/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_loop11/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_loop11/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_loop11/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_loop11/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_loop13_seq/model.onnx create mode 100644 onnx/backend/test/data/node/test_loop13_seq/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_loop13_seq/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_loop13_seq/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_loop13_seq/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_loop16_seq_none/model.onnx create mode 100644 onnx/backend/test/data/node/test_loop16_seq_none/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_loop16_seq_none/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_loop16_seq_none/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_loop16_seq_none/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lpnormalization_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_lpnormalization_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lpnormalization_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_1d_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_lppool_1d_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_1d_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_lppool_2d_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_dilations/model.onnx create mode 100644 onnx/backend/test/data/node/test_lppool_2d_dilations/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_dilations/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_pads/model.onnx create mode 100644 onnx/backend/test/data/node/test_lppool_2d_pads/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_pads/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_same_lower/model.onnx create mode 100644 onnx/backend/test/data/node/test_lppool_2d_same_lower/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_same_lower/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_same_upper/model.onnx create mode 100644 onnx/backend/test/data/node/test_lppool_2d_same_upper/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_same_upper/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_strides/model.onnx create mode 100644 onnx/backend/test/data/node/test_lppool_2d_strides/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_2d_strides/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_3d_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_lppool_3d_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lppool_3d_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lrn/model.onnx create mode 100644 onnx/backend/test/data/node/test_lrn/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lrn/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lrn_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_lrn_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lrn_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lstm_batchwise/model.onnx create mode 100644 onnx/backend/test/data/node/test_lstm_batchwise/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lstm_batchwise/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_lstm_batchwise/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_lstm_batchwise/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lstm_batchwise/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_lstm_defaults/model.onnx create mode 100644 onnx/backend/test/data/node/test_lstm_defaults/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lstm_defaults/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_lstm_defaults/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_lstm_defaults/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_initial_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_lstm_with_initial_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_initial_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_initial_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_initial_bias/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_initial_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_peepholes/model.onnx create mode 100644 onnx/backend/test/data/node/test_lstm_with_peepholes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_peepholes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_peepholes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_peepholes/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_peepholes/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_peepholes/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_peepholes/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_peepholes/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_lstm_with_peepholes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_1d_1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_matmul_1d_1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_1d_1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_matmul_1d_1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_1d_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_matmul_1d_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_1d_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_matmul_1d_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_matmul_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_matmul_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_matmul_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_matmul_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_matmul_4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_matmul_4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_4d_1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_matmul_4d_1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_4d_1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_matmul_4d_1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_bcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_matmul_bcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_matmul_bcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_matmul_bcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_matmulinteger/model.onnx create mode 100644 onnx/backend/test/data/node/test_matmulinteger/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_matmulinteger/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_matmulinteger/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_matmulinteger/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_matmulinteger/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_example/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_max_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_float16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_float32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_float64/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_float64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_float64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_float64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_int32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_int64/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_int64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_int64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_int64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_one_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_one_input/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_one_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_two_inputs/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_two_inputs/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_two_inputs/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_two_inputs/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_max_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_max_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_max_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_max_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_1d_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_1d_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_1d_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_ceil/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_ceil/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_ceil/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_ceil_output_size_reduce_by_one/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_ceil_output_size_reduce_by_one/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_ceil_output_size_reduce_by_one/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_dilations/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_dilations/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_dilations/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_pads/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_pads/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_pads/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_precomputed_pads/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_precomputed_pads/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_precomputed_pads/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_precomputed_same_upper/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_precomputed_same_upper/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_precomputed_same_upper/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_precomputed_strides/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_precomputed_strides/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_precomputed_strides/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_same_lower/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_same_lower/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_same_lower/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_same_upper/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_same_upper/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_same_upper/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_strides/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_strides/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_strides/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_2d_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_dilations/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_dilations/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_dilations/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_dilations_use_ref_impl/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_dilations_use_ref_impl/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_dilations_use_ref_impl/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_dilations_use_ref_impl_large/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_dilations_use_ref_impl_large/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_3d_dilations_use_ref_impl_large/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_with_argmax_2d_precomputed_pads/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_with_argmax_2d_precomputed_pads/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_with_argmax_2d_precomputed_pads/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_with_argmax_2d_precomputed_pads/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_with_argmax_2d_precomputed_strides/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxpool_with_argmax_2d_precomputed_strides/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_with_argmax_2d_precomputed_strides/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxpool_with_argmax_2d_precomputed_strides/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_maxunpool_export_with_output_shape/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxunpool_export_with_output_shape/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxunpool_export_with_output_shape/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_maxunpool_export_with_output_shape/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_maxunpool_export_with_output_shape/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_maxunpool_export_without_output_shape/model.onnx create mode 100644 onnx/backend/test/data/node/test_maxunpool_export_without_output_shape/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_maxunpool_export_without_output_shape/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_maxunpool_export_without_output_shape/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mean_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_mean_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mean_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mean_example/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_mean_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mean_one_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_mean_one_input/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mean_one_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mean_two_inputs/model.onnx create mode 100644 onnx/backend/test/data/node/test_mean_two_inputs/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mean_two_inputs/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mean_two_inputs/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_melweightmatrix/model.onnx create mode 100644 onnx/backend/test/data/node/test_melweightmatrix/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_melweightmatrix/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_melweightmatrix/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_melweightmatrix/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_melweightmatrix/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_melweightmatrix/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_example/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_min_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_float16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_float32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_float64/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_float64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_float64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_float64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_int32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_int64/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_int64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_int64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_int64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_one_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_one_input/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_one_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_two_inputs/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_two_inputs/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_two_inputs/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_two_inputs/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_min_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_min_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_min_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_min_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mish/model.onnx create mode 100644 onnx/backend/test/data/node/test_mish/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mish/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mish_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_mish_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mish_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_broadcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_broadcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_broadcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_broadcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_int64_fmod/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_int64_fmod/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_int64_fmod/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_int64_fmod/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float64/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_float64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int64/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_mixed_sign_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_mod_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mod_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_momentum/model.onnx create mode 100644 onnx/backend/test/data/node/test_momentum/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_momentum/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_momentum/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_momentum/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_momentum/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_momentum/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_momentum/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/model.onnx create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_momentum_multiple/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_mul/model.onnx create mode 100644 onnx/backend/test/data/node/test_mul/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mul/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mul/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_bcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_mul_bcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_bcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mul_bcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_mul_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mul_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_mul_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mul_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_mul_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mul_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_mul_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_mul_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_mul_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_mul_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_mul_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mvn/model.onnx create mode 100644 onnx/backend/test/data/node/test_mvn/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mvn/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mvn_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_mvn_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mvn_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_mvn_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_mvn_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_mvn_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_neg/model.onnx create mode 100644 onnx/backend/test/data/node/test_neg/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_neg/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_neg_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_neg_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_neg_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nesterov_momentum/model.onnx create mode 100644 onnx/backend/test/data/node/test_nesterov_momentum/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nesterov_momentum/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nesterov_momentum/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nesterov_momentum/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nesterov_momentum/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nesterov_momentum/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nesterov_momentum/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NC/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NC/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NC/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NC/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NC_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NC_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NC_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NC_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_mean_weight_negative_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_mean_weight_negative_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_mean_weight_negative_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_mean_weight_negative_ii/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_mean_weight_negative_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_mean_weight_negative_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_mean_weight_negative_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_mean_weight_negative_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_mean_weight_negative_ii_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_mean_weight_negative_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_ii/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_ii_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1_weight_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_no_weight_reduction_mean_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_no_weight_reduction_mean_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_no_weight_reduction_mean_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_no_weight_reduction_mean_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_no_weight_reduction_mean_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_no_weight_reduction_mean_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_no_weight_reduction_mean_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_no_weight_reduction_mean_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_mean/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_mean/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_mean/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_mean/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_mean_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_mean_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_mean_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_mean_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_sum/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_sum/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_sum/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_sum/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_sum_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_sum_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_sum_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_reduction_sum_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_mean/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_mean/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_mean/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_mean/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_mean/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_mean_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_mean_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_mean_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_mean_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_mean_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_ii/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_ii_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2_with_weight_reduction_sum_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_none_no_weight_negative_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_none_no_weight_negative_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_none_no_weight_negative_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_none_no_weight_negative_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_none_no_weight_negative_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_none_no_weight_negative_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_none_no_weight_negative_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_none_no_weight_negative_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_sum_weight_high_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_sum_weight_high_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_sum_weight_high_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_sum_weight_high_ii/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_sum_weight_high_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_sum_weight_high_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_sum_weight_high_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_sum_weight_high_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_sum_weight_high_ii_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3_sum_weight_high_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_mean_weight/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_mean_weight/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_mean_weight/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_mean_weight/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_mean_weight/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_mean_weight_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_mean_weight_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_mean_weight_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_mean_weight_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_mean_weight_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_none_no_weight/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_none_no_weight/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_none_no_weight/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_none_no_weight/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_none_no_weight_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_none_no_weight_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_none_no_weight_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nllloss_NCd1d2d3d4d5_none_no_weight_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_center_point_box_format/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_center_point_box_format/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_center_point_box_format/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_center_point_box_format/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_center_point_box_format/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_center_point_box_format/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_center_point_box_format/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_flipped_coordinates/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_flipped_coordinates/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_flipped_coordinates/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_flipped_coordinates/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_flipped_coordinates/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_flipped_coordinates/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_flipped_coordinates/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_identical_boxes/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_identical_boxes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_identical_boxes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_identical_boxes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_identical_boxes/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_identical_boxes/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_identical_boxes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_iou_threshold_boundary/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_iou_threshold_boundary/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_iou_threshold_boundary/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_iou_threshold_boundary/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_iou_threshold_boundary/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_iou_threshold_boundary/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_iou_threshold_boundary/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_limit_output_size/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_limit_output_size/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_limit_output_size/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_limit_output_size/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_limit_output_size/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_limit_output_size/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_limit_output_size/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_single_box/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_single_box/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_single_box/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_single_box/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_single_box/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_single_box/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_single_box/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU_and_scores/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU_and_scores/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU_and_scores/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU_and_scores/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU_and_scores/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU_and_scores/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_suppress_by_IOU_and_scores/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_batches/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_batches/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_batches/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_batches/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_batches/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_batches/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_batches/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_classes/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_classes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_classes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_classes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_classes/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_classes/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_nonmaxsuppression_two_classes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_nonzero_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_nonzero_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_nonzero_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_not_2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_not_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_not_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_not_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_not_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_not_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_not_4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_not_4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_not_4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_onehot_negative_indices/model.onnx create mode 100644 onnx/backend/test/data/node/test_onehot_negative_indices/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_onehot_negative_indices/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_onehot_negative_indices/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_onehot_negative_indices/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_onehot_out_of_range_indices/model.onnx create mode 100644 onnx/backend/test/data/node/test_onehot_out_of_range_indices/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_onehot_out_of_range_indices/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_onehot_out_of_range_indices/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_onehot_out_of_range_indices/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_onehot_with_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_onehot_with_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_onehot_with_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_onehot_with_axis/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_onehot_with_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_onehot_with_negative_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_onehot_with_negative_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_onehot_with_negative_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_onehot_with_negative_axis/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_onehot_with_negative_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_onehot_without_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_onehot_without_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_onehot_without_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_onehot_without_axis/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_onehot_without_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_get_element_optional_sequence/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_get_element_optional_sequence/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_get_element_optional_sequence/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_get_element_optional_tensor/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_get_element_optional_tensor/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_get_element_optional_tensor/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_get_element_sequence/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_get_element_sequence/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_get_element_sequence/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_get_element_tensor/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_get_element_tensor/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_get_element_tensor/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_no_input_name_optional_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_no_input_name_optional_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_no_input_name_tensor_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_no_input_name_tensor_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_no_input_optional_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_no_input_optional_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_no_input_tensor_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_no_input_tensor_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_optional_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_optional_input/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_has_element_empty_optional_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_has_element_optional_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_has_element_optional_input/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_has_element_optional_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_has_element_tensor_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_optional_has_element_tensor_input/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_optional_has_element_tensor_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_or2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_or2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_or2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_or2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_or3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_or3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_or3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_or3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_or4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_or4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_or4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_or4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast3v1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_or_bcast3v1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast3v1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast3v1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast3v2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_or_bcast3v2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast3v2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast3v2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast4v2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_or_bcast4v2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast4v2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast4v2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast4v3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_or_bcast4v3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast4v3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast4v3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast4v4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_or_bcast4v4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast4v4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_or_bcast4v4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_bcast_array/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_bcast_array/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_bcast_array/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_bcast_array/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_bcast_scalar/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_bcast_scalar/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_bcast_scalar/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_bcast_scalar/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_int32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_int64/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_int64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_int64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_int64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_float32_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int32_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_types_int32_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int32_float32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int32_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int32_int32/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_types_int32_int32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int32_int32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int32_int32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int64_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_types_int64_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int64_float32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int64_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int64_int64/model.onnx create mode 100644 onnx/backend/test/data/node/test_pow_types_int64_int64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int64_int64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_pow_types_int64_int64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_prelu_broadcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_prelu_broadcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_prelu_broadcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_prelu_broadcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_prelu_broadcast_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_prelu_broadcast_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_prelu_broadcast_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_prelu_broadcast_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_prelu_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_prelu_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_prelu_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_prelu_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_prelu_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_prelu_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_prelu_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_prelu_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearconv/model.onnx create mode 100644 onnx/backend/test/data/node/test_qlinearconv/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearconv/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_qlinearconv/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_qlinearconv/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_qlinearconv/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_qlinearconv/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_qlinearconv/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_qlinearconv/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_qlinearconv/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float16/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float16/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float16/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float16/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float16/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float32/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float32/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float32/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float32/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float32/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float32/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_int8_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float16/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float16/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float16/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float16/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float16/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float32/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float32/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float32/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float32/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float32/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float32/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_2D_uint8_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float16/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float16/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float16/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float16/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float16/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float32/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float32/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float32/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float32/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float32/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float32/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_int8_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float16/model.onnx create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float16/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float16/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float16/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float16/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float16/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float32/model.onnx create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float32/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float32/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float32/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float32/test_data_set_0/input_5.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float32/test_data_set_0/input_6.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float32/test_data_set_0/input_7.pb create mode 100644 onnx/backend/test/data/node/test_qlinearmatmul_3D_uint8_float32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_axis/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_blocked_asymmetric/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_blocked_asymmetric/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_blocked_asymmetric/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_blocked_asymmetric/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_blocked_asymmetric/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_blocked_symmetric/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_blocked_symmetric/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_blocked_symmetric/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_blocked_symmetric/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_e4m3fn/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_e4m3fn/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_e4m3fn/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_e4m3fn/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_e4m3fn/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_e5m2/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_e5m2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_e5m2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_e5m2/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_e5m2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_float4e2m1/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_float4e2m1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_float4e2m1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_float4e2m1/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_float4e2m1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int2/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int2/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int4/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int4/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_int4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint16/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint2/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint2/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint4/model.onnx create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint4/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_quantizelinear_uint4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_range_bfloat16_type_positive_delta/model.onnx create mode 100644 onnx/backend/test/data/node/test_range_bfloat16_type_positive_delta/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_range_bfloat16_type_positive_delta/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_range_bfloat16_type_positive_delta/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_range_bfloat16_type_positive_delta/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_range_bfloat16_type_positive_delta_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_range_bfloat16_type_positive_delta_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_range_bfloat16_type_positive_delta_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_range_bfloat16_type_positive_delta_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_range_bfloat16_type_positive_delta_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_range_float16_type_positive_delta/model.onnx create mode 100644 onnx/backend/test/data/node/test_range_float16_type_positive_delta/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_range_float16_type_positive_delta/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_range_float16_type_positive_delta/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_range_float16_type_positive_delta/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_range_float16_type_positive_delta_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_range_float16_type_positive_delta_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_range_float16_type_positive_delta_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_range_float16_type_positive_delta_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_range_float16_type_positive_delta_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_range_float_type_positive_delta/model.onnx create mode 100644 onnx/backend/test/data/node/test_range_float_type_positive_delta/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_range_float_type_positive_delta/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_range_float_type_positive_delta/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_range_float_type_positive_delta/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_range_float_type_positive_delta_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_range_float_type_positive_delta_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_range_float_type_positive_delta_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_range_float_type_positive_delta_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_range_float_type_positive_delta_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_range_int32_type_negative_delta/model.onnx create mode 100644 onnx/backend/test/data/node/test_range_int32_type_negative_delta/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_range_int32_type_negative_delta/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_range_int32_type_negative_delta/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_range_int32_type_negative_delta/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_range_int32_type_negative_delta_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_range_int32_type_negative_delta_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_range_int32_type_negative_delta_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_range_int32_type_negative_delta_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_range_int32_type_negative_delta_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reciprocal/model.onnx create mode 100644 onnx/backend/test/data/node/test_reciprocal/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reciprocal/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reciprocal_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reciprocal_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reciprocal_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_empty_set/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_empty_set/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_empty_set/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_empty_set/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_empty_set_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_empty_set_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_empty_set_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_empty_set_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_keep_dims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_default_axes_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_do_not_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_empty_set/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_empty_set/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_empty_set/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_empty_set/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_empty_set_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_empty_set_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_empty_set_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_empty_set_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_keep_dims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_l2_negative_axes_keep_dims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_asc_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_asc_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_asc_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_asc_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_asc_axes_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_asc_axes_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_asc_axes_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_asc_axes_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_default/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_default_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_default_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_default_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_default_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_desc_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_desc_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_desc_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_desc_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_desc_axes_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_desc_axes_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_desc_axes_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_desc_axes_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_empty_set/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_empty_set/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_empty_set/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_empty_set/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_empty_set_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_empty_set_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_empty_set_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_empty_set_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_default_axes_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_do_not_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_empty_set/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_empty_set/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_empty_set/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_empty_set/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_empty_set_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_empty_set_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_empty_set_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_empty_set_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_exp_negative_axes_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_negative_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_negative_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_negative_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_negative_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_negative_axes_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_negative_axes_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_negative_axes_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_log_sum_negative_axes_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_bool_inputs/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_max_bool_inputs/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_bool_inputs/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_bool_inputs/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_default_axes_keepdim_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_max_default_axes_keepdim_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_default_axes_keepdim_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_default_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_max_default_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_default_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_do_not_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_max_do_not_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_do_not_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_do_not_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_do_not_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_max_do_not_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_do_not_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_do_not_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_empty_set/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_max_empty_set/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_empty_set/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_empty_set/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_max_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_max_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_negative_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_max_negative_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_negative_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_negative_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_negative_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_max_negative_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_negative_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_max_negative_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_default_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_mean_default_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_default_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_default_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_default_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_mean_default_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_default_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_default_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_do_not_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_mean_do_not_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_do_not_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_do_not_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_do_not_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_mean_do_not_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_do_not_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_do_not_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_mean_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_mean_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_negative_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_mean_negative_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_negative_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_negative_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_negative_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_mean_negative_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_negative_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_mean_negative_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_bool_inputs/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_min_bool_inputs/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_bool_inputs/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_bool_inputs/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_default_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_min_default_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_default_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_default_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_min_default_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_default_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_do_not_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_min_do_not_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_do_not_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_do_not_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_do_not_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_min_do_not_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_do_not_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_do_not_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_empty_set/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_min_empty_set/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_empty_set/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_empty_set/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_min_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_min_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_negative_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_min_negative_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_negative_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_negative_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_negative_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_min_negative_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_negative_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_min_negative_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_default_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_prod_default_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_default_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_default_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_prod_default_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_default_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_do_not_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_prod_do_not_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_do_not_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_do_not_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_do_not_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_prod_do_not_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_do_not_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_do_not_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_empty_set/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_prod_empty_set/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_empty_set/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_empty_set/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_prod_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_prod_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_negative_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_prod_negative_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_negative_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_negative_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_negative_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_prod_negative_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_negative_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_prod_negative_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_default_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_default_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_default_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_default_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_default_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_default_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_default_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_default_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_do_not_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_do_not_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_do_not_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_do_not_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_do_not_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_do_not_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_do_not_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_do_not_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_axes_input_noop/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_axes_input_noop/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_axes_input_noop/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_axes_input_noop/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_axes_input_noop_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_axes_input_noop_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_axes_input_noop_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_axes_input_noop_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_set/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_set/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_set/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_set/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_set_non_reduced_axis_zero/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_set_non_reduced_axis_zero/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_set_non_reduced_axis_zero/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_empty_set_non_reduced_axis_zero/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_negative_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_negative_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_negative_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_negative_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_negative_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_negative_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_negative_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_negative_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_default_axes_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_do_not_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_empty_set/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_empty_set/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_empty_set/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_empty_set/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_empty_set_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_empty_set_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_empty_set_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_empty_set_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_example_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_random/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_random/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_random/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_random/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_random_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_random_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_random_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reduce_sum_square_negative_axes_keepdims_random_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reflect_pad/model.onnx create mode 100644 onnx/backend/test/data/node/test_reflect_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reflect_pad/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reflect_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_regex_full_match_basic/model.onnx create mode 100644 onnx/backend/test/data/node/test_regex_full_match_basic/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_regex_full_match_basic/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_regex_full_match_email_domain/model.onnx create mode 100644 onnx/backend/test/data/node/test_regex_full_match_email_domain/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_regex_full_match_email_domain/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_regex_full_match_empty/model.onnx create mode 100644 onnx/backend/test/data/node/test_regex_full_match_empty/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_regex_full_match_empty/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_relu/model.onnx create mode 100644 onnx/backend/test/data/node/test_relu/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_relu/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_relu_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_relu_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_relu_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_allowzero_reordered/model.onnx create mode 100644 onnx/backend/test/data/node/test_reshape_allowzero_reordered/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_allowzero_reordered/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reshape_allowzero_reordered/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_extended_dims/model.onnx create mode 100644 onnx/backend/test/data/node/test_reshape_extended_dims/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_extended_dims/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reshape_extended_dims/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_negative_dim/model.onnx create mode 100644 onnx/backend/test/data/node/test_reshape_negative_dim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_negative_dim/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reshape_negative_dim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_negative_extended_dims/model.onnx create mode 100644 onnx/backend/test/data/node/test_reshape_negative_extended_dims/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_negative_extended_dims/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reshape_negative_extended_dims/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_one_dim/model.onnx create mode 100644 onnx/backend/test/data/node/test_reshape_one_dim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_one_dim/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reshape_one_dim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_reduced_dims/model.onnx create mode 100644 onnx/backend/test/data/node/test_reshape_reduced_dims/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_reduced_dims/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reshape_reduced_dims/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_reordered_all_dims/model.onnx create mode 100644 onnx/backend/test/data/node/test_reshape_reordered_all_dims/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_reordered_all_dims/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reshape_reordered_all_dims/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_reordered_last_dims/model.onnx create mode 100644 onnx/backend/test/data/node/test_reshape_reordered_last_dims/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_reordered_last_dims/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reshape_reordered_last_dims/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_zero_and_negative_dim/model.onnx create mode 100644 onnx/backend/test/data/node/test_reshape_zero_and_negative_dim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_zero_and_negative_dim/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reshape_zero_and_negative_dim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_zero_dim/model.onnx create mode 100644 onnx/backend/test/data/node/test_reshape_zero_dim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reshape_zero_dim/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reshape_zero_dim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_A_n0p5_exclude_outside/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_A_n0p5_exclude_outside/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_A_n0p5_exclude_outside/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_A_n0p5_exclude_outside/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_align_corners/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_align_corners/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_align_corners/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_align_corners/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_antialias/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_antialias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_antialias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_cubic_antialias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_align_corners/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_align_corners/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_align_corners/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_align_corners/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_antialias/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_antialias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_antialias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_antialias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_half_pixel_symmetric/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_half_pixel_symmetric/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_half_pixel_symmetric/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_linear_half_pixel_symmetric/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_nearest/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_nearest/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_nearest/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_scales_nearest/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_cubic/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_cubic/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_cubic/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_cubic/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_cubic_antialias/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_cubic_antialias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_cubic_antialias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_cubic_antialias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_linear_antialias/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_linear_antialias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_linear_antialias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_linear_antialias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_linear_pytorch_half_pixel/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_linear_pytorch_half_pixel/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_linear_pytorch_half_pixel/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_linear_pytorch_half_pixel/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest_not_larger/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest_not_larger/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest_not_larger/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest_not_larger/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest_not_smaller/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest_not_smaller/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest_not_smaller/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_downsample_sizes_nearest_not_smaller/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_axes_2_3/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_axes_2_3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_axes_2_3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_axes_2_3/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_axes_2_3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_axes_3_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_axes_3_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_axes_3_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_axes_3_2/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_axes_3_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_extrapolation_value/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_extrapolation_value/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_extrapolation_value/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_extrapolation_value/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_resize_tf_crop_and_resize_extrapolation_value/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_A_n0p5_exclude_outside/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_A_n0p5_exclude_outside/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_A_n0p5_exclude_outside/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_A_n0p5_exclude_outside/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_align_corners/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_align_corners/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_align_corners/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_align_corners/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_asymmetric/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_asymmetric/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_asymmetric/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_cubic_asymmetric/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear_align_corners/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear_align_corners/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear_align_corners/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear_align_corners/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear_half_pixel_symmetric/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear_half_pixel_symmetric/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear_half_pixel_symmetric/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_linear_half_pixel_symmetric/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest_axes_2_3/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest_axes_2_3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest_axes_2_3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest_axes_2_3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest_axes_3_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest_axes_3_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest_axes_3_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_scales_nearest_axes_3_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_cubic/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_cubic/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_cubic/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_cubic/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_axes_2_3/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_axes_2_3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_axes_2_3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_axes_2_3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_axes_3_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_axes_3_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_axes_3_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_axes_3_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_ceil_half_pixel/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_ceil_half_pixel/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_ceil_half_pixel/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_ceil_half_pixel/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_floor_align_corners/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_floor_align_corners/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_floor_align_corners/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_floor_align_corners/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_not_larger/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_not_larger/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_not_larger/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_not_larger/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_not_smaller/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_not_smaller/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_not_smaller/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_not_smaller/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_round_prefer_ceil_asymmetric/model.onnx create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_round_prefer_ceil_asymmetric/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_round_prefer_ceil_asymmetric/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_resize_upsample_sizes_nearest_round_prefer_ceil_asymmetric/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reversesequence_batch/model.onnx create mode 100644 onnx/backend/test/data/node/test_reversesequence_batch/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reversesequence_batch/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reversesequence_batch/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_reversesequence_time/model.onnx create mode 100644 onnx/backend/test/data/node/test_reversesequence_time/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_reversesequence_time/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_reversesequence_time/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis0/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis0_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis0_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis0_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis0_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis1/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_2d_axis_negative_2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis0_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis0_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis0_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis0_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis0_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis0_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis0_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis0_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis1_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis1_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis1_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis1_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis1_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis1_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis1_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis1_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis2_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis2_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis2_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis2_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis2_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis2_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis2_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis2_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_1_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_1_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_1_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_1_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_1_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_1_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_1_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_1_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_2_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_2_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_2_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_2_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_2_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_2_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_2_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_2_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_3_epsilon/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_3_epsilon/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_3_epsilon/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_3_epsilon/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_3_epsilon_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_3_epsilon_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_3_epsilon_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_3d_axis_negative_3_epsilon_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis0/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis0_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis0_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis0_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis0_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis1/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis2/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis3/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis3_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis3_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis3_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis3_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_1_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_2_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_3/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_3_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_3_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_3_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_3_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_4/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_4_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_4_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_4_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_4d_axis_negative_4_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_default_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_default_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_default_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_default_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_default_axis_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rms_normalization_default_axis_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_default_axis_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rms_normalization_default_axis_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rnn_seq_length/model.onnx create mode 100644 onnx/backend/test/data/node/test_rnn_seq_length/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rnn_seq_length/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rnn_seq_length/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rnn_seq_length/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rnn_seq_length/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_roialign_aligned_false/model.onnx create mode 100644 onnx/backend/test/data/node/test_roialign_aligned_false/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_roialign_aligned_false/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_roialign_aligned_false/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_roialign_aligned_false/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_roialign_aligned_true/model.onnx create mode 100644 onnx/backend/test/data/node/test_roialign_aligned_true/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_roialign_aligned_true/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_roialign_aligned_true/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_roialign_aligned_true/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_roialign_mode_max/model.onnx create mode 100644 onnx/backend/test/data/node/test_roialign_mode_max/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_roialign_mode_max/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_roialign_mode_max/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_roialign_mode_max/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_3d_input_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_interleaved_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_interleaved/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_interleaved/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_interleaved/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_interleaved/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_interleaved/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_interleaved_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_interleaved_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_interleaved_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_interleaved_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_interleaved_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_rotary_dim/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_rotary_dim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_rotary_dim/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_rotary_dim/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_rotary_dim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_rotary_dim_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_rotary_dim_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_rotary_dim_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_rotary_dim_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_no_position_ids_rotary_dim_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_interleaved_rotary_dim_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim_expanded/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_rotary_embedding_with_rotary_dim_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_round/model.onnx create mode 100644 onnx/backend/test/data/node/test_round/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_round/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scan9_multi_state/model.onnx create mode 100644 onnx/backend/test/data/node/test_scan9_multi_state/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scan9_multi_state/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scan9_multi_state/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scan9_multi_state/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scan9_multi_state/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_scan9_multi_state/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_scan9_scalar/model.onnx create mode 100644 onnx/backend/test/data/node/test_scan9_scalar/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scan9_scalar/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scan9_scalar/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scan9_scalar/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_scan9_sum/model.onnx create mode 100644 onnx/backend/test/data/node/test_scan9_sum/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scan9_sum/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scan9_sum/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scan9_sum/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_scan_sum/model.onnx create mode 100644 onnx/backend/test/data/node/test_scan_sum/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scan_sum/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scan_sum/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scan_sum/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_axis/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_duplicate_indices/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_duplicate_indices/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_duplicate_indices/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_duplicate_indices/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_duplicate_indices/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_negative_indices/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_negative_indices/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_negative_indices/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_negative_indices/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_negative_indices/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_max/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_max/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_max/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_max/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_max/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_min/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_min/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_min/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_min/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_min/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_mul/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_mul/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_mul/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_mul/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_with_reduction_mul/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_without_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatter_elements_without_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_without_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_without_axis/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatter_elements_without_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_with_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatter_with_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_with_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatter_with_axis/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatter_with_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_without_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatter_without_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatter_without_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatter_without_axis/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatter_without_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatternd/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatternd/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatternd/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_add/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatternd_add/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_add/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_add/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_add/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_max/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatternd_max/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_max/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_max/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_max/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_max_with_element_indices/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatternd_max_with_element_indices/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_max_with_element_indices/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_max_with_element_indices/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_max_with_element_indices/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_min/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatternd_min/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_min/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_min/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_min/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_min_with_element_indices/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatternd_min_with_element_indices/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_min_with_element_indices/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_min_with_element_indices/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_min_with_element_indices/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_multiply/model.onnx create mode 100644 onnx/backend/test/data/node/test_scatternd_multiply/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_multiply/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_multiply/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_scatternd_multiply/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1_mean_weight_negative_ii_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3_sum_weight_high_ii_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_mean_weight_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_NCd1d2d3d4d5_none_no_weight_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_3d_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_3d_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_4d_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_no_weight_ii_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_3d_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_4d_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_ii_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_mean_weight_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_none/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_none_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_none_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_none_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_none_weights/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob_expanded/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_none_weights_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_sum/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_sum_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_log_prob/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_sum_log_prob/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_log_prob/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_log_prob/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_log_prob/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_log_prob_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sce_sum_log_prob_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_log_prob_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_log_prob_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sce_sum_log_prob_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_selu/model.onnx create mode 100644 onnx/backend/test/data/node/test_selu/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_selu/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_selu_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_selu_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_selu_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_selu_default_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_selu_default_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_selu_default_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_selu_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_selu_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_selu_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_selu_example_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_selu_example_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_selu_example_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_selu_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_selu_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_selu_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_insert_at_back/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_insert_at_back/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_insert_at_back/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_insert_at_back/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_insert_at_front/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_insert_at_front/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_insert_at_front/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_insert_at_front/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sequence_insert_at_front/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_1_sequence_1_tensor/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_1_sequence_1_tensor/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_1_sequence_1_tensor/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_1_sequence_1_tensor/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_1_sequence_1_tensor_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_1_sequence_1_tensor_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_1_sequence_1_tensor_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_1_sequence_1_tensor_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_2_sequences/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_2_sequences/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_2_sequences/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_2_sequences/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_2_sequences_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_2_sequences_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_2_sequences_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_add_2_sequences_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_extract_shapes/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_extract_shapes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_extract_shapes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_extract_shapes_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_extract_shapes_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_extract_shapes_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_1_tensor/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_1_tensor/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_1_tensor/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_1_tensor/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_1_tensor/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_1_tensor_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_1_tensor_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_1_tensor_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_1_tensor_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_1_tensor_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_1_sequence_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_2_sequences/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_2_sequences/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_2_sequences/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_2_sequences/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_2_sequences/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_2_sequences_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_2_sequences_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_2_sequences_expanded/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_2_sequences_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sequence_map_identity_2_sequences_expanded/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_shape/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_clip_end/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape_clip_end/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_clip_end/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_clip_start/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape_clip_start/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_clip_start/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_end_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape_end_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_end_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_end_negative_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape_end_negative_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_end_negative_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_start_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape_start_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_start_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_start_1_end_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape_start_1_end_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_start_1_end_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_start_1_end_negative_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape_start_1_end_negative_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_start_1_end_negative_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_start_greater_than_end/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape_start_greater_than_end/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_start_greater_than_end/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_start_negative_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_shape_start_negative_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shape_start_negative_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shrink_hard/model.onnx create mode 100644 onnx/backend/test/data/node/test_shrink_hard/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shrink_hard/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shrink_hard_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_shrink_hard_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shrink_hard_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shrink_soft/model.onnx create mode 100644 onnx/backend/test/data/node/test_shrink_soft/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shrink_soft/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_shrink_soft_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_shrink_soft_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_shrink_soft_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sigmoid/model.onnx create mode 100644 onnx/backend/test/data/node/test_sigmoid/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sigmoid/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sigmoid_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_sigmoid_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sigmoid_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sign/model.onnx create mode 100644 onnx/backend/test/data/node/test_sign/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sign/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_batchwise/model.onnx create mode 100644 onnx/backend/test/data/node/test_simple_rnn_batchwise/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_batchwise/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_batchwise/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_batchwise/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_batchwise/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_defaults/model.onnx create mode 100644 onnx/backend/test/data/node/test_simple_rnn_defaults/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_defaults/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_defaults/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_defaults/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_with_initial_bias/model.onnx create mode 100644 onnx/backend/test/data/node/test_simple_rnn_with_initial_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_with_initial_bias/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_with_initial_bias/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_with_initial_bias/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_simple_rnn_with_initial_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sin/model.onnx create mode 100644 onnx/backend/test/data/node/test_sin/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sin/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sin_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_sin_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sin_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sinh/model.onnx create mode 100644 onnx/backend/test/data/node/test_sinh/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sinh/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sinh_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_sinh_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sinh_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_size/model.onnx create mode 100644 onnx/backend/test/data/node/test_size/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_size/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_size_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_size_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_size_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_slice/model.onnx create mode 100644 onnx/backend/test/data/node/test_slice/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_slice/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_slice/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_slice/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_slice/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_slice/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_default_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_slice_default_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_default_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_slice_default_axes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_slice_default_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_default_steps/model.onnx create mode 100644 onnx/backend/test/data/node/test_slice_default_steps/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_default_steps/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_slice_default_steps/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_slice_default_steps/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_slice_default_steps/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_end_out_of_bounds/model.onnx create mode 100644 onnx/backend/test/data/node/test_slice_end_out_of_bounds/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_end_out_of_bounds/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_slice_end_out_of_bounds/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_slice_end_out_of_bounds/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_slice_end_out_of_bounds/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_slice_end_out_of_bounds/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg/model.onnx create mode 100644 onnx/backend/test/data/node/test_slice_neg/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg_steps/model.onnx create mode 100644 onnx/backend/test/data/node/test_slice_neg_steps/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg_steps/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg_steps/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg_steps/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg_steps/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_slice_neg_steps/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_negative_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_slice_negative_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_negative_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_slice_negative_axes/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_slice_negative_axes/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_slice_negative_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_start_out_of_bounds/model.onnx create mode 100644 onnx/backend/test/data/node/test_slice_start_out_of_bounds/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_slice_start_out_of_bounds/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_slice_start_out_of_bounds/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_slice_start_out_of_bounds/test_data_set_0/input_3.pb create mode 100644 onnx/backend/test/data/node/test_slice_start_out_of_bounds/test_data_set_0/input_4.pb create mode 100644 onnx/backend/test/data/node/test_slice_start_out_of_bounds/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_0_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_axis_0_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_0_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_0_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_axis_0_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_0_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_axis_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_1_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_axis_1_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_1_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_1_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_axis_1_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_1_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_axis_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_2_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_axis_2_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_2_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_2_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_axis_2_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_axis_2_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_default_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_default_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_default_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_default_axis_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_default_axis_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_default_axis_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_default_axis_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_default_axis_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_default_axis_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_example_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_example_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_example_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_example_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_example_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_example_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_large_number/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_large_number/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_large_number/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_large_number_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_large_number_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_large_number_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_large_number_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_large_number_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_large_number_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_negative_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_negative_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_negative_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_negative_axis_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_negative_axis_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_negative_axis_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_negative_axis_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softmax_negative_axis_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softmax_negative_axis_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softplus/model.onnx create mode 100644 onnx/backend/test/data/node/test_softplus/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softplus/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softplus_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_softplus_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softplus_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softplus_example_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softplus_example_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softplus_example_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softplus_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softplus_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softplus_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softsign/model.onnx create mode 100644 onnx/backend/test/data/node/test_softsign/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softsign/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softsign_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_softsign_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softsign_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softsign_example_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softsign_example_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softsign_example_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_softsign_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_softsign_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_softsign_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_spacetodepth/model.onnx create mode 100644 onnx/backend/test/data/node/test_spacetodepth/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_spacetodepth/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_spacetodepth_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_spacetodepth_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_spacetodepth_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_1d_uneven_split_opset18/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_1d_uneven_split_opset18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_1d_uneven_split_opset18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_1d_uneven_split_opset18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_1d_uneven_split_opset18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_split_1d_uneven_split_opset18/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_split_2d_uneven_split_opset18/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_2d_uneven_split_opset18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_2d_uneven_split_opset18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_2d_uneven_split_opset18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_2d_uneven_split_opset18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_1d_opset13/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_1d_opset13/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_1d_opset13/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_1d_opset13/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_1d_opset13/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_1d_opset18/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_1d_opset18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_1d_opset18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_1d_opset18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_1d_opset18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_2d/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_2d_opset13/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_2d_opset13/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_2d_opset13/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_2d_opset13/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_default_axis_opset13/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_default_axis_opset13/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_default_axis_opset13/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_default_axis_opset13/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_default_axis_opset13/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_default_axis_opset18/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_default_axis_opset18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_default_axis_opset18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_default_axis_opset18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_equal_parts_default_axis_opset18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_nokeepdims/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_nokeepdims/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_to_sequence_nokeepdims/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_1d_opset13/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_1d_opset13/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_1d_opset13/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_1d_opset13/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_1d_opset13/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_1d_opset18/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_1d_opset18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_1d_opset18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_1d_opset18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_1d_opset18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_2d_opset13/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_2d_opset13/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_2d_opset13/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_2d_opset13/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_2d_opset13/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_2d_opset18/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_2d_opset18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_2d_opset18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_2d_opset18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_2d_opset18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_default_axis_opset13/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_default_axis_opset13/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_default_axis_opset13/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_default_axis_opset13/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_default_axis_opset13/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_default_axis_opset18/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_default_axis_opset18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_default_axis_opset18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_default_axis_opset18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_variable_parts_default_axis_opset18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset13/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset13/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset13/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset13/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset13/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset13/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset18/model.onnx create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset18/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset18/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_split_zero_size_splits_opset18/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_sqrt/model.onnx create mode 100644 onnx/backend/test/data/node/test_sqrt/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sqrt/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sqrt_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_sqrt_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sqrt_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_squeeze/model.onnx create mode 100644 onnx/backend/test/data/node/test_squeeze/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_squeeze/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_squeeze/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_squeeze_negative_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_squeeze_negative_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_squeeze_negative_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_squeeze_negative_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_stft/model.onnx create mode 100644 onnx/backend/test/data/node/test_stft/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_stft/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_stft/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_stft/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_stft_with_window/model.onnx create mode 100644 onnx/backend/test/data/node/test_stft_with_window/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_stft_with_window/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_stft_with_window/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_stft_with_window/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_concat/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_concat/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_concat/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_string_concat/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_broadcasting/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_concat_broadcasting/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_broadcasting/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_broadcasting/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_empty_string/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_concat_empty_string/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_empty_string/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_empty_string/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_utf8/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_concat_utf8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_utf8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_utf8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_zero_dimensional/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_concat_zero_dimensional/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_zero_dimensional/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_string_concat_zero_dimensional/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_basic/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_split_basic/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_basic/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_basic/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_string_split_consecutive_delimiters/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_split_consecutive_delimiters/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_consecutive_delimiters/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_consecutive_delimiters/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_string_split_empty_string_delimiter/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_split_empty_string_delimiter/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_empty_string_delimiter/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_empty_string_delimiter/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_string_split_empty_tensor/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_split_empty_tensor/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_empty_tensor/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_empty_tensor/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_string_split_maxsplit/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_split_maxsplit/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_maxsplit/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_maxsplit/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_string_split_no_delimiter/model.onnx create mode 100644 onnx/backend/test/data/node/test_string_split_no_delimiter/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_no_delimiter/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_string_split_no_delimiter/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_casesensintive_lower/model.onnx create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_casesensintive_lower/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_casesensintive_lower/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_casesensintive_nochangecase/model.onnx create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_casesensintive_nochangecase/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_casesensintive_nochangecase/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_casesensintive_upper/model.onnx create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_casesensintive_upper/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_casesensintive_upper/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_empty_output/model.onnx create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_empty_output/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_empty_output/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_insensintive_upper_twodim/model.onnx create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_insensintive_upper_twodim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_export_monday_insensintive_upper_twodim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_nostopwords_nochangecase/model.onnx create mode 100644 onnx/backend/test/data/node/test_strnormalizer_nostopwords_nochangecase/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_strnormalizer_nostopwords_nochangecase/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sub/model.onnx create mode 100644 onnx/backend/test/data/node/test_sub/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sub/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sub/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_bcast/model.onnx create mode 100644 onnx/backend/test/data/node/test_sub_bcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_bcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sub_bcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_sub_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sub_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_int16/model.onnx create mode 100644 onnx/backend/test/data/node/test_sub_int16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_int16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sub_int16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_int8/model.onnx create mode 100644 onnx/backend/test/data/node/test_sub_int8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_int8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sub_int8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint16/model.onnx create mode 100644 onnx/backend/test/data/node/test_sub_uint16/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint16/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint16/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint32/model.onnx create mode 100644 onnx/backend/test/data/node/test_sub_uint32/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint32/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint32/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_sub_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint8/model.onnx create mode 100644 onnx/backend/test/data/node/test_sub_uint8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sub_uint8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sum_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_sum_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sum_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sum_example/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_sum_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sum_one_input/model.onnx create mode 100644 onnx/backend/test/data/node/test_sum_one_input/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sum_one_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_sum_two_inputs/model.onnx create mode 100644 onnx/backend/test/data/node/test_sum_two_inputs/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_sum_two_inputs/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_sum_two_inputs/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_swish/model.onnx create mode 100644 onnx/backend/test/data/node/test_swish/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_swish/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_swish_expanded/model.onnx create mode 100644 onnx/backend/test/data/node/test_swish_expanded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_swish_expanded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tan/model.onnx create mode 100644 onnx/backend/test/data/node/test_tan/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tan/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tan_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_tan_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tan_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tanh/model.onnx create mode 100644 onnx/backend/test/data/node/test_tanh/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tanh/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tanh_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_tanh_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tanh_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter/model.onnx create mode 100644 onnx/backend/test/data/node/test_tensorscatter/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_tensorscatter_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter_3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter_3d/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter_circular/model.onnx create mode 100644 onnx/backend/test/data/node/test_tensorscatter_circular/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter_circular/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter_circular/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_tensorscatter_circular/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_batch_onlybigrams_skip0/model.onnx create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_batch_onlybigrams_skip0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_batch_onlybigrams_skip0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_batch_onlybigrams_skip5/model.onnx create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_batch_onlybigrams_skip5/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_batch_onlybigrams_skip5/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_batch_uniandbigrams_skip5/model.onnx create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_batch_uniandbigrams_skip5/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_batch_uniandbigrams_skip5/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_only_bigrams_skip0/model.onnx create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_only_bigrams_skip0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_only_bigrams_skip0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_onlybigrams_levelempty/model.onnx create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_onlybigrams_levelempty/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_onlybigrams_levelempty/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_onlybigrams_skip5/model.onnx create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_onlybigrams_skip5/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_onlybigrams_skip5/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_uniandbigrams_skip5/model.onnx create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_uniandbigrams_skip5/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tfidfvectorizer_tf_uniandbigrams_skip5/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu/model.onnx create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_default_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_default_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_default_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_example_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_example_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_example_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_expanded_ver18/model.onnx create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_expanded_ver18/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_thresholdedrelu_expanded_ver18/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tile/model.onnx create mode 100644 onnx/backend/test/data/node/test_tile/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tile/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tile/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tile_precomputed/model.onnx create mode 100644 onnx/backend/test/data/node/test_tile_precomputed/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tile_precomputed/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tile_precomputed/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k/model.onnx create mode 100644 onnx/backend/test/data/node/test_top_k/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_negative_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_top_k_negative_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_negative_axis/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_negative_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_negative_axis/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values/model.onnx create mode 100644 onnx/backend/test/data/node/test_top_k_same_values/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values_2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_top_k_same_values_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values_2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values_2d/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values_largest/model.onnx create mode 100644 onnx/backend/test/data/node/test_top_k_same_values_largest/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values_largest/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values_largest/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_same_values_largest/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_smallest/model.onnx create mode 100644 onnx/backend/test/data/node/test_top_k_smallest/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_smallest/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_smallest/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_smallest/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_uint64/model.onnx create mode 100644 onnx/backend/test/data/node/test_top_k_uint64/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_uint64/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_top_k_uint64/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_top_k_uint64/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout/model.onnx create mode 100644 onnx/backend/test/data/node/test_training_dropout/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_training_dropout_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_default/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_default/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_default_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_training_dropout_default_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_default_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_default_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_default_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_default_mask/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_training_dropout_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_mask/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio/model.onnx create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio_mask/model.onnx create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio_mask/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio_mask/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio_mask/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio_mask/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_training_dropout_zero_ratio_mask/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_3/model.onnx create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_4/model.onnx create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_5/model.onnx create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_5/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_all_permutations_5/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_default/model.onnx create mode 100644 onnx/backend/test/data/node/test_transpose_default/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_transpose_default/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tril/model.onnx create mode 100644 onnx/backend/test/data/node/test_tril/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tril/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_neg/model.onnx create mode 100644 onnx/backend/test/data/node/test_tril_neg/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_neg/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tril_neg/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_one_row_neg/model.onnx create mode 100644 onnx/backend/test/data/node/test_tril_one_row_neg/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_one_row_neg/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_out_neg/model.onnx create mode 100644 onnx/backend/test/data/node/test_tril_out_neg/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_out_neg/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tril_out_neg/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_out_pos/model.onnx create mode 100644 onnx/backend/test/data/node/test_tril_out_pos/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_out_pos/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tril_out_pos/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_pos/model.onnx create mode 100644 onnx/backend/test/data/node/test_tril_pos/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_pos/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tril_pos/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_square/model.onnx create mode 100644 onnx/backend/test/data/node/test_tril_square/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_square/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_square_neg/model.onnx create mode 100644 onnx/backend/test/data/node/test_tril_square_neg/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_square_neg/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tril_square_neg/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_zero/model.onnx create mode 100644 onnx/backend/test/data/node/test_tril_zero/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_tril_zero/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_tril_zero/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_triu/model.onnx create mode 100644 onnx/backend/test/data/node/test_triu/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_triu/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_neg/model.onnx create mode 100644 onnx/backend/test/data/node/test_triu_neg/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_neg/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_triu_neg/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_one_row/model.onnx create mode 100644 onnx/backend/test/data/node/test_triu_one_row/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_one_row/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_triu_one_row/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_out_neg_out/model.onnx create mode 100644 onnx/backend/test/data/node/test_triu_out_neg_out/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_out_neg_out/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_triu_out_neg_out/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_out_pos/model.onnx create mode 100644 onnx/backend/test/data/node/test_triu_out_pos/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_out_pos/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_triu_out_pos/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_pos/model.onnx create mode 100644 onnx/backend/test/data/node/test_triu_pos/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_pos/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_triu_pos/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_square/model.onnx create mode 100644 onnx/backend/test/data/node/test_triu_square/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_square/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_square_neg/model.onnx create mode 100644 onnx/backend/test/data/node/test_triu_square_neg/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_square_neg/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_triu_square_neg/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_zero/model.onnx create mode 100644 onnx/backend/test/data/node/test_triu_zero/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_triu_zero/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_triu_zero/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_length_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_unique_length_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_length_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_length_1/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_unique_length_1/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_unique_length_1/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_unique_not_sorted_without_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_unique_not_sorted_without_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_not_sorted_without_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_not_sorted_without_axis/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_unique_not_sorted_without_axis/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_unique_not_sorted_without_axis/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis_3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis_3d/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis_3d/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_axis_3d/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_negative_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_negative_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_negative_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_negative_axis/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_negative_axis/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_with_negative_axis/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_without_axis/model.onnx create mode 100644 onnx/backend/test/data/node/test_unique_sorted_without_axis/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_without_axis/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_without_axis/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_without_axis/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/node/test_unique_sorted_without_axis/test_data_set_0/output_3.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_0/model.onnx create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_0/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_0/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_0/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_1/model.onnx create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_2/model.onnx create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_axis_2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_negative_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_unsqueeze_negative_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_negative_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_negative_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_three_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_unsqueeze_three_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_three_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_three_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_two_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_unsqueeze_two_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_two_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_two_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_unsorted_axes/model.onnx create mode 100644 onnx/backend/test/data/node/test_unsqueeze_unsorted_axes/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_unsorted_axes/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_unsqueeze_unsorted_axes/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_upsample_nearest/model.onnx create mode 100644 onnx/backend/test/data/node/test_upsample_nearest/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_upsample_nearest/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_upsample_nearest/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_where_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_where_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_where_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_where_example/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_where_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_where_long_example/model.onnx create mode 100644 onnx/backend/test/data/node/test_where_long_example/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_where_long_example/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_where_long_example/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/node/test_where_long_example/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_wrap_pad/model.onnx create mode 100644 onnx/backend/test/data/node/test_wrap_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_wrap_pad/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_wrap_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_xor2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_xor2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_xor2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_xor2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_xor3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_xor3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_xor3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_xor3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_xor4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_xor4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_xor4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_xor4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast3v1d/model.onnx create mode 100644 onnx/backend/test/data/node/test_xor_bcast3v1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast3v1d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast3v1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast3v2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_xor_bcast3v2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast3v2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast3v2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v2d/model.onnx create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v2d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v3d/model.onnx create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v3d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v4d/model.onnx create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v4d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v4d/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/node/test_xor_bcast4v4d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool1d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool1d_stride/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool1d_stride/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool1d_stride/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool2d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool2d_stride/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool2d_stride/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool2d_stride/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool3d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool3d_stride/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool3d_stride/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool3d_stride/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool3d_stride1_pad0_gpu_input/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool3d_stride1_pad0_gpu_input/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_AvgPool3d_stride1_pad0_gpu_input/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm1d_3d_input_eval/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm1d_3d_input_eval/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm1d_3d_input_eval/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm2d_eval/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm2d_eval/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm2d_eval/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm2d_momentum_eval/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm2d_momentum_eval/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm2d_momentum_eval/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm3d_eval/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm3d_eval/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm3d_eval/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm3d_momentum_eval/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm3d_momentum_eval/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_BatchNorm3d_momentum_eval/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ConstantPad2d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_ConstantPad2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ConstantPad2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_dilated/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_dilated/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_dilated/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_groups/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_groups/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_groups/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1size1/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1size1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1size1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2size1/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2size1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2size1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_padded/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_padded/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_padded/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_strided/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_strided/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_strided/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_with_multiplier/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_with_multiplier/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_with_multiplier/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_dilated/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_dilated/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_dilated/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_groups/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_groups/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_groups/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_groups_thnn/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_groups_thnn/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_groups_thnn/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_no_bias/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_no_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_no_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_padding/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_strided/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_strided/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv2d_strided/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_dilated/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_dilated/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_dilated/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_dilated_strided/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_dilated_strided/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_dilated_strided/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_groups/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_groups/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_groups/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_no_bias/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_no_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_no_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_stride/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_stride/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_stride/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_stride_padding/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_stride_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Conv3d_stride_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ConvTranspose2d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_ConvTranspose2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ConvTranspose2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ConvTranspose2d_no_bias/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_ConvTranspose2d_no_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ConvTranspose2d_no_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ELU/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_ELU/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ELU/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Embedding/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Embedding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Embedding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Embedding_sparse/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Embedding_sparse/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Embedding_sparse/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_GLU/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_GLU/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_GLU/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_GLU_dim/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_GLU_dim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_GLU_dim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_LeakyReLU/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_LeakyReLU/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_LeakyReLU/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_LeakyReLU_with_negval/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_LeakyReLU_with_negval/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_LeakyReLU_with_negval/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Linear/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Linear/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Linear/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Linear_no_bias/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Linear_no_bias/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Linear_no_bias/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_LogSoftmax/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_LogSoftmax/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_LogSoftmax/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool1d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool1d_stride/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool1d_stride/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool1d_stride/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool1d_stride_padding_dilation/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool1d_stride_padding_dilation/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool1d_stride_padding_dilation/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool2d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool2d_stride_padding_dilation/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool2d_stride_padding_dilation/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool2d_stride_padding_dilation/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool3d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool3d_stride/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool3d_stride/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool3d_stride/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool3d_stride_padding/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool3d_stride_padding/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_MaxPool3d_stride_padding/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_1d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_1d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_1d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_1d_multiparam/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_1d_multiparam/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_1d_multiparam/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_2d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_2d_multiparam/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_2d_multiparam/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_2d_multiparam/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_3d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_3d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_3d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_3d_multiparam/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_3d_multiparam/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PReLU_3d_multiparam/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PixelShuffle/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_PixelShuffle/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PixelShuffle/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PoissonNLLLLoss_no_reduce/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_PoissonNLLLLoss_no_reduce/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_PoissonNLLLLoss_no_reduce/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ReLU/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_ReLU/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ReLU/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ReflectionPad2d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_ReflectionPad2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ReflectionPad2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ReplicationPad2d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_ReplicationPad2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ReplicationPad2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_SELU/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_SELU/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_SELU/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Sigmoid/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Sigmoid/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Sigmoid/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softmax/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softmax/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softmax/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softmin/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softmin/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softmin/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softplus/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softplus/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softplus/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softsign/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softsign/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Softsign/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Tanh/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_Tanh/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_Tanh/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ZeroPad2d/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_ZeroPad2d/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_ZeroPad2d/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_log_softmax_dim3/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_log_softmax_dim3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_log_softmax_dim3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_log_softmax_lastdim/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_log_softmax_lastdim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_log_softmax_lastdim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_softmax_functional_dim3/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_softmax_functional_dim3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_softmax_functional_dim3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_softmax_lastdim/model.onnx create mode 100644 onnx/backend/test/data/pytorch-converted/test_softmax_lastdim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-converted/test_softmax_lastdim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_broadcast/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_broadcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_broadcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_broadcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_broadcast/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_broadcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_broadcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_broadcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_right_broadcast/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_right_broadcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_right_broadcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_right_broadcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_singleton_broadcast/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_singleton_broadcast/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_singleton_broadcast/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_add_size1_singleton_broadcast/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_addconstant/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_addconstant/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_addconstant/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_addmm/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_addmm/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_addmm/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_addmm/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_addmm/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_basic/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_basic/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_basic/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_basic/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_chunk/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_chunk/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_chunk/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_chunk/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_clip/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_clip/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_clip/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_concat2/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_concat2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_concat2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_concat2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_conv/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_conv/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_conv/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_convtranspose/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_convtranspose/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_convtranspose/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_exp/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_exp/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_exp/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_flatten/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_flatten/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_flatten/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_index/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_index/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_index/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_max/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_max/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_max/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_max/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_maxpool/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_maxpool/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_maxpool/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_min/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_min/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_min/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_min/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_mm/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_mm/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_mm/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_mm/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_non_float_params/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_non_float_params/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_non_float_params/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_pad/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_pad/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_pad/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_params/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_params/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_params/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_permute2/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_permute2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_permute2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_pow/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_pow/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_pow/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_pow/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_mean/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_mean/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_mean/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_mean_keepdim/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_mean_keepdim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_mean_keepdim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_sum/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_sum/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_sum/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_sum_keepdim/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_sum_keepdim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_reduced_sum_keepdim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_repeat/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_repeat/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_repeat/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_repeat_dim_overflow/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_repeat_dim_overflow/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_repeat_dim_overflow/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_selu/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_selu/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_selu/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_sqrt/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_sqrt/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_sqrt/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_symbolic_override/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_symbolic_override/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_symbolic_override/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_symbolic_override_nested/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_symbolic_override_nested/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_symbolic_override_nested/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_symbolic_override_nested/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_symbolic_override_nested/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_symbolic_override_nested/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_symbolic_override_nested/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_view/model.onnx create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_view/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/pytorch-operator/test_operator_view/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/real/test_bvlc_alexnet/data.json create mode 100644 onnx/backend/test/data/real/test_densenet121/data.json create mode 100644 onnx/backend/test/data/real/test_inception_v1/data.json create mode 100644 onnx/backend/test/data/real/test_inception_v2/data.json create mode 100644 onnx/backend/test/data/real/test_resnet50/data.json create mode 100644 onnx/backend/test/data/real/test_shufflenet/data.json create mode 100644 onnx/backend/test/data/real/test_squeezenet/data.json create mode 100644 onnx/backend/test/data/real/test_vgg19/data.json create mode 100644 onnx/backend/test/data/real/test_zfnet512/data.json create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model1/model.onnx create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model2/model.onnx create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model3/model.onnx create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model4/model.onnx create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_expand_shape_model4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add/model.onnx create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add_and_mul/model.onnx create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add_and_mul/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add_and_mul/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add_and_mul/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add_and_mul/test_data_set_0/output_1.pb create mode 100644 onnx/backend/test/data/simple/test_gradient_of_add_and_mul/test_data_set_0/output_2.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model1/model.onnx create mode 100644 onnx/backend/test/data/simple/test_sequence_model1/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model1/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model1/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model1/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model2/model.onnx create mode 100644 onnx/backend/test/data/simple/test_sequence_model2/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model2/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model2/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model2/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model3/model.onnx create mode 100644 onnx/backend/test/data/simple/test_sequence_model3/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model3/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model3/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model3/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model4/model.onnx create mode 100644 onnx/backend/test/data/simple/test_sequence_model4/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model4/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model4/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model4/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model5/model.onnx create mode 100644 onnx/backend/test/data/simple/test_sequence_model5/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model5/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model5/test_data_set_0/input_2.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model5/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model6/model.onnx create mode 100644 onnx/backend/test/data/simple/test_sequence_model6/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model6/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model7/model.onnx create mode 100644 onnx/backend/test/data/simple/test_sequence_model7/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model7/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model8/model.onnx create mode 100644 onnx/backend/test/data/simple/test_sequence_model8/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model8/test_data_set_0/input_1.pb create mode 100644 onnx/backend/test/data/simple/test_sequence_model8/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_shrink/model.onnx create mode 100644 onnx/backend/test/data/simple/test_shrink/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_shrink/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_sign_model/model.onnx create mode 100644 onnx/backend/test/data/simple/test_sign_model/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_sign_model/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_single_relu_model/model.onnx create mode 100644 onnx/backend/test/data/simple/test_single_relu_model/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_single_relu_model/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_casesensintive_lower/model.onnx create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_casesensintive_lower/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_casesensintive_lower/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_casesensintive_nochangecase/model.onnx create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_casesensintive_nochangecase/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_casesensintive_nochangecase/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_casesensintive_upper/model.onnx create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_casesensintive_upper/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_casesensintive_upper/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_empty_output/model.onnx create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_empty_output/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_empty_output/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_insensintive_upper_twodim/model.onnx create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_insensintive_upper_twodim/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_monday_insensintive_upper_twodim/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_nostopwords_nochangecase/model.onnx create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_nostopwords_nochangecase/test_data_set_0/input_0.pb create mode 100644 onnx/backend/test/data/simple/test_strnorm_model_nostopwords_nochangecase/test_data_set_0/output_0.pb create mode 100644 onnx/backend/test/loader/__init__.py create mode 100644 onnx/backend/test/report/__init__.py create mode 100644 onnx/backend/test/report/base.py create mode 100644 onnx/backend/test/report/coverage.py create mode 100644 onnx/backend/test/runner/__init__.py create mode 100644 onnx/backend/test/runner/item.py create mode 100644 onnx/backend/test/stat_coverage.py create mode 100644 onnx/bin/__init__.py create mode 100644 onnx/bin/checker.py create mode 100644 onnx/checker.cc create mode 100644 onnx/checker.h create mode 100644 onnx/checker.py create mode 100644 onnx/common/CMakeLists.txt create mode 100644 onnx/common/array_ref.h create mode 100644 onnx/common/assertions.cc create mode 100644 onnx/common/assertions.h create mode 100644 onnx/common/common.h create mode 100644 onnx/common/constants.h create mode 100644 onnx/common/file_utils.h create mode 100644 onnx/common/graph_node_list.h create mode 100644 onnx/common/interned_strings.cc create mode 100644 onnx/common/interned_strings.h create mode 100644 onnx/common/ir.h create mode 100644 onnx/common/ir_pb_converter.cc create mode 100644 onnx/common/ir_pb_converter.h create mode 100644 onnx/common/model_helpers.cc create mode 100644 onnx/common/model_helpers.h create mode 100644 onnx/common/path.h create mode 100644 onnx/common/platform_helpers.h create mode 100644 onnx/common/proto_util.h create mode 100644 onnx/common/safe_math.h create mode 100644 onnx/common/scoped_resource.h create mode 100644 onnx/common/status.cc create mode 100644 onnx/common/status.h create mode 100644 onnx/common/tensor.h create mode 100644 onnx/common/version.h.in create mode 100644 onnx/common/visitor.h create mode 100644 onnx/compose.py create mode 100644 onnx/cpp2py_export.cc create mode 100644 onnx/defs/CMakeLists.txt create mode 100644 onnx/defs/__init__.py create mode 100644 onnx/defs/attr_proto_util.cc create mode 100644 onnx/defs/attr_proto_util.h create mode 100644 onnx/defs/controlflow/CMakeLists.txt create mode 100644 onnx/defs/controlflow/defs.cc create mode 100644 onnx/defs/controlflow/old.cc create mode 100644 onnx/defs/controlflow/utils.cc create mode 100644 onnx/defs/controlflow/utils.h create mode 100644 onnx/defs/data_propagators.h create mode 100644 onnx/defs/data_type_utils.cc create mode 100644 onnx/defs/data_type_utils.h create mode 100644 onnx/defs/doc_strings.cc create mode 100644 onnx/defs/doc_strings.h create mode 100644 onnx/defs/function.cc create mode 100644 onnx/defs/function.h create mode 100644 onnx/defs/gen_doc.py create mode 100644 onnx/defs/gen_shape_inference_information.py create mode 100644 onnx/defs/generator/CMakeLists.txt create mode 100644 onnx/defs/generator/defs.cc create mode 100644 onnx/defs/generator/old.cc create mode 100644 onnx/defs/generator/utils.cc create mode 100644 onnx/defs/generator/utils.h create mode 100644 onnx/defs/image/CMakeLists.txt create mode 100644 onnx/defs/image/defs.cc create mode 100644 onnx/defs/logical/CMakeLists.txt create mode 100644 onnx/defs/logical/defs.cc create mode 100644 onnx/defs/logical/old.cc create mode 100644 onnx/defs/math/CMakeLists.txt create mode 100644 onnx/defs/math/defs.cc create mode 100644 onnx/defs/math/old.cc create mode 100644 onnx/defs/math/utils.cc create mode 100644 onnx/defs/math/utils.h create mode 100644 onnx/defs/nn/CMakeLists.txt create mode 100644 onnx/defs/nn/defs.cc create mode 100644 onnx/defs/nn/old.cc create mode 100644 onnx/defs/nn/utils.cc create mode 100644 onnx/defs/nn/utils.h create mode 100644 onnx/defs/object_detection/CMakeLists.txt create mode 100644 onnx/defs/object_detection/defs.cc create mode 100644 onnx/defs/object_detection/old.cc create mode 100644 onnx/defs/operator_sets.h create mode 100644 onnx/defs/operator_sets_ml.h create mode 100644 onnx/defs/operator_sets_preview.h create mode 100644 onnx/defs/operator_sets_training.h create mode 100644 onnx/defs/optional/CMakeLists.txt create mode 100644 onnx/defs/optional/defs.cc create mode 100644 onnx/defs/optional/old.cc create mode 100644 onnx/defs/parser.cc create mode 100644 onnx/defs/parser.h create mode 100644 onnx/defs/preview/CMakeLists.txt create mode 100644 onnx/defs/preview/defs.cc create mode 100644 onnx/defs/printer.cc create mode 100644 onnx/defs/printer.h create mode 100644 onnx/defs/quantization/CMakeLists.txt create mode 100644 onnx/defs/quantization/defs.cc create mode 100644 onnx/defs/quantization/old.cc create mode 100644 onnx/defs/reduction/CMakeLists.txt create mode 100644 onnx/defs/reduction/defs.cc create mode 100644 onnx/defs/reduction/old.cc create mode 100644 onnx/defs/reduction/utils.cc create mode 100644 onnx/defs/reduction/utils.h create mode 100644 onnx/defs/rnn/CMakeLists.txt create mode 100644 onnx/defs/rnn/defs.cc create mode 100644 onnx/defs/rnn/old.cc create mode 100644 onnx/defs/schema.cc create mode 100644 onnx/defs/schema.h create mode 100644 onnx/defs/sequence/CMakeLists.txt create mode 100644 onnx/defs/sequence/defs.cc create mode 100644 onnx/defs/sequence/old.cc create mode 100644 onnx/defs/sequence/utils.cc create mode 100644 onnx/defs/sequence/utils.h create mode 100644 onnx/defs/shape_inference.cc create mode 100644 onnx/defs/shape_inference.h create mode 100644 onnx/defs/tensor/CMakeLists.txt create mode 100644 onnx/defs/tensor/defs.cc create mode 100644 onnx/defs/tensor/old.cc create mode 100644 onnx/defs/tensor/utils.cc create mode 100644 onnx/defs/tensor/utils.h create mode 100644 onnx/defs/tensor_proto_util.cc create mode 100644 onnx/defs/tensor_proto_util.h create mode 100644 onnx/defs/tensor_util.cc create mode 100644 onnx/defs/tensor_util.h create mode 100644 onnx/defs/text/CMakeLists.txt create mode 100644 onnx/defs/text/defs.cc create mode 100644 onnx/defs/traditionalml/CMakeLists.txt create mode 100644 onnx/defs/traditionalml/defs.cc create mode 100644 onnx/defs/traditionalml/old.cc create mode 100644 onnx/defs/traditionalml/utils.h create mode 100644 onnx/defs/training/CMakeLists.txt create mode 100644 onnx/defs/training/defs.cc create mode 100644 onnx/defs/type_builders.h create mode 100644 onnx/external_data_helper.py create mode 100644 onnx/frontend/__init__.py create mode 100644 onnx/fuzz/README.md create mode 100644 onnx/fuzz/fuzz_checker.py create mode 100644 onnx/fuzz/fuzz_compose.py create mode 100644 onnx/fuzz/fuzz_model_loader.py create mode 100644 onnx/fuzz/fuzz_parser.py create mode 100644 onnx/fuzz/fuzz_shape_inference.py create mode 100644 onnx/fuzz/fuzz_version_converter.py create mode 100644 onnx/fuzz/make_seed_corpus.py create mode 100755 onnx/gen_proto.py create mode 100644 onnx/helper.py create mode 100644 onnx/inliner.py create mode 100644 onnx/inliner/CMakeLists.txt create mode 100644 onnx/inliner/inliner.cc create mode 100644 onnx/inliner/inliner.h create mode 100644 onnx/model_container.py create mode 100644 onnx/numpy_helper.py create mode 100644 onnx/onnx-data.in.proto create mode 100644 onnx/onnx-data.proto create mode 100644 onnx/onnx-data.proto3 create mode 100644 onnx/onnx-ml.proto create mode 100644 onnx/onnx-ml.proto3 create mode 100644 onnx/onnx-operators-ml.proto create mode 100644 onnx/onnx-operators-ml.proto3 create mode 100644 onnx/onnx-operators.in.proto create mode 100644 onnx/onnx-operators.proto create mode 100644 onnx/onnx-operators.proto3 create mode 100644 onnx/onnx-operators_pb.h create mode 100644 onnx/onnx.in.proto create mode 100644 onnx/onnx.proto create mode 100644 onnx/onnx.proto3 create mode 100644 onnx/onnx_cpp2py_export/__init__.pyi create mode 100644 onnx/onnx_cpp2py_export/checker.pyi create mode 100644 onnx/onnx_cpp2py_export/defs.pyi create mode 100644 onnx/onnx_cpp2py_export/inliner.pyi create mode 100644 onnx/onnx_cpp2py_export/parser.pyi create mode 100644 onnx/onnx_cpp2py_export/printer.pyi create mode 100644 onnx/onnx_cpp2py_export/shape_inference.pyi create mode 100644 onnx/onnx_cpp2py_export/version_converter.pyi create mode 100644 onnx/onnx_pb.h create mode 100644 onnx/parser.py create mode 100644 onnx/printer.py create mode 100644 onnx/proto_utils.h create mode 100644 onnx/py.typed create mode 100644 onnx/py_utils.h create mode 100644 onnx/reference/README.md create mode 100644 onnx/reference/__init__.py create mode 100644 onnx/reference/op_run.py create mode 100644 onnx/reference/ops/__init__.py create mode 100644 onnx/reference/ops/_helpers.py create mode 100644 onnx/reference/ops/_op.py create mode 100644 onnx/reference/ops/_op_common_indices.py create mode 100644 onnx/reference/ops/_op_common_pool.py create mode 100644 onnx/reference/ops/_op_common_random.py create mode 100644 onnx/reference/ops/_op_common_window.py create mode 100644 onnx/reference/ops/_op_list.py create mode 100644 onnx/reference/ops/_quant_utils.py create mode 100644 onnx/reference/ops/aionnx_preview/__init__.py create mode 100644 onnx/reference/ops/aionnx_preview/_op_list.py create mode 100644 onnx/reference/ops/aionnx_preview/op_flex_attention.py create mode 100644 onnx/reference/ops/aionnx_preview_training/__init__.py create mode 100644 onnx/reference/ops/aionnx_preview_training/_op_list.py create mode 100644 onnx/reference/ops/aionnx_preview_training/_op_run_training.py create mode 100644 onnx/reference/ops/aionnx_preview_training/op_adagrad.py create mode 100644 onnx/reference/ops/aionnx_preview_training/op_adam.py create mode 100644 onnx/reference/ops/aionnx_preview_training/op_momentum.py create mode 100644 onnx/reference/ops/aionnxml/__init__.py create mode 100644 onnx/reference/ops/aionnxml/_common_classifier.py create mode 100644 onnx/reference/ops/aionnxml/_op_list.py create mode 100644 onnx/reference/ops/aionnxml/_op_run_aionnxml.py create mode 100644 onnx/reference/ops/aionnxml/op_array_feature_extractor.py create mode 100644 onnx/reference/ops/aionnxml/op_binarizer.py create mode 100644 onnx/reference/ops/aionnxml/op_dict_vectorizer.py create mode 100644 onnx/reference/ops/aionnxml/op_feature_vectorizer.py create mode 100644 onnx/reference/ops/aionnxml/op_imputer.py create mode 100644 onnx/reference/ops/aionnxml/op_label_encoder.py create mode 100644 onnx/reference/ops/aionnxml/op_linear_classifier.py create mode 100644 onnx/reference/ops/aionnxml/op_linear_regressor.py create mode 100644 onnx/reference/ops/aionnxml/op_normalizer.py create mode 100644 onnx/reference/ops/aionnxml/op_one_hot_encoder.py create mode 100644 onnx/reference/ops/aionnxml/op_scaler.py create mode 100644 onnx/reference/ops/aionnxml/op_svm_classifier.py create mode 100644 onnx/reference/ops/aionnxml/op_svm_helper.py create mode 100644 onnx/reference/ops/aionnxml/op_svm_regressor.py create mode 100644 onnx/reference/ops/aionnxml/op_tree_ensemble.py create mode 100644 onnx/reference/ops/aionnxml/op_tree_ensemble_classifier.py create mode 100644 onnx/reference/ops/aionnxml/op_tree_ensemble_helper.py create mode 100644 onnx/reference/ops/aionnxml/op_tree_ensemble_regressor.py create mode 100644 onnx/reference/ops/experimental/__init__.py create mode 100644 onnx/reference/ops/experimental/_op_list.py create mode 100644 onnx/reference/ops/experimental/_op_run_experimental.py create mode 100644 onnx/reference/ops/experimental/op_im2col.py create mode 100644 onnx/reference/ops/op_abs.py create mode 100644 onnx/reference/ops/op_acos.py create mode 100644 onnx/reference/ops/op_acosh.py create mode 100644 onnx/reference/ops/op_add.py create mode 100644 onnx/reference/ops/op_affine_grid.py create mode 100644 onnx/reference/ops/op_and.py create mode 100644 onnx/reference/ops/op_argmax.py create mode 100644 onnx/reference/ops/op_argmin.py create mode 100644 onnx/reference/ops/op_asin.py create mode 100644 onnx/reference/ops/op_asinh.py create mode 100644 onnx/reference/ops/op_atan.py create mode 100644 onnx/reference/ops/op_atanh.py create mode 100644 onnx/reference/ops/op_attention.py create mode 100644 onnx/reference/ops/op_attribute_has_value.py create mode 100644 onnx/reference/ops/op_average_pool.py create mode 100644 onnx/reference/ops/op_batch_normalization.py create mode 100644 onnx/reference/ops/op_bernoulli.py create mode 100644 onnx/reference/ops/op_bitcast.py create mode 100644 onnx/reference/ops/op_bitshift.py create mode 100644 onnx/reference/ops/op_bitwise_and.py create mode 100644 onnx/reference/ops/op_bitwise_not.py create mode 100644 onnx/reference/ops/op_bitwise_or.py create mode 100644 onnx/reference/ops/op_bitwise_xor.py create mode 100644 onnx/reference/ops/op_blackman_window.py create mode 100644 onnx/reference/ops/op_cast.py create mode 100644 onnx/reference/ops/op_cast_like.py create mode 100644 onnx/reference/ops/op_causal_conv_with_state.py create mode 100644 onnx/reference/ops/op_ceil.py create mode 100644 onnx/reference/ops/op_celu.py create mode 100644 onnx/reference/ops/op_center_crop_pad.py create mode 100644 onnx/reference/ops/op_clip.py create mode 100644 onnx/reference/ops/op_col2im.py create mode 100644 onnx/reference/ops/op_compress.py create mode 100644 onnx/reference/ops/op_concat.py create mode 100644 onnx/reference/ops/op_concat_from_sequence.py create mode 100644 onnx/reference/ops/op_constant.py create mode 100644 onnx/reference/ops/op_constant_of_shape.py create mode 100644 onnx/reference/ops/op_conv.py create mode 100644 onnx/reference/ops/op_conv_integer.py create mode 100644 onnx/reference/ops/op_conv_transpose.py create mode 100644 onnx/reference/ops/op_cos.py create mode 100644 onnx/reference/ops/op_cosh.py create mode 100644 onnx/reference/ops/op_cum_prod.py create mode 100644 onnx/reference/ops/op_cum_sum.py create mode 100644 onnx/reference/ops/op_deform_conv.py create mode 100644 onnx/reference/ops/op_depth_to_space.py create mode 100644 onnx/reference/ops/op_dequantize_linear.py create mode 100644 onnx/reference/ops/op_det.py create mode 100644 onnx/reference/ops/op_dft.py create mode 100644 onnx/reference/ops/op_div.py create mode 100644 onnx/reference/ops/op_dropout.py create mode 100644 onnx/reference/ops/op_dynamic_quantize_linear.py create mode 100644 onnx/reference/ops/op_einsum.py create mode 100644 onnx/reference/ops/op_elu.py create mode 100644 onnx/reference/ops/op_equal.py create mode 100644 onnx/reference/ops/op_erf.py create mode 100644 onnx/reference/ops/op_exp.py create mode 100644 onnx/reference/ops/op_expand.py create mode 100644 onnx/reference/ops/op_eyelike.py create mode 100644 onnx/reference/ops/op_flatten.py create mode 100644 onnx/reference/ops/op_floor.py create mode 100644 onnx/reference/ops/op_gather.py create mode 100644 onnx/reference/ops/op_gather_elements.py create mode 100644 onnx/reference/ops/op_gathernd.py create mode 100644 onnx/reference/ops/op_gemm.py create mode 100644 onnx/reference/ops/op_global_average_pool.py create mode 100644 onnx/reference/ops/op_global_max_pool.py create mode 100644 onnx/reference/ops/op_greater.py create mode 100644 onnx/reference/ops/op_greater_or_equal.py create mode 100644 onnx/reference/ops/op_grid_sample.py create mode 100644 onnx/reference/ops/op_gru.py create mode 100644 onnx/reference/ops/op_hamming_window.py create mode 100644 onnx/reference/ops/op_hann_window.py create mode 100644 onnx/reference/ops/op_hard_sigmoid.py create mode 100644 onnx/reference/ops/op_hardmax.py create mode 100644 onnx/reference/ops/op_identity.py create mode 100644 onnx/reference/ops/op_if.py create mode 100644 onnx/reference/ops/op_image_decoder.py create mode 100644 onnx/reference/ops/op_instance_normalization.py create mode 100644 onnx/reference/ops/op_isinf.py create mode 100644 onnx/reference/ops/op_isnan.py create mode 100644 onnx/reference/ops/op_layer_normalization.py create mode 100644 onnx/reference/ops/op_leaky_relu.py create mode 100644 onnx/reference/ops/op_less.py create mode 100644 onnx/reference/ops/op_less_or_equal.py create mode 100644 onnx/reference/ops/op_linear_attention.py create mode 100644 onnx/reference/ops/op_log.py create mode 100644 onnx/reference/ops/op_log_softmax.py create mode 100644 onnx/reference/ops/op_loop.py create mode 100644 onnx/reference/ops/op_lp_normalization.py create mode 100644 onnx/reference/ops/op_lp_pool.py create mode 100644 onnx/reference/ops/op_lrn.py create mode 100644 onnx/reference/ops/op_lstm.py create mode 100644 onnx/reference/ops/op_matmul.py create mode 100644 onnx/reference/ops/op_matmul_integer.py create mode 100644 onnx/reference/ops/op_max.py create mode 100644 onnx/reference/ops/op_max_pool.py create mode 100644 onnx/reference/ops/op_max_unpool.py create mode 100644 onnx/reference/ops/op_mean.py create mode 100644 onnx/reference/ops/op_mel_weight_matrix.py create mode 100644 onnx/reference/ops/op_min.py create mode 100644 onnx/reference/ops/op_mod.py create mode 100644 onnx/reference/ops/op_mul.py create mode 100644 onnx/reference/ops/op_neg.py create mode 100644 onnx/reference/ops/op_negative_log_likelihood_loss.py create mode 100644 onnx/reference/ops/op_non_max_suppression.py create mode 100644 onnx/reference/ops/op_non_zero.py create mode 100644 onnx/reference/ops/op_not.py create mode 100644 onnx/reference/ops/op_one_hot.py create mode 100644 onnx/reference/ops/op_optional.py create mode 100644 onnx/reference/ops/op_optional_get_element.py create mode 100644 onnx/reference/ops/op_optional_has_element.py create mode 100644 onnx/reference/ops/op_or.py create mode 100644 onnx/reference/ops/op_pad.py create mode 100644 onnx/reference/ops/op_pool_common.py create mode 100644 onnx/reference/ops/op_pow.py create mode 100644 onnx/reference/ops/op_prelu.py create mode 100644 onnx/reference/ops/op_qlinear_conv.py create mode 100644 onnx/reference/ops/op_qlinear_matmul.py create mode 100644 onnx/reference/ops/op_quantize_linear.py create mode 100644 onnx/reference/ops/op_random_normal.py create mode 100644 onnx/reference/ops/op_random_normal_like.py create mode 100644 onnx/reference/ops/op_random_uniform.py create mode 100644 onnx/reference/ops/op_random_uniform_like.py create mode 100644 onnx/reference/ops/op_range.py create mode 100644 onnx/reference/ops/op_reciprocal.py create mode 100644 onnx/reference/ops/op_reduce_l1.py create mode 100644 onnx/reference/ops/op_reduce_l2.py create mode 100644 onnx/reference/ops/op_reduce_log_sum.py create mode 100644 onnx/reference/ops/op_reduce_log_sum_exp.py create mode 100644 onnx/reference/ops/op_reduce_max.py create mode 100644 onnx/reference/ops/op_reduce_mean.py create mode 100644 onnx/reference/ops/op_reduce_min.py create mode 100644 onnx/reference/ops/op_reduce_prod.py create mode 100644 onnx/reference/ops/op_reduce_sum.py create mode 100644 onnx/reference/ops/op_reduce_sum_square.py create mode 100644 onnx/reference/ops/op_regex_full_match.py create mode 100644 onnx/reference/ops/op_relu.py create mode 100644 onnx/reference/ops/op_reshape.py create mode 100644 onnx/reference/ops/op_resize.py create mode 100644 onnx/reference/ops/op_reverse_sequence.py create mode 100644 onnx/reference/ops/op_rms_normalization.py create mode 100644 onnx/reference/ops/op_rnn.py create mode 100644 onnx/reference/ops/op_roi_align.py create mode 100644 onnx/reference/ops/op_rotary_embedding.py create mode 100644 onnx/reference/ops/op_round.py create mode 100644 onnx/reference/ops/op_scan.py create mode 100644 onnx/reference/ops/op_scatter_elements.py create mode 100644 onnx/reference/ops/op_scatternd.py create mode 100644 onnx/reference/ops/op_selu.py create mode 100644 onnx/reference/ops/op_sequence_at.py create mode 100644 onnx/reference/ops/op_sequence_construct.py create mode 100644 onnx/reference/ops/op_sequence_empty.py create mode 100644 onnx/reference/ops/op_sequence_erase.py create mode 100644 onnx/reference/ops/op_sequence_insert.py create mode 100644 onnx/reference/ops/op_sequence_length.py create mode 100644 onnx/reference/ops/op_sequence_map.py create mode 100644 onnx/reference/ops/op_shape.py create mode 100644 onnx/reference/ops/op_shrink.py create mode 100644 onnx/reference/ops/op_sigmoid.py create mode 100644 onnx/reference/ops/op_sign.py create mode 100644 onnx/reference/ops/op_sin.py create mode 100644 onnx/reference/ops/op_sinh.py create mode 100644 onnx/reference/ops/op_size.py create mode 100644 onnx/reference/ops/op_slice.py create mode 100644 onnx/reference/ops/op_softmax.py create mode 100644 onnx/reference/ops/op_softmax_cross_entropy_loss.py create mode 100644 onnx/reference/ops/op_softplus.py create mode 100644 onnx/reference/ops/op_softsign.py create mode 100644 onnx/reference/ops/op_space_to_depth.py create mode 100644 onnx/reference/ops/op_split.py create mode 100644 onnx/reference/ops/op_split_to_sequence.py create mode 100644 onnx/reference/ops/op_sqrt.py create mode 100644 onnx/reference/ops/op_squeeze.py create mode 100644 onnx/reference/ops/op_stft.py create mode 100644 onnx/reference/ops/op_string_concat.py create mode 100644 onnx/reference/ops/op_string_normalizer.py create mode 100644 onnx/reference/ops/op_string_split.py create mode 100644 onnx/reference/ops/op_sub.py create mode 100644 onnx/reference/ops/op_sum.py create mode 100644 onnx/reference/ops/op_swish.py create mode 100644 onnx/reference/ops/op_tan.py create mode 100644 onnx/reference/ops/op_tanh.py create mode 100644 onnx/reference/ops/op_tensor_scatter.py create mode 100644 onnx/reference/ops/op_tfidf_vectorizer.py create mode 100644 onnx/reference/ops/op_thresholded_relu.py create mode 100644 onnx/reference/ops/op_tile.py create mode 100644 onnx/reference/ops/op_topk.py create mode 100644 onnx/reference/ops/op_transpose.py create mode 100644 onnx/reference/ops/op_trilu.py create mode 100644 onnx/reference/ops/op_unique.py create mode 100644 onnx/reference/ops/op_unsqueeze.py create mode 100644 onnx/reference/ops/op_upsample.py create mode 100644 onnx/reference/ops/op_where.py create mode 100644 onnx/reference/ops/op_xor.py create mode 100644 onnx/reference/ops_optimized/__init__.py create mode 100644 onnx/reference/ops_optimized/op_conv_optimized.py create mode 100644 onnx/reference/reference_evaluator.py create mode 100644 onnx/serialization.py create mode 100644 onnx/shape_inference.py create mode 100644 onnx/shape_inference/CMakeLists.txt create mode 100644 onnx/shape_inference/attribute_binder.h create mode 100644 onnx/shape_inference/implementation.cc create mode 100644 onnx/shape_inference/implementation.h create mode 100644 onnx/string_utils.h create mode 100644 onnx/test/basic_test.py create mode 100644 onnx/test/checker_test.py create mode 100644 onnx/test/cmake/CMakeLists.txt create mode 100644 onnx/test/cmake/main.cc create mode 100644 onnx/test/compose_test.py create mode 100644 onnx/test/cpp/checker_test.cc create mode 100644 onnx/test/cpp/data_propagation_test.cc create mode 100644 onnx/test/cpp/function_context_test.cc create mode 100644 onnx/test/cpp/function_get_test.cc create mode 100644 onnx/test/cpp/function_verify_test.cc create mode 100644 onnx/test/cpp/inliner_test.cc create mode 100644 onnx/test/cpp/ir_test.cc create mode 100644 onnx/test/cpp/op_reg_test.cc create mode 100644 onnx/test/cpp/parser_test.cc create mode 100644 onnx/test/cpp/schema_registration_test.cc create mode 100644 onnx/test/cpp/shape_inference_test.cc create mode 100644 onnx/test/cpp/tensor_test.cc create mode 100644 onnx/test/cpp/test_main.cc create mode 100644 onnx/test/cpp/type_builders_test.cc create mode 100644 onnx/test/cpp/utf8_conversion_test.cc create mode 100644 onnx/test/data_propagation_test.py create mode 100644 onnx/test/function_inference_test.py create mode 100644 onnx/test/function_test.py create mode 100644 onnx/test/helper_test.py create mode 100644 onnx/test/inference_function_test.py create mode 100644 onnx/test/inliner_test.py create mode 100644 onnx/test/model_container_refeval_test.py create mode 100644 onnx/test/model_container_test.py create mode 100644 onnx/test/model_inference_test.py create mode 100644 onnx/test/node_shape_inference_test.py create mode 100644 onnx/test/numpy_helper_test.py create mode 100644 onnx/test/parser_test.py create mode 100644 onnx/test/printer_test.py create mode 100644 onnx/test/reference_evaluator_ml_test.py create mode 100644 onnx/test/reference_evaluator_model_test.py create mode 100644 onnx/test/reference_evaluator_test.py create mode 100644 onnx/test/schema_test.py create mode 100644 onnx/test/serialization_test.py create mode 100644 onnx/test/shape_inference_test.py create mode 100644 onnx/test/symbolic_shape_test.py create mode 100644 onnx/test/test_backend_reference.py create mode 100644 onnx/test/test_backend_test.py create mode 100644 onnx/test/test_external_data.py create mode 100644 onnx/test/tools_test.py create mode 100644 onnx/test/training_tool_test.py create mode 100644 onnx/test/utils_test.py create mode 100644 onnx/test/version_converter/automatic_conversion_test_base.py create mode 100644 onnx/test/version_converter/automatic_downgrade_test.py create mode 100644 onnx/test/version_converter/automatic_upgrade_test.py create mode 100644 onnx/test/version_converter_test.py create mode 100644 onnx/test/version_utils.py create mode 100644 onnx/tools/__init__.py create mode 100644 onnx/tools/replace_constants.py create mode 100644 onnx/tools/update_model_dims.py create mode 100644 onnx/utils.py create mode 100644 onnx/version.py create mode 100644 onnx/version_converter.py create mode 100644 onnx/version_converter/BaseConverter.h create mode 100644 onnx/version_converter/CMakeLists.txt create mode 100644 onnx/version_converter/adapters/Attention_24_23.h create mode 100644 onnx/version_converter/adapters/CMakeLists.txt create mode 100644 onnx/version_converter/adapters/adapter.h create mode 100644 onnx/version_converter/adapters/axes_attribute_to_input.h create mode 100644 onnx/version_converter/adapters/axes_input_to_attribute.h create mode 100644 onnx/version_converter/adapters/axis_attribute_to_input.h create mode 100644 onnx/version_converter/adapters/axis_input_to_attribute.h create mode 100644 onnx/version_converter/adapters/batch_normalization_13_14.h create mode 100644 onnx/version_converter/adapters/broadcast_backward_compatibility.h create mode 100644 onnx/version_converter/adapters/broadcast_forward_compatibility.h create mode 100644 onnx/version_converter/adapters/cast_9_8.h create mode 100644 onnx/version_converter/adapters/clip_10_11.h create mode 100644 onnx/version_converter/adapters/compatible.h create mode 100644 onnx/version_converter/adapters/dropout_11_12.h create mode 100644 onnx/version_converter/adapters/extend_supported_types.h create mode 100644 onnx/version_converter/adapters/gemm_6_7.h create mode 100644 onnx/version_converter/adapters/gemm_7_6.h create mode 100644 onnx/version_converter/adapters/gridsample_19_20.h create mode 100644 onnx/version_converter/adapters/group_normalization_20_21.h create mode 100644 onnx/version_converter/adapters/maxpool_8_7.h create mode 100644 onnx/version_converter/adapters/no_previous_version.h create mode 100644 onnx/version_converter/adapters/pad_10_11.h create mode 100644 onnx/version_converter/adapters/q_dq_21_20.h create mode 100644 onnx/version_converter/adapters/range_27_26.h create mode 100644 onnx/version_converter/adapters/remove_consumed_inputs.h create mode 100644 onnx/version_converter/adapters/reshape_4_5.h create mode 100644 onnx/version_converter/adapters/reshape_5_4.h create mode 100644 onnx/version_converter/adapters/resize_10_11.h create mode 100644 onnx/version_converter/adapters/scan_8_9.h create mode 100644 onnx/version_converter/adapters/scan_9_8.h create mode 100644 onnx/version_converter/adapters/scatter_10_11.h create mode 100644 onnx/version_converter/adapters/scatter_16_15.h create mode 100644 onnx/version_converter/adapters/scatter_18_17.h create mode 100644 onnx/version_converter/adapters/slice_9_10.h create mode 100644 onnx/version_converter/adapters/softmax_12_13.h create mode 100644 onnx/version_converter/adapters/softmax_13_12.h create mode 100644 onnx/version_converter/adapters/split_12_13.h create mode 100644 onnx/version_converter/adapters/split_13_12.h create mode 100644 onnx/version_converter/adapters/split_17_18.h create mode 100644 onnx/version_converter/adapters/sum_8_7.h create mode 100644 onnx/version_converter/adapters/topk_9_10.h create mode 100644 onnx/version_converter/adapters/transformers.h create mode 100644 onnx/version_converter/adapters/type_restriction.h create mode 100644 onnx/version_converter/adapters/upsample_6_7.h create mode 100644 onnx/version_converter/adapters/upsample_8_9.h create mode 100644 onnx/version_converter/adapters/upsample_9_10.h create mode 100644 onnx/version_converter/adapters/upsample_9_8.h create mode 100644 onnx/version_converter/convert.cc create mode 100644 onnx/version_converter/convert.h create mode 100644 onnx/version_converter/helper.cc create mode 100644 onnx/version_converter/helper.h create mode 100644 pixi.lock create mode 100644 pixi.toml create mode 100644 pyproject.toml create mode 100644 renovate.json5 create mode 100644 requirements-dev.txt create mode 100644 requirements-lintrunner.txt create mode 100644 requirements-release_test.txt create mode 100644 sbom.cdx.json create mode 100644 tools/check_namespace.py create mode 100755 tools/gen_coverage_report.py create mode 100644 tools/pixi-shim.py create mode 100644 tools/pretty_clang_tidy.py create mode 100644 tools/update_doc.bat create mode 100755 tools/update_doc.sh create mode 100644 typos.toml diff --git a/.agents/skills/add-function-body/SKILL.md b/.agents/skills/add-function-body/SKILL.md new file mode 100644 index 0000000..21ee68d --- /dev/null +++ b/.agents/skills/add-function-body/SKILL.md @@ -0,0 +1,120 @@ +--- +name: add-function-body +description: Add a function body definition to an ONNX operator, defining how it decomposes into simpler ops. Use when asked to make an op decomposable, add a FunctionBody, implement SetContextDependentFunctionBodyBuilder, or express an op in terms of other ONNX operators. +--- + +Follow the full guide in [docs/AddFunctionBody.md](../../../docs/AddFunctionBody.md). + +## File Locations + +| Component | File | +|-----------|------| +| Function body definition | `onnx/defs//defs.cc` (inline with schema) | +| FunctionBuilder utilities | `onnx/defs/function.h` | +| Function tests | `onnx/test/cpp/function_get_test.cc`, `onnx/test/cpp/function_verify_test.cc` | + +## Method 1: Simple String-Based Function Body + +```cpp +ONNX_OPERATOR_SET_SCHEMA( + LessOrEqual, 16, + OpSchema() + // ... inputs, outputs, type constraints ... + .TypeAndShapeInferenceFunction(inferenceFunction) + .FunctionBody(R"ONNX( + { + O1 = Less (A, B) + O2 = Equal (A, B) + C = Or (O1, O2) + } + )ONNX")); +``` + +With explicit opset version: +```cpp + .FunctionBody(R"ONNX(...)ONNX", 18) // Valid from opset 18 +``` + +Referencing attributes with `@attr_name`: +```cpp + .FunctionBody(R"ONNX( + { + Alpha = Constant () + AlphaCast = CastLike (Alpha, X) + ... + } + )ONNX") +``` + +## Method 2: Context-Dependent Function Body + +For ops whose decomposition varies based on attributes or optional inputs: + +```cpp +static bool BuildFunctionBodyMyOp( + const FunctionBodyBuildContext& ctx, + const OpSchema& schema, + FunctionProto& functionProto) { + FunctionBuilder builder(functionProto); + // Build graph based on ctx.hasInput(), ctx.getAttribute(), etc. + builder.Add("output = SomeOp (input)"); + schema.BuildFunction(functionProto); + return true; +} + +// Register: + .SetContextDependentFunctionBodyBuilder(BuildFunctionBodyMyOp) +``` + +## FunctionBuilder API + +```cpp +FunctionBuilder builder(functionProto); +builder.Add("Y = Relu (X)"); // Add node +builder.Const("alpha", std::vector{0.01f}); // Constant tensor +builder.Const1D("axes", int64_t(1)); // 1-D constant +builder.Add(R"( // Multi-line + X_Sub = Sub (X, X_Max) + X_Exp = Exp (X_Sub) +)"); +builder.AddOpset("", 18); // Opset dependency +schema.BuildFunction(functionProto); // Finalize +return true; +``` + +## Multiple Opset Versions + +```cpp + .SetContextDependentFunctionBodyBuilder(builderForOpset13) + .SetContextDependentFunctionBodyBuilder(builderForOpset18, 18) +``` + +## ONNX Function Body Syntax + +Function body strings use the ONNX text format ("onnxtxt"). See the [`onnxtxt`](../onnxtxt/SKILL.md) skill for the full syntax cheat sheet, attribute references (`@attr_name`), `Constant ` forms, `CastLike` vs `Cast`, body-subgraph idioms, and parser tests. Quick reminders specific to function bodies: + +- Variable names are local intermediates; input/output names must match the schema's declared names. +- Reference enclosing-op attributes with `@attr_name` — and only those declared in `.Attr(...)` calls. +- Use `CastLike` (not `Cast`) when the target type depends on another input. + +## Code Style: Prefer Named Functions + +Define context-dependent function body builders as **separate named functions** rather than inline lambdas within `ONNX_OPERATOR_SET_SCHEMA`. The macro expansion makes setting breakpoints on inline lambdas unreliable in debuggers. + +Simple string-based `.FunctionBody(R"ONNX(...)ONNX")` definitions don't have this issue. + +## After Making Changes + +```bash +python onnx/defs/gen_doc.py +lintrunner -a --output oneline +``` + +## Common Mistakes + +- Forgetting `schema.BuildFunction(functionProto)` at end of context-dependent builders +- Forgetting to `return true` from the builder function +- Variable names conflicting with input/output names +- Using `Cast` instead of `CastLike` for dynamic type matching +- Function body not producing all declared outputs +- Using `@attr_name` for an attribute not declared in `.Attr()` calls diff --git a/.agents/skills/add-op/SKILL.md b/.agents/skills/add-op/SKILL.md new file mode 100644 index 0000000..952b682 --- /dev/null +++ b/.agents/skills/add-op/SKILL.md @@ -0,0 +1,99 @@ +--- +name: add-op +description: Add a new ONNX operator or update an existing operator to a new opset version. Use when asked to define an operator schema, register an op, add inputs/outputs/attributes to an op, move an op to old.cc, or bump an op's opset version. +--- + +Follow the full procedure in [docs/AddNewOp.md](../../../docs/AddNewOp.md). + +## Files to Modify + +| Component | File | +|-----------|------| +| Schema definition | `onnx/defs//defs.cc` | +| Operator set registration | `onnx/defs/operator_sets.h` | +| Type/shape inference | Inline in schema via `.TypeAndShapeInferenceFunction(...)` | +| Function body (if applicable) | Inline in schema via `.FunctionBody(...)` | +| Reference implementation | `onnx/reference/ops/op_.py` | +| Node tests | `onnx/backend/test/case/node/.py` | +| Shape inference tests | `onnx/test/shape_inference_test.py` | +| Version converter adapter (if behavior changed) | `onnx/version_converter/adapters/__.h` | +| Upgrade/downgrade tests | `onnx/test/version_converter/automatic_upgrade_test.py` and `automatic_downgrade_test.py` | + +Domain subdirectories under `onnx/defs/`: `math/`, `nn/`, `tensor/`, `logical/`, `reduction/`, `rnn/`, `sequence/`, `image/`, `text/`, `quantization/`, `controlflow/`, `optional/`, `traditionalml/`, `training/` + +## Schema Registration Pattern + +```cpp +ONNX_OPERATOR_SET_SCHEMA( + OperatorName, + OPSET_VERSION, + OpSchema() + .SetDoc(OperatorName_verN_doc) + .Input(0, "X", "Description", "T", OpSchema::Single, true, 1, OpSchema::Differentiable) + .Output(0, "Y", "Description", "T", OpSchema::Single, true, 1, OpSchema::Differentiable) + .Attr("attr_name", "Description", AttributeProto::FLOAT, default_value) + .TypeConstraint("T", {"tensor(float)", "tensor(double)", ...}, "Description") + .TypeAndShapeInferenceFunction(InferShapeForOperatorName) + // Function body uses ONNX text format — see the onnxtxt skill for syntax/conventions. + .FunctionBody(R"ONNX( + { + ... + } + )ONNX", FUNCTION_OPSET_VERSION)); +``` + +## Updating an Existing Operator + +1. Move current schema from `defs.cc` to `old.cc` in the same domain directory +2. Create new version in `defs.cc` with the new opset version number +3. Update `onnx/defs/operator_sets.h` +4. Add a version converter adapter if behavior changed +5. Add upgrade/downgrade tests + +### Avoiding duplication between defs.cc and old.cc + +When moving a schema to `old.cc`, avoid significant code/documentation duplication: + +- Extract common logic (doc strings, type constraint lists, shape inference helpers) into shared functions in the domain's header or utils file. +- Use parameterized helpers when versions differ only slightly (e.g., expanded type list, additional optional input). +- Some duplication is acceptable when sharing would create overly complicated logic. Prefer clarity over DRY when the alternative makes either version harder to understand independently. + +## Code Style: Prefer Named Functions + +Define shape inference functions and function body builders as **separate named functions** rather than inline lambdas within `ONNX_OPERATOR_SET_SCHEMA`. The macro expansion makes setting breakpoints on inline lambdas unreliable in debuggers. + +```cpp +// PREFERRED: named function — easy to set breakpoints +static void InferShapeForMyOp(InferenceContext& ctx) { + propagateElemTypeFromInputToOutput(ctx, 0, 0); + // ... +} + +ONNX_OPERATOR_SET_SCHEMA( + MyOp, 21, + OpSchema() + // ... + .TypeAndShapeInferenceFunction(InferShapeForMyOp)); +``` + +## Writing Tests + +For test fixtures, prefer the ONNX text format via `onnx.parser` / C++ `OnnxParser`. See the [`onnxtxt`](../onnxtxt/SKILL.md) skill for the per-file recommendations table, body-subgraph idioms, the argument-order convention, and the `unk__*` materialization gotcha. (Empirical: PR #7962 cut ~58–70% of test LOC by switching.) + +## After Making Changes + +```bash +python onnx/defs/gen_doc.py +python onnx/backend/test/stat_coverage.py +python onnx/gen_proto.py # only if proto changed +lintrunner -a --output oneline +``` + +## References + +- [docs/AddNewOp.md](../../../docs/AddNewOp.md) — Full procedure for adding/updating ops +- [docs/AddFunctionBody.md](../../../docs/AddFunctionBody.md) — Function body guide +- [docs/ShapeInference.md](../../../docs/ShapeInference.md) — Shape inference guide +- [../onnxtxt/SKILL.md](../onnxtxt/SKILL.md) — ONNX text format ("onnxtxt") for tests and function bodies +- [references/node-test-pattern.md](references/node-test-pattern.md) — Node test example +- [references/reference-impl-pattern.md](references/reference-impl-pattern.md) — Reference implementation example diff --git a/.agents/skills/add-op/references/node-test-pattern.md b/.agents/skills/add-op/references/node-test-pattern.md new file mode 100644 index 0000000..d432c9a --- /dev/null +++ b/.agents/skills/add-op/references/node-test-pattern.md @@ -0,0 +1,39 @@ +# Node Test Pattern + +Example test file at `onnx/backend/test/case/node/.py`: + +```python +from __future__ import annotations + +import numpy as np + +import onnx +from onnx.backend.test.case.base import Base +from onnx.backend.test.case.node import expect + + +class OpName(Base): + @staticmethod + def export() -> None: + node = onnx.helper.make_node( + "OpName", + inputs=["x"], + outputs=["y"], + ) + x = np.random.randn(3, 4, 5).astype(np.float32) + y = np.some_operation(x) + expect(node, inputs=[x], outputs=[y], name="test_opname") + + @staticmethod + def export_with_broadcasting() -> None: + node = onnx.helper.make_node( + "OpName", + inputs=["x", "y"], + outputs=["z"], + ) + x = np.random.randn(3, 4, 5).astype(np.float32) + y = np.random.randn(5).astype(np.float32) + expect(node, inputs=[x, y], outputs=[x + y], name="test_opname_bcast") +``` + +Each `export*` static method becomes a separate test case. The `expect` helper validates the reference implementation output matches. diff --git a/.agents/skills/add-op/references/reference-impl-pattern.md b/.agents/skills/add-op/references/reference-impl-pattern.md new file mode 100644 index 0000000..6ec0480 --- /dev/null +++ b/.agents/skills/add-op/references/reference-impl-pattern.md @@ -0,0 +1,56 @@ +# Reference Implementation Pattern + +Example at `onnx/reference/ops/op_.py`: + +```python +from __future__ import annotations + +import numpy as np + +from onnx.reference.ops._op import OpRunUnaryNum + + +class OpName(OpRunUnaryNum): + def _run(self, x): + return (np.some_operation(x),) +``` + +## Base Classes + +| Base class | Use case | +|-----------|----------| +| `OpRun` | General-purpose (any signature) | +| `OpRunUnary` | Single input, single output | +| `OpRunUnaryNum` | Single numeric input, validates output dtype matches | +| `OpRunBinary` | Two inputs | +| `OpRunBinaryNumpy` | Two inputs using a numpy function directly | + +## Binary Op Example + +```python +from __future__ import annotations + +import numpy as np + +from onnx.reference.ops._op import OpRunBinaryNumpy + + +class Add(OpRunBinaryNumpy): + def __init__(self, onnx_node, run_params): + OpRunBinaryNumpy.__init__(self, np.add, onnx_node, run_params) +``` + +## General Op Example + +```python +from __future__ import annotations + +import numpy as np + +from onnx.reference.op_run import OpRun + + +class Concat(OpRun): + def _run(self, *inputs, axis=None): + return (np.concatenate(inputs, axis=axis),) +``` diff --git a/.agents/skills/add-shape-inference/SKILL.md b/.agents/skills/add-shape-inference/SKILL.md new file mode 100644 index 0000000..92601ac --- /dev/null +++ b/.agents/skills/add-shape-inference/SKILL.md @@ -0,0 +1,140 @@ +--- +name: add-shape-inference +description: Add or update type and shape inference for an ONNX operator. Use when asked to implement TypeAndShapeInferenceFunction, propagate shapes, add shape inference tests, fix shape inference bugs, or handle broadcasting logic. +--- + +See also: [docs/ShapeInference.md](../../../docs/ShapeInference.md) + +## File Locations + +| Component | File | +|-----------|------| +| Inference function | `onnx/defs//defs.cc` (inline with schema) | +| Utility functions | `onnx/defs/shape_inference.h` | +| Tests | `onnx/test/shape_inference_test.py` | + +## Type Inference vs. Shape Inference + +**Type inference** (element type) is often handled automatically by type constraints. When `"T"` is shared between input and output, the framework infers output type automatically. + +However, many existing ops still explicitly call `propagateElemTypeFromInputToOutput` as a best practice for robustness. + +Explicit type inference logic is only needed when: +- Output type is determined by an **attribute** (e.g., `Cast`) +- Output type differs from all inputs in a way not expressible via type constraints +- The operator uses **heterogeneous** variadic inputs/outputs + +### Homogeneous vs. Heterogeneous + +Applies only to variadic (repeated) inputs/outputs: +- **Homogeneous** (default): All repeated arguments share the same type. Framework propagates automatically. +- **Heterogeneous**: Each argument can differ. Used by `Loop`/`Scan`. The inference method must explicitly propagate types for each argument. + +## Common Patterns + +### Unary Element-wise + +```cpp +.TypeAndShapeInferenceFunction(propagateShapeAndTypeFromFirstInput) +``` + +### Binary with Broadcasting + +```cpp +static void InferShapeForBinaryOp(InferenceContext& ctx) { + propagateElemTypeFromInputToOutput(ctx, 0, 0); + if (hasNInputShapes(ctx, 2)) + bidirectionalBroadcastShapeInference( + ctx.getInputType(0)->tensor_type().shape(), + ctx.getInputType(1)->tensor_type().shape(), + *ctx.getOutputType(0)->mutable_tensor_type()->mutable_shape()); +} +``` + +### Shape-Changing Op + +```cpp +static void InferShapeForTranspose(InferenceContext& ctx) { + propagateElemTypeFromInputToOutput(ctx, 0, 0); + if (!hasNInputShapes(ctx, 1)) return; + + auto input_shape = ctx.getInputType(0)->tensor_type().shape(); + int rank = input_shape.dim_size(); + std::vector perm; + getRepeatedAttribute(ctx, "perm", perm); + + auto* output_shape = getOutputShape(ctx, 0); + for (int i = 0; i < rank; ++i) { + *output_shape->add_dim() = input_shape.dim(perm[i]); + } +} +``` + +## Key Utility Functions + +| Function | Purpose | +|----------|---------| +| `propagateElemTypeFromInputToOutput(ctx, in, out)` | Copy element type | +| `propagateShapeFromInputToOutput(ctx, in, out)` | Copy entire shape | +| `propagateShapeAndTypeFromFirstInput(ctx)` | Both type and shape from input 0 | +| `hasNInputShapes(ctx, n)` | Check first n inputs have shapes | +| `getOutputShape(ctx, out)` | Get mutable output shape | +| `bidirectionalBroadcastShapeInference(L, R, out)` | Numpy broadcasting | +| `getRepeatedAttribute(ctx, "name", vec)` | Get repeated attr values | +| `getAttribute(ctx, "name", default)` | Get single attr value | +| `mergeInDimensionInfo(src, dst, dim_idx)` | Merge dimension info | +| `fail_shape_inference("msg")` | Throw inference error | + +## Dimension Arithmetic + +```cpp +Dim operator*(const Dim& a, const Dim& b); +Dim operator*(const Dim& a, int64_t val); +Dim operator/(const Dim& a, int64_t divisor); +Dim multiplyDims(const TensorShapeProto& shape, int from, int upto); +``` + +## Writing Tests + +The `_make_graph` / `_assert_inferred` helpers are right for parameterized op-version sweeps: + +```python +@pytest.mark.parametrize("version", all_versions_for("OpName")) +def test_opname(self, version) -> None: + graph = self._make_graph( + [("X", TensorProto.FLOAT, (2, 3, 4))], + [make_node("OpName", ["X"], ["Y"], attr_name=attr_value)], + [], + ) + self._assert_inferred( + graph, + [make_tensor_value_info("Y", TensorProto.FLOAT, expected_shape)], + opset_imports=[helper.make_opsetid(ONNX_DOMAIN, version)], + ) +``` + +For one-off fixtures — anything with attributes, body subgraphs, or non-trivial type info — prefer the [`onnxtxt`](../onnxtxt/SKILL.md) skill's parser-based fixtures (it also covers the C++ `unk__*` materialization gotcha for free dims). + +Cover: known shapes, partial shapes (`None`), rank inference, error cases, broadcasting, attribute-dependent shapes. + +## Code Style: Prefer Named Functions + +Define inference functions as **separate named functions** rather than inline lambdas. The macro expansion makes breakpoints on inline lambdas unreliable. + +Short one-liners (e.g., `propagateShapeAndTypeFromFirstInput`) are fine as direct references. + +## Rules for Robust Inference + +1. Always check `hasNInputShapes(ctx, n)` before accessing shapes +2. Always check `has_dim_value()` before using `dim_value()` +3. Handle unknown dimensions gracefully — leave unset, don't fail +4. At minimum provide rank inference (correct number of output dims) +5. Propagate symbolic dimensions (`dim_param`) when possible + +## After Making Changes + +```bash +pytest onnx/test/shape_inference_test.py -k "test_opname" -x +python onnx/defs/gen_doc.py +lintrunner -a --output oneline +``` diff --git a/.agents/skills/onnxtxt/SKILL.md b/.agents/skills/onnxtxt/SKILL.md new file mode 100644 index 0000000..78e569f --- /dev/null +++ b/.agents/skills/onnxtxt/SKILL.md @@ -0,0 +1,121 @@ +--- +name: onnxtxt +description: Read or write ONNX text format ("onnxtxt"). Use when authoring `.FunctionBody(R"ONNX(...)")` blocks, writing tests with `onnx.parser.parse_model` / `parse_graph`, using the C++ `OnnxParser`, debugging parser errors, or interpreting `Constant ` and body-subgraph syntax. +--- + +ONNX has a compact text format implemented by `onnx/parser.py` (Python) and `onnx/defs/parser.{h,cc}` (C++). The formal grammar lives in [docs/Syntax.md](../../../docs/Syntax.md); this skill captures the practical conventions, idioms, and gotchas that matter when authoring or reviewing code that uses it. + +## Where the format appears + +| Surface | API | +|---|---| +| C++ function bodies | `.FunctionBody(R"ONNX( ... )ONNX")` and `FunctionBuilder::Add(...)` | +| Python test fixtures | `onnx.parser.parse_model("...")`, `onnx.parser.parse_graph("...")` | +| C++ tests | `OnnxParser` in `onnx/defs/parser.h` — parse a model, then call `shape_inference::InferShapes` | + +## Core syntax + +``` + = (, ) +``` + +- Variables on the LHS are local to the surrounding graph (function body, subgraph, or model main graph). +- Inputs/outputs of a function body must match the names declared in the schema's `.Input(...)` / `.Output(...)`. +- Constants: `Const = Constant ()` or `Alpha = Constant ()`. +- Use `CastLike` (not `Cast`) when the target dtype depends on another input. +- Reference enclosing-op attributes with `@attr_name` (only inside function bodies, and only for attributes declared on the schema). + +## Argument-order convention + +**Simple scalar/tensor attributes** — keep the conventional `Op(inputs)` form; reads well on one line: + +``` +Y = Transpose(X) +``` + +**Subgraph attributes** (Scan, Loop, If, ScanVarLen, …) — prefer `Op(inputs)`. The body spans multiple lines, so putting inputs first keeps the call site readable: + +``` +so, xo = Scan (s, x) < + num_scan_inputs = 1, + body = scan_body (float[1] s_in, float[1] x_in) => (float[1] s_out, float[1] x_out) { + s_out = Add(s_in, x_in) + x_out = Identity(x_in) + } +> +``` + +## Testing idioms + +| Test file | Recommendation | +|---|---| +| `onnx/test/shape_inference_test.py`, `onnx/test/reference_evaluator_test.py` | Use `onnx.parser.parse_model(...)` for one-off fixtures. | +| `onnx/backend/test/case/node/.py` | Keep the outer `helper.make_node` + `expect(...)` (it drives data generation). For body-subgraph ops, build the body with `onnx.parser.parse_graph`. | +| `onnx/test/cpp/shape_inference_test.cc` | Use `OnnxParser` (`onnx/defs/parser.h`); pair with `shape_inference::InferShapes`. | +| `onnx/test/version_converter/automatic_upgrade_test.py` and similar harnesses | Keep the established `_test_op_upgrade` / `_test_op_downgrade` style — do not rewrite. | + +Empirical: PR #7962 (ScanVarLen) cut ~58–70% of test LOC by switching to parser-based fixtures. + +### Python example — shape inference fixture + +```python +import onnx +import onnx.parser +import onnx.shape_inference + +model = onnx.parser.parse_model(""" + + g (float[2, 3, 4] X) => (float[4, 2, 3] Y) { + Y = Transpose(X) + } +""") +inferred = onnx.shape_inference.infer_shapes(model, strict_mode=True) +``` + +### Python example — body subgraph for a node test + +```python +body = onnx.parser.parse_graph(""" + b (float[1] s, float[1] xi) => (float[1] so, float[1] xo) { + so = Identity(s) + xo = Identity(xi) + } +""") +node = onnx.helper.make_node("Scan", ["s", "x"], ["so", "xo"], body=body, num_scan_inputs=1) +``` + +### C++ example — shape inference test + +```cpp +#include "onnx/defs/parser.h" +#include "onnx/shape_inference/implementation.h" + +ModelProto model; +OnnxParser parser(R"ONNX( + + g (float[2, 3, 4] X) => (Y) { + Y = Transpose(X) + } +)ONNX"); +auto status = parser.Parse(model); +ASSERT_TRUE(status.IsOK()) << status.ErrorMessage(); +shape_inference::InferShapes(model); +``` + +## Gotchas + +- **`unk__*` materialization in C++ shape-inference tests.** Under `InferShapes`, unset output dims are materialized by `MaterializeSymbolicShape` into `dim_param` names like `unk__0`, `unk__1`, … Assertions on free dims must accept either an unset dim or an `unk__*` placeholder — write (or use) a helper that treats both forms as equivalent. +- **Variable-name collisions.** Local variables in a function body must not reuse declared input/output names of the enclosing op. +- **`@attr_name` scope.** Only valid inside a function body, and only for attributes declared on the enclosing schema's `.Attr(...)` calls. +- **`CastLike` vs `Cast`.** Use `CastLike` when the desired target dtype is determined by another input; `Cast` requires a static `to` attribute. + +## References + +| Resource | Path | +|---|---| +| Formal grammar | [docs/Syntax.md](../../../docs/Syntax.md) | +| C++ parser | `onnx/defs/parser.h`, `onnx/defs/parser.cc` | +| Python parser | `onnx/parser.py` | +| C++ parser tests | `onnx/test/cpp/parser_test.cc` | +| Python parser tests | `onnx/test/parser_test.py` | +| Empirical LOC win | PR #7962 (ScanVarLen test rewrite) | diff --git a/.clang-format b/.clang-format new file mode 100644 index 0000000..c085b7b --- /dev/null +++ b/.clang-format @@ -0,0 +1,94 @@ +--- +Language: Cpp +BasedOnStyle: Google +AccessModifierOffset: -1 +AlignAfterOpenBracket: AlwaysBreak +AlignConsecutiveAssignments: false +AlignConsecutiveDeclarations: false +AlignEscapedNewlines: Left +AlignOperands: false +AlignTrailingComments: false +AllowAllParametersOfDeclarationOnNextLine: false +AllowShortBlocksOnASingleLine: false +AllowShortCaseLabelsOnASingleLine: false +AllowShortFunctionsOnASingleLine: Empty +AllowShortIfStatementsOnASingleLine: false +AllowShortLoopsOnASingleLine: false +AlwaysBreakAfterDefinitionReturnType: None +AlwaysBreakAfterReturnType: None +AlwaysBreakBeforeMultilineStrings: true +AlwaysBreakTemplateDeclarations: true +BinPackArguments: false +BinPackParameters: false +BraceWrapping: + AfterClass: false + AfterControlStatement: false + AfterEnum: false + AfterFunction: false + AfterNamespace: false + AfterObjCDeclaration: false + AfterStruct: false + AfterUnion: false + BeforeCatch: false + BeforeElse: false + IndentBraces: false +BreakBeforeBinaryOperators: None +BreakBeforeBraces: Attach +BreakBeforeInheritanceComma: false +BreakBeforeTernaryOperators: true +BreakConstructorInitializersBeforeComma: false +BreakConstructorInitializers: BeforeColon +BreakStringLiterals: false +ColumnLimit: 120 +CommentPragmas: '^ IWYU pragma:' +CompactNamespaces: false +ConstructorInitializerAllOnOneLineOrOnePerLine: true +ConstructorInitializerIndentWidth: 4 +ContinuationIndentWidth: 4 +Cpp11BracedListStyle: true +DerivePointerAlignment: false +DisableFormat: false +ExperimentalAutoDetectBinPacking: false +FixNamespaceComments: true +ForEachMacros: + - FOR_EACH_RANGE + - FOR_EACH + - BOOST_FOREACH +IncludeIsMainRegex: '(Test)?$' +IndentCaseLabels: true +IndentWidth: 2 +IndentWrappedFunctionNames: false +JavaScriptQuotes: Leave +JavaScriptWrapImports: true +KeepEmptyLinesAtTheStartOfBlocks: false +MacroBlockBegin: '' +MacroBlockEnd: '' +MaxEmptyLinesToKeep: 1 +NamespaceIndentation: None +ObjCBlockIndentWidth: 2 +ObjCSpaceAfterProperty: false +ObjCSpaceBeforeProtocolList: false +PenaltyBreakAssignment: 2 +PenaltyBreakBeforeFirstCallParameter: 1 +PenaltyBreakComment: 300 +PenaltyBreakFirstLessLess: 120 +PenaltyBreakString: 1000 +PenaltyExcessCharacter: 1000000 +PenaltyReturnTypeOnItsOwnLine: 200 +PointerAlignment: Left +ReflowComments: true +SortIncludes: true +SpaceAfterCStyleCast: false +SpaceAfterTemplateKeyword: true +SpaceBeforeAssignmentOperators: true +SpaceBeforeParens: ControlStatements +SpaceInEmptyParentheses: false +SpacesBeforeTrailingComments: 1 +SpacesInAngles: false +SpacesInContainerLiterals: true +SpacesInCStyleCastParentheses: false +SpacesInParentheses: false +SpacesInSquareBrackets: false +StatementMacros: [define_data, ONNX_ASSERT, ONNX_ASSERTM, TENSOR_ASSERTM] +TabWidth: 8 +UseTab: Never diff --git a/.clang-tidy b/.clang-tidy new file mode 100644 index 0000000..ac277c7 --- /dev/null +++ b/.clang-tidy @@ -0,0 +1,86 @@ +Checks: >- + -*, + bugprone-*, + -bugprone-easily-swappable-parameters, + -bugprone-exception-escape, + -bugprone-derived-method-shadowing-base-method, + -bugprone-lambda-function-name, + -bugprone-narrowing-conversions, + -bugprone-reserved-identifier, + -bugprone-throwing-static-initialization, + clang-diagnostic-*, + clang-analyzer-*, + cppcoreguidelines-*, + -cppcoreguidelines-avoid-c-arrays, + -cppcoreguidelines-avoid-const-or-ref-data-members, + -cppcoreguidelines-avoid-do-while, + -cppcoreguidelines-avoid-magic-numbers, + -cppcoreguidelines-avoid-non-const-global-variables, + -cppcoreguidelines-macro-usage, + -cppcoreguidelines-narrowing-conversions, + -cppcoreguidelines-non-private-member-variables-in-classes, + -cppcoreguidelines-owning-memory, + -cppcoreguidelines-pro-bounds-array-to-pointer-decay, + -cppcoreguidelines-pro-bounds-avoid-unchecked-container-access, + -cppcoreguidelines-pro-bounds-constant-array-index, + -cppcoreguidelines-pro-bounds-pointer-arithmetic, + -cppcoreguidelines-pro-type-const-cast, + -cppcoreguidelines-pro-type-cstyle-cast, + -cppcoreguidelines-pro-type-reinterpret-cast, + -cppcoreguidelines-pro-type-vararg, + -cppcoreguidelines-use-enum-class, + performance-*, + -performance-enum-size, + google-default-arguments, + google-global-names-in-headers, + google-explicit-constructor, + google-readability-casting, + misc-*, + -misc-const-correctness, + -misc-include-cleaner, + -misc-override-with-different-visibility, + -misc-no-recursion, + -misc-non-private-member-variables-in-classes, + -misc-use-anonymous-namespace, + modernize-*, + -modernize-avoid-c-arrays, + -modernize-concat-nested-namespaces, + -modernize-pass-by-value, + -modernize-return-braced-init-list, + -modernize-use-auto, + -modernize-use-constraints, + -modernize-use-integer-sign-comparison, + -modernize-use-ranges, + -modernize-use-trailing-return-type, + -modernize-use-nodiscard, + readability-*, + -readability-braces-around-statements, + -readability-container-contains, + -readability-convert-member-functions-to-static, + -readability-else-after-return, + -readability-function-cognitive-complexity, + -readability-identifier-length, + -readability-implicit-bool-conversion, + -readability-inconsistent-declaration-parameter-name, + -readability-isolate-declaration, + -readability-magic-numbers, + -readability-math-missing-parentheses, + -readability-redundant-access-specifiers, + -readability-redundant-typename, + -readability-uppercase-literal-suffix, + -readability-use-anyofallof, + +# Lint ONNX headers; exclude generated *.pb.h, third_party, deps, and hidden build/env dirs. +HeaderFilterRegex: 'onnx/' +ExcludeHeaderFilterRegex: '(\.pb\.h$|/third_party/|/_deps/|/\.)' + +CheckOptions: + # `cppcoreguidelines-special-member-functions` is enabled, refer to https://en.cppreference.com/w/cpp/language/rule_of_three + - key: cppcoreguidelines-special-member-functions.AllowSoleDefaultDtor + value: True + - key: performance-move-const-arg.CheckTriviallyCopyableMove + value: False + - key: cppcoreguidelines-special-member-functions.AllowMissingMoveFunctionsWhenCopyIsDeleted + value: True + - key: modernize-use-scoped-lock.WarnOnSingleLocks + value: False diff --git a/.claude/instructions/add-function-body.md b/.claude/instructions/add-function-body.md new file mode 100644 index 0000000..f02c3de --- /dev/null +++ b/.claude/instructions/add-function-body.md @@ -0,0 +1,13 @@ +# Adding a Function Body Definition for an Operator + +Canonical guide: [`.agents/skills/add-function-body/SKILL.md`](../../.agents/skills/add-function-body/SKILL.md). Background: [`docs/AddFunctionBody.md`](../../docs/AddFunctionBody.md). For the ONNX text format itself (syntax, `Constant `, body subgraphs, `@attr_name`, parser tests), see [`.agents/skills/onnxtxt/SKILL.md`](../../.agents/skills/onnxtxt/SKILL.md). + +## Workflow-specific reminders + +- Function body lives inline in the schema in `onnx/defs//defs.cc` via `.FunctionBody(R"ONNX(...)")` (simple) or `.SetContextDependentFunctionBodyBuilder(...)` (context-dependent). +- For context-dependent builders, always finalize with `schema.BuildFunction(functionProto)` and `return true`. +- The body must produce all declared outputs. Variable names must not collide with declared input/output names. Use `CastLike` (not `Cast`) when the target dtype depends on another input. Reference enclosing-op attributes with `@attr_name` — only for attributes declared in `.Attr(...)` calls. +- Prefer named `static bool` builder functions over inline lambdas (macro expansion breaks debugger breakpoints). Simple string-based `.FunctionBody(R"ONNX(...)")` is fine as-is. + +General build/lint/DCO/copyright conventions live in [`CLAUDE.md`](../../CLAUDE.md). + diff --git a/.claude/instructions/add-op.md b/.claude/instructions/add-op.md new file mode 100644 index 0000000..8af3354 --- /dev/null +++ b/.claude/instructions/add-op.md @@ -0,0 +1,12 @@ +# Adding or Updating an ONNX Operator + +Canonical guide: [`.agents/skills/add-op/SKILL.md`](../../.agents/skills/add-op/SKILL.md). Full procedure: [`docs/AddNewOp.md`](../../docs/AddNewOp.md). For tests using the ONNX text format / `onnx.parser`, see [`.agents/skills/onnxtxt/SKILL.md`](../../.agents/skills/onnxtxt/SKILL.md). + +## Workflow-specific reminders + +- Schema definition lives in `onnx/defs//defs.cc`; register in `onnx/defs/operator_sets.h`. +- Updating an op: move the old schema to `/old.cc` *before* adding the new version. Add a version-converter adapter in `onnx/version_converter/adapters/__.h` if behavior changed, plus upgrade/downgrade tests. +- After schema changes regenerate docs: `python onnx/defs/gen_doc.py && python onnx/backend/test/stat_coverage.py`. Do not edit `docs/Operators.md` / `docs/Changelog.md` directly. + +General build/lint/DCO/copyright conventions live in [`CLAUDE.md`](../../CLAUDE.md). + diff --git a/.claude/instructions/add-shape-inference.md b/.claude/instructions/add-shape-inference.md new file mode 100644 index 0000000..56e939d --- /dev/null +++ b/.claude/instructions/add-shape-inference.md @@ -0,0 +1,14 @@ +# Adding Type and Shape Inference for an Operator + +Canonical guide: [`.agents/skills/add-shape-inference/SKILL.md`](../../.agents/skills/add-shape-inference/SKILL.md). Background: [`docs/ShapeInference.md`](../../docs/ShapeInference.md). For test fixtures using `onnx.parser`, see [`.agents/skills/onnxtxt/SKILL.md`](../../.agents/skills/onnxtxt/SKILL.md) (also covers the C++ `unk__*` materialization gotcha for free dims). + +## Workflow-specific reminders + +- Inference function is inline in the schema via `.TypeAndShapeInferenceFunction(...)` in `onnx/defs//defs.cc`. Utility helpers live in `onnx/defs/shape_inference.h`. +- Always check `hasNInputShapes(ctx, n)` before accessing shapes and `has_dim_value()` before reading dim values. Leave unknown dims unset rather than failing. +- At minimum, provide rank inference; propagate symbolic dimensions (`dim_param`) when possible. +- Prefer named `static` inference functions over inline lambdas in `ONNX_OPERATOR_SET_SCHEMA` (macro expansion breaks debugger breakpoints). +- Tests: `onnx/test/shape_inference_test.py`. The `_make_graph` / `_assert_inferred` helpers fit parameterized op-version sweeps; for one-off fixtures prefer `onnx.parser.parse_model`. + +General build/lint/DCO/copyright conventions live in [`CLAUDE.md`](../../CLAUDE.md). + diff --git a/.editorconfig b/.editorconfig new file mode 100644 index 0000000..16df8da --- /dev/null +++ b/.editorconfig @@ -0,0 +1,7 @@ +root = true + +[*] + +trim_trailing_whitespace = true +insert_final_newline = true +indent_style = space diff --git a/.git-blame-ignore-revs b/.git-blame-ignore-revs new file mode 100644 index 0000000..5af302e --- /dev/null +++ b/.git-blame-ignore-revs @@ -0,0 +1,8 @@ +# Python clean up #3982 +35092895d9bf3592e58f4710d098f8131afef259 +# Apply clang-format #4084 +a525f98d5ed31660b629bab90c680a39658a5c08 +# Upgrade python syntax with pyupgrade #4212 +529f7cab88eed143fc9f0126147e9732585e4c5e +# Format all python code with black and isort #4427 +fddb2b6d4ea3fb3dba751d884865042503260899 diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000..255168e --- /dev/null +++ b/.gitattributes @@ -0,0 +1,7 @@ +# Set the default behavior +* text=auto eol=auto +*.bat text eol=crlf +# Shell scripts must keep LF endings on all platforms; a CRLF checkout breaks +# shellcheck (SC1017) and execution under Unix shells. +*.sh text eol=lf +*.pb -linguist-detectable diff --git a/.github/ISSUE_TEMPLATE/bug.md b/.github/ISSUE_TEMPLATE/bug.md new file mode 100644 index 0000000..b35b41f --- /dev/null +++ b/.github/ISSUE_TEMPLATE/bug.md @@ -0,0 +1,42 @@ +--- +name: Bug report +about: Create a bug report to help improve the ONNX. +title: '' +labels: 'bug' +assignees: '' + +--- +# Bug Report + +### Is the issue related to model conversion? + + +### Describe the bug + + +### System information + + + +### Reproduction instructions + + +### Expected behavior + + +### Notes + diff --git a/.github/ISSUE_TEMPLATE/config.yml b/.github/ISSUE_TEMPLATE/config.yml new file mode 100644 index 0000000..8a5e0c8 --- /dev/null +++ b/.github/ISSUE_TEMPLATE/config.yml @@ -0,0 +1,14 @@ +blank_issues_enabled: true +contact_links: + - name: ONNX Runtime Issues + url: https://github.com/microsoft/onnxruntime/issues + about: issues/questions related to ONNX Runtime + - name: ONNX Model Zoo Issues + url: https://github.com/onnx/models/issues + about: issues/questions related to ONNX Model Zoo + - name: PyTorch-ONNX exporters Issues + url: https://github.com/pytorch/pytorch/issues + about: issues/questions related to PyTorch-ONNX exporters (torch.onnx.export) + - name: TensorFlow Converter Issues + url: https://github.com/onnx/tensorflow-onnx + about: issues/questions converting TensorFlow models to ONNX (tf2onnx) diff --git a/.github/ISSUE_TEMPLATE/feature_request.yml b/.github/ISSUE_TEMPLATE/feature_request.yml new file mode 100644 index 0000000..e943212 --- /dev/null +++ b/.github/ISSUE_TEMPLATE/feature_request.yml @@ -0,0 +1,71 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Feature request +description: Create a feature request for a functionality that does not currently exist in the ONNX. +title: "[Feature request] " +labels: ["enhancement"] +body: + - type: markdown + attributes: + value: Thanks for taking the time to create a feature request! + - type: textarea + id: system-info + attributes: + label: System information + description: "ONNX version (you are using):" + validations: + required: false + - type: textarea + id: solves-problem + attributes: + label: What is the problem that this feature solves? + description: Please detail the discrepancy with our current functionality. + validations: + required: false + - type: textarea + id: alternatives + attributes: + label: Alternatives considered + description: Describe the alternatives you have considered + placeholder: A clear and concise description of any alternative solutions or features you've considered. + validations: + required: false + - type: textarea + id: feature + attributes: + label: Describe the feature + description: Why is this feature necessary? What does it accomplish? + validations: + required: false + - type: textarea + id: api-impact + attributes: + label: Will this influence the current api (Y/N)? + placeholder: If yes, how? + validations: + required: false + - type: textarea + id: feature-area + attributes: + label: Feature Area + description: Which area in ONNX does this impact? e.g., model usage, backend, best practices, converters, shape_inference, version_converter, training, test, operators, IR, ONNX Hub, data preprocessing, CI pipelines. + validations: + required: false + - type: dropdown + id: contribute + attributes: + label: "Are you willing to contribute it (Y/N)" + options: + - "Yes" + - "No" + validations: + required: false + - type: textarea + id: notes + attributes: + label: Notes + description: Any additional information + validations: + required: false diff --git a/.github/ISSUE_TEMPLATE/operator.md b/.github/ISSUE_TEMPLATE/operator.md new file mode 100644 index 0000000..968a2be --- /dev/null +++ b/.github/ISSUE_TEMPLATE/operator.md @@ -0,0 +1,24 @@ +--- +name: New Operator +about: Create a new operator that does not currently exist in the ONNX. +title: '' +labels: 'operator' +assignees: '' + + + +--- +# New Operator + +### Describe the operator + + +### Can this operator be constructed using existing onnx operators? + + +### Is this operator used by any model currently? Which one? + +### Are you willing to contribute it? (Y/N) + +### Notes + diff --git a/.github/ISSUE_TEMPLATE/question.md b/.github/ISSUE_TEMPLATE/question.md new file mode 100644 index 0000000..186afc8 --- /dev/null +++ b/.github/ISSUE_TEMPLATE/question.md @@ -0,0 +1,24 @@ +--- +name: Question +about: Ask a question about the ONNX. +title: '' +labels: 'question' +assignees: '' + + + +--- +# Ask a Question + +### Question + + +### Further information +- Relevant Area: + +- Is this issue related to a specific model? +**Model name**: +**Model opset**: + +### Notes + diff --git a/.github/copilot-instructions.md b/.github/copilot-instructions.md new file mode 100644 index 0000000..5bafbfd --- /dev/null +++ b/.github/copilot-instructions.md @@ -0,0 +1,17 @@ +We use lintrunner as the linter: + +```sh +# Display all lints and apply the fixes +lintrunner -a --output oneline +# Or apply fixes only (faster) +lintrunner f --output oneline +``` + +To build ONNX: + +```sh +python -m pip install --quiet --upgrade pip setuptools wheel +export ONNX_BUILD_TESTS=0 +export ONNX_ML=1 +python -m pip install . +``` diff --git a/.github/dependabot.yml b/.github/dependabot.yml new file mode 100644 index 0000000..c634e4c --- /dev/null +++ b/.github/dependabot.yml @@ -0,0 +1,40 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +# To get started with Dependabot version updates, you'll need to specify which +# package ecosystems to update and where the package manifests are located. +# Please see the documentation for all configuration options: +# https://docs.github.com/github/administering-a-repository/configuration-options-for-dependency-updates + +version: 2 +updates: + - package-ecosystem: "pip" # See documentation for possible values + directory: "/" # Location of package manifests + schedule: + interval: "monthly" + cooldown: + default-days: 7 + ignore: + # Only update them manually since updating them might break compatibility + - dependency-name: "numpy" + - dependency-name: "protobuf" + - dependency-name: "typing_extensions" + - dependency-name: "ml_dtypes" + open-pull-requests-limit: 10 + labels: + - "python" + - "run release CIs" + + - package-ecosystem: "github-actions" + # Workflow files stored in the + # default location of `.github/workflows` + directory: "/" + schedule: + interval: "monthly" + cooldown: + default-days: 7 + open-pull-requests-limit: 20 + labels: + - "github_actions" + - "run release CIs" diff --git a/.github/install_test.yml b/.github/install_test.yml new file mode 100644 index 0000000..27241db --- /dev/null +++ b/.github/install_test.yml @@ -0,0 +1,44 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Test (pip install -e .) + +on: + push: + branches: + - main + pull_request: + branches: + - main + workflow_dispatch: + +permissions: + contents: read + +jobs: + test-install: + runs-on: ${{ matrix.os }} + strategy: + matrix: + os: [ubuntu-24.04] + python-version: ['3.13'] + fail-fast: false + + steps: + - name: Checkout code + uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 + with: + persist-credentials: false + + - name: Set up Python + uses: actions/setup-python@v5 + with: + python-version: ${{ matrix.python-version }} + + - name: Install dependencies, build ONNX + run: | + python -m pip install -q --upgrade pip + python -m pip install scikit-build-core "protobuf==4.25.1" + pip install -e . + diff --git a/.github/instructions/onnx-defs.instructions.md b/.github/instructions/onnx-defs.instructions.md new file mode 100644 index 0000000..b049a1e --- /dev/null +++ b/.github/instructions/onnx-defs.instructions.md @@ -0,0 +1,28 @@ +--- +applyTo: "onnx/defs/**" +--- + +# ONNX `onnx/defs/` Quick Reference + +When working in `onnx/defs/**`, follow the relevant agent skill — these contain the full, canonical guidance: + +| Task | Skill | +|---|---| +| Add a new op or bump an opset | [`.agents/skills/add-op/SKILL.md`](../../.agents/skills/add-op/SKILL.md) | +| Implement type/shape inference | [`.agents/skills/add-shape-inference/SKILL.md`](../../.agents/skills/add-shape-inference/SKILL.md) | +| Add a function body decomposition | [`.agents/skills/add-function-body/SKILL.md`](../../.agents/skills/add-function-body/SKILL.md) | +| Write or interpret ONNX text format / `.FunctionBody(R"ONNX(...)")` / `onnx.parser.parse_model` | [`.agents/skills/onnxtxt/SKILL.md`](../../.agents/skills/onnxtxt/SKILL.md) | + +## Critical reminders (apply to all `onnx/defs/` work) + +- **Do not edit generated files** directly: `docs/Operators.md`, `docs/Changelog.md`, `docs/TestCoverage.md`, `*_pb2.py`, `onnx-*.in.proto`-derived `.proto` files. Edit the source, then regenerate. +- **After schema changes:** `python onnx/defs/gen_doc.py && python onnx/backend/test/stat_coverage.py`. +- **After proto changes:** `python onnx/gen_proto.py` (edit the `.in.proto` files, not the `.proto` files). +- **Lint before pushing:** `lintrunner -a --output oneline`. +- **DCO sign-off** on every commit: `git commit -s`. +- **New Python files** need `from __future__ import annotations` and the standard copyright header (`# Copyright (c) ONNX Project Contributors` + `# SPDX-License-Identifier: Apache-2.0`). +- **Updating an op:** move the old schema to `/old.cc` before adding the new version to `defs.cc`; update `onnx/defs/operator_sets.h`; add upgrade/downgrade tests. +- **Prefer named functions** over inline lambdas in `ONNX_OPERATOR_SET_SCHEMA` (macro expansion makes breakpoints on lambdas unreliable). +- **Context-dependent function bodies:** always finalize the builder with `schema.BuildFunction(functionProto)` and `return true`. + +For the full procedures (file locations, registration patterns, common idioms, test recipes), open the skills above. diff --git a/.github/pull_request_template.md b/.github/pull_request_template.md new file mode 100644 index 0000000..3123b55 --- /dev/null +++ b/.github/pull_request_template.md @@ -0,0 +1,5 @@ + + +### Motivation and Context + +Fixes # diff --git a/.github/release.yml b/.github/release.yml new file mode 100644 index 0000000..ad3de36 --- /dev/null +++ b/.github/release.yml @@ -0,0 +1,43 @@ +changelog: + exclude: + authors: + - dependabot + categories: + - title: Breaking Changes and Deprecations + labels: + - "topic: bc breaking" + - "topic: deprecation" + - title: Spec and Operator + labels: + - "module: spec" + - "topic: spec clarification" + - "topic: operator" + - "module: schema" + - title: Reference Implementation + labels: + - "module: reference implementation" + - title: Utilities and Tools + labels: + - "module: checker" + - "module: parser" + - "module: inliner" + - "module: utility" + - "module: optimizer" + - "module: version converter" + - "module: shape inference" + - "topic: partial data propagation" + - title: ONNX Hub + labels: + - "module: hub" + - title: Build, CI and Tests + labels: + - "topic: build" + - "module: CI pipelines" + - "topic: test" + - "topic: compiler warning" + - title: Documentation + labels: + - "topic: documentation" + - title: Other Changes + labels: + - "*" diff --git a/.github/workflows/check_pr_label.yml b/.github/workflows/check_pr_label.yml new file mode 100644 index 0000000..3ad545e --- /dev/null +++ b/.github/workflows/check_pr_label.yml @@ -0,0 +1,43 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Require topic/module label (prefix-only) + +on: + pull_request: + types: [opened, labeled, unlabeled, synchronize, reopened, ready_for_review] + +permissions: + contents: read + pull-requests: read + actions: read + +jobs: + check-label: + runs-on: ubuntu-latest + steps: + - name: Ensure PR has topic/module label + uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0 + with: + script: | + const author = context.payload.pull_request.user.login; + const prLabels = context.payload.pull_request.labels.map(l => l.name); + + if (author === "dependabot[bot]") { + core.info("Skipping check for dependabot"); + return; + } + + if (prLabels.length === 0) { + core.setFailed("❌ Add at least one label (topic:/module:)."); + return; + } + + const hasRequired = prLabels.some(l => l.startsWith("topic:") || l.startsWith("module:")); + + if (!hasRequired) { + core.setFailed("❌ Add a 'topic:' or 'module:' label."); + } else { + core.info("✅ Label check passed"); + } diff --git a/.github/workflows/check_urls.yml b/.github/workflows/check_urls.yml new file mode 100644 index 0000000..ead7ff6 --- /dev/null +++ b/.github/workflows/check_urls.yml @@ -0,0 +1,47 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Check URLs + +on: + push: + branches: [ "main", rel-* ] + pull_request: + paths: + - "**/*.md" + - ".github/workflows/check_urls.yml" + schedule: + # Run every month + - cron: '0 0 1 * *' + workflow_dispatch: + +permissions: + contents: read + +jobs: + build: + runs-on: ubuntu-latest + + steps: + - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + - name: Set up pixi + uses: prefix-dev/setup-pixi@5185adfbffb4bd703da3010310260805d89ebb11 # v0.9.6 + with: + environments: link-checker + + - name: Restore lychee cache + id: restore-cache + uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0 + with: + path: .lycheecache + # We don't hit the 'key' directly in subsequent + # commits. Instead we fall back to the latest 'restore-key' + # match and will store a new cached file when done. + key: cache-lychee-${{ github.sha }} + restore-keys: cache-lychee- + + - name: Check links + run: pixi run -e link-checker lychee --root-dir . --cache-exclude-status=400..599 --cache . diff --git a/.github/workflows/clang-tidy.yml b/.github/workflows/clang-tidy.yml new file mode 100644 index 0000000..9e6f2a3 --- /dev/null +++ b/.github/workflows/clang-tidy.yml @@ -0,0 +1,40 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: clang-tidy + +on: + pull_request: + push: + branches: + - main + +permissions: + contents: read + +concurrency: + group: ${{ github.workflow }}-${{ github.ref }} + cancel-in-progress: true + +jobs: + clang-tidy: + name: clang-tidy + runs-on: ubuntu-24.04-arm + strategy: + fail-fast: false + steps: + - name: Checkout branch + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + submodules: recursive + fetch-depth: 0 + persist-credentials: false + - name: Set up pixi + uses: prefix-dev/setup-pixi@5185adfbffb4bd703da3010310260805d89ebb11 # v0.9.6 + with: + environments: default clang-tools + - name: Install repository + run: pixi run install + - name: Run clang-tidy + run: pixi run run-clang-tidy diff --git a/.github/workflows/codeql.yml b/.github/workflows/codeql.yml new file mode 100644 index 0000000..0c67efc --- /dev/null +++ b/.github/workflows/codeql.yml @@ -0,0 +1,88 @@ +name: "CodeQL" + +on: + push: + branches: [ "main" ] + paths-ignore: + - 'docs/**' + - '**/*.md' + pull_request: + branches: [ "main" ] + paths-ignore: + - 'docs/**' + - '**/*.md' + schedule: + - cron: '33 6 * * 5' + workflow_dispatch: + +permissions: # set top-level default permissions as security best practice + contents: read + +concurrency: + group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name == 'workflow_dispatch' }} + cancel-in-progress: true + +jobs: + analyze: + name: Analyze + runs-on: ubuntu-latest + permissions: + actions: read + contents: read + security-events: write + + strategy: + fail-fast: false + matrix: + language: [ 'actions', 'cpp', 'python' ] + + steps: + - name: Checkout repository + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + submodules: recursive + persist-credentials: false + + - name: Set up Python + if: matrix.language != 'actions' + uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: '3.11' + + - name: Upgrade pip + if: matrix.language != 'actions' + run: python -m pip install --upgrade pip + + # Initializes the CodeQL tools for scanning. + - name: Initialize CodeQL + uses: github/codeql-action/init@8aad20d150bbac5944a9f9d289da16a4b0d87c1e # v4.36.2 + with: + languages: ${{ matrix.language }} + config: | + queries: + - uses: security-extended + - uses: security-and-quality + paths-ignore: + - '.setuptools-cmake-build/**' + - 'build/**' + - 'onnx/**/*_pb2.py' + - 'onnx/**/*_pb2.pyi' + - '**/*.pb.cc' + - '**/*.pb.h' + - 'onnx/backend/test/data/**' + query-filters: + - exclude: + id: py/import-and-import-from + + # Install onnx so that it is found by the linters + - name: Install ONNX + if: matrix.language != 'actions' + run: | + export ONNX_ML=1 + export ONNX_BUILD_TESTS=1 + pip install . + + - name: Perform CodeQL Analysis + uses: github/codeql-action/analyze@8aad20d150bbac5944a9f9d289da16a4b0d87c1e # v4.36.2 + with: + category: "/language:${{matrix.language}}" diff --git a/.github/workflows/copilot-setup-steps.yml b/.github/workflows/copilot-setup-steps.yml new file mode 100644 index 0000000..8e298fc --- /dev/null +++ b/.github/workflows/copilot-setup-steps.yml @@ -0,0 +1,47 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: "Copilot Setup Steps" + +# Automatically run the setup steps when they are changed to allow for easy validation, and +# allow manual testing through the repository's "Actions" tab +on: + workflow_dispatch: + push: + paths: + - .github/workflows/copilot-setup-steps.yml + pull_request: + paths: + - .github/workflows/copilot-setup-steps.yml + +permissions: + contents: read + +jobs: + # The job MUST be called `copilot-setup-steps` or it will not be picked up by Copilot. + copilot-setup-steps: + runs-on: ubuntu-latest + + # Set the permissions to the lowest permissions possible needed for your steps. + # Copilot will be given its own token for its operations. + permissions: + # If you want to clone the repository as part of your setup steps, for example to install dependencies, you'll need the `contents: read` permission. If you don't clone the repository in your setup steps, Copilot will do this for you automatically after the steps complete. + contents: read + # You can define any steps you want, and they will run before the agent starts. + # If you do not check out your code, Copilot will do this for you. + steps: + - name: Checkout code + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + - name: Setup Python + uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: "3.13" + - name: Install dependencies + run: | + python -m pip install lintrunner>=0.10.7 + python -m pip install -r requirements-release_test.txt + python -m pip install -r requirements-lintrunner.txt + lintrunner init diff --git a/.github/workflows/create_release.yml b/.github/workflows/create_release.yml new file mode 100644 index 0000000..ef5b909 --- /dev/null +++ b/.github/workflows/create_release.yml @@ -0,0 +1,437 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Create Releases +on: + schedule: + # Run weekly on Monday 00:00 UTC + - cron: '0 0 * * MON' + + push: + branches: [main, rel-*] + pull_request: + branches: [main, rel-*] + types: + - labeled + - synchronize + workflow_dispatch: + inputs: + publish_pypi_weekly: # only from main branch it is possible to publish to pypi-weekly (official weekly preview build) + description: 'Publish to pypi-weekly' + required: true + type: choice + options: + - 'yes' + - 'no' + default: 'no' + publish_testpypi_weekly: # only from main branch it is possible to publish to testpypi-weekly + description: 'Publish to testpypi-weekly' + required: true + type: choice + options: + - 'yes' + - 'no' + default: 'no' + publish_testpypi_release: # only from rel branch it is possible to publish to test-pypi (for rc1, rc2, etc.) + description: 'Publish to testpypi-release' + required: true + type: choice + options: + - 'yes' + - 'no' + default: 'no' + publish_pypi_release: + description: 'Caution: Publish to pypi-release' + required: true + type: choice + options: + - 'yes' + - 'no' + default: 'no' + build_mode: + description: 'Specify the build mode (release or preview)' + required: true + type: choice + options: + - 'release' + - 'preview' + default: 'preview' + +permissions: + contents: read + +concurrency: + group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name == 'workflow_dispatch' }} + cancel-in-progress: true + +jobs: + + call-linux: + if: github.event_name != 'pull_request' || contains(github.event.pull_request.labels.*.name, 'run release CIs') || contains(github.event.pull_request.labels.*.name, 'ci-slsa-provenance') + uses: ./.github/workflows/release_linux_cibw.yml + with: + build_mode: ${{ github.event.inputs.build_mode || 'preview' }} + + call-win: + if: github.event_name != 'pull_request' || contains(github.event.pull_request.labels.*.name, 'run release CIs') || contains(github.event.pull_request.labels.*.name, 'ci-slsa-provenance') + uses: ./.github/workflows/release_windows_cibw.yml + with: + build_mode: ${{ github.event.inputs.build_mode || 'preview' }} + + call-mac: + if: github.event_name != 'pull_request' || contains(github.event.pull_request.labels.*.name, 'run release CIs') || contains(github.event.pull_request.labels.*.name, 'ci-slsa-provenance') + uses: ./.github/workflows/release_macos_cibw.yml + with: + build_mode: ${{ github.event.inputs.build_mode || 'preview' }} + + call-pyodide: + if: github.event_name != 'pull_request' || contains(github.event.pull_request.labels.*.name, 'run release CIs') || contains(github.event.pull_request.labels.*.name, 'ci-slsa-provenance') + uses: ./.github/workflows/release_pyodide_cibw.yml + with: + build_mode: ${{ github.event.inputs.build_mode || 'preview' }} + + call-sdist: + if: github.event_name != 'pull_request' || contains(github.event.pull_request.labels.*.name, 'run release CIs') || contains(github.event.pull_request.labels.*.name, 'ci-slsa-provenance') + uses: ./.github/workflows/release_sdist.yml + with: + os: "macos" + build_mode: ${{ github.event.inputs.build_mode || 'preview' }} + + attest_ci_build_artifacts: + name: Attest CI build artifacts + runs-on: ubuntu-latest + needs: [call-linux, call-mac, call-win, call-pyodide, call-sdist] + if: >- + (!contains(join(needs.*.result, ' '), 'skipped')) && + (github.repository_owner == 'onnx') && + ( + (github.event_name == 'push') || + (github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'ci-slsa-provenance')) || + ( + (github.event_name == 'workflow_dispatch') && + (github.event.inputs.publish_pypi_weekly != 'yes') && + (github.event.inputs.publish_testpypi_weekly != 'yes') && + (github.event.inputs.publish_testpypi_release != 'yes') && + (github.event.inputs.publish_pypi_release != 'yes') + ) + ) + + permissions: + contents: read + id-token: write + attestations: write + + steps: + - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c + with: + pattern: wheels* + path: dist + merge-multiple: true + + - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c + with: + pattern: sdist + path: dist + merge-multiple: true + + - name: Generate SLSA Build Provenance attestations + if: hashFiles('dist/**') != '' + id: attest + uses: actions/attest-build-provenance@a2bbfa25375fe432b6a289bc6b6cd05ecd0c4c32 # v4.1.0 + with: + subject-path: dist/** + + - name: Upload attestation bundle + if: hashFiles('dist/**') != '' + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: attestation-bundle-ci + path: ${{ steps.attest.outputs.bundle-path }} + + check_for_publish_release_build_to_pypi: + name: Check for Publish release build to pypi + runs-on: ubuntu-latest + + needs: [call-linux, call-mac, call-win, call-pyodide, call-sdist] + if: (!contains(join(needs.*.result, ' '), 'skipped')) && (github.event.inputs.publish_pypi_release == 'yes') && (github.repository_owner == 'onnx') && startsWith(github.ref, 'refs/heads/rel-') && (github.event_name == 'workflow_dispatch') + + steps: + + - name: Ensure build mode is release + run: | + if [ "$BUILD_MODE" != "release" ]; then + echo "Error: build_mode must be set to 'release' to proceed." + exit 1 + fi + env: + BUILD_MODE: ${{ github.event.inputs.build_mode }} + + - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + submodules: true + + - name: Check if package_version matches branch + run: | + branch_version=${GITHUB_REF#refs/heads/rel-} + package_version=$(cat VERSION_NUMBER) + echo "Branch version: $branch_version" + echo "Package version: $package_version" + + if [[ "$package_version" != "$branch_version" && "$package_version" != "$branch_version"rc* ]]; then + echo "Error: Package version ($package_version) does not match branch version ($branch_version) or expected RC format." + exit 1 + fi + + check_for_publish_preview_build_to_testpypi_weekly: + name: Check for Publish preview build to test.pypi-weekly + runs-on: ubuntu-latest + + needs: [call-linux, call-mac, call-win, call-pyodide, call-sdist] + if: (!contains(join(needs.*.result, ' '), 'skipped')) && (github.event.inputs.publish_testpypi_weekly == 'yes') && (github.ref == 'refs/heads/main') && (github.repository_owner == 'onnx') && (github.event_name == 'workflow_dispatch') + + steps: + - name: Confirm preview publish request + run: echo "Proceeding with test.pypi-weekly publish checks." + + + publish_preview_build_to_testpypi_weekly: + name: Publish preview build to test.pypi-weekly + runs-on: ubuntu-latest + needs: [check_for_publish_preview_build_to_testpypi_weekly] + + environment: + name: testpypi-weekly + url: https://test.pypi.org/p/onnx-weekly + + permissions: + contents: read + id-token: write + attestations: write + + steps: + + - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c + if: (github.event_name == 'workflow_dispatch' ) + with: + pattern: wheels* + path: dist + merge-multiple: true + + - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c + if: (github.event_name == 'workflow_dispatch' ) + with: + pattern: sdist + path: dist + merge-multiple: true + + - name: Generate SLSA Build Provenance attestations + if: hashFiles('dist/**') != '' + id: attest + uses: actions/attest-build-provenance@a2bbfa25375fe432b6a289bc6b6cd05ecd0c4c32 # v4.1.0 + with: + subject-path: dist/** + + - name: Upload attestation bundle + if: hashFiles('dist/**') != '' + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: attestation-bundle-testpypi-weekly + path: ${{ steps.attest.outputs.bundle-path }} + + - name: Upload preview build to test.pypi + if: (github.ref == 'refs/heads/main') && (github.event.inputs.publish_testpypi_weekly == 'yes') && (github.repository_owner == 'onnx') + id: upload_preview_build_to_testpypi_weekly + + uses: pypa/gh-action-pypi-publish@cef221092ed1bacb1cc03d23a2d87d1d172e277b + with: + repository-url: https://test.pypi.org/legacy/ + verbose: true + print-hash: true + + check_for_publish_release_build_to_testpypi: + name: Check for Publish release build to test.pypi (rc-candidates) + runs-on: ubuntu-latest + + needs: [call-linux, call-mac, call-win, call-pyodide, call-sdist] + if: (!contains(join(needs.*.result, ' '), 'skipped')) && (github.event.inputs.publish_testpypi_release == 'yes') && startsWith(github.ref, 'refs/heads/rel') && (github.repository_owner == 'onnx') && (github.event_name == 'workflow_dispatch') + + steps: + - name: Confirm release publish request + run: echo "Proceeding with test.pypi release publish checks." + + publish_release_build_to_testpypi: + name: Publish release build to test.pypi + runs-on: ubuntu-latest + needs: [check_for_publish_release_build_to_testpypi] + + environment: + name: testpypi-release + url: https://test.pypi.org/p/onnx + + permissions: + contents: read + id-token: write + attestations: write + + steps: + + - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c + if: (github.event_name == 'workflow_dispatch' ) + with: + pattern: wheels* + path: dist + merge-multiple: true + + - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c + if: (github.event_name == 'workflow_dispatch' ) + with: + pattern: sdist + path: dist + merge-multiple: true + + - name: Generate SLSA Build Provenance attestations + if: hashFiles('dist/**') != '' + id: attest + uses: actions/attest-build-provenance@a2bbfa25375fe432b6a289bc6b6cd05ecd0c4c32 # v4.1.0 + with: + subject-path: dist/** + + - name: Upload attestation bundle + if: hashFiles('dist/**') != '' + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: attestation-bundle-testpypi-release + path: ${{ steps.attest.outputs.bundle-path }} + + - name: Upload release build to test.pypi + id: upload_release_build_to_testpypi + + uses: pypa/gh-action-pypi-publish@cef221092ed1bacb1cc03d23a2d87d1d172e277b + with: + repository-url: https://test.pypi.org/legacy/ + verbose: true + print-hash: true + + + check_for_publish_preview_build_to_pypi_weekly: + name: Check for Publish preview build to pypi-weekly + runs-on: ubuntu-latest + + needs: [call-linux, call-mac, call-win, call-pyodide, call-sdist] + if: (!contains(join(needs.*.result, ' '), 'skipped')) && (github.event_name == 'schedule' || github.event.inputs.publish_pypi_weekly == 'yes') && (github.repository_owner == 'onnx') + + steps: + - name: Confirm weekly publish request + run: echo "Proceeding with pypi-weekly publish checks." + + publish_preview_build_to_pypi_weekly: + name: Publish preview build to pypi-weekly + runs-on: ubuntu-latest + needs: [check_for_publish_preview_build_to_pypi_weekly] + + environment: + name: pypi-weekly + url: https://pypi.org/p/onnx-weekly + + permissions: + contents: read + id-token: write + attestations: write + + steps: + - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c + if: (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch') + with: + pattern: wheels* + path: dist + merge-multiple: true + + - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c + if: (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch') + with: + pattern: sdist + path: dist + merge-multiple: true + + - name: Generate SLSA Build Provenance attestations + if: hashFiles('dist/**') != '' + id: attest + uses: actions/attest-build-provenance@a2bbfa25375fe432b6a289bc6b6cd05ecd0c4c32 # v4.1.0 + with: + subject-path: dist/** + + - name: Upload attestation bundle + if: hashFiles('dist/**') != '' + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: attestation-bundle-weekly + path: ${{ steps.attest.outputs.bundle-path }} + + - name: Upload preview_build to pypi-weekly + id: upload_preview_build_to_pypi_weekly + if: (github.ref == 'refs/heads/main') + uses: pypa/gh-action-pypi-publish@cef221092ed1bacb1cc03d23a2d87d1d172e277b + with: + repository-url: https://upload.pypi.org/legacy/ + verbose: true + print-hash: true + + + publish_release_build_to_pypi: + name: Publish release build to pypi + runs-on: ubuntu-latest + needs: [check_for_publish_release_build_to_pypi] + + environment: + name: pypi-release + url: https://pypi.org/p/onnx + + permissions: + contents: read + id-token: write + attestations: write + + steps: + - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c + if: (github.event_name == 'workflow_dispatch') + with: + pattern: wheels* + path: dist + merge-multiple: true + + - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c + if: (github.event_name == 'workflow_dispatch') + with: + pattern: sdist + path: dist + merge-multiple: true + + - name: Generate SLSA Build Provenance attestations + if: hashFiles('dist/**') != '' + id: attest + uses: actions/attest-build-provenance@a2bbfa25375fe432b6a289bc6b6cd05ecd0c4c32 # v4.1.0 + with: + subject-path: dist/** + + - name: Upload attestation bundle + if: hashFiles('dist/**') != '' + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: attestation-bundle-release + path: ${{ steps.attest.outputs.bundle-path }} + + - name: Publish release_build to pypi + if: (github.repository_owner == 'onnx') + uses: pypa/gh-action-pypi-publish@cef221092ed1bacb1cc03d23a2d87d1d172e277b + with: + repository-url: https://upload.pypi.org/legacy/ + verbose: true + print-hash: true + + test_source_dist: + name: test source distribution + needs: [publish_preview_build_to_pypi_weekly, publish_release_build_to_testpypi] + if: (needs.publish_preview_build_to_pypi_weekly.result == 'success' || needs.publish_release_build_to_testpypi.result == 'success') + uses: ./.github/workflows/preview_source_dist_test.yml diff --git a/.github/workflows/dco_merge_group.yml b/.github/workflows/dco_merge_group.yml new file mode 100644 index 0000000..9719eca --- /dev/null +++ b/.github/workflows/dco_merge_group.yml @@ -0,0 +1,17 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: DCO +on: + merge_group: + +permissions: # set top-level default permissions as security best practice + contents: read # Check https://github.com/ossf/scorecard/blob/7ce8609469289d5f3b1bf5ee3122f42b4e3054fb/docs/checks.md#token-permissions + +jobs: + DCO: + runs-on: ubuntu-latest + if: ${{ github.actor != 'dependabot[bot]' }} + steps: + - run: echo "dummy DCO workflow (it won't run any check actually) to trigger by merge_group in order to enable merge queue" diff --git a/.github/workflows/dependency-review.yml b/.github/workflows/dependency-review.yml new file mode 100644 index 0000000..50acb44 --- /dev/null +++ b/.github/workflows/dependency-review.yml @@ -0,0 +1,22 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: 'Dependency Review' +on: [pull_request] + +permissions: + contents: read + +jobs: + dependency-review: + runs-on: ubuntu-latest + steps: + - name: 'Checkout Repository' + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + with: + submodules: recursive + persist-credentials: false + + - name: 'Dependency Review' + uses: actions/dependency-review-action@dcd589ca9f7a6ded22e224ca2e288beb6bf9846b diff --git a/.github/workflows/fuzz.yml b/.github/workflows/fuzz.yml new file mode 100644 index 0000000..80783c1 --- /dev/null +++ b/.github/workflows/fuzz.yml @@ -0,0 +1,129 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +# Runs the OSS-Fuzz harnesses under onnx/fuzz/ directly (not via OSS-Fuzz +# infrastructure) as a regression check. This is deliberately not a +# replacement for OSS-Fuzz's own continuous, long-running campaigns +# (see google/oss-fuzz#15382) — it exists to catch harness regressions +# (import errors, API drift, a harness file going missing) at PR time +# instead of silently, since OSS-Fuzz failures are not surfaced here. +# +# atheris only ships wheels for Linux and macOS, so this does not run on +# Windows. See https://github.com/onnx/onnx/blob/main/onnx/fuzz/README.md +# for the harness/toggle-byte details. + +name: Fuzz + +on: + pull_request: + paths: + - 'onnx/**' + - 'pyproject.toml' + - '.github/workflows/fuzz.yml' + push: + branches: + - main + paths: + - 'onnx/**' + - 'pyproject.toml' + - '.github/workflows/fuzz.yml' + schedule: + - cron: '0 6 * * *' # nightly: run each harness for longer than the PR smoke test + workflow_dispatch: + +permissions: + contents: read + +concurrency: + group: ${{ github.workflow }}-${{ github.ref }} + cancel-in-progress: true + +jobs: + fuzz: + name: Run fuzz harnesses (${{ github.event_name == 'schedule' && 'nightly' || 'smoke' }}) + runs-on: ubuntu-24.04 + steps: + - name: Checkout repository + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + submodules: recursive + persist-credentials: false + + - name: Set up Python + uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: '3.12' + + - name: Build and install ONNX + run: pip install -e ".[reference]" -v + env: + ONNX_ML: 1 + + - name: Install atheris + run: pip install atheris + + - name: Generate seed corpora + run: | + mkdir -p corpus + python onnx/fuzz/make_seed_corpus.py \ + corpus/version_converter.zip corpus/parser.zip \ + corpus/checker.zip corpus/shape_inference.zip corpus/compose.zip + for name in version_converter parser checker shape_inference compose; do + mkdir -p "corpus/$name" + python -m zipfile -e "corpus/$name.zip" "corpus/$name/" + done + # fuzz_model_loader.py has no dedicated seeds; it exercises the same + # load+check path as fuzz_checker.py, so reuse that corpus. + mkdir -p corpus/model_loader + cp corpus/checker/* corpus/model_loader/ + + - name: Set fuzz duration + id: duration + run: | + if [ "${{ github.event_name }}" = "schedule" ]; then + echo "seconds=300" >> "$GITHUB_OUTPUT" + else + echo "seconds=60" >> "$GITHUB_OUTPUT" + fi + + - name: Run fuzz_checker.py + run: python onnx/fuzz/fuzz_checker.py -max_total_time="$FUZZ_SECONDS" -timeout=20 corpus/checker + env: + FUZZ_SECONDS: ${{ steps.duration.outputs.seconds }} + + - name: Run fuzz_model_loader.py + run: python onnx/fuzz/fuzz_model_loader.py -max_total_time="$FUZZ_SECONDS" -timeout=20 corpus/model_loader + env: + FUZZ_SECONDS: ${{ steps.duration.outputs.seconds }} + + - name: Run fuzz_parser.py + run: python onnx/fuzz/fuzz_parser.py -max_total_time="$FUZZ_SECONDS" -timeout=20 corpus/parser + env: + FUZZ_SECONDS: ${{ steps.duration.outputs.seconds }} + + - name: Run fuzz_shape_inference.py + run: python onnx/fuzz/fuzz_shape_inference.py -max_total_time="$FUZZ_SECONDS" -timeout=20 corpus/shape_inference + env: + FUZZ_SECONDS: ${{ steps.duration.outputs.seconds }} + + - name: Run fuzz_version_converter.py + run: python onnx/fuzz/fuzz_version_converter.py -max_total_time="$FUZZ_SECONDS" -timeout=20 corpus/version_converter + env: + FUZZ_SECONDS: ${{ steps.duration.outputs.seconds }} + + - name: Run fuzz_compose.py + run: python onnx/fuzz/fuzz_compose.py -max_total_time="$FUZZ_SECONDS" -timeout=20 corpus/compose + env: + FUZZ_SECONDS: ${{ steps.duration.outputs.seconds }} + + - name: Upload crash artifacts + if: failure() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: fuzz-crashes + path: | + crash-* + timeout-* + oom-* + if-no-files-found: ignore diff --git a/.github/workflows/lint.yml b/.github/workflows/lint.yml new file mode 100644 index 0000000..35e5af4 --- /dev/null +++ b/.github/workflows/lint.yml @@ -0,0 +1,117 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Lint + +on: + push: + branches: + - main + pull_request: + merge_group: + workflow_dispatch: + +permissions: + contents: read + +concurrency: + group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name == 'workflow_dispatch' }} + cancel-in-progress: true + +jobs: + validate-sbom: + name: Validate SBOM + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: '3.12' + - run: | + pip install -q check-jsonschema + python -m check_jsonschema --schemafile "https://cyclonedx.org/schema/bom-1.7.schema.json" sbom.cdx.json + + enforce-style: + name: Enforce style + runs-on: ubuntu-latest + permissions: + security-events: write + steps: + - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + - name: Setup Python + uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: "3.13" + - name: Install ONNX + run: | + source workflow_scripts/protobuf/build_protobuf_unix.sh $(nproc) + + python -m pip install --quiet --upgrade pip setuptools wheel + + export ONNX_BUILD_TESTS=0 + export ONNX_ML=1 + export CMAKE_ARGS="-DONNXIFI_DUMMY_BACKEND=ON -DONNX_WERROR=ON" + export ONNX_NAMESPACE=ONNX_NAMESPACE_FOO_BAR_FOR_CI + + python -m pip install . + - name: Install dependencies + run: | + python -m pip install lintrunner>=0.10.7 + # Use release_test to pin package versions + python -m pip install -r requirements-release_test.txt + python -m pip install -r requirements-lintrunner.txt + lintrunner init + - name: Run lintrunner on all files + run: | + set +e + if ! lintrunner --force-color --all-files --tee-json=lint.json -v; then + echo "" + echo -e "\e[1m\e[36mYou can reproduce these results locally by using \`lintrunner\`.\e[0m" + echo -e "\e[1m\e[36mSee https://github.com/onnx/onnx/blob/main/CONTRIBUTING.md#coding-style for setup instructions.\e[0m" + exit 1 + fi + - name: Produce SARIF + if: always() + run: | + python -m lintrunner_adapters to-sarif lint.json lintrunner.sarif + - name: Upload SARIF file + # Use always() to always upload SARIF even if lintrunner returns with error code + # To toggle linter comments in the files page, press `i` on the keyboard + if: always() + continue-on-error: true + uses: github/codeql-action/upload-sarif@8aad20d150bbac5944a9f9d289da16a4b0d87c1e + with: + # Path to SARIF file relative to the root of the repository + sarif_file: lintrunner.sarif + category: lintrunner + checkout_path: ${{ github.workspace }} + - name: Upload lint results as artifacts + if: always() + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a + with: + name: lint-results + path: | + lint.json + lintrunner.sarif + retention-days: 30 + - name: Check auto-gen files are up-to-date + run: | + echo -e "\n::group:: ===> check auto-gen files are up-to-date..." + + ONNX_ML=1 python onnx/defs/gen_doc.py + python onnx/gen_proto.py -l + python onnx/gen_proto.py -l --ml + python onnx/backend/test/stat_coverage.py + + git status + git diff --exit-code -- . ':(exclude)onnx/onnx-data.proto' ':(exclude)onnx/onnx-data.proto3' + if [ $? -ne 0 ]; then + echo "git diff returned failures" + exit 1 + fi + echo -e "::endgroup::" diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml new file mode 100644 index 0000000..d9d25a8 --- /dev/null +++ b/.github/workflows/main.yml @@ -0,0 +1,280 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: CI + +env: + ASAN_OPTIONS: detect_leaks=0:symbolize=1:detect_stack_use_after_return=true:strict_init_order=true:detect_odr_violation=1:detect_container_overflow=0:check_initialization_order=true:debug=true:fast_unwind_on_malloc=1:verify_asan_link_order=0 + UBSAN_OPTIONS: print_stacktrace=1 + +on: + schedule: + - cron: '0 0 * * *' # every day at midnight for reporting code coverage to codecov + push: + branches: + - main + pull_request: + merge_group: + workflow_dispatch: + +permissions: # set top-level default permissions as security good practice + contents: read + +concurrency: + group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name == 'workflow_dispatch' }} + cancel-in-progress: true + +jobs: + test: + name: Test ${{ matrix.os }}, ${{ matrix.python_version }}, ${{ matrix.protobuf_type }}, debug=${{ matrix.debug_build }}, unity_build=${{ matrix.unity_build }}, onnx_ml=${{ matrix.onnx_ml }}, autogen=${{ matrix.autogenerate_files }} + strategy: + fail-fast: false + matrix: + os: [ubuntu-24.04, windows-latest, macos-latest] + # Test on the oldest and latest supported Python versions + autogenerate_files: [0] + debug_build: [0] + onnx_ml: [1] + protobuf_type: ['Internal'] + python_version: ['3.14t', '3.14', '3.10'] + unity_build: [0] + + include: + # Ubuntu debug build runs with sanitizer + - python_version: "3.14" + autogenerate_files: 0 + debug_build: 1 + onnx_ml: 1 + os: "ubuntu-24.04" + protobuf_type: 'Internal' + unity_build: 0 + + # Test compilation with dynamically linked protobuf. + # This is actually redundant with the pixi-tests which + # also use dynamic linking. + - python_version: "3.14" + autogenerate_files: 0 + debug_build: 0 + onnx_ml: 1 + os: "ubuntu-24.04" + protobuf_type: 'External' + unity_build: 0 + - python_version: "3.14" + autogenerate_files: 0 + debug_build: 0 + onnx_ml: 1 + os: "windows-2022" + protobuf_type: 'External' + unity_build: 0 + # build_protobuf_unix.sh appears broken on macos-arm64 + # - python_version: "3.14" + # debug_build: 0 + # unity_build: 0 + # autogenerate_files: 0 + # protobuf_type: 'External' + # os: "macos-latest" + + # Unity build AND autogenerated files + # The unity build may surface name clashes at compile time + # but produces functionally identical binaries. We can + # therefore safely reuse this build to test the + # autogenerated files. + - python_version: '3.14' + autogenerate_files: 1 + debug_build: 0 + onnx_ml: 1 + os: "ubuntu-24.04" + protobuf_type: 'External' + unity_build: 1 + + # Toggling onnx_ml should be additive. There is likely very + # different signal in testing it in relation with other + # parameters. + - python_version: '3.14' + autogenerate_files: 0 + debug_build: 0 + onnx_ml: 0 + os: "ubuntu-24.04" + protobuf_type: 'External' + unity_build: 0 + + runs-on: ${{ matrix.os }} + steps: + - name: Checkout repository + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + with: + submodules: recursive + persist-credentials: false + + - name: Set up Python + uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: ${{ matrix.python_version }} + + - name: Show versions + run: | + python --version + cmake --version + + - name: Install external protobuf - Linux + if: matrix.protobuf_type == 'External' && startsWith(matrix.os,'ubuntu') + run: | + sudo apt-get update + sudo apt-get install libprotobuf-dev protobuf-compiler + + - name: Install external protobuf - MacOS + if: matrix.protobuf_type == 'External' && matrix.os == 'macos-latest' + run: | + source workflow_scripts/protobuf/build_protobuf_unix.sh 3 $(pwd)/protobuf/protobuf_install + + - name: Set up MSBuild (x64) + if: startsWith(matrix.os,'windows') + uses: microsoft/setup-msbuild@30375c66a4eea26614e0d39710365f22f8b0af57 # v3.0.0 + with: + msbuild-architecture: x64 + + - name: Install external protobuf - Windows + if: matrix.protobuf_type == 'External' && startsWith(matrix.os, 'windows') + run: | + if ($matrix.os -like "windows-11-arm*") { + $cmake_arch = "ARM64" + } else { + $cmake_arch = "x64" + } + + workflow_scripts/protobuf/build_protobuf_win.ps1 -cmake_arch $cmake_arch + shell: pwsh + + - name: Build and install ONNX - Linux + if: startsWith(matrix.os,'ubuntu') + run: | + export SOURCE_DATE_EPOCH=$(git log -1 --pretty=%ct) + if [ "${{ matrix.python_version }}" == "3.14" ]; then + sudo apt-get update + sudo apt-get install libjpeg-dev zlib1g-dev libpng-dev + fi + + if [ "${{ matrix.protobuf_type }}" == "External" ]; then + export CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_POSITION_INDEPENDENT_CODE=ON -DONNX_USE_PROTOBUF_SHARED_LIBS=ON" + fi + pip install -e ".[reference]" -v + env: + DEBUG: ${{ matrix.debug_build }} + ONNX_ML: ${{ matrix.onnx_ml }} + ONNX_BUILD_TESTS: 1 + CMAKE_ARGS: "-DONNX_WERROR=ON -DONNX_USE_ASAN=${{ matrix.debug_build }} -DONNX_USE_UNITY_BUILD=${{ matrix.unity_build }} -DONNX_HARDENING=ON" + + - name: Build and install ONNX - MacOS + if: matrix.os == 'macos-latest' + run: | + pip install -e ".[reference]" -v + env: + DEBUG: ${{ matrix.debug_build }} + ONNX_ML: ${{ matrix.onnx_ml }} + ONNX_BUILD_TESTS: 1 + CMAKE_ARGS: "-DONNX_WERROR=ON -DONNX_USE_UNITY_BUILD=${{ matrix.unity_build }} -DONNX_HARDENING=ON" + + - name: Build and install ONNX - Windows + if: startsWith(matrix.os,'windows') + run: | + pip install -e . -v + env: + DEBUG: ${{ matrix.debug_build }} + ONNX_ML: ${{ matrix.onnx_ml }} + ONNX_BUILD_TESTS: 1 + CMAKE_ARGS: "-DONNX_WERROR=ON -DONNX_USE_PROTOBUF_SHARED_LIBS=OFF -DONNX_USE_LITE_PROTO=ON -DONNX_USE_UNITY_BUILD=${{ matrix.unity_build }} -DONNX_HARDENING=ON" + + - name: pip freeze + run: | + pip freeze + + - name: Setup GCC ASAN LD_PRELOAD + if: startsWith(matrix.os,'ubuntu') && (matrix.python_version != '3.13t') + run: | + export LD_PRELOAD="$(/usr/bin/c++ -print-file-name=libasan.so):$LD_PRELOAD" + + - name: Install test dependencies + run: | + python -m pip install -r requirements-release_test.txt + + - name: Run Python tests + # Prohibitively slow for Windows debug builds + if: ${{ !(startsWith(matrix.os, 'windows') && matrix.debug_build == 1) }} + run: | + pytest -sv --cov=onnx --cov-report=xml --cov-append --cov-branch --junitxml junit.xml -n auto --dist loadscope + - name: Run C++ tests + if: startsWith(matrix.os,'ubuntu') || matrix.os == 'macos-latest' + run: | + export LD_LIBRARY_PATH="./.setuptools-cmake-build/:$LD_LIBRARY_PATH" + ./.setuptools-cmake-build/onnx_gtests + + - name: Run C++ extension test + if: (startsWith(matrix.os,'ubuntu') || matrix.os == 'macos-latest') && matrix.debug_build == 0 && matrix.protobuf_type == 'External' + run: | + cmake -S . -B build_for_install \ + -DCMAKE_BUILD_TYPE=Release \ + -DCMAKE_INSTALL_PREFIX=/usr/local \ + -DONNX_ML=${{ matrix.onnx_ml }} \ + -DONNX_USE_PROTOBUF_SHARED_LIBS=ON + cmake --build build_for_install --parallel + sudo cmake --install build_for_install + + cmake -S onnx/test/cmake -B onnx/test/cmake/build \ + -DONNX_ML=${{ matrix.onnx_ml }} + cmake --build onnx/test/cmake/build + ./onnx/test/cmake/build/main + + - name: Upload coverage to Codecov + if: github.repository_owner == 'onnx' + uses: codecov/codecov-action@fb8b3582c8e4def4969c97caa2f19720cb33a72f # v7.0.0 + with: + token: ${{ secrets.CODECOV_TOKEN }} + + - name: Upload test results to Codecov + if: github.repository_owner == 'onnx' && !cancelled() + uses: codecov/codecov-action@fb8b3582c8e4def4969c97caa2f19720cb33a72f # v7.0.0 + with: + token: ${{ secrets.CODECOV_TOKEN }} + report_type: test_results + + # Note that the test data should be generated with numpy>=2.0. + # numpy 1.x and numpy 2.0 produce slightly different numerical values. + - name: Test backend test data + if: matrix.autogenerate_files == 1 && startsWith(matrix.os,'ubuntu') + run: | + python onnx/backend/test/cmd_tools.py generate-data --clean + git status + git diff --exit-code -- . ':!onnx/onnx-data.proto' ':!onnx/onnx-data.proto3' ':!*output_*.pb' ':!*input_*.pb' + if [ $? -ne 0 ]; then + echo "git diff for test generation returned failures. Please check updated node test files" + exit 1 + fi + git diff --exit-code --diff-filter=ADR -- . ':!onnx/onnx-data.proto' ':!onnx/onnx-data.proto3' + if [ $? -ne 0 ]; then + echo "Test generation returned failures. Please check the number of node test files (input_*.pb or output_*.pb)" + exit 1 + fi + pip uninstall -y pillow + python onnx/backend/test/cmd_tools.py generate-data --clean + git status + git diff --exit-code -- . ':!onnx/onnx-data.proto' ':!onnx/onnx-data.proto3' ':!*output_*.pb' ':!*input_*.pb' + if [ $? -ne 0 ]; then + echo "git diff for test generation without pillow returned failures. Please check updated node test files" + exit 1 + fi + + - name: Run Python tests with numpy<2.0 (win, mac) + if: (matrix.python_version == '3.10') && (matrix.os == 'windows-latest' || matrix.os == 'macos-latest') && (matrix.debug_build != 1) + run: | + pip install "numpy<2.0" pillow + pytest -s + + - name: Run Python tests with numpy<2.0 (ubuntu, python<3.13) + if: (matrix.python_version == '3.10') && startsWith(matrix.os,'ubuntu') + run: | + # 2024.10.15: Error message: The headers or library files could not be found for jpeg, a required dependency when compiling Pillow from source. + sudo apt-get update + sudo apt-get install libjpeg-dev zlib1g-dev libpng-dev + pip install --prefer-binary "numpy<2.0" pillow + pytest -s diff --git a/.github/workflows/pages.yml b/.github/workflows/pages.yml new file mode 100644 index 0000000..6450945 --- /dev/null +++ b/.github/workflows/pages.yml @@ -0,0 +1,55 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Generate and publish ONNX docs + +on: + pull_request: + branches: ["main"] + # Runs on pushes targeting the default branch + push: + branches: ["main"] + # Allows you to run this workflow manually from the Actions tab + workflow_dispatch: + +# Sets permissions of the GITHUB_TOKEN to allow deployment to GitHub Pages +permissions: + contents: read + +jobs: + build: + runs-on: ubuntu-latest + steps: + - name: Checkout + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + - name: Set up pixi + uses: prefix-dev/setup-pixi@5185adfbffb4bd703da3010310260805d89ebb11 # v0.9.6 + with: + environments: docs + - name: Install repository + run: pixi run -e docs install + - name: Build Docs + run: pixi run -e docs docs-build + - name: Upload artifact + uses: actions/upload-pages-artifact@fc324d3547104276b827a68afc52ff2a11cc49c9 # v5.0.0 + with: + path: 'docs/docsgen/build/html' + deploy: + needs: 'build' + if: github.event_name != 'pull_request' + permissions: + pages: write + id-token: write + environment: + name: github-pages + url: ${{ steps.deployment.outputs.page_url }} + runs-on: ubuntu-latest + steps: + - name: Setup Pages + uses: actions/configure-pages@45bfe0192ca1faeb007ade9deae92b16b8254a0d # v6.0.0 + - name: Deploy to GitHub Pages + id: deployment + uses: actions/deploy-pages@cd2ce8fcbc39b97be8ca5fce6e763baed58fa128 # v5.0.0 diff --git a/.github/workflows/pixi_build.yml b/.github/workflows/pixi_build.yml new file mode 100644 index 0000000..d31195e --- /dev/null +++ b/.github/workflows/pixi_build.yml @@ -0,0 +1,124 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Pixi CI + +on: + pull_request: + push: + branches: + - main + +permissions: + contents: read + +concurrency: + group: ${{ github.workflow }}-${{ github.ref }} + cancel-in-progress: true + +jobs: + pre-commit-hooks: + name: Install and lint (${{ matrix.os }}) + runs-on: ${{ matrix.os }} + strategy: + fail-fast: false + matrix: + os: + - ubuntu-24.04-arm + - windows-2022 + steps: + - name: Checkout branch + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + fetch-depth: 0 + persist-credentials: false + - name: Set up pixi + uses: prefix-dev/setup-pixi@5185adfbffb4bd703da3010310260805d89ebb11 # v0.9.6 + with: + environments: default reuse + - name: Install repository + run: pixi run install + - name: Lint + run: pixi run lint --show-diff-on-failure + + xcode-build: + # Guards against regressions in CMake's Xcode generator (issue #8053), which + # downstream consumers such as onnxruntime use for their iOS/macOS framework + # builds. The other jobs build with Ninja and so do not exercise the + # Xcode-specific code paths. + name: Xcode generator build + runs-on: macos-latest + steps: + - name: Checkout branch + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + fetch-depth: 0 + persist-credentials: false + - name: Set up pixi + uses: prefix-dev/setup-pixi@5185adfbffb4bd703da3010310260805d89ebb11 # v0.9.6 + with: + environments: default + - name: Build with the Xcode generator + run: | + pixi run cmake -G Xcode -B .xcode-cmake-build -S . -DONNX_BUILD_PYTHON=ON -DONNX_USE_PROTOBUF_SHARED_LIBS=ON + pixi run cmake --build .xcode-cmake-build + + install-and-test: + name: Install and test (${{ matrix.os }}, ${{ matrix.environment }}) + runs-on: ${{ matrix.os }} + permissions: + contents: read + issues: write # Needed to create an issue on failure + + strategy: + fail-fast: false + matrix: + os: + - ubuntu-24.04-arm + - ubuntu-latest + - macos-latest + - windows-2022 + environment: + - default + - oldies + steps: + - name: Checkout branch + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + fetch-depth: 0 + persist-credentials: false + - name: Set up pixi + uses: prefix-dev/setup-pixi@5185adfbffb4bd703da3010310260805d89ebb11 # v0.9.6 + with: + environments: ${{ matrix.environment }} + - name: Install repository + run: pixi run -e ${{ matrix.environment }} install + - name: pip check installation + run: pixi run -e ${{ matrix.environment }} pip check + - name: gtests + run: pixi run -e ${{ matrix.environment }} gtest + - name: pytest + run: pixi run -e ${{ matrix.environment }} pytest + - name: Issue on failure + uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0 + if: ${{ failure() && github.ref == 'refs/heads/main' }} + with: + script: | + github.rest.issues.listForRepo({ + owner: context.repo.owner, + repo: context.repo.repo, + state: "open", + labels: "[bot] pixi CI" + }).then((issues) => { + if (issues.data.length === 0){ + github.rest.issues.create({ + owner: context.repo.owner, + repo: context.repo.repo, + title: "Scheduled pixi CI failed", + body: "The scheduled pixi-based CI failed. See https://github.com/${{ github.repository }}/actions/runs/${{github.run_id}} for details.", + assignees: ["cbourjau"], + labels: ["[bot] pixi CI"] + }) + } + }); diff --git a/.github/workflows/preview_source_dist_test.yml b/.github/workflows/preview_source_dist_test.yml new file mode 100644 index 0000000..b08bccb --- /dev/null +++ b/.github/workflows/preview_source_dist_test.yml @@ -0,0 +1,43 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Test source dist of preview build at onnx-weekly + +on: + schedule: + - cron: '0 0 * * 2' + workflow_call: + workflow_dispatch: + +permissions: + contents: read + +jobs: + + test_sdist_preview: + + strategy: + matrix: + os: [ubuntu-24.04, windows-latest] + python-version: ['3.10', '3.12', '3.13'] + fail-fast: false + runs-on: ${{ matrix.os }} + + steps: + + - name: Checkout repository + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 + with: + persist-credentials: false + + - name: Set up Python ${{ matrix.python-version }} + uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: ${{ matrix.python-version }} + + - name: Test preview build source distribution from PyPI + run: | + python -m pip install --no-binary onnx-weekly onnx-weekly + python -m pip install pytest ml_dtypes pillow + pytest diff --git a/.github/workflows/release_linux_cibw.yml b/.github/workflows/release_linux_cibw.yml new file mode 100644 index 0000000..07e2ea9 --- /dev/null +++ b/.github/workflows/release_linux_cibw.yml @@ -0,0 +1,101 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Linux Release + +on: + workflow_call: + inputs: + build_mode: + required: false + type: string + default: "preview" + workflow_dispatch: + +permissions: + contents: read + +concurrency: + group: linux-release-${{ github.workflow }}-${{ github.ref }} + cancel-in-progress: true + +jobs: + build: + strategy: + fail-fast: false + matrix: + build: + - "cp310-manylinux_x86_64" + - "cp310-manylinux_aarch64" + - "cp311-manylinux_x86_64" + - "cp311-manylinux_aarch64" + - "cp312-manylinux_x86_64" + - "cp312-manylinux_aarch64" + - "cp314t-manylinux_x86_64" + - "cp314t-manylinux_aarch64" + + runs-on: ${{ contains(matrix.build, 'aarch64') && 'ubuntu-24.04-arm' || 'ubuntu-24.04' }} + + steps: + - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + + - name: Read protobuf version from sbom.cdx.json + run: echo "PROTOBUF_VERSION=$(jq -r '.components[] | select(.name=="protobuf") | .version' sbom.cdx.json)" >> $GITHUB_ENV + + - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: "3.12" + + - name: Set preview version + if: inputs.build_mode != 'release' + shell: bash + run: | + sed -i'' 's/name = "onnx"/name = "onnx-weekly"/' pyproject.toml + echo "$(cat VERSION_NUMBER).dev$(date -u +%Y%m%d)" > VERSION_NUMBER + + - name: Download protoc + run: | + ARCH=${{ contains(matrix.build, 'aarch64') && 'aarch_64' || 'x86_64' }} + curl -sSL "https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOBUF_VERSION}/protoc-${PROTOBUF_VERSION}-linux-${ARCH}.zip" -o protoc.zip + unzip -q protoc.zip -d protoc-host + chmod +x protoc-host/bin/protoc + + - name: Download protobuf source + run: | + curl -sSL "https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOBUF_VERSION}/protobuf-${PROTOBUF_VERSION}.tar.gz" -o protobuf.tar.gz + tar -xf protobuf.tar.gz + + - name: Set SOURCE_DATE_EPOCH for reproducible builds + run: echo "SOURCE_DATE_EPOCH=$(git log -1 --pretty=%ct)" >> $GITHUB_ENV + + - name: Build wheels + uses: pypa/cibuildwheel@294735312765b09d24a2fbec22660ce817587d55 # v4.1.0 + with: + output-dir: dist/ + only: ${{ matrix.build }} + env: + CIBW_ENVIRONMENT: >- + CMAKE_ARGS=" + -DFETCHCONTENT_SOURCE_DIR_PROTOBUF=/project/protobuf-${{ env.PROTOBUF_VERSION }} + -DONNX_CUSTOM_PROTOC_EXECUTABLE=/project/protoc-host/bin/protoc + -DONNX_HARDENING=ON + -DONNX_USE_LITE_PROTO=ON + -DONNX_WERROR=ON + " + + - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: wheels-linux-${{ matrix.build }} + path: dist/*.whl + + - name: Validate wheel + run: | + python -m pip install -q abi3audit check-wheel-contents + for whl in dist/*.whl; do + echo "Checking $whl" + check-wheel-contents "$whl" + python -m abi3audit -v "$whl" + done diff --git a/.github/workflows/release_macos_cibw.yml b/.github/workflows/release_macos_cibw.yml new file mode 100644 index 0000000..339c226 --- /dev/null +++ b/.github/workflows/release_macos_cibw.yml @@ -0,0 +1,97 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: macOS Release + +on: + workflow_call: + inputs: + build_mode: + required: false + type: string + default: "preview" + workflow_dispatch: + +permissions: + contents: read + +concurrency: + group: macos-release-${{ github.workflow }}-${{ github.ref }} + cancel-in-progress: true + +jobs: + build: + strategy: + fail-fast: false + matrix: + build: + - "cp310-macosx_universal2" + - "cp311-macosx_universal2" + - "cp312-macosx_universal2" + - "cp314t-macosx_universal2" + + runs-on: macos-14 + + steps: + - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + + - name: Read protobuf version from sbom.cdx.json + run: echo "PROTOBUF_VERSION=$(jq -r '.components[] | select(.name=="protobuf") | .version' sbom.cdx.json)" >> $GITHUB_ENV + + - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: "3.12" + + - name: Set preview version + if: inputs.build_mode != 'release' + shell: bash + run: | + sed -i '' 's/name = "onnx"/name = "onnx-weekly"/' pyproject.toml + echo "$(cat VERSION_NUMBER).dev$(date -u +%Y%m%d)" > VERSION_NUMBER + + - name: Download protoc + run: | + curl -sSL "https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOBUF_VERSION}/protoc-${PROTOBUF_VERSION}-osx-universal_binary.zip" -o protoc.zip + unzip -q protoc.zip -d protoc-host + chmod +x protoc-host/bin/protoc + + - name: Download protobuf source + run: | + curl -sSL "https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOBUF_VERSION}/protobuf-${PROTOBUF_VERSION}.tar.gz" -o protobuf.tar.gz + tar -xf protobuf.tar.gz + + - name: Set SOURCE_DATE_EPOCH for reproducible builds + run: echo "SOURCE_DATE_EPOCH=$(git log -1 --pretty=%ct)" >> $GITHUB_ENV + + - name: Build wheels + uses: pypa/cibuildwheel@294735312765b09d24a2fbec22660ce817587d55 # v4.1.0 + with: + output-dir: dist/ + only: ${{ matrix.build }} + env: + CIBW_ENVIRONMENT: >- + MACOSX_DEPLOYMENT_TARGET=12.0 + CMAKE_ARGS=" + -DFETCHCONTENT_SOURCE_DIR_PROTOBUF=${{ github.workspace }}/protobuf-${{ env.PROTOBUF_VERSION }} + -DONNX_CUSTOM_PROTOC_EXECUTABLE=${{ github.workspace }}/protoc-host/bin/protoc + -DONNX_HARDENING=ON + -DONNX_USE_LITE_PROTO=ON + -DONNX_WERROR=ON + " + + - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: wheels-macos-${{ matrix.build }} + path: dist/*.whl + + - name: Validate wheel + run: | + python -m pip install -q abi3audit check-wheel-contents + for whl in dist/*.whl; do + echo "Checking $whl" + check-wheel-contents "$whl" + python -m abi3audit -v "$whl" + done diff --git a/.github/workflows/release_pyodide_cibw.yml b/.github/workflows/release_pyodide_cibw.yml new file mode 100644 index 0000000..8aaaaac --- /dev/null +++ b/.github/workflows/release_pyodide_cibw.yml @@ -0,0 +1,83 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Pyodide Build + +on: + workflow_call: + inputs: + build_mode: + required: false + type: string + default: "preview" + workflow_dispatch: + +permissions: + contents: read + +concurrency: + group: pyodide-release-${{ github.workflow }}-${{ github.ref }} + cancel-in-progress: true + +jobs: + build: + runs-on: ubuntu-24.04 + + steps: + - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + + - name: Read protobuf version from sbom.cdx.json + run: echo "PROTOBUF_VERSION=$(jq -r '.components[] | select(.name=="protobuf") | .version' sbom.cdx.json)" >> $GITHUB_ENV + + - name: Set up Python + uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: "3.12" + + - name: Set preview version + if: inputs.build_mode != 'release' + shell: bash + run: | + sed -i'' 's/name = "onnx"/name = "onnx-weekly"/' pyproject.toml + echo "$(cat VERSION_NUMBER).dev$(date -u +%Y%m%d)" > VERSION_NUMBER + + - name: Download protoc for host + run: | + curl -sSL "https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOBUF_VERSION}/protoc-${PROTOBUF_VERSION}-linux-x86_64.zip" -o protoc.zip + unzip -q protoc.zip -d protoc-host + chmod +x protoc-host/bin/protoc + + - name: Download protobuf source + run: | + curl -sSL "https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOBUF_VERSION}/protobuf-${PROTOBUF_VERSION}.tar.gz" -o protobuf.tar.gz + tar -xf protobuf.tar.gz + + - name: Set SOURCE_DATE_EPOCH for reproducible builds + run: echo "SOURCE_DATE_EPOCH=$(git log -1 --pretty=%ct)" >> $GITHUB_ENV + + - name: Build wheels + uses: pypa/cibuildwheel@294735312765b09d24a2fbec22660ce817587d55 # v4.1.0 + with: + output-dir: dist/ + env: + CIBW_PLATFORM: pyodide + CIBW_TEST_COMMAND: >- + python -c "from onnx import NodeProto; n = NodeProto(); n.op_type = 'Add'; print(n)" + CIBW_ENVIRONMENT: >- + CMAKE_ARGS=" + -DFETCHCONTENT_SOURCE_DIR_PROTOBUF=${{ github.workspace }}/protobuf-${{ env.PROTOBUF_VERSION }} + -DONNX_CUSTOM_PROTOC_EXECUTABLE=${{ github.workspace }}/protoc-host/bin/protoc + -DONNX_HARDENING=ON + -DONNX_PYODIDE_BUILD=ON + -DONNX_USE_LITE_PROTO=ON + -DONNX_WERROR=ON + -DCMAKE_CXX_FLAGS='-Wno-error=deprecated-builtins' + " + + - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: wheels-pyodide + path: dist/*.whl diff --git a/.github/workflows/release_sdist.yml b/.github/workflows/release_sdist.yml new file mode 100644 index 0000000..08b5de0 --- /dev/null +++ b/.github/workflows/release_sdist.yml @@ -0,0 +1,66 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: sdistRelease + +on: + workflow_call: + inputs: + os: + required: true + type: string + build_mode: + required: true + type: string + +permissions: + contents: read + +concurrency: + group: sdist-release-${{ github.workflow }}-${{ github.ref }} + +jobs: + build: + if: github.event_name != 'pull_request' || startsWith( github.base_ref, 'rel-') || contains( github.event.pull_request.labels.*.name, 'run release CIs') + runs-on: ubuntu-24.04 + strategy: + matrix: + python-version: ['3.10'] + target-architecture: ['arm64'] + + steps: + - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + + - name: Set up Python ${{ matrix.python-version }} + uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: ${{ matrix.python-version }} + + - name: Install Python dependencies + run: | + python -m pip install -q --upgrade pip + python -m pip install -q build scikit-build-core + + - name: Build source distribution (preview build / weekly) + if: ${{ inputs.build_mode != 'release' }} + run: | + git clean -xdf + sed -i 's/name = "onnx"/name = "onnx-weekly"/' 'pyproject.toml' + echo "$(cat VERSION_NUMBER).dev$(date -u +%Y%m%d)" > VERSION_NUMBER + python -m build --sdist + + - name: Build source distribution (for release) + if: ${{ inputs.build_mode == 'release' }} + run: | + git clean -xdf + python -m build --sdist + + - name: Upload sdist + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a + with: + name: sdist + path: | + ./dist/*.tar.gz diff --git a/.github/workflows/release_windows_cibw.yml b/.github/workflows/release_windows_cibw.yml new file mode 100644 index 0000000..e7f877f --- /dev/null +++ b/.github/workflows/release_windows_cibw.yml @@ -0,0 +1,108 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Windows Release + +on: + workflow_call: + inputs: + build_mode: + required: false + type: string + default: "preview" + workflow_dispatch: + +permissions: + contents: read + +concurrency: + group: windows-release-${{ github.workflow }}-${{ github.ref }} + cancel-in-progress: true + +jobs: + build: + strategy: + fail-fast: false + matrix: + build: + - "cp310-win_amd64" + - "cp310-win32" + - "cp311-win_amd64" + - "cp311-win32" + - "cp311-win_arm64" + - "cp312-win_amd64" + - "cp312-win32" + - "cp312-win_arm64" + - "cp314t-win_amd64" + - "cp314t-win_arm64" + runs-on: ${{ contains(matrix.build, 'arm64') && 'windows-11-arm' || 'windows-2022' }} + + steps: + - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + + - name: Read protobuf version from sbom.cdx.json + shell: bash + run: echo "PROTOBUF_VERSION=$(jq -r '.components[] | select(.name=="protobuf") | .version' sbom.cdx.json)" >> $GITHUB_ENV + + - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: "3.12" + + - name: Set preview version + if: inputs.build_mode != 'release' + shell: bash + run: | + sed -i'' 's/name = "onnx"/name = "onnx-weekly"/' pyproject.toml + echo "$(cat VERSION_NUMBER).dev$(date -u +%Y%m%d)" > VERSION_NUMBER + + - name: Download protoc + run: | + curl -sSL "https://github.com/protocolbuffers/protobuf/releases/download/v${env:PROTOBUF_VERSION}/protoc-${env:PROTOBUF_VERSION}-win64.zip" -o protoc.zip + Expand-Archive protoc.zip -DestinationPath protoc-host + + - name: Download protobuf source + run: | + curl -sSL "https://github.com/protocolbuffers/protobuf/releases/download/v${env:PROTOBUF_VERSION}/protobuf-${env:PROTOBUF_VERSION}.tar.gz" -o protobuf.tar.gz + tar -xf protobuf.tar.gz + + - name: Set paths + shell: bash + run: | + echo "PROTOC_PATH=$(cygpath -m "${{ github.workspace }}/protoc-host/bin/protoc.exe")" >> $GITHUB_ENV + echo "PROTOBUF_SRC=$(cygpath -m "${{ github.workspace }}/protobuf-${PROTOBUF_VERSION}")" >> $GITHUB_ENV + + - name: Set SOURCE_DATE_EPOCH for reproducible builds + shell: bash + run: echo "SOURCE_DATE_EPOCH=$(git log -1 --pretty=%ct)" >> $GITHUB_ENV + + - name: Build wheels + uses: pypa/cibuildwheel@294735312765b09d24a2fbec22660ce817587d55 # v4.1.0 + with: + output-dir: dist/ + only: ${{ matrix.build }} + env: + CIBW_ENVIRONMENT: >- + CMAKE_ARGS=" + -DFETCHCONTENT_SOURCE_DIR_PROTOBUF=${{ env.PROTOBUF_SRC }} + -DONNX_CUSTOM_PROTOC_EXECUTABLE=${{ env.PROTOC_PATH }} + -DONNX_HARDENING=ON + -DONNX_USE_LITE_PROTO=ON + -DONNX_WERROR=ON + " + + - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: wheels-windows-${{ matrix.build }} + path: dist/*.whl + + - name: Validate wheel + run: | + python -m pip install -q abi3audit check-wheel-contents + Get-ChildItem -Path dist/*.whl | ForEach-Object { + echo "Checking $($_.Name)" + check-wheel-contents $_.FullName + python -m abi3audit -v $_.FullName + } diff --git a/.github/workflows/scorecard.yml b/.github/workflows/scorecard.yml new file mode 100644 index 0000000..af9ab3b --- /dev/null +++ b/.github/workflows/scorecard.yml @@ -0,0 +1,48 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Scorecard supply-chain security +on: + branch_protection_rule: + schedule: + - cron: '24 10 * * 6' + push: + branches: [ "main" ] + +permissions: read-all + +jobs: + analysis: + name: Scorecard analysis + runs-on: ubuntu-latest + permissions: + # Needed to upload the results to code-scanning dashboard. + security-events: write + # Needed to publish results and get a badge (see publish_results below). + id-token: write + steps: + - name: "Checkout code" + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + persist-credentials: false + + - name: "Run analysis" + uses: ossf/scorecard-action@4eaacf0543bb3f2c246792bd56e8cdeffafb205a + with: + results_file: results.sarif + results_format: sarif + publish_results: true + + - name: "Upload artifact" + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a + with: + name: SARIF file + path: results.sarif + retention-days: 5 + + # Upload the results to GitHub's code scanning dashboard. + - name: "Upload to code-scanning" + uses: github/codeql-action/upload-sarif@8aad20d150bbac5944a9f9d289da16a4b0d87c1e + with: + sarif_file: results.sarif diff --git a/.github/workflows/stale.yml b/.github/workflows/stale.yml new file mode 100644 index 0000000..a74e97c --- /dev/null +++ b/.github/workflows/stale.yml @@ -0,0 +1,38 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +# This workflow warns and then closes issues and PRs that have had no activity for a specified amount of time. +# +# You can adjust the behavior by modifying this file. +# For more information, see: +# https://github.com/actions/stale +name: Mark stale issues and pull requests + +on: + schedule: + - cron: '39 6 * * *' + +permissions: # set top-level default permissions as security best practice + contents: read + +jobs: + stale: + + runs-on: ubuntu-latest + permissions: + issues: write + pull-requests: write + + steps: + - uses: actions/stale@eb5cf3af3ac0a1aa4c9c45633dd1ae542a27a899 # v10.3.0 + with: + repo-token: ${{ secrets.GITHUB_TOKEN }} + days-before-stale: 365 + days-before-close: 21 + ascending: true + exempt-issue-labels: bug,no-stale + exempt-pr-labels: no-stale,contributions welcome + remove-issue-stale-when-updated: true + remove-pr-stale-when-updated: true + exempt-all-milestones: true diff --git a/.github/workflows/win_no_exception_ci.yml b/.github/workflows/win_no_exception_ci.yml new file mode 100644 index 0000000..c9ddd85 --- /dev/null +++ b/.github/workflows/win_no_exception_ci.yml @@ -0,0 +1,81 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +name: Windows_No_Exception_CI + +on: + push: + branches: [ main, rel-* ] + pull_request: + branches: [ main, rel-* ] + +permissions: # set top-level default permissions as security best practice + contents: read + +concurrency: + group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name == 'workflow_dispatch' }} + cancel-in-progress: true + +jobs: + build: + runs-on: windows-2022 + strategy: + matrix: + python-version: ['3.10'] + architecture: ['x64'] + steps: + - name: Checkout ONNX + uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0 + with: + path: ./onnx + submodules: 'recursive' + persist-credentials: false + + - name: Set up Python ${{ matrix.python-version }} + uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0 + with: + python-version: ${{ matrix.python-version }} + architecture: ${{ matrix.architecture }} + cache: 'pip' + + - name: Add msbuild to PATH + uses: microsoft/setup-msbuild@30375c66a4eea26614e0d39710365f22f8b0af57 # v3.0.0 + with: + msbuild-architecture: ${{ matrix.architecture }} + + - name: Install dependencies + run: | + python -m pip install --upgrade pip + python -m pip install cmake + + - name: Build and test ONNX binaries + run: | + . .\onnx\workflow_scripts\protobuf\build_protobuf_win.ps1 -cmake_arch ${{ matrix.architecture }} + + cd onnx + echo "Build ONNX" + cmake -G "Visual Studio 17 2022" -A ${{ matrix.architecture }} -DONNX_USE_PROTOBUF_SHARED_LIBS=OFF -DONNX_USE_LITE_PROTO=ON -DONNX_WERROR=ON -DONNX_DISABLE_EXCEPTIONS=ON -DCMAKE_BUILD_TYPE=Release -DONNX_USE_MSVC_STATIC_RUNTIME=OFF -DONNX_ML=1 -DONNX_BUILD_TESTS=ON -S . -B .setuptools-cmake-build\ + cd .setuptools-cmake-build\ + cmake --build . --config Release + + echo "Run gtests" + Release\onnx_gtests.exe + if($lastexitcode -ne 0) { + EXIT 1 + } + + cd .. + git clean -xdf + set ONNX_BUILD_TESTS=1 + echo "Build ONNX with non-static registration for testing selective ONNX schema loading" + cmake -G "Visual Studio 17 2022" -A ${{ matrix.architecture }} -DONNX_USE_PROTOBUF_SHARED_LIBS=OFF -DONNX_USE_LITE_PROTO=ON -DONNX_WERROR=ON -DCMAKE_BUILD_TYPE=Release -DONNX_USE_MSVC_STATIC_RUNTIME=OFF -DONNX_ML=1 -DONNX_BUILD_TESTS=ON -DONNX_DISABLE_STATIC_REGISTRATION=ON -S . -B .setuptools-cmake-build\ + + cd .setuptools-cmake-build\ + cmake --build . --config Release + + echo "Only test selective ONNX schema loading" + Release\onnx_gtests.exe --gtest_filter="SchemaRegistrationTest*" + if($lastexitcode -ne 0) { + EXIT 1 + } diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..15aa2af --- /dev/null +++ b/.gitignore @@ -0,0 +1,118 @@ +## General + +# Compiled Object files +*.slo +*.lo +*.o +*.cuo + +# Compiled Dynamic libraries +*.so +*.dylib +*.pyd + +# Compiled Static libraries +*.lai +*.la +*.a + +# Compiled protocol buffers +*.pb.h +*.pb.cc +onnx/*_pb2.py +onnx/*_pb.py +onnx/*_pb2.pyi +onnx/*_pb.pyi + +# Compiled python +*.pyc + +# Compiled MATLAB +*.mex* + +# IPython notebook checkpoints +.ipynb_checkpoints + +# Editor temporaries +*.swn +*.swo +*.swp +*~ + +# Sublime Text settings +*.sublime-workspace +*.sublime-project + +# Eclipse Project settings +*.*project +.settings + +# QtCreator files +*.user + +# PyCharm files +.idea + +# Visual Studio Code files +.vscode +!/.vscode/settings.json + +# OSX dir files +.DS_Store + +## ONNX + +# build, distribute, and bins (+ python proto bindings) +build +build_* +.build_debug/* +.build_release/* +.setuptools-cmake-build/* + +# setup.py intermediates +.eggs +dist +onnx.egg-info +*.ninja +.ninja_deps +.ninja_log +compile_commands.json + +# generated files +compile_commands.json + +# test generated files +.cache +.coverage +onnx/examples/.coverage.nbval +.pytest_cache +test_report +test-output.xml + +# autocomplete +.ycm_extra_conf.py + +# test coverage data files +*.gcov + +.mypy_cache +virtualenv +venv + +# direnv, posh-direnv +.envrc +.psenvrc + +# documentation +docs/docsgen/source/onnx-api/modules/ +docs/docsgen/source/operators/ +docs/docsgen/**/*.onnx +docs/docsgen/**/*.pb + +# PyEnv files +.python-version +.pixi/** +=0.* +_codeql_build_dir/ +_codeql_detected_source_root +/.lycheecache diff --git a/.gitmodules b/.gitmodules new file mode 100644 index 0000000..e69de29 diff --git a/.lintrunner.toml b/.lintrunner.toml new file mode 100644 index 0000000..6dcc689 --- /dev/null +++ b/.lintrunner.toml @@ -0,0 +1,221 @@ +# Configuration for lintrunner https://github.com/suo/lintrunner +# You can install the dependencies and initialize with +# +# ```sh +# pip install lintrunner lintrunner-adapters +# lintrunner init +# ``` +# +# This will install lintrunner on your system and download all the necessary +# dependencies to run linters locally. +# If you want to see what lintrunner init will install, run +# `lintrunner init --dry-run`. +# +# To lint local changes: +# +# ```bash +# lintrunner +# ``` +# +# To lint all files: +# +# ```bash +# lintrunner --all-files +# ``` +# +# To format files: +# +# ```bash +# lintrunner -a +# ``` +# +# To read more about lintrunner, see [wiki](https://github.com/pytorch/pytorch/wiki/lintrunner). +# To update an existing linting rule or create a new one, modify this file or create a +# new adapter following examples in https://github.com/justinchuby/lintrunner-adapters. +merge_base_with = 'main' + +[[linter]] +code = 'RUFF' +include_patterns = [ + '**/*.py', + '**/*.pyi', +] +exclude_patterns = [ + '*_pb2*', + '.setuptools-cmake-build/*', + 'docs/**', +] +command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'ruff_linter', + '--config=pyproject.toml', + '@{{PATHSFILE}}' +] +init_command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'pip_init', + '--dry-run={{DRYRUN}}', + '--requirement=requirements-lintrunner.txt', +] +is_formatter = true + +[[linter]] +code = 'MYPY' +include_patterns = [ + 'onnx/**/*.py', + 'tools/**/*.py', +] +exclude_patterns = [ + 'onnx/backend/test/**', + 'onnx/reference/ops/**', # FIXME: Enable this once typing is fixed + 'onnx/test/**', # Disable mypy for tests + 'onnx/reference/reference_evaluator.py', +] +command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'mypy_linter', + '--config=pyproject.toml', + '--show-disable', + '--', + '@{{PATHSFILE}}' +] +init_command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'pip_init', + '--dry-run={{DRYRUN}}', + '--requirement=requirements-lintrunner.txt', +] + +[[linter]] +code = 'RUFF-FORMAT' +include_patterns = [ + '**/*.py', +] +exclude_patterns = [ + '*_pb2*', + '.setuptools-cmake-build/*', + 'cmake/**', + 'docs/**', +] +command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'ruff_format_linter', + '--', + '@{{PATHSFILE}}' +] +init_command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'pip_init', + '--dry-run={{DRYRUN}}', + '--requirement=requirements-lintrunner.txt', +] +is_formatter = true + +[[linter]] +code = 'NAMESPACE' +include_patterns = ['**/*.cc', '**/*.h'] +exclude_patterns = [] +command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'grep_linter', + '--pattern=namespace onnx|onnx::', + '--linter-name=NAMESPACE', + '--error-name=hardcoded onnx namespace', + """--error-description=\ + Do not hardcode onnx's namespace in the c++ source code, so that \ + other libraries that statically link with onnx can hide onnx symbols \ + in a private namespace.\ + """, + '--', + '@{{PATHSFILE}}' +] + +[[linter]] +code = 'CLANGFORMAT' +include_patterns = [ + 'onnx/**/*.h', + 'onnx/**/*.cc', +] +exclude_patterns = [ +] +command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'clangformat_linter', + '--binary=clang-format', + '--fallback', + '--', + '@{{PATHSFILE}}' +] +init_command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'pip_init', + '--dry-run={{DRYRUN}}', + '--requirement=requirements-lintrunner.txt', +] +is_formatter = true + +[[linter]] +code = 'EDITORCONFIG-CHECKER' +include_patterns=[ + '**/*.py', + '**/*.pyi', + '**/*.cc', + '**/*.h', + '**/*.md', + '**/*.cpp', + '**/*.yml', +] +exclude_patterns = [ + '*_pb2*', + '.setuptools-cmake-build/*', + 'cmake/**', + 'docs/Changelog*', + 'docs/Operators*', + 'docs/TestCoverage*', + 'community/sc-election-guidelines.md', +] +command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'editorconfig_checker_linter', + '--', + '@{{PATHSFILE}}' +] +init_command = [ + 'python', + '-m', + 'lintrunner_adapters', + 'run', + 'pip_init', + '--dry-run={{DRYRUN}}', + '--requirement=requirements-lintrunner.txt', +] diff --git a/.lycheeignore b/.lycheeignore new file mode 100644 index 0000000..8ad34aa --- /dev/null +++ b/.lycheeignore @@ -0,0 +1,15 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +# Used in RFC template as an example +https://github.com/onnx/onnx/pull/0000 + +# Behind cloudflare +https://trademarks.justia.com/877/25/onnx-87725026.html +https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf + +# slack blocks automated requests. This is problematic since the invites expire after ~30 days, but lychee is not going to fix that +https://join.slack.com/t/lfaifoundation/shared_invite/zt-3wx5vohc3-MeSYi3_dscb~u~cqs7zlPg +# Also blocked: +https://en.cppreference.com/c/language/identifier \ No newline at end of file diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml new file mode 100644 index 0000000..e6552c3 --- /dev/null +++ b/.pre-commit-config.yaml @@ -0,0 +1,102 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +# Dependencies are centrally managed by Pixi. Using Pixi is the +# recommended way to obtain linter-results that match what runs on +# CI. A fall-back that looks for the tools in PATH is provided, but +# not recommended. +# +# The fall-back logic hinges on the assumption that every supported +# system provides a `python3` interpreter in PATH. The executed script +# then looks for Pixi. If found the hooks are executed through its +# default environment. Otherwise, they are looked for in PATH directly +# (user-managed). +repos: + - repo: builtin + hooks: + - id: trailing-whitespace + exclude: "^docs/docsgen/source/_static/diff2html" + # TODO: Enable this step which creates some churn + # - id: end-of-file-fixer + # exclude: "^docs/docsgen/source/_static/diff2html" + - id: check-merge-conflict + - id: check-yaml + - id: check-added-large-files + + - repo: local + hooks: + - id: ruff-lint + name: ruff lint + entry: python3 tools/pixi-shim.py ruff check --force-exclude --fix + language: system + types_or: [python, pyi] + exclude: '(_pb2|\.setuptools-cmake-build/|^docs/)' + require_serial: true + + - id: ruff-format + name: ruff format + entry: python3 tools/pixi-shim.py ruff format --force-exclude + language: system + types_or: [python, pyi] + exclude: '(_pb2|\.setuptools-cmake-build/|^cmake/|^docs/)' + require_serial: true + + - id: mypy + name: mypy + entry: python3 tools/pixi-shim.py mypy --config-file=pyproject.toml + language: system + types: [python] + files: "^(onnx|tools)/" + exclude: '(^onnx/backend/test/|^onnx/reference/ops/|^onnx/test/|^onnx/reference/reference_evaluator\.py)' + require_serial: true + + - id: clang-format + name: clang-format + entry: python3 tools/pixi-shim.py clang-format -i + language: system + types_or: [c, c++] + files: "^onnx/" + + - id: namespace-check + name: namespace check (no hardcoded onnx namespace) + entry: python3 tools/pixi-shim.py python tools/check_namespace.py + language: system + types_or: [c, c++] + exclude: "^third_party/" + + # TODO: Enable this step which creates some churn + # - id: prettier + # name: prettier + # entry: pixi run prettier --write --list-different + # language: system + # types: [text] + # files: \.(md|yml|yaml)$ + + - id: typos + name: typos + entry: python3 tools/pixi-shim.py typos --force-exclude + language: system + types: [text] + require_serial: true + exclude: "^docs/docsgen/" + + # REUSE license compliance + - id: reuse + name: reuse lint + entry: python3 tools/pixi-shim.py reuse lint + language: system + pass_filenames: false + + - id: shellcheck + name: shellcheck + entry: python3 tools/pixi-shim.py shellcheck + language: system + types: [shell] + + - id: zizmor + name: zizmor + entry: python3 tools/pixi-shim.py zizmor --fix . + language: system + types: [yaml] + pass_filenames: false diff --git a/CLAUDE.md b/CLAUDE.md new file mode 100644 index 0000000..d06cf00 --- /dev/null +++ b/CLAUDE.md @@ -0,0 +1,81 @@ + + +# CLAUDE.md — ONNX Project Guide + +ONNX (Open Neural Network Exchange) — open-source standard format for AI models. Python + C++ codebase using protobuf for serialization. Builds and runs on **Linux, macOS, and Windows** — keep all three platforms in mind when making changes. + +Also follow the shared AI assistant guidelines in `.github/copilot-instructions.md`. + +## Project Norms + +- Follow the [ONNX Code of Conduct](https://onnx.ai/codeofconduct.html). All generated code, comments, commit messages, and PR descriptions must be professional, welcoming, and free of hostile, discriminatory, or demeaning language. +- ONNX is an **open standard** — changes to operator definitions, proto schemas, or the IR spec affect the entire ML ecosystem. Be conservative and deliberate with spec-level changes. +- Stay **vendor-neutral**. Do not favor any specific framework, runtime, or hardware in code or comments. +- Preserve **backward compatibility**. Breaking changes to the spec or public API have outsized impact across the ecosystem. +- Match existing code patterns and conventions — read surrounding code before making changes. +- Keep PRs focused. Do not bundle unrelated changes or refactor code outside the scope of the task. +- New operators must follow the process in `docs/AddNewOp.md`. + +## Build + +```bash +pip install -e . -v # Development install +ONNX_BUILD_TESTS=1 pip install -e . -v # With C++ tests +``` + +Pure Python changes take effect immediately in editable installs. C++ changes require rebuild. + +## Testing + +```bash +pytest # All Python tests + +# C++ tests (build with ONNX_BUILD_TESTS=1 first) +# Linux/macOS: +LD_LIBRARY_PATH=./.setuptools-cmake-build/ .setuptools-cmake-build/onnx_gtests +# Windows: +.setuptools-cmake-build\Release\onnx_gtests.exe +``` + +Tests live in `onnx/test/` with `*_test.py` naming. + +## Linting + +```bash +lintrunner init # First-time setup +lintrunner # Lint changed files +lintrunner -a # Auto-fix +``` + +Runs ruff, mypy, clang-format, editorconfig-checker, and a namespace checker. + +## Code Conventions + +- All Python files require `from __future__ import annotations` +- No relative imports — use absolute imports from `onnx` +- Copyright header on all files: `# Copyright (c) ONNX Project Contributors` + `# SPDX-License-Identifier: Apache-2.0` +- DCO sign-off required on all commits (`git commit -s`) + +## Auto-Generated Files (Do Not Edit) + +Edit the source, then regenerate. CI verifies these are up to date. + +| Generated files | Source of truth | Regenerate with | +|---|---|---| +| `docs/Operators.md`, `docs/Changelog.md`, `docs/TestCoverage.md` | Op schemas in `onnx/defs/` | `python onnx/defs/gen_doc.py` | +| `onnx/*_pb2.py`, `onnx/*_pb.h`, `onnx/onnx_data.proto` | `onnx/onnx.in.proto`, `onnx/onnx-ml.in.proto` | `python onnx/gen_proto.py` | +| `onnx/backend/test/data/` node tests | Op schemas + reference impl | `python onnx/backend/test/stat_coverage.py` | + +Edit `.in.proto` files, **not** `.proto` files. When adding/changing operator schemas, run all three scripts. + +## C++/Python Boundary + +Core validation (`checker`), shape inference, and version conversion are C++ exposed via nanobind (`onnx_cpp2py_export/`). Operator schemas are defined in C++ under `onnx/defs/`. Helper utilities, reference implementation, parser, and compose are pure Python. + +**ONNX_ML flag** (on by default): controls traditional ML types (sequences, maps, sparse tensors). When enabled, builds use `onnx-ml.in.proto` instead of `onnx.in.proto`. + +Build artifacts go to `.setuptools-cmake-build/`. diff --git a/CMakeLists.txt b/CMakeLists.txt new file mode 100644 index 0000000..6f02a5a --- /dev/null +++ b/CMakeLists.txt @@ -0,0 +1,681 @@ +# Minimum CMake required +cmake_minimum_required(VERSION 3.26) +# Project +project(onnx LANGUAGES CXX) + +include(cmake/Utils.cmake) +# Honor visibility properties for all target types. +cmake_policy(SET CMP0063 NEW) +# Modules FindPython3 and FindPython use LOCATION for lookup strategy. +cmake_policy(SET CMP0094 NEW) + +# Set default build type +get_property(isMultiConfig GLOBAL PROPERTY GENERATOR_IS_MULTI_CONFIG) +if(NOT isMultiConfig AND NOT CMAKE_BUILD_TYPE) + message(STATUS "Build type not set - defaulting to Release") + set( + CMAKE_BUILD_TYPE "Release" + CACHE + STRING + "Choose the type of build from: Debug Release RelWithDebInfo MinSizeRel Coverage." + FORCE) +endif() + +# https://reproducible-builds.org/docs/source-date-epoch/ +# cmake's string(TIMESTAMP ...) command reads SOURCE_DATE_EPOCH automatically, +# which avoids shelling out to `date` (whose flags differ between GNU and BSD). +string(TIMESTAMP BUILD_DATE "%Y-%m-%d" UTC) + +if(NOT BUILD_SHARED_LIBS) + # by default, cmake builds static libraries + set(BUILD_SHARED_LIBS OFF) +endif() + + +option(ONNX_BUILD_PYTHON "Build Python binaries" OFF) +option(ONNX_BUILD_CUSTOM_PROTOBUF "Build and use ONNX's own protobuf" OFF) +option(ONNX_USE_PROTOBUF_SHARED_LIBS "Build ONNX using protobuf shared library." OFF) +option(ONNX_GEN_PB_TYPE_STUBS "Generate protobuf python type stubs" ON) +option(ONNX_WERROR "Build with Werror" OFF) +option(ONNX_COVERAGE "Build with coverage instrumentation" OFF) +option(ONNX_BUILD_TESTS "Build ONNX C++ APIs Tests" OFF) +option(ONNX_USE_ASAN "Build ONNX with ASAN" OFF) +option(ONNX_HARDENING "Build with compiler hardening flags (OpenSSF guidelines)" OFF) +option(ONNX_USE_LITE_PROTO "Use lite protobuf instead of full." OFF) +option(ONNX_DISABLE_EXCEPTIONS "Disable exception handling." OFF) +option(ONNX_DISABLE_STATIC_REGISTRATION "Disable static registration for ONNX operator schemas." OFF) +option(ONNX_USE_UNITY_BUILD "Enable Unity (Jumbo) build for" OFF) +option(ONNX_PYODIDE_BUILD "Build ONNX for Pyodide/Emscripten (wasm32)." OFF) +option(ONNX_INSTALL "Install ONNX targets, headers, and CMake config files" ON) +if(WIN32) + option(ONNX_USE_MSVC_STATIC_RUNTIME "Build with MSVC static runtime. Ignored if CMAKE_MSVC_RUNTIME_LIBRARY is defined." OFF) +endif() + +set(CMAKE_EXPORT_COMPILE_COMMANDS ON) + +if(NOT DEFINED ONNX_ML) + if(DEFINED ENV{ONNX_ML}) + set(DEFAULT_ONNX_ML $ENV{ONNX_ML}) + else() + set(DEFAULT_ONNX_ML ON) + endif() + option(ONNX_ML "Enable traditional ML API." ${DEFAULT_ONNX_ML}) +endif() + +if(NOT DEFINED ONNX_VERIFY_PROTO3) + if(DEFINED ENV{ONNX_VERIFY_PROTO3}) + set(PROTO3_ENABLED $ENV{ONNX_VERIFY_PROTO3}) + else() + set(PROTO3_ENABLED OFF) + endif() + option(ONNX_VERIFY_PROTO3 "Generate code by proto3" ${PROTO3_ENABLED}) +endif() + +if(NOT DEFINED CMAKE_CXX_STANDARD) + set(CMAKE_CXX_STANDARD 17) +else() + if(CMAKE_CXX_STANDARD VERSION_LESS 17) + message(FATAL_ERROR "At least C++17 is required.") + endif() +endif() + +include(GNUInstallDirs) + +set(ONNX_ROOT ${onnx_SOURCE_DIR}) + +# Read ONNX version +file(READ "${ONNX_ROOT}/VERSION_NUMBER" ONNX_VERSION) +string(STRIP "${ONNX_VERSION}" ONNX_VERSION) + +# Parse version components for version macros (strip any pre-release suffix like "rc1") +string(REGEX MATCH "^([0-9]+)\\.([0-9]+)\\.([0-9]+)" _ ${ONNX_VERSION}) +set(ONNX_VERSION_MAJOR ${CMAKE_MATCH_1}) +set(ONNX_VERSION_MINOR ${CMAKE_MATCH_2}) +set(ONNX_VERSION_PATCH ${CMAKE_MATCH_3}) + +if(NOT MSVC) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wnon-virtual-dtor") + set(CMAKE_C_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -O0") + set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -O0") + if(ONNX_COVERAGE) + set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fprofile-arcs -ftest-coverage") + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fprofile-arcs -ftest-coverage") + endif() +endif() + +if(NOT ONNX_NAMESPACE) + set(ONNX_NAMESPACE "onnx") +endif() + +if(MSVC) + if(NOT ONNX_DISABLE_EXCEPTIONS) + string(APPEND CMAKE_CXX_FLAGS " /EHsc /wd26812") + string(APPEND CMAKE_C_FLAGS " /EHsc /wd26812") + endif() + add_compile_options(/MP /utf-8 /nologo) + add_compile_options( + /wd5287 # https://developercommunity.visualstudio.com/t/warning-C5287:-operands-are-different-e/10877942?sort=newest + /Zc:lambda # https://developercommunity.visualstudio.com/t/fatal--error-C1001:-Internal-compiler-er/10906076 + ) +endif() + +if(ONNX_DISABLE_EXCEPTIONS) + add_compile_definitions("ONNX_NO_EXCEPTIONS") + # Disable C++ exceptions. + if(MSVC) + string(REGEX REPLACE "/EHsc" "/EHs-c-" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}") + add_definitions(-D_HAS_EXCEPTIONS=0) + else() + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-exceptions -fno-unwind-tables -fno-asynchronous-unwind-tables") + endif() +endif() + +if(ONNX_BUILD_PYTHON AND ONNX_DISABLE_EXCEPTIONS) + message(FATAL_ERROR + "ONNX_DISABLE_EXCEPTIONS=ON is incompatible with ONNX_BUILD_PYTHON=ON. " + "The Python binding requires exceptions to be enabled.") +endif() + +if(ONNX_BUILD_PYTHON) + set(python_dev_component Development.Module Development.SABIModule) +endif() + +if(CMAKE_CROSSCOMPILING) + # When cross-compiling, the interpreter and the compiling/linking steps + # must use a different package. See the discussion about this at + # https://gitlab.kitware.com/cmake/cmake/-/issues/25145 + if(ONNX_BUILD_PYTHON) + find_package(Python3 REQUIRED COMPONENTS ${python_dev_component}) + endif() + find_package(Python REQUIRED COMPONENTS Interpreter) + set(ONNX_PYTHON_INTERPRETER Python::Interpreter) + if(ONNX_BUILD_PYTHON) + # nanobind >= 2.13 only auto-runs its own find_package(Python ...) when + # neither Python::Interpreter nor Python::Module exist yet. Since + # Python::Interpreter is already provided above (host interpreter, while + # Python3::Module above is the cross-compilation target's dev libs), + # nanobind would otherwise skip that step and fail to find Python::Module. + if(NOT TARGET Python::Module) + add_library(Python::Module ALIAS Python3::Module) + endif() + if(TARGET Python3::SABIModule AND NOT TARGET Python::SABIModule) + add_library(Python::SABIModule ALIAS Python3::SABIModule) + endif() + endif() +else() + find_package(Python3 REQUIRED COMPONENTS Interpreter ${python_dev_component}) + # Find Python for nanobind + set(Python_EXECUTABLE ${Python3_EXECUTABLE}) + find_package(Python REQUIRED COMPONENTS Interpreter ${python_dev_component}) + set(ONNX_PYTHON_INTERPRETER Python3::Interpreter) +endif() + +if(CMAKE_SYSTEM_NAME STREQUAL "AIX") + set(CMAKE_NO_SYSTEM_FROM_IMPORTED 1) +endif() + +# Build the libraries with -fPIC including the protobuf lib. +if(NOT DEFINED CMAKE_POSITION_INDEPENDENT_CODE) + set(CMAKE_POSITION_INDEPENDENT_CODE ON) +endif() + +list(APPEND CMAKE_MODULE_PATH ${ONNX_ROOT}/cmake/external) +if(NOT ONNX_BUILD_CUSTOM_PROTOBUF) + if((ONNX_USE_LITE_PROTO AND TARGET protobuf::libprotobuf-lite) OR ((NOT ONNX_USE_LITE_PROTO) AND TARGET protobuf::libprotobuf)) + # Sometimes we need to use protoc compiled for host architecture while linking + # libprotobuf against target architecture. See https://github.com/caffe2/caffe + # 2/blob/96f35ad75480b25c1a23d6e9e97bccae9f7a7f9c/cmake/ProtoBuf.cmake#L92-L99 + if(EXISTS "${ONNX_CUSTOM_PROTOC_EXECUTABLE}") + message(STATUS "Using custom protoc executable") + set(ONNX_PROTOC_EXECUTABLE ${ONNX_CUSTOM_PROTOC_EXECUTABLE}) + else() + if(TARGET protobuf::protoc) + set(ONNX_PROTOC_EXECUTABLE $) + endif() + endif() + else() + # Customized version of find Protobuf. We need to avoid situations mentioned + # in https://github.com/caffe2/caffe2/blob/b7d983f255ef5496474f1ea188edb5e0ac4 + # 42761/cmake/ProtoBuf.cmake#L82-L92 The following section is stolen from + # cmake/ProtoBuf.cmake in Caffe2 + find_program(Protobuf_PROTOC_EXECUTABLE + NAMES protoc + DOC "The Google Protocol Buffers Compiler") + + # Only if protoc was found, seed the include directories and libraries. We + # assume that protoc is installed at PREFIX/bin. We use get_filename_component + # to resolve PREFIX. + if(Protobuf_PROTOC_EXECUTABLE) + set(ONNX_PROTOC_EXECUTABLE ${Protobuf_PROTOC_EXECUTABLE}) + get_filename_component(_PROTOBUF_INSTALL_PREFIX + ${Protobuf_PROTOC_EXECUTABLE} DIRECTORY) + get_filename_component(_PROTOBUF_INSTALL_PREFIX + ${_PROTOBUF_INSTALL_PREFIX}/.. REALPATH) + find_library(Protobuf_PROTOC_LIBRARY + NAMES protoc + PATHS ${_PROTOBUF_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBDIR} + NO_DEFAULT_PATH) + if(ONNX_USE_LITE_PROTO) + find_library(Protobuf_LITE_LIBRARY + NAMES protobuf-lite + PATHS ${_PROTOBUF_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBDIR} + NO_DEFAULT_PATH) + else() + find_library(Protobuf_LIBRARY + NAMES protobuf + PATHS ${_PROTOBUF_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBDIR} + NO_DEFAULT_PATH) + endif(ONNX_USE_LITE_PROTO) + find_path(Protobuf_INCLUDE_DIR google/protobuf/service.h + PATHS ${_PROTOBUF_INSTALL_PREFIX}/include + NO_DEFAULT_PATH) + if(ONNX_USE_PROTOBUF_SHARED_LIBS) + set(Protobuf_USE_STATIC_LIBS OFF) + else() + set(Protobuf_USE_STATIC_LIBS ON) + endif() + find_package(Protobuf) + if(Protobuf_FOUND) + set(PROTOBUF_DIR "${_PROTOBUF_INSTALL_PREFIX}") + set(Build_Protobuf OFF) + if("${Protobuf_VERSION}" VERSION_GREATER_EQUAL "4.22.0") + # There are extra dependencies for protobuf. + find_package(absl REQUIRED) + find_package(utf8_range) + message(STATUS "absl_VERSION: ${absl_VERSION}") + set(protobuf_ABSL_USED_TARGETS + absl::absl_check + absl::absl_log + absl::algorithm + absl::base + absl::bind_front + absl::bits + absl::btree + absl::cleanup + absl::cord + absl::core_headers + absl::debugging + absl::die_if_null + absl::dynamic_annotations + absl::flags + absl::flat_hash_map + absl::flat_hash_set + absl::function_ref + absl::hash + absl::layout + absl::log_initialize + absl::log_severity + absl::memory + absl::node_hash_map + absl::node_hash_set + absl::optional + absl::span + absl::status + absl::statusor + absl::strings + absl::synchronization + absl::time + absl::type_traits + absl::utility + absl::variant + utf8_range::utf8_range + utf8_range::utf8_validity + ) + endif() + endif() + endif() + endif() +endif() + +if(ONNX_PYODIDE_BUILD) + # Prevent absl from using EM_JS-based symbolize_emscripten.inc which is + # incompatible with Emscripten SIDE_MODULE builds (Pyodide). + add_compile_definitions(STANDALONE_WASM) +endif() + +if(NOT ONNX_PROTOC_EXECUTABLE) + set(Build_Protobuf ON) + set(protobuf_MSVC_STATIC_RUNTIME ${ONNX_USE_MSVC_STATIC_RUNTIME}) + + include(FetchContent) + set(ABSL_PROPAGATE_CXX_STD 1) + set(ONNX_BUILD_SHARED_LIBS ${BUILD_SHARED_LIBS}) + set(ONNX_CMAKE_CXX_FLAGS ${CMAKE_CXX_FLAGS}) + # Use this setting to build third-party libs. + set(BUILD_SHARED_LIBS ${ONNX_USE_PROTOBUF_SHARED_LIBS}) + # Read dependency URLs/hashes from sbom.cdx.json (single source of truth). + sbom_get_dep("abseil-cpp" _absl) + sbom_get_dep("protobuf" _protobuf) + # Assign ${Protobuf_VERSION} to the version read from sbom.cdx.json like `find_package(Protobuf)` would do. + set(Protobuf_VERSION ${_protobuf_VERSION}) + if(NOT DEFINED ABSL_ENABLE_INSTALL) + if(ONNX_BUILD_PYTHON) + set(ABSL_ENABLE_INSTALL OFF CACHE BOOL "Enable installation of Abseil targets") + else() + set(ABSL_ENABLE_INSTALL ON CACHE BOOL "Enable installation of Abseil targets") + endif() + endif() + FetchContent_Declare( + absl + URL ${_absl_URL} + URL_HASH SHA256=${_absl_SHA256} + ) + FetchContent_MakeAvailable(absl) + FetchContent_Declare( + Protobuf + URL ${_protobuf_URL} + URL_HASH SHA256=${_protobuf_SHA256} + ) + if(NOT DEFINED protobuf_BUILD_TESTS) + set(protobuf_BUILD_TESTS OFF CACHE BOOL "Build protobuf tests") + endif() + if(EXISTS "${ONNX_CUSTOM_PROTOC_EXECUTABLE}") + # We already have a host protoc binary; skip building libprotoc and protoc + # (saves compiling code generators for Java, C#, Kotlin, ObjC, etc.) + set(protobuf_BUILD_PROTOC_BINARIES OFF CACHE BOOL "Skip building protoc") + # libupb is only needed by protoc's code generators; skip it too. + set(protobuf_BUILD_LIBUPB OFF CACHE BOOL "Skip building libupb") + endif() + set(protobuf_WITH_ZLIB OFF CACHE BOOL "Skip zlib support") + if(ONNX_BUILD_PYTHON) + set(protobuf_INSTALL OFF CACHE BOOL "Skip protobuf install rules") + endif() + message(STATUS "Download and build Protobuf from ${_protobuf_URL}") + FetchContent_MakeAvailable(Protobuf) + # utf8_validity.h is included by protobuf headers but its include dir + # is only added via include_directories() inside protobuf's directory scope. + # Attach it to the protobuf target so ONNX targets can compile .pb.cc files + # without introducing a global include path. + if(EXISTS "${protobuf_SOURCE_DIR}/third_party/utf8_range") + # protobuf::libprotobuf and protobuf::libprotobuf-lite are ALIAS targets + # and CMake does not allow target_include_directories on aliases; + # resolve to the real targets before adding the utf8_range include path. + foreach(_pb_alias libprotobuf libprotobuf-lite) + if(TARGET protobuf::${_pb_alias}) + get_target_property(_pb_real protobuf::${_pb_alias} ALIASED_TARGET) + if(NOT _pb_real) + set(_pb_real protobuf::${_pb_alias}) + endif() + target_include_directories( + ${_pb_real} + INTERFACE + $ + ) + endif() + endforeach() + endif() + if(EXISTS "${ONNX_CUSTOM_PROTOC_EXECUTABLE}") + set(ONNX_PROTOC_EXECUTABLE ${ONNX_CUSTOM_PROTOC_EXECUTABLE}) + else() + set(ONNX_PROTOC_EXECUTABLE $) + endif() + # Change back the BUILD_SHARED_LIBS to control the onnx project. + set(BUILD_SHARED_LIBS ${ONNX_BUILD_SHARED_LIBS}) + set(PROTOBUF_DIR "${protobuf_BINARY_DIR}") + set(CMAKE_CXX_FLAGS ${ONNX_CMAKE_CXX_FLAGS}) +endif() +message(STATUS "ONNX_PROTOC_EXECUTABLE: ${ONNX_PROTOC_EXECUTABLE}") + +# function(RELATIVE_PROTOBUF_GENERATE_CPP SRCS HDRS ROOT_DIR) from +# https://github.com/tensorflow/tensorflow/blob/d2c3b873/tensorflow/contrib/cmake/tf_core_framework.cmake +# to solve the problem that customized dir can't be specified when +# calling PROTOBUF_GENERATE_CPP. +function(RELATIVE_PROTOBUF_GENERATE_CPP SRCS) + if(NOT ARGN) + message( + SEND_ERROR + "Error: RELATIVE_PROTOBUF_GENERATE_CPP() called without any proto files" + ) + return() + endif() + + set(${SRCS}) + set(_py_files) + + set(GEN_PROTO_PY "${ONNX_ROOT}/onnx/gen_proto.py") + set(GENERATED_FILES) + foreach(INFILE ${ARGN}) + set(ABS_FILE "${ONNX_ROOT}/${INFILE}") + get_filename_component(FILE_DIR ${ABS_FILE} DIRECTORY) + get_filename_component(FILE_WE ${INFILE} NAME_WE) + # "onnx-data" check is because we do not want to create/compile an "onnx-data-ml.proto" file + if(ONNX_ML AND NOT(FILE_WE STREQUAL "onnx-data")) + if(ONNX_NAMESPACE STREQUAL "onnx") + set(GENERATED_FILE_WE "${FILE_WE}-ml") + else() + set(GENERATED_FILE_WE "${FILE_WE}_${ONNX_NAMESPACE}-ml") + endif() + else() + if(ONNX_NAMESPACE STREQUAL "onnx") + set(GENERATED_FILE_WE "${FILE_WE}") + else() + set(GENERATED_FILE_WE "${FILE_WE}_${ONNX_NAMESPACE}") + endif() + endif() + file(RELATIVE_PATH REL_DIR "${ONNX_ROOT}" "${FILE_DIR}") + set(OUTPUT_PROTO_DIR "${CMAKE_CURRENT_BINARY_DIR}/${REL_DIR}") + + set(OUTPUT_PB_SRC "${OUTPUT_PROTO_DIR}/${GENERATED_FILE_WE}.pb.cc") + set(GENERATED_PROTO "${OUTPUT_PROTO_DIR}/${GENERATED_FILE_WE}.proto") + list(APPEND ${SRCS} "${OUTPUT_PB_SRC}") + + if(NOT EXISTS "${OUTPUT_PROTO_DIR}") + file(MAKE_DIRECTORY "${OUTPUT_PROTO_DIR}") + endif() + + set(GEN_PROTO_ARGS + -p + "${ONNX_NAMESPACE}" + -o + "${OUTPUT_PROTO_DIR}" + "${FILE_WE}") + if(ONNX_ML) + list(APPEND GEN_PROTO_ARGS -m) + endif() + if(ONNX_USE_LITE_PROTO) + list(APPEND GEN_PROTO_ARGS -l) + endif() + if(ONNX_VERIFY_PROTO3) + if(NOT ONNX_PROTOC_EXECUTABLE) + message(FATAL_ERROR "Protobuf compiler not found") + endif() + list(APPEND GEN_PROTO_ARGS --protoc_path) + list(APPEND GEN_PROTO_ARGS "${ONNX_PROTOC_EXECUTABLE}") + endif() + + # Use add_custom_command to avoid re-generate of PROTO files + set(deps ${INFILE}) + if(TARGET protobuf::protoc) + list(APPEND deps protobuf::protoc) + endif() + set(_gen_proto_outputs "${GENERATED_PROTO}") + if(ONNX_BUILD_PYTHON) + # gen_proto.py also produces a _pb.py re-export wrapper + string(REPLACE "-" "_" _stem_py "${FILE_WE}") + set(_pb_py "${OUTPUT_PROTO_DIR}/${_stem_py}_pb.py") + list(APPEND _gen_proto_outputs "${_pb_py}") + list(APPEND _py_files "${_pb_py}") + endif() + + add_custom_command(OUTPUT ${_gen_proto_outputs} + COMMAND ${ONNX_PYTHON_INTERPRETER} "${GEN_PROTO_PY}" ${GEN_PROTO_ARGS} + DEPENDS ${deps} + COMMENT "Running gen_proto.py on ${INFILE}") + message("Generated: ${GENERATED_PROTO}") + + set(PROTOC_ARGS + ${GENERATED_PROTO} + -I + ${CMAKE_CURRENT_BINARY_DIR} + --cpp_out + ${CMAKE_CURRENT_BINARY_DIR}) + if(ONNX_BUILD_PYTHON) + list(APPEND PROTOC_ARGS --python_out) + if(ONNX_GEN_PB_TYPE_STUBS) + list(APPEND PROTOC_ARGS pyi_out:${CMAKE_CURRENT_BINARY_DIR}) + else() + list(APPEND PROTOC_ARGS ${CMAKE_CURRENT_BINARY_DIR}) + endif() + endif() + list(APPEND GENERATED_FILES "${GENERATED_PROTO}") + + set(_protoc_outputs "${OUTPUT_PB_SRC}") + if(ONNX_BUILD_PYTHON) + string(REPLACE "-" "_" _py_stem "${GENERATED_FILE_WE}") + set(_pb2_py "${OUTPUT_PROTO_DIR}/${_py_stem}_pb2.py") + list(APPEND _protoc_outputs "${_pb2_py}") + list(APPEND _py_files "${_pb2_py}") + if(ONNX_GEN_PB_TYPE_STUBS) + set(_pb2_pyi "${OUTPUT_PROTO_DIR}/${_py_stem}_pb2.pyi") + list(APPEND _protoc_outputs "${_pb2_pyi}") + list(APPEND _py_files "${_pb2_pyi}") + endif() + endif() + + add_custom_command(OUTPUT ${_protoc_outputs} + COMMAND "${ONNX_PROTOC_EXECUTABLE}" ${PROTOC_ARGS} + DEPENDS ${GENERATED_FILES} + COMMENT "Running C++ protocol buffer compiler on ${GENERATED_PROTO}") + endforeach() + + set(${SRCS} ${${SRCS}} PARENT_SCOPE) + set(ONNX_PROTO_PY_FILES ${_py_files} PARENT_SCOPE) + set(ONNX_GENERATED_PROTO_FILES ${GENERATED_FILES} PARENT_SCOPE) +endfunction() + +relative_protobuf_generate_cpp(ONNX_PROTO_SRCS + onnx/onnx.in.proto + onnx/onnx-operators.in.proto + onnx/onnx-data.in.proto) + +# Serialises the gen_proto.py step so parallel builds don't race writing the +# generated .proto files; protoc and C++ compilation still overlap afterwards. +add_custom_target(onnx_proto_gen DEPENDS ${ONNX_GENERATED_PROTO_FILES}) + +set(LINKED_PROTOBUF_TARGET protobuf::libprotobuf) +if(ONNX_USE_LITE_PROTO) + if(TARGET protobuf::libprotobuf-lite) + set(LINKED_PROTOBUF_TARGET protobuf::libprotobuf-lite) + endif() +endif() + +# Compile the generated .pb.cc sources in exactly one target: Xcode's build +# system rejects a protoc custom command shared across several targets. +add_library(onnx_proto ${ONNX_PROTO_SRCS}) +add_onnx_global_defines(onnx_proto) +add_dependencies(onnx_proto onnx_proto_gen) + +# Core library. Hand-written sources come from target_sources(onnx ...) in the +# subdirectories; the protobuf objects are pulled in from onnx_proto. +add_library(onnx $) +add_subdirectory(onnx) +add_dependencies(onnx onnx_proto) +set_target_properties(onnx PROPERTIES CXX_VISIBILITY_PRESET hidden) +add_onnx_global_defines(onnx) + +if(ONNX_BUILD_PYTHON) + if(BUILD_SHARED_LIBS) + # The extension links ONNX internals not on the public ONNX_API surface, + # which a hidden-visibility shared libonnx does not export. The wheel always + # builds onnx statically; a shared libonnx is for C++ use only. + message(FATAL_ERROR + "ONNX_BUILD_PYTHON=ON requires a static onnx; do not combine it with " + "BUILD_SHARED_LIBS=ON. Use ONNX_BUILD_PYTHON=OFF to build a shared libonnx.") + endif() + # find system nanobind + if(NOT DEFINED nanobind_DIR OR "${nanobind_DIR}" STREQUAL "") + execute_process( + COMMAND ${Python_EXECUTABLE} -m nanobind --cmake_dir + RESULT_VARIABLE NANOBIND_CMAKE_RESULT + OUTPUT_VARIABLE NANOBIND_CMAKE_DIR + OUTPUT_STRIP_TRAILING_WHITESPACE + ) + if(NANOBIND_CMAKE_RESULT EQUAL 0 AND EXISTS "${NANOBIND_CMAKE_DIR}") + set(nanobind_DIR "${NANOBIND_CMAKE_DIR}") + endif() + endif() + find_package(nanobind CONFIG QUIET) + if(NOT nanobind_FOUND) + if(FETCHCONTENT_FULLY_DISCONNECTED) + message(FATAL_ERROR + "nanobind not found while FETCHCONTENT_FULLY_DISCONNECTED is ON. " + "Provide nanobind locally (nanobind_DIR/CMAKE_PREFIX_PATH) or use a " + "FetchContent dependency provider for first-run offline builds.") + endif() + include(FetchContent) + sbom_get_dep("nanobind" _nanobind) + FetchContent_Declare( + nanobind + GIT_REPOSITORY ${_nanobind_URL} + GIT_TAG v${_nanobind_VERSION} + GIT_SHALLOW TRUE + ) + FetchContent_MakeAvailable(nanobind) + endif() + + # Configure nanobind: https://nanobind.readthedocs.io/en/latest/api_cmake.html + nanobind_add_module( + onnx_cpp2py_export + NB_STATIC NB_DOMAIN onnx STABLE_ABI FREE_THREADED LTO + "${ONNX_ROOT}/onnx/cpp2py_export.cc") + # Force-load the whole (static) archive so the operator schemas' static + # initializers are not dropped as apparently-unused. $ + # is the portable --whole-archive / -force_load / /WHOLEARCHIVE spelling. + target_link_libraries(onnx_cpp2py_export PRIVATE + $) +endif() + +add_onnx_compile_options(onnx_proto) +add_onnx_compile_options(onnx) +if(TARGET onnx_cpp2py_export) + add_onnx_global_defines(onnx_cpp2py_export) + add_onnx_compile_options(onnx_cpp2py_export) +endif() +if(ONNX_USE_ASAN AND NOT MSVC) + find_package(Sanitizer REQUIRED) + if(TARGET Sanitizer::address) + target_link_libraries(onnx PRIVATE Sanitizer::address) + message(STATUS "Use ASAN") + endif() + if(TARGET Sanitizer::undefined) + target_link_libraries(onnx PRIVATE Sanitizer::undefined) + message(STATUS "Use UBSAN") + endif() +endif() + +if(ONNX_INSTALL) + install(DIRECTORY ${ONNX_ROOT}/onnx + DESTINATION ${CMAKE_INSTALL_INCLUDEDIR} + FILES_MATCHING + PATTERN "*.h" + PATTERN "backend/test/case" EXCLUDE + PATTERN "backend/test/data" EXCLUDE) + install(DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/onnx + DESTINATION ${CMAKE_INSTALL_INCLUDEDIR} + FILES_MATCHING + PATTERN "*.h") +endif() + +configure_file( + ${PROJECT_SOURCE_DIR}/cmake/ONNXConfigVersion.cmake.in + ${PROJECT_BINARY_DIR}/ONNXConfigVersion.cmake + @ONLY) +configure_file( + ${PROJECT_SOURCE_DIR}/cmake/ONNXConfig.cmake.in + ${PROJECT_BINARY_DIR}/ONNXConfig.cmake + @ONLY) +if(ONNX_INSTALL) + install(FILES + ${PROJECT_BINARY_DIR}/ONNXConfigVersion.cmake + ${PROJECT_BINARY_DIR}/ONNXConfig.cmake + DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/ONNX + COMPONENT dev) +endif() + +if(ONNX_USE_UNITY_BUILD) + # If ONNX_USE_UNITY_BUILD is set to ON, set onnx to use Unity builds. + set_target_properties(onnx + PROPERTIES + UNITY_BUILD ON + ) + + # With unity build object file could get big, need this switch in MSVC. + if(MSVC) + target_compile_options(onnx PRIVATE /bigobj) + endif() +# should be enabled for onnx_proto when protobuf can support Unity builds +endif() + +if(ONNX_BUILD_TESTS) + find_package(GTest) + if(NOT GTest_FOUND) + include(googletest) + endif() +endif() + +if(ONNX_INSTALL) + install(TARGETS + onnx onnx_proto + EXPORT ONNXTargets DESTINATION ${CMAKE_INSTALL_LIBDIR}) + + install(EXPORT ONNXTargets + DESTINATION "${CMAKE_INSTALL_LIBDIR}/cmake/ONNX" + NAMESPACE ONNX:: + ) +endif() + +if(ONNX_BUILD_PYTHON) + install(TARGETS onnx_cpp2py_export LIBRARY DESTINATION onnx) + install(FILES ${ONNX_PROTO_PY_FILES} DESTINATION onnx) + # PEP 770: embed SBOM in dist-info/sboms/ inside the wheel. + if(DEFINED SKBUILD_METADATA_DIR) + install(FILES "${CMAKE_SOURCE_DIR}/sbom.cdx.json" DESTINATION "${SKBUILD_METADATA_DIR}/sboms") + endif() +endif() + +if(ONNX_BUILD_TESTS) + include(${ONNX_ROOT}/cmake/unittest.cmake) +endif() + +include(cmake/summary.cmake) +onnx_print_configuration_summary() diff --git a/CODEOWNERS b/CODEOWNERS new file mode 100644 index 0000000..3379773 --- /dev/null +++ b/CODEOWNERS @@ -0,0 +1,18 @@ +# ONNX Project Code Owners +# See https://github.com/orgs/onnx/teams for team structure +# +# https://github.com/onnx/sigs/blob/main/CONTRIBUTORS + +* @onnx/sig-archinfra-approvers +/community @onnx/steering-committee +/onnx/defs @onnx/sig-operators-approvers +/onnx/defs/parser.* @onnx/sig-archinfra-approvers +/onnx/defs/printer.* @onnx/sig-archinfra-approvers +/onnx/backend/test @onnx/sig-operators-approvers +/onnx/reference/ops @onnx/sig-operators-approvers +/docs/AddNewOp.md @onnx/sig-operators-approvers +/docs/TestCoverage*.md @onnx/sig-operators-approvers +/docs/Operators*.md @onnx/sig-operators-approvers +/docs/OpConventions.md @onnx/sig-operators-approvers +/docs/Broadcasting.md @onnx/sig-operators-approvers +/docs/Changelog*.md @onnx/sig-operators-approvers diff --git a/CODE_OF_CONDUCT.md b/CODE_OF_CONDUCT.md new file mode 100644 index 0000000..f282992 --- /dev/null +++ b/CODE_OF_CONDUCT.md @@ -0,0 +1 @@ +The ONNX Code Of Conduct is described at https://onnx.ai/codeofconduct.html diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md new file mode 100644 index 0000000..d8a0d47 --- /dev/null +++ b/CONTRIBUTING.md @@ -0,0 +1,271 @@ + + +# ONNX Community Involvement and Contribution Guidelines + +ONNX is a community project and we welcome your contributions! In addition to contributing code, you can also contribute in many other ways: + +- Meetings and Discussions + + Join SIGS, Working Groups, Community meetings to learn about what is needed and then where there is a good fit to interest and areas of expertise, find ways to actively contribute. Participate in [ONNX technical discussions](https://github.com/onnx/onnx/discussions) on GitHub. Join the ONNX Slack channels at LF AI and Data, help answer questions and welcome new members. + +- Use Cases and Tools + + Develop use cases for ONNX and advocate for ONNX in developer conferences and meetups. Develop tools that import and export using the ONNX spec, and help grow the community of ONNX users. Become a champion for ONNX in your company or organization. + +- Roadmap and Features + + Understand the ONNX roadmap document, feature priorities, and help implement them. Become an ONNX code and documentation contributor, and work towards committer status on important repos. + +- Releases and Model Zoo + + Help in achieving a release of ONNX, including increasing the number of models in the ONNX Model Zoo that exercise ONNX features. + +- Publications and Blogs + + Add to the growing number of arXiv papers that refer to ONNX. Create blogs, presentations, books, articles and other materials that help increase the adoption of ONNX, and grow the community of users and contributors. + +- Steering Committee + + Attend ONNX Steering Committee meetings - they are open to all in the community. Help out where needed and appropriate on SC to-do items. Note that SIG and Working Groups leaders as well as others with demonstrated commitment and contributions to ONNX community may want to self-nominate during the annual SC election cycle. + +## Adding a new operator or creating a new version of an existing operator + +ONNX is an open standard, and we encourage developers to contribute high +quality operators to ONNX specification. + +Before proposing a new operator, please read [the tutorial](docs/AddNewOp.md). + +## Contributing code + +You can submit a pull request (PR) with your code. The [SIG](community/sigs.md) or [Working Group](community/working-groups.md) that is responsible for the area of the project your PR touches will review it and merge once any comments are addressed. + +### Pixi-based development + +The easiest and most reproducible developer experience for this project is provided through [Pixi](https://pixi.prefix.dev/latest/) - a cross-platform package manager based on conda-forge. + +Running + +```bash +pixi run install +``` + +In the root of this repository compiles and installs the onnx package editably in a pixi-managed environment. +Optionally, users may subsequently call `ln -s .setuptools-cmake-build/compile_commands.json` on Unix systems to create a `compile_commands.json` symlink where tools such as `clangd` (code navigation) and `clang-tidy` expect it. + +Pre-commit hooks, that are identical to those run on CI, can be installed by running: + +```bash +pixi run pre-commit-install +``` + +A list of all pixi task is available by running `pixi run`. +The following is a list of the most common ones: + +- `pixi run gen-all` to regenerate all auto-generated files +- `pixi run gtest` to run the googletest suite +- `pixi run pytest` to run the Python test suite +- `pixi run lint` to run the pre-commit hooks on all files +- `pixi run docs-build` to build the documentation locally (may require a prior `rm -rf .setuptools-cmake-build && pixi run -e docs install`) + +The following section provide guidance for developing onnx without Pixi and is not relevant to Pixi users. + +### Development + +To build ONNX from source please follow the instructions listed [here](https://github.com/onnx/onnx/blob/main/INSTALL.md#build-onnx-from-source). + +Then, after you have made changes to Python and C++ files: + +- `Python files`: The changes are effective immediately in your installation. You don't need to install these again. +- `C++ files`: You need to install these again to trigger the native extension build. + +Assuming build succeed in the initial step, simply running + +```sh +pip install -e . -v +``` + +from onnx root dir should work. + +### Folder structure + +- `onnx/`: the main folder that all code lies under + - `onnx.proto`: the protobuf that contains all the structures + - `checker.py`: a utility to check whether a serialized ONNX proto is legal + - `shape_inference.py`: a utility to infer types and shapes for ONNX models + - `version_converter.py`: a utility to upgrade or downgrade version for ONNX models + - `parser.py`: a utility to create an ONNX model or graph from a textual representation + - `compose.py`: a utility to merge ONNX models + - `helper.py`: tools for graph operation + - `defs/`: a subfolder that defines the ONNX operators + - `test/`: test files + +### Auto generated files + +Various files in this project are auto generated and may have to be updated in a PR. + +#### Generate operator documentation + +Operator docs ([Operators.md](docs/Operators.md), [Operators-ml.md](docs/Operators-ml.md)) and Changelog docs ([Changelog.md](docs/Changelog.md), [Changelog-ml.md](docs/Changelog-ml.md)) are automatically generated based on C++ operator definitions and backend Python snippets. To refresh all these docs, run the following commands from the repo root and commit the results by setting "ONNX_ML=1". By contrast, setting `ONNX_ML=0` will only update `Operators.md` and `Changelog.md`. + +```pwsh +# Windows +set ONNX_ML=1 +``` + +```sh +# UNIX +export ONNX_ML=1 +pip install -e . -v +python onnx/defs/gen_doc.py +``` + +#### Generate test coverage report + + + +The test coverage report can be generated by running: + +```sh +python onnx/backend/test/stat_coverage.py +``` + +#### Generate protobuf definitions + +Variants of the `.proto` files are generated for the default and ml opset using the following commands: + +``` +python onnx/gen_proto.py -l +python onnx/gen_proto.py -l --ml +``` + +#### Generate test data + +Test ONNX models, their inputs, and expected outputs can be regenerated after updating test cases by running: + +``` +python onnx/backend/test/cmd_tools.py generate-data --diff +``` + +This will only re-generate test data for test cases that were updated in the current feature branch compared to main. + + +### Coding style + +We adopted the [Google Python Style Guide](https://google.github.io/styleguide/pyguide.html) and [Google C++ Style Guide](https://google.github.io/styleguide/cppguide.html) for this project. + +We use `lintrunner` to drive multiple linters defined in `.lintrunner.toml` to lint the codebase. + +To run these checks locally, install `lintrunner` and the linters with + +```sh +pip install lintrunner lintrunner-adapters +lintrunner init +``` + +Then lint with + +```sh +lintrunner +``` + +format with + +```sh +# Display all lints and apply the fixes +lintrunner -a +# Or apply fixes only (faster) +lintrunner f +``` + +Run `lintrunner --help` and see the `.lintrunner.toml` file for more usage examples, as well as instructions on how to adopt new linters. + +#### Naming for legacy helper functions in `onnx/defs/*/old.cc` + +When adding or renaming helper functions in legacy schema files (`old.cc`), prefer explicit opset-based names over numeric suffixes. This keeps intent clear and avoids ambiguity. + +- Prefer `..._opsetN` for a single opset. +- Prefer `..._opsetN_to_M` for helpers shared across a contiguous opset range. +- Avoid names like `...1`, `...2`, `..._9`, or `...Opset9`. + +Examples: `RNNShapeInference_opset7_to_13`, `PoolOpSchemaGenerator_opset10_to_11`. + +### Testing + +ONNX uses [pytest](https://docs.pytest.org) as a test driver. To run tests, you'll first need to install pytest: + +```sh +pip install pytest +``` + +After installing pytest, run from the root of the repo: + +```sh +pytest +``` + +to run the tests. + +#### Cpp tests (googletest) + +Some functionalities are tested with googletest. Those tests are listed in `test/cpp`, and include tests for shape inference, data propagation, parser, and others. + +To run them, first build ONNX with `-DONNX_BUILD_TESTS=1` or `ONNX_BUILD_TESTS=1 pip install -e . -v`. + +##### Linux and MacOS + +The cpp tests require dynamically linking to built libraries. + +```sh +export LD_LIBRARY_PATH="./.setuptools-cmake-build/:$LD_LIBRARY_PATH" +.setuptools-cmake-build/onnx_gtests +``` + +##### Windows + +```pwsh +# If you set DEBUG=1, use `.setuptools-cmake-build\Debug\onnx_gtests.exe` instead +.setuptools-cmake-build\Release\onnx_gtests.exe +``` + +## DCO + +ONNX has adopted the [DCO](https://en.wikipedia.org/wiki/Developer_Certificate_of_Origin). All code repositories under ONNX require a DCO. (ONNX previously used a CLA, which is being replaced with the DCO.) + +DCO is provided by including a sign-off-by line in commit messages. Using the `-s` flag for `git commit` will automatically append this line. For example, running `git commit -s -m 'commit info.'` it will produce a commit that has the message `commit info. Signed-off-by: My Name `. The DCO bot will ensure commits are signed with an email address that matches the commit author before they are eligible to be merged. + +If you are using a GUI like the GitHub web site or GitHub Desktop, you'll need to append the `Signed-off-by: My Name ` manually to each commit message. For the onnx organization [sign-off](https://github.blog/changelog/2022-06-08-admins-can-require-sign-off-on-web-based-commits/) for web based commits is enabled. When this is activated you will see "Sign off and propose changes" instead of "Propose changes" when you are editing files directly at github. It is recommended to set this setting for your own fork as well. Since in the review process commits are made on this fork. + +NOTE: the sign-off is needed for each commit in the PR, not at the PR level. + +If you have old commits that are not signed, use the following commands to squash the old PR (original branch) into a single commit. This is an easier way to signoff old commits in old PR. + +```bash +git checkout main +git checkout -b temporary_patch # create a new branch as temporary +git merge --squash original_patch # copy from old branch +git branch -d original_patch # remove old branch +git checkout -b original_patch # create a new branch with the same name (override) +git commit -m 'type your own commit msg' -s # signoff that single commit +git push origin original_patch -f # forcibly override the old branch` +``` + +## CI Pipelines + +Every PR needs to pass CIs before merge. CI pipelines details are [here](docs/CIPipelines.md). + +## Other developer documentation + +- [How to implement ONNX backend (ONNX to something converter)](docs/ImplementingAnOnnxBackend.md) +- [Backend test infrastructure and how to add tests](docs/OnnxBackendTest.md) + +## License + +[Apache License v2.0](/LICENSE) + +## Code of Conduct + +[ONNX Open Source Code of Conduct](http://onnx.ai/codeofconduct.html) diff --git a/INSTALL.md b/INSTALL.md new file mode 100644 index 0000000..fa71def --- /dev/null +++ b/INSTALL.md @@ -0,0 +1,261 @@ + + +# Installation + +## Official Python packages + +ONNX released packages are published in PyPi. + +```sh +pip install onnx # or pip install onnx[reference] for optional reference implementation dependencies +``` + +[ONNX weekly packages](https://pypi.org/project/onnx-weekly/) are published in PyPI to enable experimentation and early testing. + +## vcpkg packages + +ONNX is in the maintenance list of [vcpkg](https://github.com/microsoft/vcpkg), you can easily use vcpkg to build and install it. + +```sh +git clone https://github.com/microsoft/vcpkg.git +cd vcpkg +./bootstrap-vcpkg.bat # For powershell +./bootstrap-vcpkg.sh # For bash +./vcpkg install onnx +``` + +## Conda packages + +A binary build of ONNX is available from [Conda](https://conda.io), in [conda-forge](https://conda-forge.org/): + +```sh +conda install -c conda-forge onnx +``` + +### Building ONNX from source in a conda-forge-based environment + +A conda-forge-based development environment is provided through the cross-platform [Pixi package manager](https://prefix.dev/). +The `pixi.toml` file in the root directory defines various tasks for common development workflows. +Running +```sh +pixi run install +``` +builds the C++ component and installs it as an editable Python package. + +After the installation has completed one can run the gtest and pytest suites via the pixi-tasks of the same name: + +```sh +pixi run gtest +``` + +and + +```sh +pixi run pytest +``` + + +## Build ONNX from Source with manually managed dependencies + +Before building from source uninstall any existing versions of ONNX via `pip uninstall onnx`. + +C++17 or higher C++ compiler version is required to build ONNX from source. Still, users can specify their own `CMAKE_CXX_STANDARD` version for building ONNX. + +Protobuf is required for ONNX. If you don't have Protobuf installed, ONNX will internally download and build Protobuf for ONNX build. + +Or, you can manually install [Protobuf C/C++ libraries and tools](https://github.com/protocolbuffers/protobuf) with specified version before proceeding forward. Then depending on how you installed Protobuf, you need to set environment variable CMAKE_ARGS to "-DONNX_USE_PROTOBUF_SHARED_LIBS=ON" or "-DONNX_USE_PROTOBUF_SHARED_LIBS=OFF". For example, you may need to run the following command: + +Linux or Mac: + +```sh +export CMAKE_ARGS="-DONNX_USE_PROTOBUF_SHARED_LIBS=ON" +``` + +Windows: + +```bat +set CMAKE_ARGS="-DONNX_USE_PROTOBUF_SHARED_LIBS=ON" +``` + +The ON/OFF depends on what kind of Protobuf library you have. Shared libraries are files ending with \*.dll/\*.so/\*.dylib. Static libraries are files ending with \*.a/\*.lib. This option depends on how you get your Protobuf library and how it was built. Because its default value is OFF, you don't need to run the commands above if you'd prefer to use a static Protobuf library. + +### Windows + +``` +git clone https://github.com/onnx/onnx.git +cd onnx +git submodule update --init --recursive +# prefer lite proto +set CMAKE_ARGS='-DONNX_USE_LITE_PROTO=ON -DONNX_USE_PROTOBUF_SHARED_LIBS=ON' +pip install -e . -v +``` + + +#### Old instructions + +If you are building ONNX from source, it is recommended that you also build Protobuf locally as a static library. The version distributed with conda-forge is a DLL, but ONNX expects it to be a static library. Building Protobuf locally also lets you control the version of Protobuf. The tested and recommended version is 5.29.2. + +The instructions in this README assume you are using Visual Studio 2019. It is recommended that you run all the commands from a shell started from "x64 Native Tools Command Prompt for VS 2019" and keep the build system generator for cmake (e.g., cmake -G "Visual Studio 16 2019") consistent while building Protobuf as well as ONNX. + +You can build Protobuf from source by running the following commands: + +```bat +git clone https://github.com/protocolbuffers/protobuf.git +cd protobuf +git checkout v5.29.2 +git submodule update --init --recursive +cmake -G "Visual Studio 16 2019" -A x64 -DCMAKE_INSTALL_PREFIX= -Dprotobuf_MSVC_STATIC_RUNTIME=OFF -Dprotobuf_BUILD_SHARED_LIBS=OFF -Dprotobuf_BUILD_TESTS=OFF -Dprotobuf_BUILD_EXAMPLES=OFF +cmake --build . --config Release --target install +``` + +Then it will be built as a static library and installed to . Please add the bin directory(which contains protoc.exe) to your PATH. + +```bat +set CMAKE_PREFIX_PATH=;%CMAKE_PREFIX_PATH% +``` + +Please note: if your protobuf_install_dir contains spaces, **do not** add quotation marks around it. + +Alternative: if you have local Protobuf executable and want to use it for ONNX, you can set ONNX_PROTOC_EXECUTABLE instead. + +```bat +set CMAKE_ARGS=-DONNX_PROTOC_EXECUTABLE= +``` + +Then you can build ONNX as: + +``` +git clone https://github.com/onnx/onnx.git +cd onnx +git submodule update --init --recursive +# prefer lite proto +set CMAKE_ARGS=-DONNX_USE_LITE_PROTO=ON +pip install -e . -v +``` + +### Linux + +First, you need to install Protobuf. The minimum Protobuf compiler (protoc) version required by ONNX is 6.31.1. + +Ubuntu 20.04 (and newer) users may choose to install Protobuf (which is usually lower than required) via + +```sh +apt-get install python3-pip python3-dev libprotobuf-dev protobuf-compiler +``` +In this case, ONNX is able to detect and use the system Profobuf. Users of other Linux distributions can use their system package manager to install Profobuf libraries similarly. + +A better way is to build and install the required Protobuf version from source. See the instructions below for more details. + +
+ Installing Protobuf from source + +```sh + git clone https://github.com/protocolbuffers/protobuf.git + cd protobuf + git checkout v5.29.2 + git submodule update --init --recursive + mkdir build_source && cd build_source + cmake -Dprotobuf_BUILD_SHARED_LIBS=OFF -DCMAKE_INSTALL_PREFIX=/usr -Dprotobuf_BUILD_TESTS=OFF -DCMAKE_BUILD_TYPE=Release -DCMAKE_POSITION_INDEPENDENT_CODE=ON .. + cmake --build . --target install +``` + + Here "-DCMAKE_POSITION_INDEPENDENT_CODE=ON" is crucial. By default static libraries are built without "-fPIC" flag, they are not position independent code. But shared libraries must be position independent code. Python C/C++ extensions(like ONNX) are shared libraries. So if a static library was not built with "-fPIC", it can't be linked to such a shared library. + + Once build is successful, update PATH to include Protobuf paths so that ONNX can find Protobuf. + +
+ +Then you can build ONNX as: + +```sh +git clone https://github.com/onnx/onnx.git +cd onnx +git submodule update --init --recursive +# Optional: prefer lite proto +export CMAKE_ARGS=-DONNX_USE_LITE_PROTO=ON +pip install -e . -v +``` + +### Mac + +```sh +brew update +brew install cmake +git clone https://github.com/protocolbuffers/protobuf.git +cd protobuf +git checkout v5.29.2 +git submodule update --init --recursive +mkdir build_source && cd build_source +cmake -Dprotobuf_BUILD_SHARED_LIBS=OFF -Dprotobuf_BUILD_TESTS=OFF -DCMAKE_BUILD_TYPE=Release -DCMAKE_POSITION_INDEPENDENT_CODE=ON .. +cmake --build . --target install +``` + +Once build is successful, update PATH to include Protobuf paths so that ONNX can find Protobuf. + +Then you can build ONNX as: + +```sh +git clone --recursive https://github.com/onnx/onnx.git +cd onnx +# Optional: prefer lite proto +set CMAKE_ARGS=-DONNX_USE_LITE_PROTO=ON +pip install -e . -v +``` + +## Verify Installation + +After installation, run + +```sh +python -c "import onnx" +``` + +to verify it works. + +## Common Build Options + +For full list refer to CMakeLists.txt + +### Environment variables + +* `USE_MSVC_STATIC_RUNTIME` should be 1 or 0, not ON or OFF. When set to 1 ONNX links statically to runtime library. +**Default**: `USE_MSVC_STATIC_RUNTIME=0` + +* `DEBUG` should be 0 or 1. When set to 1 ONNX is built in debug mode. For debug versions of the dependencies, you need to open the [CMakeLists file](https://github.com/onnx/onnx/blob/main/CMakeLists.txt) and append a letter `d` at the end of the package name lines. For example, `NAMES protobuf-lite` would become `NAMES protobuf-lited`. +**Default**: `Debug=0` + +### CMake variables + +* `ONNX_USE_PROTOBUF_SHARED_LIBS` should be `ON` or `OFF`. +**Default**: `ONNX_USE_PROTOBUF_SHARED_LIBS=OFF USE_MSVC_STATIC_RUNTIME=0` +`ONNX_USE_PROTOBUF_SHARED_LIBS` determines how ONNX links to Protobuf libraries. + * When set to `ON` - ONNX will dynamically link to Protobuf shared libs, PROTOBUF_USE_DLLS will be defined as described [here](https://github.com/protocolbuffers/protobuf/blob/main/cmake/README.md#dlls-vs-static-linking). + * When set to `OFF` - ONNX will link statically to Protobuf. + +* `ONNX_USE_LITE_PROTO` should be `ON` or `OFF`. When set to `ON` ONNX uses lite Protobuf instead of full Protobuf. +**Default**: `ONNX_USE_LITE_PROTO=OFF` + +* `ONNX_WERROR` should be `ON` or `OFF`. When set to `ON` warnings are treated as errors. +**Default**: `ONNX_WERROR=OFF` in local builds, `ON` in CI and release pipelines. + +* `nanobind_DIR` can be set to the directory that contains `nanobindConfig.cmake` (for example, + `python -m nanobind --cmake_dir`) if CMake cannot find nanobind. You can also set + `CMAKE_PREFIX_PATH` instead. + +* `FETCHCONTENT_FULLY_DISCONNECTED` is intended for subsequent re-configures after + dependencies are already populated. It does not prevent network access on the initial + configure; for fully offline first-run builds, prefer a + [dependency provider](https://cmake.org/cmake/help/latest/module/FetchContent.html#dependency-providers) + or provide dependencies locally (for example, via `nanobind_DIR` or `CMAKE_PREFIX_PATH`). + +## Common Errors + +* Note: the `import onnx` command does not work from the source checkout directory; in this case you'll see `ModuleNotFoundError: No module named 'onnx.onnx_cpp2py_export'`. Change into another directory to fix this error. + +* If you run into any issues while building Protobuf as a static library, please ensure that shared Protobuf libraries, like libprotobuf, are not installed on your device or in the conda environment. If these shared libraries exist, either remove them to build Protobuf from source as a static library, or skip the Protobuf build from source to use the shared version directly. + +* If you run into any issues while building ONNX from source, and your error message reads, `Could not find pythonXX.lib`, ensure that you have consistent Python versions for common commands, such as `python` and `pip`. Clean all existing build files and rebuild ONNX again. diff --git a/LICENSE b/LICENSE new file mode 100644 index 0000000..d645695 --- /dev/null +++ b/LICENSE @@ -0,0 +1,202 @@ + + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "[]" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright [yyyy] [name of copyright owner] + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. diff --git a/LICENSES/Apache-2.0.txt b/LICENSES/Apache-2.0.txt new file mode 100644 index 0000000..137069b --- /dev/null +++ b/LICENSES/Apache-2.0.txt @@ -0,0 +1,73 @@ +Apache License +Version 2.0, January 2004 +http://www.apache.org/licenses/ + +TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + +1. Definitions. + +"License" shall mean the terms and conditions for use, reproduction, and distribution as defined by Sections 1 through 9 of this document. + +"Licensor" shall mean the copyright owner or entity authorized by the copyright owner that is granting the License. + +"Legal Entity" shall mean the union of the acting entity and all other entities that control, are controlled by, or are under common control with that entity. For the purposes of this definition, "control" means (i) the power, direct or indirect, to cause the direction or management of such entity, whether by contract or otherwise, or (ii) ownership of fifty percent (50%) or more of the outstanding shares, or (iii) beneficial ownership of such entity. + +"You" (or "Your") shall mean an individual or Legal Entity exercising permissions granted by this License. + +"Source" form shall mean the preferred form for making modifications, including but not limited to software source code, documentation source, and configuration files. + +"Object" form shall mean any form resulting from mechanical transformation or translation of a Source form, including but not limited to compiled object code, generated documentation, and conversions to other media types. + +"Work" shall mean the work of authorship, whether in Source or Object form, made available under the License, as indicated by a copyright notice that is included in or attached to the work (an example is provided in the Appendix below). + +"Derivative Works" shall mean any work, whether in Source or Object form, that is based on (or derived from) the Work and for which the editorial revisions, annotations, elaborations, or other modifications represent, as a whole, an original work of authorship. For the purposes of this License, Derivative Works shall not include works that remain separable from, or merely link (or bind by name) to the interfaces of, the Work and Derivative Works thereof. + +"Contribution" shall mean any work of authorship, including the original version of the Work and any modifications or additions to that Work or Derivative Works thereof, that is intentionally submitted to Licensor for inclusion in the Work by the copyright owner or by an individual or Legal Entity authorized to submit on behalf of the copyright owner. For the purposes of this definition, "submitted" means any form of electronic, verbal, or written communication sent to the Licensor or its representatives, including but not limited to communication on electronic mailing lists, source code control systems, and issue tracking systems that are managed by, or on behalf of, the Licensor for the purpose of discussing and improving the Work, but excluding communication that is conspicuously marked or otherwise designated in writing by the copyright owner as "Not a Contribution." + +"Contributor" shall mean Licensor and any individual or Legal Entity on behalf of whom a Contribution has been received by Licensor and subsequently incorporated within the Work. + +2. Grant of Copyright License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form. + +3. Grant of Patent License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable (except as stated in this section) patent license to make, have made, use, offer to sell, sell, import, and otherwise transfer the Work, where such license applies only to those patent claims licensable by such Contributor that are necessarily infringed by their Contribution(s) alone or by combination of their Contribution(s) with the Work to which such Contribution(s) was submitted. If You institute patent litigation against any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Work or a Contribution incorporated within the Work constitutes direct or contributory patent infringement, then any patent licenses granted to You under this License for that Work shall terminate as of the date such litigation is filed. + +4. Redistribution. You may reproduce and distribute copies of the Work or Derivative Works thereof in any medium, with or without modifications, and in Source or Object form, provided that You meet the following conditions: + + (a) You must give any other recipients of the Work or Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works that You distribute, all copyright, patent, trademark, and attribution notices from the Source form of the Work, excluding those notices that do not pertain to any part of the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its distribution, then any Derivative Works that You distribute must include a readable copy of the attribution notices contained within such NOTICE file, excluding those notices that do not pertain to any part of the Derivative Works, in at least one of the following places: within a NOTICE text file distributed as part of the Derivative Works; within the Source form or documentation, if provided along with the Derivative Works; or, within a display generated by the Derivative Works, if and wherever such third-party notices normally appear. The contents of the NOTICE file are for informational purposes only and do not modify the License. You may add Your own attribution notices within Derivative Works that You distribute, alongside or as an addendum to the NOTICE text from the Work, provided that such additional attribution notices cannot be construed as modifying the License. + + You may add Your own copyright statement to Your modifications and may provide additional or different license terms and conditions for use, reproduction, or distribution of Your modifications, or for any such Derivative Works as a whole, provided Your use, reproduction, and distribution of the Work otherwise complies with the conditions stated in this License. + +5. Submission of Contributions. Unless You explicitly state otherwise, any Contribution intentionally submitted for inclusion in the Work by You to the Licensor shall be under the terms and conditions of this License, without any additional terms or conditions. Notwithstanding the above, nothing herein shall supersede or modify the terms of any separate license agreement you may have executed with Licensor regarding such Contributions. + +6. Trademarks. This License does not grant permission to use the trade names, trademarks, service marks, or product names of the Licensor, except as required for reasonable and customary use in describing the origin of the Work and reproducing the content of the NOTICE file. + +7. Disclaimer of Warranty. Unless required by applicable law or agreed to in writing, Licensor provides the Work (and each Contributor provides its Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, including, without limitation, any warranties or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. You are solely responsible for determining the appropriateness of using or redistributing the Work and assume any risks associated with Your exercise of permissions under this License. + +8. Limitation of Liability. In no event and under no legal theory, whether in tort (including negligence), contract, or otherwise, unless required by applicable law (such as deliberate and grossly negligent acts) or agreed to in writing, shall any Contributor be liable to You for damages, including any direct, indirect, special, incidental, or consequential damages of any character arising as a result of this License or out of the use or inability to use the Work (including but not limited to damages for loss of goodwill, work stoppage, computer failure or malfunction, or any and all other commercial damages or losses), even if such Contributor has been advised of the possibility of such damages. + +9. Accepting Warranty or Additional Liability. While redistributing the Work or Derivative Works thereof, You may choose to offer, and charge a fee for, acceptance of support, warranty, indemnity, or other liability obligations and/or rights consistent with this License. However, in accepting such obligations, You may act only on Your own behalf and on Your sole responsibility, not on behalf of any other Contributor, and only if You agree to indemnify, defend, and hold each Contributor harmless for any liability incurred by, or claims asserted against, such Contributor by reason of your accepting any such warranty or additional liability. + +END OF TERMS AND CONDITIONS + +APPENDIX: How to apply the Apache License to your work. + +To apply the Apache License to your work, attach the following boilerplate notice, with the fields enclosed by brackets "[]" replaced with your own identifying information. (Don't include the brackets!) The text should be enclosed in the appropriate comment syntax for the file format. We also recommend that a file or class name and description of purpose be included on the same "printed page" as the copyright notice for easier identification within third-party archives. + +Copyright [yyyy] [name of copyright owner] + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + +http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. diff --git a/LICENSES/BSD-2-Clause.txt b/LICENSES/BSD-2-Clause.txt new file mode 100644 index 0000000..eb3c575 --- /dev/null +++ b/LICENSES/BSD-2-Clause.txt @@ -0,0 +1,9 @@ +Copyright (c) + +Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: + +1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. + +2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/NOTICE b/NOTICE new file mode 100644 index 0000000..07e3d4e --- /dev/null +++ b/NOTICE @@ -0,0 +1,59 @@ +ONNX is licensed under the Apache License, Version 2.0 (see LICENSE). + +This product includes software from the following third-party projects, which +are linked into ONNX wheel builds. Their copyright notices are reproduced below. + +--- + +## abseil-cpp + +Copyright 2017 The Abseil Authors. + +Version: 20250127.0 +Source: https://github.com/abseil/abseil-cpp +License: Apache-2.0 (see LICENSE) + +--- + +## protobuf + +Copyright 2008 Google Inc. + +Version: 25.1 +Source: https://github.com/protocolbuffers/protobuf +License: Apache-2.0 (see LICENSE) + +--- + +## nanobind + +Copyright (c) 2022 Wenzel Jakob + +Version: 2.12.0 +Source: https://github.com/wjakob/nanobind +License: BSD-3-Clause + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + +1. Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + +2. Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the documentation + and/or other materials provided with the distribution. + +3. Neither the name of the copyright holder nor the names of its contributors + may be used to endorse or promote products derived from this software + without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/README.md b/README.md new file mode 100644 index 0000000..24b9fa4 --- /dev/null +++ b/README.md @@ -0,0 +1,141 @@ + + +

+ +[![PyPI - Version](https://img.shields.io/pypi/v/onnx.svg)](https://pypi.org/project/onnx) +[![CI](https://github.com/onnx/onnx/actions/workflows/main.yml/badge.svg)](https://github.com/onnx/onnx/actions/workflows/main.yml) +[![CII Best Practices](https://bestpractices.coreinfrastructure.org/projects/3313/badge)](https://bestpractices.coreinfrastructure.org/projects/3313) +[![OpenSSF Scorecard](https://api.securityscorecards.dev/projects/github.com/onnx/onnx/badge)](https://api.securityscorecards.dev/projects/github.com/onnx/onnx) +[![SLSA 2](https://slsa.dev/images/gh-badge-level2.svg)](https://slsa.dev) +[![REUSE compliant](https://api.reuse.software/badge/github.com/onnx/onnx)](https://api.reuse.software/info/github.com/onnx/onnx) +[![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json)](https://github.com/astral-sh/ruff) +[![abi3 compatible](https://img.shields.io/badge/abi3-compatible-brightgreen)](https://docs.python.org/3/c-api/stable.html) + +[Open Neural Network Exchange (ONNX)](https://onnx.ai) is an open ecosystem that empowers AI developers +to choose the right tools as their project evolves. ONNX provides an open source format for AI models, both deep learning and traditional ML. It defines an extensible computation graph model, as well as definitions of built-in operators and standard +data types. Currently we focus on the capabilities needed for inferencing (scoring). + +ONNX is [widely supported](http://onnx.ai/supported-tools) and can be found in many frameworks, tools, and hardware. Enabling interoperability between different frameworks and streamlining the path from research to production helps increase the speed of innovation in the AI community. We invite the community to join us and further evolve ONNX. + + +# Use ONNX + +* [Documentation of ONNX Python Package](https://onnx.ai/onnx/) +* [Tutorials for creating ONNX models](https://github.com/onnx/tutorials) +* [Pre-trained ONNX models](https://github.com/onnx/models) + +# Learn about the ONNX spec + +* [Overview](https://github.com/onnx/onnx/blob/main/docs/Overview.md) +* [ONNX intermediate representation spec](https://github.com/onnx/onnx/blob/main/docs/IR.md) +* [Versioning principles of the spec](https://github.com/onnx/onnx/blob/main/docs/Versioning.md) +* [Operators documentation](https://github.com/onnx/onnx/blob/main/docs/Operators.md) +* [Operators documentation](https://onnx.ai/onnx/operators/index.html) (latest release) +* [Python API Overview](https://github.com/onnx/onnx/blob/main/docs/PythonAPIOverview.md) + +# Programming utilities for working with ONNX Graphs + +* [Shape and Type Inference](https://github.com/onnx/onnx/blob/main/docs/ShapeInference.md) +* [Graph Optimization](https://github.com/onnx/optimizer) +* [Opset Version Conversion](https://github.com/onnx/onnx/blob/main/docs/docsgen/source/api/version_converter.md) + +# Contribute + +ONNX is a community project and the open governance model is described [here](https://github.com/onnx/onnx/blob/main/community/readme.md). We encourage you to join the effort and contribute feedback, ideas, and code. You can participate in the [Special Interest Groups](https://github.com/onnx/onnx/blob/main/community/sigs.md) and [Working Groups](https://github.com/onnx/onnx/blob/main/community/working-groups.md) to shape the future of ONNX. + +Check out our [contribution guide](https://github.com/onnx/onnx/blob/main/CONTRIBUTING.md) to get started. + +If you think some operator should be added to ONNX specification, please read +[this document](https://github.com/onnx/onnx/blob/main/docs/AddNewOp.md). + +# Community meetings + +The schedules of the regular meetings of the Steering Committee, the working groups and the SIGs can be found [here](https://onnx.ai/calendar) + +Community Meetups are held at least once a year. Content from previous community meetups are at: + +* 2020.04.09 +* 2020.10.14 +* 2021.03.24 +* 2021.10.21 +* 2022.06.24 +* 2023.06.28 + +# Discuss + +We encourage you to open [Issues](https://github.com/onnx/onnx/issues), or use [Slack](https://lfaifoundation.slack.com/) (If you have not joined yet, please use this [link](https://join.slack.com/t/lfaifoundation/shared_invite/zt-3wx5vohc3-MeSYi3_dscb~u~cqs7zlPg) to join the group) for more real-time discussion. + +# Follow Us + +Stay up to date with the latest ONNX news. [[Facebook](https://www.facebook.com/onnxai/)] [[Twitter/X](https://twitter.com/onnxai)] + +# Roadmap + +A roadmap process takes place every year. More details can be found [here](https://github.com/onnx/steering-committee/tree/main/roadmap) + +# Installation + +ONNX released packages are published in PyPi. + +```sh +pip install onnx # or pip install onnx[reference] for optional reference implementation dependencies +``` + +[ONNX weekly packages](https://pypi.org/project/onnx-weekly/) are published in PyPI to enable experimentation and early testing. + +Detailed install instructions, including Common Build Options and Common Errors can be found [here](https://github.com/onnx/onnx/blob/main/INSTALL.md) + +# Python ABI3 Compatibility + +This package provides [abi3](https://docs.python.org/3/c-api/stable.html)-compatible wheels, allowing a single binary wheel to work across multiple Python versions (from 3.12 onwards). + + +# Testing + +ONNX uses [pytest](https://docs.pytest.org) as test driver. In order to run tests, you will first need to install `pytest`: + +```sh +pip install pytest +``` + +After installing pytest, use the following command to run tests. + +```sh +pytest +``` + +# Development + +Check out the [contributor guide](https://github.com/onnx/onnx/blob/main/CONTRIBUTING.md) for instructions. + +# Reproducible Builds (Linux) + +This project provides reproducible builds for Linux. + +A *reproducible build* means that the same source code will always produce identical binary outputs, no matter who builds it or where it is built. + +To achieve this, we use the [`SOURCE_DATE_EPOCH`](https://reproducible-builds.org/docs/source-date-epoch/) standard. This ensures that build timestamps and other time-dependent information are fixed, making the output bit-for-bit identical across different environments. + +### Why this matters +- **Transparency**: Anyone can verify that the distributed binaries were created from the published source code. +- **Security**: Prevents tampering or hidden changes in the build process. +- **Trust**: Users can be confident that the binaries they download are exactly what the maintainers intended. + +If you prefer, you can use the prebuilt reproducible binaries instead of building from source yourself. + +# License + +[Apache License v2.0](LICENSE) + +# Trademark +Checkout [https://trademarks.justia.com](https://trademarks.justia.com/877/25/onnx-87725026.html) for the trademark. + +[General rules of the Linux Foundation on Trademark usage](https://www.linuxfoundation.org/legal/trademark-usage) + +# Code of Conduct + +[ONNX Open Source Code of Conduct](https://onnx.ai/codeofconduct.html) diff --git a/README.wehub.md b/README.wehub.md new file mode 100644 index 0000000..7d52a62 --- /dev/null +++ b/README.wehub.md @@ -0,0 +1,7 @@ +# WeHub 来源说明 + +- 原始项目：`onnx/onnx` +- 原始仓库：https://github.com/onnx/onnx +- 导入方式：上游默认分支的最新快照 +- 原作者、版权和许可证信息以原始仓库及本仓库 LICENSE 为准 +- 本文件仅用于记录来源，不代表 WeHub 是原项目作者 diff --git a/RELEASE-MANAGEMENT.md b/RELEASE-MANAGEMENT.md new file mode 100644 index 0000000..473621a --- /dev/null +++ b/RELEASE-MANAGEMENT.md @@ -0,0 +1,129 @@ + + +# ONNX release management + +This describes the process by which versions of ONNX are officially released to the public. + +Release Cadence +--------------- + +Branch cuts for a new release are planned every 3 months. However, the times can be changed as required. + +| Minor Version | Release branch cut | Release date | First patch release date | +| --- | --- | --- | --- | +| 1.17.0 | XYZ | XYZ | Not planned | +| 1.18.0 | Mar 2025 | May 2025 | Not planned | +| 1.19.0 | 31. July 2025 | 27. August 2025 | 9. October 2025 | +| 1.20.0 | 31. October 2025 | November 2025 | January 2026 | +| 1.21.0 | 25. February 2026 | 28. March 2026 | - | +| 1.22.0 (tbd) | 1st June 2026 | June 2026 | tbd | + +## Long-Term Support (LTS) + +ONNX does not currently maintain LTS branches. Each minor release is superseded by the next; only the latest release receives ongoing attention. If your environment cannot upgrade and you need a backported security or bug fix, please +[open an issue](https://github.com/onnx/onnx/issues/new) describing your situation — we will assess feasibility on a case-by-case basis. Community contributions for backport PRs are welcome and will be reviewed. + +Release Compatibility Matrix +---------------------------- + +*Support for a Python version that went eol will be discontinued in the following ONNX release.* +*ONNX does NOT follow https://scientific-python.org/specs/spec-0000/ or https://protobuf.dev/support/version-support/* + +Changes are discussed in the community. Please do not hesitate to contact us if you have any requests. +Planned changes for future releases as listed in the table below are subject to change. + +|ONNX version | Python wheels | C++ | Min CMake Version | Min Protobuf | manylinux | +| --- | --- | --- | --- | --- | --- | +| 1.10 | 3.6-3.9 | 11 | 3.1 | --- | manylinux2010 | +| 1.11 | 3.6-3.9 | 11 | 3.1 | 3.12.2 | manylinux2010 | +| 1.12 | 3.7-3.10 | 11 | 3.1 | 3.12.2 | manylinux2014 | +| 1.13 | 3.7-3.11 | 11 | 3.1 | 3.20.2 | manylinux2014 | +| 1.14 | 3.7-3.11 | 11 | 3.1 | 3.20.2 | manylinux2014 | +| 1.15 | 3.8-3.11 | 14 | 3.1 | 3.20.2 | manylinux2014 | +| 1.16 | 3.8-3.12 | 17 | 3.1 | 3.20.2 | manylinux2014 | +| 1.17 | 3.8-3.12 | 17 | 3.14 | 3.20.2 | manylinux2014 | +| 1.18 | 3.9-3.13, 3.13t (win, mac) | 17 | 3.18 | v25.1 | manylinux2014 | +| 1.19 | 3.9-3.13, 3.13t (win, mac, linux) | 17 | 3.24 | v25.1 | manylinux2014 | +| 1.19.1 | 3.9-3.13, 3.13t (win, mac, linux) | 17 | 3.24 | v25.1 | manylinux2014 | +| 1.20 | 3.10-3.13, 3.13t (win, mac, linux), 3.14 (mac) | 17 | 3.26 | v25.1 | manylinux_2_28 | +| 1.21 | 3.10-3.13, 3.13t, 3.14, 3.14t | 17 | 3.26 | v25.1 | manylinux_2_28 | +| *1.22* | 3.10, 3.11, 3.12-abi3, 3.14 (win, mac, linux), 3.14 pyodide | 17 | 3.24 | v25.1 | manylinux_2_28 | + + +Releases +-------- + +Releases are versioned according to [ONNX Versioning](docs/Versioning.md). This describes IR and operator versioning policies, as well as propose how models themselves should be versioned. + +On a regular basis, new versions of ONNX are published, representing the aggregate of changes in the IR and operator sets. Such releases use semantic versioning to describe the progression of the standard. + +The GitHub repo for ONNX provides release branches where the project is stabilized as per the process described here. Release notes are used to communicate the stability and status of a release. The main branch will be used to continue work for subsequent releases. + +Major, minor and patch releases will have branch names and version numbers reflecting the nature of the change as per semantic versioning definitions. + +Workflow +-------- + +The following workflow describes the steps taken to release an update of ONNX, +and can be undertaken regardless of whether a major, minor or patch release is +to be produced. + +- The trigger for the workflow will typically be a time-based trigger based on + elapsed time (say every three months). + +- The release manager will announce the intent of the process (to produce + major, minor or patch update) and the overall timeline (documented in our + wiki: [example](https://github.com/onnx/onnx/wiki/Logistics-for-ONNX-Release-1.19.0). + A release branch is created with the name rel-major#.minor#(.patch#), and any version + references in build scripts or version checks are updated. + +- The release manager announces the initial commit for testing. The first + period lasts two weeks; any regressions found should be fixed, typically via + the main branch. Incomplete features should be done or excised during this + period. A distribution can be made available with an -RC1 suffix. + +- The release manager announces a second round of testing (unless it's only a + patch update with no regressions found). Only critical bugs are fixed at + this point, or those introduced by patches from the first week. A third + week may be introduced at the release manager's discretion if significant + fixes need to be taken. Distributions with -RCn suffixes can be made + available if convenient. + +- Release notes are updated with final changes, and a file with sources is + provided along with a release on the GitHub project page. + +Testing +------- + +The release process really consists of communicating, testing to establish a +known state of the project, and distributing files. This section deals with +the second task. + +At the very least, the tests that are part of the /test folder should be run +under a variety of configurations. Issues fixed should ensure coverage in this +suite to avoid regressions. + +Send a Pull Request for updates to this section to include a configuration you +can help test if you care about one that's missing. + +The community is encouraged to perform additional testing during the test +periods. Bugs and issues should be filed in the ONNX GitHub repo. + +# ONNX Weekly Builds on PyPI + +In addition to stable releases, we publish **weekly development builds** to a separate PyPI package: [`onnx-weekly`](https://pypi.org/project/onnx-weekly/). + +## Why a Separate Package? + +- **Avoid accidental installs:** Pre-release versions can be installed unintentionally; `onnx-weekly` ensures stable users are unaffected. +- **Enable safe testing:** Try upcoming features without impacting stable installs. Both packages can coexist. +- **Simplify automation:** Weekly builds are pushed automatically from `main` without polluting the main release history. + +## Installation + +```bash +pip install onnx-weekly diff --git a/REUSE.toml b/REUSE.toml new file mode 100644 index 0000000..7abf599 --- /dev/null +++ b/REUSE.toml @@ -0,0 +1,236 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +version = 1 +SPDX-PackageName = "onnx" +SPDX-PackageSupplier = "onnx-technical-discuss@lists.lfaidata.foundation" +SPDX-PackageDownloadLocation = "onnx.ai" + +[[annotations]] +path = "codecov.yml" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = [".github/**/**.md", ".github/pull_request_template.md"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ".github/**/**.yml" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/simple/**/test_data_set_0/**pb" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/node/**/test_data_set_0/**.pb" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/simple/**/**.onnx" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/pytorch-operator/**/test_data_set_0/**.pb" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/pytorch-operator/**/**.onnx" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/pytorch-operator/**/**.pb" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/node/**/**.onnx" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/pytorch-converted/**/test_data_set_0/**.pb" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/light/**.onnx" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/light/**.pb" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/real/**/**.json" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/pytorch-converted/**/**.onnx" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/backend/test/data/light/light/**.onnx" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["onnx/onnx-operators**", "onnx/onnx-ml**", "onnx/onnx-data**", "onnx/onnx**"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["docs/docsgen/source/api/**md", "docs/docsgen/source/_static/**", "docs/docsgen/source/onnx-favicon.png"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["onnx/backend/test/stat_coverage.py", "onnx/defs/gen_doc.py"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "docs/Change**.md" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "docs/Test**.md" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "docs/Operator**.md" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = [".gitignore", "**/**/.gitignore", ".gitmodules", ".lintrunner.toml", ".gitattributes", ".git-blame-ignore-revs", ".clang**", ".editorconfig"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["docs/docsgen/source/intro/images/**", "docs/docsgen/source/intro/**md"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["MANIFEST.in", "VERSION_NUMBER", "CODE_OF_CONDUCT.md", "CODEOWNERS", "NOTICE", "pyproject**.toml", "requirements**.txt"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["onnx/onnx_cpp2py_export/**pyi", "onnx/py.typed"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["onnx/defs/**/**.cc", "onnx/defs/**/**.h"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "onnx/reference/ops/aionnxml/**py" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["onnx/defs/**.cc", "onnx/defs/**.h"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["cmake/ONNXConfig**.in", "CMakeLists.txt", "cmake/**cmake", "cmake/", "onnx/test/cmake/CMakeLists.txt"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = [".vscode/settings.json", "docs/docsgen/source/_templates/layout.html", "docs/docsgen/source/_templates/sidebar-nav-bs.html", "docs/images/onnx_hub_arch.svg", "docs/onnx-horizontal-color.png", "tools/protoc-gen-mypy.sh.in"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["onnx/gen_proto.py", "tools/protoc-gen-mypy.bat", "onnx/reference/op_run.py"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["onnx/backend/sample/ops/abs.py", "onnx/gen_proto.py"] +precedence = "override" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["docs/proposals/images/composing_broadcast_axes.png"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = ["pixi.lock"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "docs/docsgen/source/technical/**.png" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = [".agents/**", ".claude/**", "uv.lock"] +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" + +[[annotations]] +path = "sbom.cdx.json" +precedence = "aggregate" +SPDX-FileCopyrightText = "Copyright (c) ONNX Project Contributors" +SPDX-License-Identifier = "Apache-2.0" diff --git a/ROADMAP.md b/ROADMAP.md new file mode 100644 index 0000000..bb88a3e --- /dev/null +++ b/ROADMAP.md @@ -0,0 +1,141 @@ + + +# ONNX 12-Month Roadmap (Q3 2026 – Q2 2027) + +> This is a living document and a starting point — not yet a complete picture of all ONNX SIGs and Working Groups. +> Every group is invited to add or update their milestones below. +> Milestones are **estimations**, not commitments, unless stated otherwise. + +--- + +## How to contribute + +1. Find your SIG/WG section below (or add a new one following the template). +2. Fill in your objectives per quarter. +3. Submit a PR or send your input to the roadmap coordinator. + +--- + +## SIGs & Working Groups + +--- + +### SIG Architecture & Infrastructure + +**Lead(s):** Andreas Fehlner (TRUMPF Laser), Christian Bourjau (QuantCo) + +**Last updated:** 2026-06-24 + +#### Q3 2026 +- Remove test files from release packages (reduce package size and install surface) +- Automate and improve SBOM generation and publication +- Immutable releases (tamper-evident, verifiable artifacts) +- Convert deprecated branch protection rules to Rules within Github +- Newer minimal protobuf + +#### Q4 2026 +- Move to C++20 +- Begin C++ hardening (compiler flags, static analysis integration) +- Integrate fuzz testing into CI +- Define a stable C-API + +#### Q1 2027 +- OpenSSF Gold Badge +- Achieve SLSA Build Level 3 compliance + +#### Q2 2027 +- Improved conformance test suite (parameterized, filterable, usable from non-Python runtimes) +- Continued C++ hardening improvements + +--- + +### Operator SIG + +**Lead(s):** G. Ramalingam, Michal Karzynski + +**Last updated:** 2026-06-08 + +Note: The items below may be reprioritized as needed. It may be better to view the following as +an unprioritized list of items for the next year. + +#### Q3 2026: +- ~~Attention op (fix causal-mask position-anchoring issue)~~ (landing in [#8068](https://github.com/onnx/onnx/pull/8068)) +- Attention op (add support for local window) +- Attention op (add support for pre-softcap additive bias) +- Support symbolic shape inference + +#### Q4 2026: +- Variable length scan support +- LinearAttention (support for Gated DeltaNet-2) + +#### Q1 2027: +- Support Mixture-of-Exports (possibly via Grouped MatMul operator) +- RotaryEmbedding for visual models (2D) + +#### Q2 2027: +- Support for ternary-value quantization and single-bit quantization +- FlexAttention + +### SONNX Working Group + +**Lead(s):** Eric JENN, Jean SOUYRIS +**Last updated:** 2026-06-01 + +#### Q3 2026 +- First subset of ONNX operators (i) fully specified in compliance with the SONNX guidelines, (ii) fully formally specified and proved, (iii) implemented, (iv) tested and (v) integrated in the AIDGE framework. +- Final version of the informal specification guidelines. +- First version of the formal specification guidelines. +- First version of the verification guidelines (testing). +- First version of the graph execution semantics formally specified (excluding control flow ops). + +#### Q4 2026 +- Second subset of ONNX operators (i) fully specified in compliance with the SONNX guidelines, (ii) fully formally specified and proved, (iii) implemented, (iv) tested and (v) integrated in the AIDGE framework. +- Final version of the formal specification guidelines. +- First version of the numerical accuracy analysis guidelines. + +#### Q1 2027 +- Third subset of ONNX operators (i) fully specified in compliance with the SONNX guidelines, (ii) fully formally specified and proved, (iii) implemented, (iv) tested and (v) integrated in the AIDGE framework. +- Final version of the numerical accuracy analysis guidelines. +- Final version of the verification guidelines (all techniques). +- First subset of operators with numerical accuracy analysis. +- Final version of the graph execution formal specification (including control flow ops). + +#### Q2 2027 +- Fourth subset of ONNX operators (i) fully specified in compliance with the SONNX guidelines, (ii) fully formally specified and proved, (iii) implemented, (iv) tested and (v) integrated in the AIDGE framework. +- Second subset of operators with numerical accuracy analysis. + +--- + + +### [SIG/WG Name] + +**Lead(s):** _TBD_ +**Last updated:** _YYYY-MM-DD_ + +#### Q3 2026 +- _Objective 1_ +- _Objective 2_ +- _Objective 3_ + +#### Q4 2026 +- _Objective 1_ +- _Objective 2_ +- _Objective 3_ + +#### Q1 2027 +- _Objective 1_ +- _Objective 2_ +- _Objective 3_ + +#### Q2 2027 +- _Objective 1_ +- _Objective 2_ +- _Objective 3_ + +--- + + diff --git a/SECURITY.md b/SECURITY.md new file mode 100644 index 0000000..344a0db --- /dev/null +++ b/SECURITY.md @@ -0,0 +1,73 @@ + + +# Security Policy + +## Reporting a Vulnerability + +Bugs, even safety-critical ones, that are easily discovered using widely available tooling are considered publicly known. +Please open a public issue or PR if you discovered an issue in this manner. +If you believe you have discovered a non-trivial security vulnerability in ONNX that does not fall into the above category, please report it privately using GitHub Security Advisories. + +👉 Open a private report: https://github.com/onnx/onnx/security/advisories/new + +This allows maintainers to triage the issue, collaborate on a fix, and coordinate disclosure. + +If you are unable to use GitHub for reporting, you may contact the maintainers at onnx-security@lists.lfaidata.foundation as a fallback. + +After your report is received, a maintainer will acknowledge it, work with you to understand impact and remediation, and keep you informed about progress toward a fix and public disclosure. We aim to provide an initial response within 14 business days. ONNX is a volunteer-driven open-source project, so response times may vary. + +Please do not disclose the vulnerability publicly until a fix and advisory have been released. + +Reporters are credited in the published advisory unless they request to remain anonymous. + +## Response Process + +Once a report is received, maintainers follow this process: + +1. **Confirm.** Verify the report describes a genuine security issue (not an ordinary bug or feature request) and assign an incident lead from the security team (the GitHub team with access to private advisories). The GitHub Security Advisory draft serves as the private coordination channel. +2. **Triage.** Assess severity case by case using [CVSS](https://www.first.org/cvss/) (v3.1 is preferred, but v4.0 is also accepted). +The security team decides per incident whether the fix warrants an out-of-cycle patch release or can be included in the next scheduled release. Not every report results in a CVE — a CVE is issued only when there is a confirmed, exploitable vulnerability with real-world impact. +Reports describing expected behavior, unrealistic preconditions, or issues outside the project's threat model may be closed without a CVE. +3. **Fix.** A patch is developed in a private fork or Security Advisory draft and reviewed by a second maintainer. +4. **Disclose.** Merge the fix and release the patched version, then publish the GitHub Security Advisory — this requests a CVE if applicable and serves as the public announcement. + +Out-of-cycle releases are triggered for confirmed Critical/High vulnerabilities or active exploitation. + +## Security announcements + +Security advisories are published via GitHub Security Advisories. Users depending on ONNX will be notified automatically via GitHub's dependency graph. + +## Security Requirements + +Open Neural Network Exchange (ONNX) manages reported vulnerabilities according to its documented security policy and delivers remediations in maintained releases. The project employs established secure development practices such as automated testing, continuous integration, and tooling intended to identify defects during development. Third-party dependencies and build components are periodically reviewed and updated to address known issues and to mitigate supply-chain risk. ONNX does not guarantee that models or inputs are trustworthy, and operators are responsible for validating provenance and applying appropriate isolation, resource limits, and runtime safeguards when executing untrusted workloads. + +For the project's threat model, secure design principles, and the weaknesses they mitigate, see the [Security Assurance Case](docs/AssuranceCase.md). + +## Supply Chain Security + +ONNX release artifacts (wheels and source distributions) meet [SLSA Build Level 2](https://slsa.dev/spec/v1.0/levels#build-l2). Artifacts published to PyPI from this repository's GitHub Actions workflows have a corresponding signed provenance attestation generated by GitHub Actions and stored in GitHub's attestation store. + +### Verifying attestations + +Install the [GitHub CLI](https://cli.github.com/) and run: + +```bash +gh attestation verify --owner onnx +``` + +For example: + +```bash +pip download onnx --no-deps -d ./dist +gh attestation verify ./dist/onnx-*.whl --owner onnx +``` + +A successful verification confirms that the artifact was built by GitHub Actions in the `onnx/onnx` repository and has not been tampered with since it was built. + +### Software Bill of Materials (SBOM) + +Each wheel also embeds a [CycloneDX 1.7](https://cyclonedx.org/) SBOM (`.cdx.json`) listing the bundled third-party components shipped inside the wheel (e.g. statically linked C++ libraries). diff --git a/VERSION_NUMBER b/VERSION_NUMBER new file mode 100644 index 0000000..a6c2798 --- /dev/null +++ b/VERSION_NUMBER @@ -0,0 +1 @@ +1.23.0 diff --git a/cmake/ONNXConfig.cmake.in b/cmake/ONNXConfig.cmake.in new file mode 100644 index 0000000..aa1ef47 --- /dev/null +++ b/cmake/ONNXConfig.cmake.in @@ -0,0 +1,12 @@ +# - Config file for the ONNX package +# It defines ONNX targets for other cmake libraries to use. + +# library version information +set(ONNX_VERSION "@ONNX_VERSION@") + +if(NOT @ONNX_USE_PROTOBUF_SHARED_LIBS@) + find_package(Protobuf REQUIRED CONFIG) +endif() + +# import targets +include ("${CMAKE_CURRENT_LIST_DIR}/ONNXTargets.cmake") diff --git a/cmake/ONNXConfigVersion.cmake.in b/cmake/ONNXConfigVersion.cmake.in new file mode 100644 index 0000000..a25c168 --- /dev/null +++ b/cmake/ONNXConfigVersion.cmake.in @@ -0,0 +1,11 @@ +set(PACKAGE_VERSION "@ONNX_VERSION@") + +# Check whether the requested PACKAGE_FIND_VERSION is compatible +if(PACKAGE_VERSION VERSION_LESS PACKAGE_FIND_VERSION) + set(PACKAGE_VERSION_COMPATIBLE FALSE) +else() + set(PACKAGE_VERSION_COMPATIBLE TRUE) + if (PACKAGE_VERSION VERSION_EQUAL PACKAGE_FIND_VERSION) + set(PACKAGE_VERSION_EXACT TRUE) + endif() +endif() diff --git a/cmake/Utils.cmake b/cmake/Utils.cmake new file mode 100644 index 0000000..44e306a --- /dev/null +++ b/cmake/Utils.cmake @@ -0,0 +1,208 @@ +# SPDX-License-Identifier: Apache-2.0 + +# Look up a dependency by name in sbom.cdx.json and set URL, SHA256, and VERSION +# variables in the caller's scope. For example: +# sbom_get_dep("abseil-cpp" _absl) +# sets _absl_URL, _absl_SHA256, _absl_VERSION. +function(sbom_get_dep dep_name prefix) + file(READ "${CMAKE_CURRENT_SOURCE_DIR}/sbom.cdx.json" _sbom) + string(JSON _count LENGTH "${_sbom}" "components") + foreach(_i RANGE 0 ${_count}) + if(_i EQUAL _count) + break() + endif() + string(JSON _name GET "${_sbom}" "components" ${_i} "name") + if(_name STREQUAL "${dep_name}") + string(JSON _url GET "${_sbom}" "components" ${_i} "externalReferences" 0 "url") + string(JSON _version GET "${_sbom}" "components" ${_i} "version") + set(${prefix}_URL "${_url}" PARENT_SCOPE) + set(${prefix}_VERSION "${_version}" PARENT_SCOPE) + # SHA256 is optional (e.g. git-only deps like nanobind). + string(JSON _hash_count ERROR_VARIABLE _err LENGTH "${_sbom}" "components" ${_i} "hashes") + if(_hash_count AND _hash_count GREATER 0) + string(JSON _sha256 GET "${_sbom}" "components" ${_i} "hashes" 0 "content") + set(${prefix}_SHA256 "${_sha256}" PARENT_SCOPE) + else() + set(${prefix}_SHA256 "" PARENT_SCOPE) + endif() + return() + endif() + endforeach() + message(FATAL_ERROR "Dependency '${dep_name}' not found in sbom.cdx.json") +endfunction() + +# Compiler hardening flags based on OpenSSF guidelines: +# https://best.openssf.org/Compiler-Hardening-Guides/Compiler-Options-Hardening-Guide-for-C-and-C++.html +function(add_onnx_hardening_flags target) + if(NOT ONNX_HARDENING) + return() + endif() + + if(MSVC) + # MSVC hardening compile flags + target_compile_options(${target} PRIVATE + /GS # Buffer security checks + /guard:cf # Control Flow Guard + /sdl # Security Development Lifecycle checks + ) + # MSVC hardening linker flags + target_link_options(${target} PRIVATE + /DYNAMICBASE # ASLR + /NXCOMPAT # Data Execution Prevention + /guard:cf # Control Flow Guard (linker side) + ) + # CET shadow stack requires VS 2019 (MSVC 19.20+) and is x86/x64 only + if(MSVC_VERSION GREATER_EQUAL 1920 AND CMAKE_SYSTEM_PROCESSOR MATCHES "x86|x64|X86|AMD64") + target_link_options(${target} PRIVATE /CETCOMPAT) + endif() + else() + # GCC/Clang hardening compile flags + target_compile_options(${target} PRIVATE + -Wformat + -Wformat=2 + -Wimplicit-fallthrough + -Werror=format-security + -fstack-protector-strong + -fno-delete-null-pointer-checks + -fno-strict-overflow + -fno-strict-aliasing + ) + + # Zero-initialize uninitialized stack variables (requires compiler support) + include(CheckCXXCompilerFlag) + check_cxx_compiler_flag(-ftrivial-auto-var-init=zero COMPILER_SUPPORTS_AUTO_VAR_INIT) + if(COMPILER_SUPPORTS_AUTO_VAR_INIT) + target_compile_options(${target} PRIVATE -ftrivial-auto-var-init=zero) + endif() + + # _FORTIFY_SOURCE requires optimization and conflicts with sanitizers + if(NOT ONNX_USE_ASAN) + target_compile_options(${target} PRIVATE + -U_FORTIFY_SOURCE + -D_FORTIFY_SOURCE=3 + ) + endif() + + # C++ standard library assertions + target_compile_definitions(${target} PRIVATE _GLIBCXX_ASSERTIONS) + + # Stack clash protection (Linux only - not supported on macOS) + if(CMAKE_SYSTEM_NAME STREQUAL "Linux") + include(CheckCXXCompilerFlag) + check_cxx_compiler_flag(-fstack-clash-protection COMPILER_SUPPORTS_STACK_CLASH) + if(COMPILER_SUPPORTS_STACK_CLASH) + target_compile_options(${target} PRIVATE -fstack-clash-protection) + endif() + endif() + + # Control-flow protection for x86_64 (Linux only - not supported on macOS) + if(CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64|amd64|AMD64") + include(CheckCXXCompilerFlag) + check_cxx_compiler_flag(-fcf-protection=full COMPILER_SUPPORTS_CF_PROTECTION) + if(COMPILER_SUPPORTS_CF_PROTECTION) + target_compile_options(${target} PRIVATE -fcf-protection=full) + endif() + endif() + + # Branch protection for AArch64 (Linux only - macOS uses different mechanism) + if(CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64|arm64|ARM64") + include(CheckCXXCompilerFlag) + check_cxx_compiler_flag(-mbranch-protection=standard COMPILER_SUPPORTS_BRANCH_PROTECTION) + if(COMPILER_SUPPORTS_BRANCH_PROTECTION) + target_compile_options(${target} PRIVATE -mbranch-protection=standard) + endif() + endif() + + # Linker hardening flags (Linux only, not macOS) + if(CMAKE_SYSTEM_NAME STREQUAL "Linux") + target_link_options(${target} PRIVATE + -Wl,-z,noexecstack + -Wl,-z,relro + -Wl,-z,now + ) + endif() + endif() +endfunction() + +# Adjusts the MSVC_RUNTIME_LIBRARY property for a given target. +# If CMAKE_MSVC_RUNTIME_LIBRARY is defined, we assume the user wants to explicitly use that value. +# If not, we respect ONNX_USE_MSVC_STATIC_RUNTIME and delegate to the static/dynamic lib respectively, +# with Debug builds using the debug libs. +function(add_msvc_runtime_flag lib) + if(DEFINED CMAKE_MSVC_RUNTIME_LIBRARY) + # Don't do anything here and respect parent-project provided value. + # CMake will have already associated the default value with our target. + message(STATUS "Ignoring ONNX_USE_MSVC_STATIC_RUNTIME since CMAKE_MSVC_RUNTIME_LIBRARY is defined.") + else() + if(ONNX_USE_MSVC_STATIC_RUNTIME) + set_target_properties(${lib} PROPERTIES + MSVC_RUNTIME_LIBRARY "MultiThreaded$<$:Debug>" + ) + else() + set_target_properties(${lib} PROPERTIES + MSVC_RUNTIME_LIBRARY "MultiThreaded$<$:Debug>DLL" + ) + endif() + endif() +endfunction() + +function(add_onnx_global_defines target) + target_compile_definitions(${target} + PUBLIC "ONNX_NAMESPACE=${ONNX_NAMESPACE}") + + if(ONNX_ML) + target_compile_definitions(${target} PUBLIC "ONNX_ML=1") + endif() + + if(ONNX_USE_LITE_PROTO) + target_compile_definitions(${target} PUBLIC "ONNX_USE_LITE_PROTO=1") + endif() + + if(ONNX_DISABLE_STATIC_REGISTRATION) + target_compile_definitions(${target} + PUBLIC "__ONNX_DISABLE_STATIC_REGISTRATION") + endif() +endfunction() + +function(add_onnx_compile_options target) + if(MSVC) + add_msvc_runtime_flag(${target}) + if(ONNX_WERROR) + target_compile_options(${target} PRIVATE "/WX") + endif() + else() + target_compile_options(${target} PRIVATE -Wall -Wextra) + if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_CXX_COMPILER_VERSION + VERSION_GREATER_EQUAL 13) + target_compile_options(${target} PRIVATE "-Wno-stringop-overflow") + endif() + if(CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang") + target_compile_options(${target} PRIVATE "-Wno-shorten-64-to-32") + endif() + if(ONNX_WERROR) + target_compile_options(${target} PRIVATE "-Werror") + endif() + endif() + target_include_directories( + ${target} + PUBLIC $ $ + $) + target_link_libraries(${target} PUBLIC ${LINKED_PROTOBUF_TARGET}) + get_target_property(_onnx_linked_protobuf_is_imported ${LINKED_PROTOBUF_TARGET} IMPORTED) + foreach(ABSL_USED_TARGET IN LISTS protobuf_ABSL_USED_TARGETS) + if(TARGET ${ABSL_USED_TARGET}) + target_link_libraries(${target} PUBLIC ${ABSL_USED_TARGET}) + if(NOT _onnx_linked_protobuf_is_imported) + add_dependencies(${LINKED_PROTOBUF_TARGET} ${ABSL_USED_TARGET}) + endif() + endif() + endforeach() + # Prevent "undefined symbol: _ZNSt10filesystem7__cxx114path14_M_split_cmptsEv" + # (std::filesystem::__cxx11::path::_M_split_cmpts()) on gcc 8 + if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS 9.0) + target_link_libraries(${target} PRIVATE "-lstdc++fs") + endif() + + # Apply hardening flags if enabled + add_onnx_hardening_flags(${target}) +endfunction() diff --git a/cmake/external/FindSanitizer.cmake b/cmake/external/FindSanitizer.cmake new file mode 100644 index 0000000..56be1aa --- /dev/null +++ b/cmake/external/FindSanitizer.cmake @@ -0,0 +1,166 @@ +# SPDX-License-Identifier: Apache-2.0 + +# Find sanitizers +# +# This module sets the following targets: +# Sanitizer::address +# Sanitizer::thread +# Sanitizer::undefined +# Sanitizer::memory +include_guard(GLOBAL) + +option(UBSAN_FLAGS "additional UBSAN flags" OFF) +option(MSAN_FLAGS "additional MSAN flags" OFF) + +get_property(languages GLOBAL PROPERTY ENABLED_LANGUAGES) + +set(_source_code + [==[ + #include + int main() { + printf("hello world!"); + return 0; + } + ]==]) + +set(_bug_address_code + [==[ +#include +int main(int argc, char **argv) { + int *array = (int*)malloc(100*sizeof(int)); + array[0] = 0; + int res = array[argc + 100]; // BOOM + free(array); + return res; +} +]==]) + +set(_bug_undefined_code + [==[ +int main(int argc, char **argv) { + int k = 0x7fffffff; + k += argc; + return 0; +} +]==]) + +include(CMakePushCheckState) +foreach(lang IN LISTS languages) + if(lang STREQUAL C) + include(CheckCSourceCompiles) + include(CheckCSourceRuns) + elseif(lang STREQUAL CXX) + include(CheckCXXSourceCompiles) + include(CheckCXXSourceRuns) + else() + continue() + endif() + foreach(sanitizer_name IN ITEMS address thread undefined memory) + if(TARGET Sanitizer::${sanitizer_name}_${lang}) + continue() + endif() + if(CMAKE_${lang}_COMPILER_ID STREQUAL "MSVC") + if(sanitizer_name STREQUAL "address") + set(SANITIZER_FLAGS "/fsanitize=${sanitizer_name}") + else() + continue() + endif() + else() + set(SANITIZER_FLAGS + "-fsanitize=${sanitizer_name};-fno-omit-frame-pointer") + endif() + if(sanitizer_name STREQUAL "undefined" AND UBSAN_FLAGS) + list(APPEND SANITIZER_FLAGS "${UBSAN_FLAGS}") + endif() + if(sanitizer_name STREQUAL "memory") + list(APPEND SANITIZER_FLAGS "-fsanitize-memory-track-origins=2") + if(MSAN_FLAGS) + list(APPEND SANITIZER_FLAGS "${MSAN_FLAGS}") + endif() + endif() + cmake_push_check_state(RESET) + set(CMAKE_REQUIRED_QUIET ON) + string(REPLACE ";" " " CMAKE_REQUIRED_FLAGS "${SANITIZER_FLAGS}") + + set(SANITIZER_LINK_FLAGS) + if(CMAKE_${lang}_COMPILER_ID STREQUAL "MSVC") + list(APPEND SANITIZER_LINK_FLAGS "/INCREMENTAL:NO") + else() + list(APPEND SANITIZER_LINK_FLAGS "-fsanitize=${sanitizer_name}") + endif() + set(CMAKE_REQUIRED_LINK_OPTIONS "${SANITIZER_LINK_FLAGS}") + + unset(__res CACHE) + if(lang STREQUAL C) + if(CMAKE_${lang}_COMPILER_ID STREQUAL "MSVC") + check_c_source_compiles("${_source_code}" __res) + else() + check_c_source_runs("${_source_code}" __res) + endif() + else() + if(CMAKE_${lang}_COMPILER_ID STREQUAL "MSVC") + check_cxx_source_compiles("${_source_code}" __res) + else() + check_cxx_source_runs("${_source_code}" __res) + endif() + endif() + if(NOT __res) + message(WARNING "Can't find ${sanitizer_name} in ${lang}") + cmake_pop_check_state() + continue() + endif() + + unset(__res CACHE) + if(NOT CMAKE_${lang}_COMPILER_ID STREQUAL "MSVC" AND (sanitizer_name STREQUAL "address") OR (sanitizer_name STREQUAL "undefined")) + set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} -fno-sanitize-recover=all") + if(lang STREQUAL C) + check_c_source_runs("${_bug_${sanitizer_name}_code}" __res) + else() + check_cxx_source_runs("${_bug_${sanitizer_name}_code}" __res) + endif() + if(__res) + message(WARNING "Buffer overflow bug is not detected in ${lang} ${sanitizer_name}") + cmake_pop_check_state() + continue() + endif() + endif() + + add_library(Sanitizer::${sanitizer_name}_${lang} INTERFACE IMPORTED GLOBAL) + if(NOT TARGET Sanitizer::${sanitizer_name}) + add_library(Sanitizer::${sanitizer_name} INTERFACE IMPORTED GLOBAL) + endif() + target_link_libraries(Sanitizer::${sanitizer_name} + INTERFACE Sanitizer::${sanitizer_name}_${lang}) + foreach(SANITIZER_FLAG IN LISTS SANITIZER_FLAGS) + target_compile_options( + Sanitizer::${sanitizer_name}_${lang} + INTERFACE $<$:${SANITIZER_FLAG}>) + endforeach() + foreach(SANITIZER_FLAG IN LISTS SANITIZER_LINK_FLAGS) + target_link_options(Sanitizer::${sanitizer_name}_${lang} INTERFACE + $<$:${SANITIZER_FLAG}>) + endforeach() + + if(CMAKE_${lang}_COMPILER_ID STREQUAL "Clang") + target_compile_options( + Sanitizer::${sanitizer_name}_${lang} + INTERFACE $<$:-shared-libsan>) + endif() + + if(sanitizer_name STREQUAL "address" AND lang STREQUAL CXX) + if(CMAKE_${lang}_COMPILER_ID STREQUAL "MSVC") + target_compile_definitions( + Sanitizer::${sanitizer_name}_${lang} + INTERFACE $<$:_DISABLE_VECTOR_ANNOTATION> + $<$:_DISABLE_STRING_ANNOTATION>) + else() + target_compile_definitions( + Sanitizer::${sanitizer_name}_${lang} + INTERFACE + $<$:_GLIBCXX_SANITIZE_VECTOR> + $<$:_GLIBCXX_SANITIZE_STD_ALLOCATOR>) + endif() + endif() + cmake_pop_check_state() + endforeach() +endforeach() diff --git a/cmake/external/googletest.cmake b/cmake/external/googletest.cmake new file mode 100644 index 0000000..b984283 --- /dev/null +++ b/cmake/external/googletest.cmake @@ -0,0 +1,14 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + +include(FetchContent) +FetchContent_Declare( + googletest + # Specify the commit you depend on and update it regularly. + URL https://github.com/google/googletest/releases/download/v1.17.0/googletest-1.17.0.tar.gz + URL_HASH SHA256=65fab701d9829d38cb77c14acdc431d2108bfdbf8979e40eb8ae567edf10b27c +) +# For Windows: Prevent overriding the parent project's compiler/linker settings +set(gtest_force_shared_crt ON CACHE BOOL "" FORCE) +FetchContent_MakeAvailable(googletest) diff --git a/cmake/summary.cmake b/cmake/summary.cmake new file mode 100644 index 0000000..f262a3b --- /dev/null +++ b/cmake/summary.cmake @@ -0,0 +1,90 @@ +# SPDX-License-Identifier: Apache-2.0 + +# Prints accumulated ONNX configuration summary +function(onnx_print_configuration_summary) + message(STATUS "") + message(STATUS "******** Summary ********") + message(STATUS " CMake version : ${CMAKE_VERSION}") + message(STATUS " CMake command : ${CMAKE_COMMAND}") + message(STATUS " System : ${CMAKE_SYSTEM_NAME}") + message(STATUS " C++ compiler : ${CMAKE_CXX_COMPILER}") + message(STATUS " C++ compiler version : ${CMAKE_CXX_COMPILER_VERSION}") + message(STATUS " Build type : ${CMAKE_BUILD_TYPE}") + message(STATUS " CMAKE_INSTALL_PREFIX : ${CMAKE_INSTALL_PREFIX}") + if(CMAKE_MODULE_PATH) + message(STATUS " CMAKE_MODULE_PATH : ${CMAKE_MODULE_PATH}") + endif() + message(STATUS "") + message(STATUS " ONNX version : ${ONNX_VERSION}") + message(STATUS " ONNX NAMESPACE : ${ONNX_NAMESPACE}") + message(STATUS " ONNX_USE_LITE_PROTO : ${ONNX_USE_LITE_PROTO}") + message(STATUS " ONNX_USE_PROTOBUF_SHARED_LIBS : ${ONNX_USE_PROTOBUF_SHARED_LIBS}") + message(STATUS " ONNX_DISABLE_EXCEPTIONS : ${ONNX_DISABLE_EXCEPTIONS}") + message(STATUS " ONNX_DISABLE_STATIC_REGISTRATION : ${ONNX_DISABLE_STATIC_REGISTRATION}") + message(STATUS " ONNX_WERROR : ${ONNX_WERROR}") + message(STATUS " ONNX_HARDENING : ${ONNX_HARDENING}") + message(STATUS " ONNX_BUILD_TESTS : ${ONNX_BUILD_TESTS}") + message(STATUS " ONNX_USE_UNITY_BUILD : ${ONNX_USE_UNITY_BUILD}") + message(STATUS " BUILD_SHARED_LIBS : ${BUILD_SHARED_LIBS}") + message(STATUS "") + + get_target_property(tmp onnx COMPILE_OPTIONS) + message(STATUS " onnx compile options : ${tmp}") + get_target_property(tmp onnx_proto COMPILE_OPTIONS) + if(tmp) + message(STATUS " onnx_proto compile options : ${tmp}") + endif() + get_target_property(tmp onnx COMPILE_DEFINITIONS) + message(STATUS " onnx compile definitions : ${tmp}") + get_target_property(tmp onnx_proto COMPILE_DEFINITIONS) + message(STATUS " onnx_proto compile definitions : ${tmp}") + get_target_property(tmp onnx LINK_OPTIONS) + if(tmp) + message(STATUS " onnx link options : ${tmp}") + endif() + get_target_property(tmp onnx_proto LINK_OPTIONS) + if(tmp) + message(STATUS " onnx_proto link options : ${tmp}") + endif() + get_target_property(tmp onnx LINK_LIBRARIES) + if(tmp) + message(STATUS " onnx link libraries : ${tmp}") + endif() + get_target_property(tmp onnx_proto LINK_LIBRARIES) + if(tmp) + message(STATUS " onnx_proto link libraries : ${tmp}") + endif() + + message(STATUS "") + message(STATUS " Protobuf version : ${Protobuf_VERSION}") + if(EXISTS "${ONNX_PROTOC_EXECUTABLE}") + message(STATUS " Protobuf compiler : ${ONNX_PROTOC_EXECUTABLE}") + else() + if(TARGET protobuf::protoc) + get_target_property(tmp protobuf::protoc IMPORTED_LOCATION) + if(tmp) + message(STATUS " Protobuf compiler : ${tmp}") + endif() + endif() + endif() + if(TARGET ${LINKED_PROTOBUF_TARGET}) + get_target_property(tmp ${LINKED_PROTOBUF_TARGET} IMPORTED_LOCATION) + if(tmp) + message(STATUS " Protobuf libraries : ${tmp}") + endif() + endif() + message(STATUS " ONNX_BUILD_PYTHON : ${ONNX_BUILD_PYTHON}") + if(ONNX_BUILD_PYTHON) + message(STATUS " Python3 version : ${Python3_VERSION}") + message(STATUS " Python3 executable : ${Python3_EXECUTABLE}") + message(STATUS " Python3 includes : ${Python3_INCLUDE_DIRS}") + message(STATUS " Python3 libraries : ${Python3_LIBRARIES}") + if(Python3_PyPy_VERSION) + message(STATUS " Python3 PyPy version : ${Python3_PyPy_VERSION}") + endif() + message(STATUS " Python3 interpreter ID : ${Python3_INTERPRETER_ID}") + if(Python_SOABI) + message(STATUS " Python SOABI : ${Python_SOABI}") + endif() + endif() +endfunction() diff --git a/cmake/unittest.cmake b/cmake/unittest.cmake new file mode 100644 index 0000000..13aa74d --- /dev/null +++ b/cmake/unittest.cmake @@ -0,0 +1,39 @@ +# SPDX-License-Identifier: Apache-2.0 + +include(CTest) + +set(ONNX_ROOT ${PROJECT_SOURCE_DIR}) +set(UT_NAME ${PROJECT_NAME}_gtests) +set(test_src + ${ONNX_ROOT}/onnx/test/cpp/checker_test.cc + ${ONNX_ROOT}/onnx/test/cpp/data_propagation_test.cc + ${ONNX_ROOT}/onnx/test/cpp/function_context_test.cc + ${ONNX_ROOT}/onnx/test/cpp/function_get_test.cc + ${ONNX_ROOT}/onnx/test/cpp/function_verify_test.cc + ${ONNX_ROOT}/onnx/test/cpp/inliner_test.cc + ${ONNX_ROOT}/onnx/test/cpp/ir_test.cc + ${ONNX_ROOT}/onnx/test/cpp/op_reg_test.cc + ${ONNX_ROOT}/onnx/test/cpp/parser_test.cc + ${ONNX_ROOT}/onnx/test/cpp/schema_registration_test.cc + ${ONNX_ROOT}/onnx/test/cpp/shape_inference_test.cc + ${ONNX_ROOT}/onnx/test/cpp/test_main.cc + ${ONNX_ROOT}/onnx/test/cpp/utf8_conversion_test.cc +) +add_executable(${UT_NAME} ${test_src}) +find_package(Threads REQUIRED) +target_link_libraries(${UT_NAME} PRIVATE onnx Threads::Threads) +if(TARGET GTest::gtest) + target_link_libraries(${UT_NAME} PRIVATE GTest::gtest) +else() + target_link_libraries(${UT_NAME} PRIVATE gtest) +endif() +set(TEST_ARGS) +if(ONNX_GENERATE_TEST_REPORTS) + # generate a report file next to the test program + list( + APPEND + TEST_ARGS + "--gtest_output=xml:$.$.results.xml>") +endif() + +add_test(NAME ${UT_NAME} COMMAND ${UT_NAME} ${TEST_ARGS}) diff --git a/codecov.yml b/codecov.yml new file mode 100644 index 0000000..bfdc987 --- /dev/null +++ b/codecov.yml @@ -0,0 +1,8 @@ +coverage: + status: + project: + default: + informational: true + patch: + default: + informational: true diff --git a/community/logo_request.md b/community/logo_request.md new file mode 100644 index 0000000..3fe317e --- /dev/null +++ b/community/logo_request.md @@ -0,0 +1,16 @@ + + +## Member Company logos + +Member Companies as defined [here](readme.md#community-roles) can request their logo be displayed on https://onnx.ai and other materials. + +To have your logo displayed, submit a PR to https://github.com/onnx/onnx.github.io with the following: +1. Text of the PR must include written permission indicating the logo can be used on the onnx.ai website as well as in presentations showing ONNX Member Companies +2. A high quality logo file with transparent background needs to be committed in the "assets" directory. The image file should be 300dpi (best if physical printing is ever required) or a vector file which can be saved in any resolution. +3. The URL of your company or product web page. Ideally the page mentions ONNX. + +Member Companies may ask for their logo to be removed at any time and their status as Member Company rescinded. The Steering Committee also can vote to remove a Member Company of their status. diff --git a/community/readme.md b/community/readme.md new file mode 100644 index 0000000..0e690fc --- /dev/null +++ b/community/readme.md @@ -0,0 +1,168 @@ + + +# ONNX Open Governance + +## TL;DR + +ONNX is rolling out open governance to encourage broader participation beyond the founding companies. We hope this will make the decision making process more transparent, enable better technical decisions with consideration of more viewpoints, and share the work of maintenance. We want ONNX to be the standard the whole community rallies to without reservations. + +ONNX open governance creates 3 roles: Member, Contributor and Approver. 3 structures are also created: Steering Committee, Special Interest Groups (SIGs), Working Groups. Contributors and Approvers can vote for the Steering Committee members. The Steering Committee charters SIGs and appoints SIG chairs. Every piece of ONNX belongs to some SIG. Contributors and Approvers participate in one or more SIGs. Our governance structure is based on the successful model of Kubernetes. + +The effort is bootstrapped with an initial Steering Committee and set of SIGs with the first elections to occur after 1 year. + +## Principles + +The ONNX community adheres to the following principles: + +* __Open__: ONNX is open source. See repository guidelines and DCO, below. +* __Welcoming and respectful__: See Code of Conduct, below. +* __Transparent and accessible__: Work and collaboration should be done in public. See SIG governance, below. +* __Merit__: Ideas and contributions are accepted according to their technical merit and alignment with project objectives, scope and design principles. Engineering investment >> corporate sponsorship +* __Speed__: Contributing the time and effort to ensure fast decision-making is key to ensuring that the specifications produced is aligned to the fast iteration of machine learning technologies. + +## Community Roles + +### Members +Members are individuals who are interested in or participate in the ONNX community. Members are able to follow and participate in all public modes of communication used by the ONNX community including but not limited to GitHub, Slack, Stack Overflow, email announcements and discussion aliases. Members are expected to adhere to the Code of Conduct but do not have any specific responsibilities. + +### Contributors +Contributors are Members who are active contributors to the community. They can have issues and PRs assigned to them. They also have voting privileges. Contributors can be active in many ways including but not limited to: + +* Authoring or reviewing PRs on GitHub +* Filing or commenting on issues on GitHub +* Contributing to SIG, subproject, or community discussions (e.g. Slack, meetings, email discussion forums, Stack Overflow, etc) +* Creator of content, promoting and advocating the ONNX specification + +A Member can become a Contributor by being sponsored by 2 existing Approvers from different companies. Contributors who are not active in the last 12 months will be removed. + +### Approvers +Approvers are Contributors who are experienced with some aspect of the project and with general software engineering principles. Approvers are responsible for reviewing contributions for acceptance by considering not just code quality but also holistic impact of the contribution including compatibility, performance, and interactions with other areas. + +Approvers need to be active Contributors for at least 3 months and be sponsored by a SIG chair with no objections from other SIG chairs. + +### Member Companies +Member Companies are organizations that support ONNX in one or more of the following ways: +* Having employees participate in SIGs, Working Groups, or the Steering Committee +* Hosting a workshop or meetup for ONNX +* Providing resources for building or hosting ONNX assets +* Doing media or PR activities to promote ONNX +* Shipping a product that supports ONNX + +Member Companies do not have any voting rights, except via their employees who are Contributors. Affiliates and subsidiaries are considered part of the Member Company and not as separate organizations. Being a Member Company does not by itself confer any compliance or certification to the Member Company's products. + +Member Companies can request their logo be displayed on the website and other materials by following these [instructions](logo_request.md). + +## Organizational Structure + +The ONNX community is organized in the following manner, with all governance and execution being planned and coordinated as follows: + +* **Steering Committee** is made up of a set number of people whose charter it is to define and iterate on the vision, goals, and governance process of the ONNX community. +* **Special Interest Groups (SIGs)** are persistent groups that are responsible for specific parts of the project. SIGs must have open and transparent proceedings. Anyone is welcome to participate and contribute provided they follow the Code of Conduct. The purpose of a SIG is to develop a set of goals to be achieved over a set period of time, and then to gather input, drive consensus and closure, implement code contributions, and other related activities to achieve the goal. SIGs are also responsible for ongoing maintenance of the code in their areas. +* **Working Groups** are temporary groups that are formed to address issues that cross SIG boundaries. Working groups do not own any code ownership or other long term artifacts. Working groups can report back and act through involved SIGs. + +### Steering Committee + +#### Role + +The Steering Committee has a set of rights and responsibilities including the following: + +* Define, evolve, and defend the vision, values, mission, and scope of the project. +* Define, evolve, and defend a Code of Conduct, which must include a neutral, unbiased process for resolving conflicts. +* Define and evolve project governance structures and policies, including how members become contributors, approvers, SIG chairs, etc. +* Charter and refine policy for defining new community groups (Special Interest Groups, Working Groups, and any future possible defined structure), and establish transparency and accountability policies for such groups. +* Decide, for the purpose of elections, who is a member of standing of the ONNX project, and what privileges that entails. +* Decide which functional areas and scope are part of the ONNX project, including accepting new or pruning old SIGs and Working Groups. +* Decide how and when official releases of ONNX artifacts are made and what they include. +* Declare releases when quality/feature/other requirements are met. +* Control access to, establish processes regarding, and provide a final escalation path for any ONNX repository, which currently includes all repositories under the ONNX GitHub organizations +* Control and delegate access to and establish processes regarding other project resources/assets, including artifact repositories, build and test infrastructure, web sites and their domains, blogs, social-media accounts, etc. +* Define any certification process. +* Manage the ONNX brand and any outbound marketing. +* Make decisions by majority vote if consensus cannot be reached. + +#### Structure + +The Steering Committee consists of 5 individuals. No single Member Company may have more than 1 representative. Members serve 1 year terms. + +The starting composition will be individuals from Microsoft, Facebook, Amazon, and 2 other Member Companies, who have been picked by the three founding members based on contributions and experience. + +After the initial term of each Steering Committee representative is completed, their seat will be open for any contributor in the community to be elected into the seat via a community vote. Only contributors may vote, but would be restricted to one vote per Member Company. + +If a member of the Steering Committee changes companies, by default they retain and may continue on with the role. If the employment change results in a single Member Company having more than one representative, then one of them must resign. When there is a vacancy on the Steering Committee, the remaining members can appoint a new representative for the remainder of the term until the next election. + +The Steering Committee will decide on and publish an election process within 3 months of formalizing this organizational structure. This will cover voting eligibility, eligibility for candidacy, election process and schedule. During this time period, the Steering Committee will also establish SIGs and Working Groups. + +A Steering Committee member can be removed due to Code of Conduct violations. + +### SIG - Special Interest Groups + +#### Role + +The ONNX project is organized primarily into Special Interest Groups, or SIGs. Each SIG is comprised of individuals from multiple companies and organizations, with a common purpose of advancing the project with respect to a specific topic. + +Our goal is to enable a distributed decision structure and code ownership, as well as providing focused forums for getting work done, making decisions, and on-boarding new contributors. Every identifiable part of the project (e.g., repository, subdirectory, API, test, issue, PR, Slack channel) is intended to be owned by some SIG. At the time of inception of this organizational structure, the following SIGs will be present: + +* Architecture & Infra + * This SIG is responsible for defining and maintaining the core ONNX format, the build and CI/CD systems for ONNX repositories, publishing release packages for ONNX, and creating tools to help integrate with and test against the ONNX standard. This SIG is also the defacto owner of files in the main ONNX repository unless explicitly owned by another SIG. +* Operator Standardization + * This SIG is responsible for determining the operators that are part of the ONNX spec (ONNX and ONNX-ML domains), ensuring high quality operator definitions and documentation, establishing criteria for adding new operators, managing ops domains and compliance tiers, and enforcing versioning mechanisms. +* Converters + * This SIG is responsible for developing and maintaining the various converter repositories under ONNX. +* Model zoo and tutorials + * This SIG is responsible for the respective repositories with the charter of providing a comprehensive collection of state of the art ONNX models from a variety of sources and making it easy for users to get started with ONNX and the ecosystem around it. + +#### Structure + +SIGs must have at least one, and may have up to two SIG chairs at any given time. SIG chairs are intended to be organizers and facilitators, responsible for the operation of the SIG and for communication and coordination with the other SIGs, the Steering Committee, and the broader community. All SIG chairs are appointed by the Steering Committee. If there are more than two contributors being considered for a particular SIG, the Steering Committee will vote on and resolve who the chairs would be. Candidates need to be Approvers. + +Each SIG must have a charter that specifies its scope (topics, subsystems, code repos and directories), responsibilities, and areas of authority. Charters are submitted to the ONNX GitHub via PR for review and approval by the Steering Committee who will be looking to ensure the scope of the SIG as represented in the charter is reasonable. All SIGs are expected to follow the standards established by the Steering Committee for how Contributors are roles of authority/leadership are selected/granted, how decisions are made, and how conflicts are resolved. + +A primary reason that SIGs exist is as forums for collaboration. Much work in a SIG should stay local within that SIG. However, SIGs must communicate in the open, ensure other SIGs and community members can find meeting notes, discussions, designs, and decisions, and periodically communicate a high-level summary of the SIG's work to the community. SIGs are also responsible to: + +* Meet regularly, at least monthly +* Keep up-to-date meeting notes, linked from the SIG's page in the community repo +* Announce meeting agenda and minutes after each meeting, on their SIG mailing list and/or Slack channel +* Ensure the SIG's mailing list is archived (i.e on GitHub) +* Report activity in overall ONNX community meetings +* Participate in release planning meetings, retrospectives, etc (if relevant) +* Actively triage issues, PRs, test failures, etc. related to code and tests owned by the SIG +* Use the above forums as the primary means of working, communicating, and collaborating, as opposed to private emails and meetings + +#### Decision making + +When it is time to formalize the work-product from a SIG, votes are taken from every contributor who participates in the SIG. The list of active contributors is determined by the one (or two) SIG leads to ensure that only those who have actively participated in the SIG can vote. At this time there are no restrictions on how many contributors from any one Member Company can participate (and hence vote). The Steering Committee will monitor how the community behaves and apply constraints if needed in the future. + +While most work shouldn’t require expensive coordination with other SIGs, there will be efforts (features, refactoring, etc.) that cross SIG boundaries. In this case, it is expected that the SIGs coordinate with each other and come to mutually agreed solutions. In some cases, it may make sense to form a Working Group for joint work. Cross-SIG coordination will naturally require more time and implies a certain amount of overhead. This is intentional to encourage changes to be well encapsulated whenever possible. + +### WG - Working Groups + +Working Groups (WGs) are primarily used to facilitate topics of discussion that cross SIG lines, or are topics which are short-lived and require a limited set of decisions to be agreed upon. Working groups: + +* do not own code +* have a clear goal measured through specific deliverables +* will be disbanded after the goal is achieved + +Working Groups can create specifications, recommendations, or implementations for submission to the relevant SIGs for approval and acceptance. + +A list of all active, inactive, and completed working groups can be found in the [working-groups repository](https://github.com/onnx/working-groups) + +Working Groups are formed by submitting a proposal via PR to the Steering Committee. The proposal should cover: + +* what is the exact problem being worked on +* what are the exit criteria +* who are the chairs (up to 2) +* what are the meeting and discussion mechanics + +Working Groups are disbanded when there is no activity for more than *3 months* or when the chair informs the Steering Committee. + +## Repository Guidelines + +The current guidelines for all repos under ONNX github.org could be found [here](repo_guidelines.md). + +## CLA / DCO + +As of October 2020, the CLA (https://cla-assistant.io/onnx/onnx) has been retired. All commits are subject to the DCO (https://www.developercertificate.com/) and need to be signed. diff --git a/community/repo_guidelines.md b/community/repo_guidelines.md new file mode 100644 index 0000000..99f9d87 --- /dev/null +++ b/community/repo_guidelines.md @@ -0,0 +1,40 @@ + + +# Repositories under ONNX GitHub organization + +The ONNX GitHub organization contains a number of repositories. Every repository is owned by a SIG and the Steering Committee is responsible for managing these repos. Requests for creating, transferring, modifying, or archiving repositories can be made by filing an issue a request against https://github.com/onnx/steering-committee. + +## Rules for all repos + +* Must be owned and managed by one of the ONNX SIGs or the Steering Committee +* Must be actively maintained +* Must adopt the ONNX Code of Conduct +* Must adopt the standard ONNX license(s) [All code projects use the Apache 2.0 license. Documentation repositories must use the Creative Commons License version 4.0.] +* Must adopt the ONNX DCO bot +* Must adopt all ONNX automation (like static code analysis) +* Must have CI or other automation in place for repos containing code to ensure quality +* All OWNERS must be members of standing as defined by ability to vote in Steering Committee elections. + +## Requirements for new, contributed repos + +We are happy to accept contributions as repos under the ONNX organization of new projects that meet the following requirements: + +* Project is closely related to ONNX +* Adds value to the ONNX ecosystem +* Determined to need a new repo rather than a folder in an existing repo +* Applicable and usable by a wide set of ONNX users (for example, implemented support for multiple hardware backends at time of contribution or commitment to do so soon after) +* All contributors must have signed the ONNX DCO +* Licenses of dependencies must be acceptable +* Commitment to maintain the repo +* Approval of the SIG that will own the repo +* Approval of the Steering Committee + +If you want to contribute a repository, you should first work with the SIG that will own it. Then the SIG can work with the Steering Committee to finalize. + +## Archiving repos + +Repositories that are inactive or unneeded will be archived. The SIG that owns the repo is responsible for deciding when it should be archived. SIGs should regularly validate the repos they own are still active and necessary. diff --git a/community/sc-election-guidelines.md b/community/sc-election-guidelines.md new file mode 100644 index 0000000..09146ba --- /dev/null +++ b/community/sc-election-guidelines.md @@ -0,0 +1,71 @@ + + +# ONNX Steering Committee election guideline + +## Introduction + +To encourage community participation and wider adoption in the industry, ONNX has introduced [open governance](https://github.com/onnx/onnx/wiki/Expanded-ONNX-Steering-Committee-Announced!) in March 2018. The governance has three defined structures to propel the development of ONNX project forward: [Steering Committee](/community/readme.md#steering-committee), [Special Interest Groups (SIGs)](/community/readme.md#sig---special-interest-groups), and [Working Groups (WGs)](/community/readme.md#wg---working-groups). While SIGs and WGs primarily focus on the technical roadmap of ONNX, the Steering Committee is responsible for setting the vision and governance process of the ONNX community. + +For the first year of its ONNX open governance, representatives from Facebook, Microsoft, AWS, Intel and Nvidia are chosen to serve as the ONNX Steering Committee to help guide the project. The Steering Committee will be elected by the [Contributors](/community/readme.md#community-roles) in its second year and will be re-elected every year. + +This document is created to provide guidelines for the election process to ensure maximum transparency and fairness. + + +## Timeline + +Candidate applications will be accepted in April, and the election will be held in May. The new term for Steering Committee begins on June 1st of the corresponding year. The following table outlines the schedule for the election process. + +| Schedule | Event | +|:-------------|:--------------------| +| 1st Monday of April| Application for Steering Committee candidates open. | +| 3rd Monday of April| Candidates and their applications posted on github.| +| 1st Monday of May| Election begins. | +| 2nd Monday of May| Election closes, and votes counted. Election results announced in the same week.| +| 3rd Monday of May| Previous Steering Committee to meet the newly elected Committee for official transition.| +| June 1 | New term begins with elected Steering Committee. Steering Committee Emeritus members help with the transition for the month of June. | + + +## Eligibility + +### Eligibility for Steering Committee candidacy +Candidates will be self-nominated, and they do not necessarily need to be a [Contributor](/community/readme.md#community-roles) to the ONNX project. The duties of the Steering Committee extend beyond simply contributing code to the ONNX project. + + +### Eligibility for voting + +To participate in the Steering committee election, you must be a Contributor to the ONNX project. As defined in the community guideline, Contributor is sponsored by 2 approvers from different companies. + +Contributors are further required to submit their github handle, email address, and affiliated company name to be eligible for voting. Any Contributor who has not submitted their information by before April 31st will not be able to participate in the election. The Steering Committee is currently reviewing options for collecting contributor information, and the best option will be notified to the Contributors shortly. + +## Candidacy process + +## Voting process + +### General election procedure +In order to promote fairness, the Steering Committee has decided to limit 1 vote per Member Company. Contributors will be able to vote individually, but their votes will be rolled up to represent the vote of associated Member Company. This procedure will prevent large companies with lots of Contributors from dominating the election results. + +### Voting mechanics and algorithm + +The election will use [Condorcet ranking](https://en.wikipedia.org/wiki/Condorcet_method) with [Schulze method](https://en.wikipedia.org/wiki/Schulze_method). Condorcet ranking allows voters to indicate ranked preference for candidates, and Schultz method provides an algorithm to tally the overall preference. + +For ONNX Steering Committee election, the Condorcet ranking with Schulze method will be performed twice. The individual Contributor votes gets tallied first to Member Companies, and the results of the Member Company votes are ranked again using the same method. + +### Voting platform +We will use Condorcet Internet Voting Service ([civs.cs.cornell.edu](http://civs1.civs.us/)) to collect votes from Contributors. + +After votes are casted, the results of individual votes will be uploaded to ONNX Github election directory to ensure transparency. + +## Election officers and Steering Committee emeritus members + +### Election officers +Two election officers will be chosen from the current Steering committee to oversee the election process. They are responsible for overseeing the progress of the election and ensure the process is correctly implemented. Their duties include coordinating election as shown in the timeline above, tallying votes and announcing results for the ONNX community. + +### Steering Committee emeritus members +Two Steering Committee members will remain as emeritus members for the newly elected Committee to help with transition process for 1 month. If previous Steering Committee members are reelected, then they will guide the transition for the new members, and there will not be a separate Steering Committee emeritus members. + + + diff --git a/community/sigs.md b/community/sigs.md new file mode 100644 index 0000000..f8b38b1 --- /dev/null +++ b/community/sigs.md @@ -0,0 +1,19 @@ + + +# SIGs - Special Interest Groups + +As described in the ONNX [governance](/community/readme.md#sig---special-interest-groups), Special Interest Groups (SIGs) are persistent groups responsible for specific parts of the project. SIGs have open and transparent proceedings to develop goals and implement code contributions. SIGs are also responsible for ongoing maintenance of the code in their areas. + +## Joining a SIG + +If you are interested in participating, please [join the discussion](https://join.slack.com/t/lfaifoundation/shared_invite/zt-3wx5vohc3-MeSYi3_dscb~u~cqs7zlPg) in the respective Slack channels. Details about any upcoming meetings will also be shared in the Slack channels. SIG artifacts can be found in the [sigs repository](https://github.com/onnx/sigs). + +You can find the schedule of SIG meetings on the [LFX calendar](https://zoom-lfx.platform.linuxfoundation.org/meetings/lfai-onnx?view=month) + +## Current SIGs + +The list of current sig is found [here](https://github.com/onnx/sigs#current-sigs). diff --git a/community/working-groups.md b/community/working-groups.md new file mode 100644 index 0000000..29e9ea3 --- /dev/null +++ b/community/working-groups.md @@ -0,0 +1,29 @@ + + +# Working Groups + +As described in the ONNX [governance](/community/readme.md#wg---working-groups), Working Groups (WGs) are temporary groups formed to address issues that cross SIG boundaries. Working Groups have a have a clear goal measured through specific deliverables and disband after the goal is achieved. Working groups do not own artifacts long term; they create specifications, recommendations, and/or code implementations for submission to the relevant SIGs for approval and acceptance. + +## Proposing a new working group +New Working Groups are created when there is sufficient interest in a topic area and someone volunteers to be the chair for the group and submits a proposal to the steering committee. The chair facilitates the discussion and helps synthesize proposals and decisions. + +## Joining a working group +Working Groups have most of their discussions on Slack. If you are interested in participating, please join the discussion in the respective Slack channels. Details about any upcoming meetings will also be shared in the Slack channel. Working Group artifacts can be found in the [working-groups repository](https://github.com/onnx/working-groups). + +You can find the schedule of meetings on the [ONNX calendar](https://onnx.ai/calendar) + +## Active working groups + +The list of active working group is found [here](https://github.com/onnx/working-groups#active-working-groups). + +## Completed working groups + +The list of completed working group is found [here](https://github.com/onnx/working-groups#completed-working-groups). + +## Inactive working groups + +The list of inactive working group is found [here](https://github.com/onnx/working-groups#inactive-working-groups). diff --git a/docs/AddFunctionBody.md b/docs/AddFunctionBody.md new file mode 100644 index 0000000..a1ab573 --- /dev/null +++ b/docs/AddFunctionBody.md @@ -0,0 +1,263 @@ + + +# Adding a Function Body Definition for an Operator + +A function body defines how an ONNX operator can be decomposed into simpler ONNX operators. This enables runtimes that don't natively support the operator to still execute it by expanding it into its constituent operations. + +## Table of Contents + +- [Adding a Function Body Definition for an Operator](#adding-a-function-body-definition-for-an-operator) + - [Table of Contents](#table-of-contents) + - [When to use a function body](#when-to-use-a-function-body) + - [File locations](#file-locations) + - [Simple function body (string-based)](#simple-function-body-string-based) + - [Referencing attributes](#referencing-attributes) + - [Context-dependent function body](#context-dependent-function-body) + - [FunctionBuilder API](#functionbuilder-api) + - [Multiple opset versions](#multiple-opset-versions) + - [ONNX function body syntax](#onnx-function-body-syntax) + - [Testing](#testing) + +## When to use a function body + +- The operator can be expressed in terms of other ONNX operators +- You want to provide a reference decomposition that any runtime can use +- The operator is being proposed as a "function" rather than a new primitive (see [Adding New Operator](AddNewOp.md) Step 1) + +If an operator can be split into new primitives, prefer proposing those primitives and making the operator a function. + +## File locations + +| Component | File | +|-----------|------| +| Function body definition | `onnx/defs//defs.cc` (inline with the schema) | +| FunctionBuilder utilities | `onnx/defs/function.h` | +| Function tests (C++) | `onnx/test/cpp/function_get_test.cc`, `onnx/test/cpp/function_verify_test.cc` | + +## Simple function body (string-based) + +For operators whose decomposition is the same regardless of attributes or optional inputs, use the `.FunctionBody()` method with an ONNX-format string: + +```cpp +ONNX_OPERATOR_SET_SCHEMA( + LessOrEqual, + 16, + OpSchema() + .SetDoc(LessOrEqual_ver16_doc) + .Input(0, "A", "First input", "T", OpSchema::Single, true, 1, OpSchema::NonDifferentiable) + .Input(1, "B", "Second input", "T", OpSchema::Single, true, 1, OpSchema::NonDifferentiable) + .Output(0, "C", "Result", "T1", OpSchema::Single, true, 1, OpSchema::NonDifferentiable) + .TypeConstraint("T", OpSchema::all_numeric_types_ir4(), "...") + .TypeConstraint("T1", {"tensor(bool)"}, "...") + .TypeAndShapeInferenceFunction(binaryLogicOpInference) + .FunctionBody(R"ONNX( + { + O1 = Less (A, B) + O2 = Equal (A, B) + C = Or (O1, O2) + } + )ONNX")); +``` + +You can optionally specify the minimum opset version for which the function body is valid: + +```cpp + .FunctionBody(R"ONNX( + { + Zero = Constant () + ZeroCast = CastLike (Zero, X) + Y = Max (X, ZeroCast) + } + )ONNX", 18) // This function body is valid from opset 18 onward +``` + +## Referencing attributes + +Use `@attr_name` syntax to reference the operator's declared attributes inside the function body: + +```cpp +ONNX_OPERATOR_SET_SCHEMA( + LeakyRelu, + 16, + OpSchema() + .Attr("alpha", "Coefficient of leakage.", AttributeProto::FLOAT, 0.01f) + .SetDoc(LeakyRelu_ver16_doc) + .Input(0, "X", "Input tensor", "T", ...) + .Output(0, "Y", "Output tensor", "T", ...) + .TypeConstraint("T", {"tensor(bfloat16)", "tensor(float16)", "tensor(float)", "tensor(double)"}, "...") + .TypeAndShapeInferenceFunction(propagateShapeAndTypeFromFirstInput) + .FunctionBody(R"ONNX( + { + Alpha = Constant () + AlphaCast = CastLike (Alpha, X) + Zero = Constant () + ZeroCast = CastLike(Zero, X) + XLessThanZero = Less(X, ZeroCast) + AlphaMulX = Mul (AlphaCast, X) + Y = Where (XLessThanZero, AlphaMulX, X) + } + )ONNX")); +``` + +The attribute must be declared in the schema's `.Attr()` call for `@attr_name` to work. + +## Context-dependent function body + +When the decomposition depends on which optional inputs are present, attribute values, or input types, use a context-dependent function body builder: + +```cpp +static bool BuildContextDependentFunctionBodyClip( + const FunctionBodyBuildContext& ctx, + const OpSchema& schema, + FunctionProto& functionProto) { + bool has_min = ctx.hasInput(1); + bool has_max = ctx.hasInput(2); + + FunctionBuilder builder(functionProto); + if (!has_min && !has_max) { + builder.Add("output = Identity (input)"); + } else if (has_min && !has_max) { + builder.Add("input_less_than_min = Less (input, min)"); + builder.Add("output = Where (input_less_than_min, min, input)"); + } else if (!has_min && has_max) { + builder.Add("input_large_than_max = Less (max, input)"); + builder.Add("output = Where (input_large_than_max, max, input)"); + } else { + builder.Add("input_less_than_min = Less (input, min)"); + builder.Add("tmp = Where (input_less_than_min, min, input)"); + builder.Add("output_large_than_max = Less (max, tmp)"); + builder.Add("output = Where (output_large_than_max, max, tmp)"); + } + + schema.BuildFunction(functionProto); + return true; +} +``` + +Register it with the schema: + +```cpp +ONNX_OPERATOR_SET_SCHEMA( + Clip, 13, + OpSchema() + .Input(0, "input", "...", "T", OpSchema::Single, ...) + .Input(1, "min", "...", "T", OpSchema::Optional, ...) + .Input(2, "max", "...", "T", OpSchema::Optional, ...) + .Output(0, "output", "...", "T", OpSchema::Single, ...) + .TypeConstraint("T", OpSchema::all_numeric_types_ir4(), "...") + .SetContextDependentFunctionBodyBuilder(BuildContextDependentFunctionBodyClip) + .TypeAndShapeInferenceFunction(propagateShapeAndTypeFromFirstInput)); +``` + +### FunctionBodyBuildContext API + +The context object provides information about the specific instantiation: + +```cpp +struct FunctionBodyBuildContext { + const AttributeProto* getAttribute(const std::string& name) const; // nullptr if not set + bool hasInput(int inputIndex) const; // Is optional input present? + bool hasOutput(int outputIndex) const; // Is optional output present? + const TypeProto* getInputType(int inputIndex) const; // Input type info +}; +``` + +## FunctionBuilder API + +The `FunctionBuilder` class (from `onnx/defs/function.h`) provides a fluent API for constructing function bodies: + +```cpp +FunctionBuilder builder(functionProto); + +// Add nodes using ONNX text format +builder.Add("Y = Relu (X)"); + +// Add with inline attributes +builder.Add("X_ReduceMax = ReduceMax (input, axes)"); + +// Add constants +builder.Const("alpha", std::vector{0.01f}); // Tensor constant +builder.Const1D("axes", int64_t(1)); // 1-D tensor constant + +// Multi-line additions +builder.Add(R"( + X_Sub = Sub (input, X_ReduceMax) + X_Exp = Exp (X_Sub) + X_ReduceSum = ReduceSum (X_Exp, axes) + output = Div (X_Exp, X_ReduceSum) +)"); + +// Add opset dependency +builder.AddOpset("", 18); // default domain, version 18 + +// Always finalize with: +schema.BuildFunction(functionProto); +return true; +``` + +## Multiple opset versions + +When the function body must change across opset versions (e.g., because a sub-op's signature changed), register multiple builders with explicit version numbers: + +```cpp +ONNX_OPERATOR_SET_SCHEMA( + Softmax, 13, + OpSchema() + // ... + .SetContextDependentFunctionBodyBuilder(builderForOpset13) // default (since_version) + .SetContextDependentFunctionBodyBuilder(builderForOpset18, 18) // opset 18+ +); +``` + +The runtime selects the appropriate function body based on the opset version in the model. + +## ONNX function body syntax + +The text format for function bodies uses this grammar: + +``` +output_var = OpName (input1, input2, ...) +``` + +Rules: +- **Variable names** are local intermediates within the function +- **Input/output names** must match the schema's declared `.Input()` and `.Output()` names exactly +- **Constants** are created with the `Constant` op (e.g., `Constant ()`) +- **Type matching** — use `CastLike` instead of `Cast` when the target type depends on an input +- **Attributes** are referenced with `@attr_name` for the enclosing op's attributes + +For the formal grammar, see [Syntax.md](Syntax.md). The parser implementation and its tests provide additional examples: + +| Resource | File | +|----------|------| +| Formal syntax specification | [docs/Syntax.md](Syntax.md) | +| C++ parser implementation | `onnx/defs/parser.h`, `onnx/defs/parser.cc` | +| Python parser | `onnx/parser.py` | +| C++ parser tests | `onnx/test/cpp/parser_test.cc` | +| Python parser tests | `onnx/test/parser_test.py` | + +## Testing + +Function bodies are tested in the C++ test suite: + +- **`onnx/test/cpp/function_get_test.cc`** — verifies `HasFunction()` and `GetFunction()` return correct results +- **`onnx/test/cpp/function_verify_test.cc`** — verifies function body type constraints and correctness + +To run: + +```bash +# Build with tests enabled +ONNX_BUILD_TESTS=1 pip install -e . -v + +# Run C++ tests (Linux/macOS) +LD_LIBRARY_PATH=./.setuptools-cmake-build/ .setuptools-cmake-build/onnx_gtests --gtest_filter="*Function*" + +# Run C++ tests (Windows) +.setuptools-cmake-build\Release\onnx_gtests.exe --gtest_filter="*Function*" +``` + +The node backend tests (in `onnx/backend/test/case/node/`) also implicitly test function body correctness when the reference implementation uses function expansion. diff --git a/docs/AddNewOp.md b/docs/AddNewOp.md new file mode 100644 index 0000000..eac4188 --- /dev/null +++ b/docs/AddNewOp.md @@ -0,0 +1,190 @@ + + +# Adding New Operator or Function to ONNX + +Or updating an existing operator to a new Opset version. + +## Table of Contents + +- [Adding New Operator or Function to ONNX](#adding-new-operator-or-function-to-onnx) + - [Table of Contents](#table-of-contents) + - [Proposing and submitting a new operator or function to ONNX](#proposing-and-submitting-a-new-operator-or-function-to-onnx) + - [4 steps to add an operator](#4-steps-to-add-an-operator) + - [Step 1: Proposing a new operator/function](#step-1-proposing-a-new-operatorfunction) + - [Step 2: Submit PR](#step-2-submit-pr) + - [Example to Follow](#example-to-follow) + - [Step 3: PR Review by Operators SIG](#step-3-pr-review-by-operators-sig) + - [Sign-off](#sign-off) + - [Step 4: ONNX release](#step-4-onnx-release) + - [Updating an existing operator](#updating-an-existing-operator) + - [Checklist](#checklist) + - [Removing operator or function](#removing-operator-or-function) + - [Removing operator](#removing-operator) + - [Removing function](#removing-function) + - [Document removing operator or function](#document-removing-operator-or-function) + +## Proposing and submitting a new operator or function to ONNX + +Operators are the basic building blocks used to define ONNX models. With a rich set of operators, ONNX can describe most DNN and ML models from various frameworks. Functions enable expressing complex operators in terms of more primitive operators. The ONNX specification includes a core set of operators that enable many models. It is a non-goal to add all possible operators, however more operators are added as needed to cover evolving needs. + +In this document, we describe the process of accepting a new proposed operator and how to properly submit a new operator as part of ONNX standard. The goal is to improve on what we currently have based on our experience, learning and feedbacks we gathered from the community. + +## 4 steps to add an operator + +1. Decide what to propose +2. Submit PR for new operator/function +3. Review of PR by Operators SIG +4. Merging of PR and inclusion in next ONNX release + +### Step 1: Proposing a new operator/function + +In order to propose a new operator/function, the following is needed: + +1. If the operator can be expressed in terms of other ONNX operators, then it should be a function and not an operator (we have a function in ONNX : MeanVarianceNormalization). +2. If the operators can be split to new primitives, propose those primitives instead and make the operator a function. +3. Based on a model. This will help us understand the usage and that it solves an actual problem. For the case of the model being private or IP and can't be shared, the operator doesn't belong to the standard and should be implemented as custom OP. +4. The operator needs to be implemented by at-least one (well-known) framework. This help us to understand the actual behavior of the operator and its usage. +5. Operator signature and behavior: + 1. If the operator is available in numpy, prefer numpy semantics. + 2. If the operator is available in more than one frameworks, make sure that your design is general and cover those frameworks. +6. Prefer attributes over inputs. +7. The operator should not be made more complex than is required by the use-cases. However, the operator +should be made as general as possible, as long as it does not make the implementation more complex. +This requires carefully balancing generality and complexity. For example, generalizing from 3-D tensors to +N-D tensors is straight-forward (implementation-wise) for some operators, but complex for other operators. +The choice in such cases will be made based on the complexity of such a generalization. + +### Step 2: Submit PR + +Once the criteria of proposing new operator/function has been satisfied, you will need to submit a PR for the new operator/function. Here the expectation of what the PR should include. The reviewer is expected to verify the completeness of the PR before signoff. + +#### Files to modify + +| Component | File location | +|-----------|--------------| +| Schema definition | `onnx/defs//defs.cc` | +| Operator set registration | `onnx/defs/operator_sets.h` | +| Type/shape inference | Inline in schema via `.TypeAndShapeInferenceFunction(...)` | +| Function body (if applicable) | Inline in schema via `.FunctionBody(...)` — see [AddFunctionBody.md](AddFunctionBody.md) | +| Reference implementation | `onnx/reference/ops/op_.py` | +| Node tests | `onnx/backend/test/case/node/.py` | +| Shape inference tests | `onnx/test/shape_inference_test.py` | +| Upgrade/downgrade tests | `onnx/test/version_converter/automatic_upgrade_test.py` and `automatic_downgrade_test.py` | + +Domain subdirectories under `onnx/defs/`: `math/`, `nn/`, `tensor/`, `logical/`, `reduction/`, `rnn/`, `sequence/`, `image/`, `text/`, `quantization/`, `controlflow/`, `optional/`, `traditionalml/`, `training/` + +1. Description: + 1. Write a detailed description about the operator, and its expected behavior. Pretty much, the description should be clear enough to avoid confusion between implementors. + 2. Add an example in the description to illustrate the usage. + 3. Add reference to the source of the operator in the corresponding framework in the description (if possible). + 4. Write the mathematical formula or a pseudocode in the description. The core algorithm needs to be very clear. +2. Write a reference implementation in Python, this reference implementation should cover all the expected behavior of the operator. Only in extremely rare case, we will waive this requirement. +3. Operator version: check out our +[versioning doc](/docs/Versioning.md#operator-versioning) +4. Write unit test, that cover main usage and corner cases. + 1. The testing examples will be extracted to the doc. + 2. We also generate binary data for it. + 3. Example: [onnx/backend/test/case/node/abs.py](/onnx/backend/test/case/node/abs.py) +5. Write upgrade and downgrade tests: + 1. Add at least one automatic upgrade test for your operator in [onnx/test/version_converter/automatic_upgrade_test.py](/onnx/test/version_converter/automatic_upgrade_test.py) using `_test_op_upgrade`. These tests create a given operator at a given opset version (usually the version the operator was introduced in) and test that the version converter is able to convert them to the highest available version. So for a new operator `_test_op_upgrade` will not test anything, but as soon as the operator gets updated in a future opset the test will automatically become nontrivial. + 2. Similarly add at least one automatic downgrade test for your operator in [onnx/test/version_converter/automatic_downgrade_test.py](/onnx/test/version_converter/automatic_downgrade_test.py) using `_test_op_downgrade`. Specifying the current version so that once the op is updated at a higher opset version the test will ensure downward conversion is validated. + +6. Update the documentation and generate the test data. + 1. Running [the script](/tools/update_doc.sh). If you have files under `onnx/backend/test/data/node` which cannot be generated by the scripts from `onnx/backend/test/case/node`, please further use `python onnx/backend/test/cmd_tools.py generate-data --clean` to cleanup the directory and only preserve needed test data. +to update the doc and generate the test data. +7. Shape Inference function + 1. Please provide a shape inference function in cases where it is meaningful and applicable. + 2. In cases where shape inference is not possible, it must have logic to perform +rank inference at the very least (adding right amount of dimensions to the output shape) + 3. Shape inference functions must be accompanied by unit tests ([onnx/test/shape_inference_test.py](/onnx/test/shape_inference_test.py)). + 4. You can refer to the shape inference function for the `TopK` operator while implementing your own function ([onnx/defs/math/defs.cc](/onnx/defs/math/defs.cc)) + 5. See [ShapeInference.md](ShapeInference.md) for details on the shape inference API and utility functions. +8. Function body (if applicable) + 1. If the operator can be expressed in terms of other ONNX operators, provide a function body definition. + 2. See [AddFunctionBody.md](AddFunctionBody.md) for the full guide on defining function bodies. + +#### Example to Follow + +[PR 1959](https://github.com/onnx/onnx/pull/1959) is a good example to follow. + +### Step 3: PR Review by Operators SIG + +The [Operators SIG](https://github.com/onnx/sigs/tree/main/operators) is responsible for the operators/functions in the ONNX specification. The SIG regularly meets and reviews PRs. + +#### Sign-off + +At least two sign-off from the Operators SIG [contributors](https://github.com/onnx/onnx/tree/main/community#community-roles). + +### Step 4: ONNX release + +Once the PR is reviewed and signed off by the Operators SIG, it will be merged. Your new operator/function will be part of the main branch and available to anyone building from source. These are not official releases. ONNX periodically releases official new versions that are a snapshot of the main branch. Your new operator/function will be part of that release. + +## Updating an existing operator + +The definition of an existing operator may need to be updated when e.g. there are new scenarios or input types to support. The process is largely similar to that for creating a new operator. + +### Steps for updating + +1. **Move the current schema** from `onnx/defs//defs.cc` to `onnx/defs//old.cc`. This preserves the previous version for the version converter. +2. **Create the new version** in `defs.cc` using `ONNX_OPERATOR_SET_SCHEMA(OpName, NEW_VERSION, ...)` with the new opset version number. +3. **Update operator set registration** in `onnx/defs/operator_sets.h` to reference the new version. +4. **Add a version converter adapter** if the behavior or signature changed (see `onnx/version_converter/adapters/` for examples). +5. **Update the reference implementation** in `onnx/reference/ops/op_.py` if behavior changed. +6. **Add upgrade/downgrade tests** using `_test_op_upgrade` and `_test_op_downgrade`. +7. **Regenerate documentation** by running `python onnx/defs/gen_doc.py`. + +### Avoiding duplication between defs.cc and old.cc + +When moving an operator schema to `old.cc`, avoid significant duplication of code or documentation between the old and new versions. Use these strategies: + +- **Shared utility functions**: Extract common logic (e.g., doc strings, type constraint lists, shape inference helpers) into shared functions in the domain's `utils.cc`/`utils.h` or a `defs.h` header. +- **Parameterized functions**: When the old and new versions differ only slightly (e.g., an expanded type list or an additional optional input), use parameterized helper functions that accept the differences as arguments. +- **Use judgment**: Some duplication is acceptable when the alternative would be overly complicated shared logic. If sharing the code makes it harder to understand either version independently, prefer clarity over DRY. + +### Additional files for updates + +| Component | File location | +|-----------|--------------| +| Previous schema version | `onnx/defs//old.cc` | +| Version converter adapter | `onnx/version_converter/adapters/__.h` | +| Upgrade tests | `onnx/test/version_converter/automatic_upgrade_test.py` | +| Downgrade tests | `onnx/test/version_converter/automatic_downgrade_test.py` | + +### Checklist + +Use this checklist when updating an existing operator: https://github.com/onnx/onnx/wiki/Checklist-for-updating-an-existing-operator + +## Removing operator or function + +There are a lot of reasons for removing existing ONNX operator or function, such us being replaced with different operator or can be decomposed by a set of other operators. This document describes the criteria of removing an existing ONNX operator from the standard. + +### Removing operator + +Any operator in ONNX was added because it was required by a model and/or framework. In order to deprecate such an operator we need to do the following. + +- Operator can’t be deprecated unless there is a replacement. + - Replacement can be a more general operator that supersedes the old one. + - Or a set of primitive operators that together can implement the same functionality and behavior of the deprecated operator (Function). +- If the deprecated operator can be decomposed by existing operators then it must be converted to a function. +- If replacement isn’t in ONNX standard yet, then add the replacement operator or set of operators first. +- Add a version adapter which turns the operator into its replacement for the version converter. Example: [onnx/version_converter/adapters/upsample_9_10.h](/onnx/version_converter/adapters/upsample_9_10.h) +- No grace period is needed for deprecated operators. + +### Removing function + +Function, by definition, is composed of ONNX primitives; however, function could have been accelerated by framework or runtime that support ONNX. So, removing function is not recommended, with the exception of adding another single function which supersedes its functionality. + +### Document removing operator or function + +To make sure everyone is aware of the deprecation, the following need to happen: + +- Any removed operator or function from ONNX need to be mentioned in the release note. +- Their old documentation needs to be updated to show the new replacement and the mapping between the old to the new. + - Only `def.cc` need to be remove, `old.cc` will remain. + - `old.cc` need to be updated with the mapping to the replacement. +- ONNX checker need to be updated to error with a proper message. +- All removed operators need to be appended at the end of the `operator.md` file. diff --git a/docs/AssuranceCase.md b/docs/AssuranceCase.md new file mode 100644 index 0000000..18d3df4 --- /dev/null +++ b/docs/AssuranceCase.md @@ -0,0 +1,111 @@ + + +# ONNX Security Assurance Case + +**Version:** 1.1 +**Date:** June 2026 +**Project:** ONNX (Open Neural Network Exchange) +**Scope:** ONNX Core (`onnx/onnx`) and the produced Python wheel + +This document provides the security assurance case for ONNX Core, supporting the OpenSSF Best Practices Badge application. + +## General scope and assurances + +The onnx package aims to provide memory-safe parsing of untrusted protobuf bytes. +Using shape/type inference, version update utilities, and model validation is also considered memory-safe. +Resource exhaustion, however, may be triggered from within these utilities and users are advised to guard against this accordingly. + +Validation utilities such as `onnx.checker.check_model` are provided on a best-effort basis (e.g. a validated `ModelProto` object may contain `NodeProto` objects that do not adhere to the ONNX specification). + +The onnx reference implementation is not yet considered safe for production use on untrusted inputs. + + +## Threat Model + +### Malicious model file + +The attacker supplies a malicious ONNX/protobuf file to a user who parses, validates, or runs type/shape inference or version-conversion on it. + +- **In scope:** memory safety while parsing, type/shape inference, version conversion, and validation of untrusted model bytes. +- **Out of scope:** resource exhaustion (DoS) from those utilities, and the reference runtime executing untrusted models. + +### Supply chain + +The attacker compromises a dependency, the build pipeline, or the published artifact — so that a user installing `onnx` (e.g. from PyPI) receives malicious code. + +- **In scope:** integrity of the published wheels and statically compiled dependencies. +- **Out of scope:** compromise of a user's own machine or CI, and vulnerabilities in transitive dependencies' upstream code itself. + +### External data references + +A malicious model references external tensor data via attacker-controlled file paths, attempting to read files outside the model's directory. + +- **In scope:** external-data paths are validated and normalized; no resolution outside the model directory. +- **Out of scope:** files the user has explicitly granted the model directory access to. + + +## Secure Design Principles (Saltzer & Schroeder) + +| Principle | Application in ONNX Core | +|-----------|--------------------------| +| Economy of Mechanism | Protocol Buffers for serialization; validation centralized in checker.cc; minimal dependencies | +| Fail-Safe Defaults | Validation on by default; must opt out with `check_model=False`; unknown protobuf fields rejected | +| Complete Mediation | Every model load goes through the validation pipeline; all operator inputs are type- and shape-checked | +| Least Privilege | No elevated privileges required; no network access; file I/O restricted to explicitly specified paths | +| Separation of Privilege | External data loading requires both model reference and file system access; releases require SLSA attestation | +| Least Common Mechanism | No global mutable state; validation is stateless; each API call operates independently | +| Psychological Acceptability | Secure defaults need no configuration; clear validation error messages; type-annotated Python API | + + +## Common Weaknesses Mitigated + +| CWE | Mitigation | +|-----|-----------| +| CWE-787/125 Out-of-bounds R/W | Modern C++ (std::vector, RAII); ASan in CI | +| CWE-20 Input Validation | Comprehensive model validation on load; protobuf schema enforcement; operator shape/type checking | +| CWE-416 Use After Free | RAII/smart pointers (unique_ptr, shared_ptr); ASan in CI; code review | +| CWE-190 Integer Overflow | Checked size arithmetic in tensor allocation; UBSan in CI | +| CWE-22 Path Traversal | External data paths validated and normalized; no auto-resolution outside model directory | +| CWE-78 Command Injection | No shell execution in ONNX Core; no system()/exec() usage; enforced by code review and static analysis | +| OWASP A06 Supply Chain | Dependabot; Sigstore signing; minimal dependency footprint; SBOM generation | +| CWE-79/89/352/434 | Not applicable — ONNX Core is not a web application or database | + + +## Security Testing + +| Method | Details | +|--------|---------| +| Static analysis | CodeQL (GitHub Advanced Security), Clang Static Analyzer, sonarcloud | +| Dynamic analysis | ASan, MSan, UBSan, TSan in CI build matrix | +| Fuzzing | Early stage — OSS-Fuzz harnesses ([onnx/fuzz](https://github.com/onnx/onnx/tree/main/onnx/fuzz)) cover the checker, model loader, text parser, shape inference, version converter, and compose; reference evaluator and external-data parsing are not yet covered. A short smoke run of these harnesses is part of this repo's CI ([fuzz.yml](https://github.com/onnx/onnx/blob/main/.github/workflows/fuzz.yml)); the full OSS-Fuzz continuous campaigns run separately and are not surfaced here | +| Dependency scanning | Dependabot, OpenSSF Scorecard | + + +## Security Processes + +**Vulnerability disclosure**: Reports via GitHub Security Advisories (preferred) or onnx-security@lists.lfaidata.foundation as a fallback; CVE assignment through Linux Foundation CNA. See [SECURITY.md](https://github.com/onnx/onnx/blob/main/SECURITY.md). + +**Code review**: All changes require maintainer review; security-sensitive changes require Architecture SIG review; one approval for dependency updates; automated checks must pass before merge ([CODEOWNERS](https://github.com/onnx/onnx/blob/main/CODEOWNERS)). + +**Build & distribution**: artifacts signed with Sigstore; PyPI Trusted Publishing with 2FA required for maintainers; SHA256 checksums published; actions pinned to SHA in CI. + + +## References + +- ONNX Security Policy: https://github.com/onnx/onnx/blob/main/SECURITY.md +- ONNX IR Spec: https://github.com/onnx/onnx/blob/main/docs/IR.md +- Model Checker: https://github.com/onnx/onnx/blob/main/onnx/checker.cc +- OpenSSF Best Practices: https://bestpractices.coreinfrastructure.org/ +- SLSA Framework: https://slsa.dev/ +- CWE Top 25: https://cwe.mitre.org/top25/ +- Saltzer & Schroeder Principles: https://web.mit.edu/Saltzer/www/publications/protection/ + +--- + +**Document Maintainer**: ONNX Architecture & Infrastructure SIG +**Last Updated**: June 2026 +**Review Cycle**: Annual (or upon significant architectural changes) diff --git a/docs/Broadcasting.md b/docs/Broadcasting.md new file mode 100644 index 0000000..9541972 --- /dev/null +++ b/docs/Broadcasting.md @@ -0,0 +1,79 @@ + + +# Broadcasting in ONNX + +In ONNX, element-wise operators can take inputs with different shape, +as long as the input tensors are broadcastable to the same shape. +ONNX supports two types of broadcasting: multidirectional broadcasting and +unidirectional broadcasting. We will introduce these two types of broadcasting +respectively in the following sections. + +## Multidirectional Broadcasting + +In ONNX, a set of tensors are multidirectional broadcastable to the same shape +if one of the following is true: + +- The tensors all have exactly the same shape. +- The tensors all have the same number of dimensions and the length of +each dimensions is either a common length or 1. +- The tensors that have too few dimensions can have their shapes prepended +with a dimension of length 1 to satisfy property 2. + +For example, the following tensor shapes are supported by multidirectional broadcasting: + +- shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar ==> shape(result) = (2, 3, 4, 5) +- shape(A) = (2, 3, 4, 5), shape(B) = (5,), ==> shape(result) = (2, 3, 4, 5) +- shape(A) = (4, 5), shape(B) = (2, 3, 4, 5), ==> shape(result) = (2, 3, 4, 5) +- shape(A) = (1, 4, 5), shape(B) = (2, 3, 1, 1), ==> shape(result) = (2, 3, 4, 5) +- shape(A) = (3, 4, 5), shape(B) = (2, 1, 1, 1), ==> shape(result) = (2, 3, 4, 5) + +Multidirectional broadcasting is the same as [Numpy's broadcasting](https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html#general-broadcasting-rules). + +Multidirectional broadcasting is supported by the following operators in ONNX: + +- [Add](Operators.md#Add) +- [And](Operators.md#And) +- [Div](Operators.md#Div) +- [Equal](Operators.md#Equal) +- [Greater](Operators.md#Greater) +- [Less](Operators.md#Less) +- [Max](Operators.md#Max) +- [Mean](Operators.md#Mean) +- [Min](Operators.md#Min) +- [Mul](Operators.md#Mul) +- [Or](Operators.md#Or) +- [Pow](Operators.md#Pow) +- [Sub](Operators.md#Sub) +- [Sum](Operators.md#Sum) +- [Where](Operators.md#Where) +- [Xor](Operators.md#Xor) + +## Unidirectional Broadcasting + +In ONNX, tensor B is unidirectional broadcastable to tensor A +if one of the following is true: + +- Tensor A and B both have exactly the same shape. +- Tensor A and B all have the same number of dimensions and the length of +each dimensions is either a common length or B's length is 1. +- Tensor B has too few dimensions, and B can have its shapes prepended +with a dimension of length 1 to satisfy property 2. + +When unidirectional broadcasting happens, the output's shape is the same as +the shape of A (i.e., the larger shape of two input tensors). + +In the following examples, tensor B is unidirectional broadcastable to tensor A: + +- shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar ==> shape(result) = (2, 3, 4, 5) +- shape(A) = (2, 3, 4, 5), shape(B) = (5,), ==> shape(result) = (2, 3, 4, 5) +- shape(A) = (2, 3, 4, 5), shape(B) = (2, 1, 1, 5), ==> shape(result) = (2, 3, 4, 5) +- shape(A) = (2, 3, 4, 5), shape(B) = (1, 3, 1, 5), ==> shape(result) = (2, 3, 4, 5) + +Unidirectional broadcasting is supported by the following operators in ONNX: + +- [Gemm](Operators.md#Gemm) +- [PRelu](Operators.md#PRelu) diff --git a/docs/CIPipelines.md b/docs/CIPipelines.md new file mode 100644 index 0000000..0eb8b5e --- /dev/null +++ b/docs/CIPipelines.md @@ -0,0 +1,59 @@ + + +# ONNX CI Pipelines + +## Core CI + +| Workflow | When it runs | What it does | +|---|---|---| +| [CI](/.github/workflows/main.yml) | Every PR, merge_group, push to main, daily (midnight UTC) | C++ and Python tests across Linux, Windows, macOS; Python 3.10–3.14 (including free-threading variants); doc generation; proto generation; node test generation; daily run reports code coverage to Codecov | +| [Windows\_No\_Exception\_CI](/.github/workflows/win_no_exception_ci.yml) | Push and PR to main and rel-\* | C++ tests compiled without exceptions; selective schema loading | +| [Lint / Enforce style](/.github/workflows/lint.yml) | Every PR | Required — runs lintrunner (ruff, mypy, clang-format, etc.) and verifies auto-generated files are up to date | +| [Require label](/.github/workflows/check_pr_label.yml) | Every PR | Requires at least one `topic:` or `module:` label (skipped for Dependabot PRs) | +| [DCO](/.github/workflows/dco_merge_group.yml) | merge\_group | Placeholder DCO job required to enable the GitHub merge queue | + +## Release Builds (1) + +| Workflow | When it runs | What it does | +|---|---|---| +| [Create Releases](/.github/workflows/create_release.yml) | Push to main/rel-\*, PRs targeting rel-\* or labeled "run release CIs", weekly (Monday 00:00 UTC), workflow\_dispatch | Orchestrator — calls WindowsRelease, LinuxRelease, MacRelease, PyodideRelease, and sdistRelease as reusable workflows | +| [WindowsRelease](/.github/workflows/release_windows_cibw.yml) | Called by Create Releases | Builds Windows wheels for x64, x86, and arm64; verifies with minimum supported packages (2)(3) | +| [LinuxRelease](/.github/workflows/release_linux_cibw.yml) | Called by Create Releases | Builds Linux wheels for x86\_64 (manylinux\_2\_28) and aarch64; verifies with minimum supported packages (3) | +| [MacRelease](/.github/workflows/release_macos_cibw.yml) | Called by Create Releases | Builds macOS wheels (macos-14, MACOSX\_DEPLOYMENT\_TARGET=12.0); verifies with minimum supported packages (3) | +| [PyodideRelease](/.github/workflows/release_pyodide_cibw.yml) | Called by Create Releases and on every push | Builds a Pyodide (WebAssembly) wheel on Ubuntu using `cibuildwheel` with a pre-downloaded host `protoc` and protobuf source; runs a basic import test (3) | +| [sdistRelease](/.github/workflows/release_sdist.yml) | Called by Create Releases | Builds and tests source distribution | + +## Security and Supply Chain + +| Workflow | When it runs | What it does | +|---|---|---| +| [CodeQL](/.github/workflows/codeql.yml) | Every PR, push to main/rel-\*, weekly (Friday) | Static analysis of C++ and Python for security vulnerabilities | +| [Scorecard](/.github/workflows/scorecard.yml) | Push to main, weekly (Saturday) | OpenSSF supply-chain security scorecard; publishes results to code-scanning dashboard | +| [Dependency Review](/.github/workflows/dependency-review.yml) | Every PR | Flags vulnerable or license-incompatible dependencies introduced by a PR | + +## Documentation and Maintenance + +| Workflow | When it runs | What it does | +|---|---|---| +| [Pages](/.github/workflows/pages.yml) | PRs to main, push to main | Builds and publishes ONNX documentation to GitHub Pages | +| [Pixi CI](/.github/workflows/pixi_build.yml) | Weekly (Sunday 23:59 UTC) and on PRs | Builds, lints, and tests with the [pixi](https://pixi.sh/) environment manager on Linux, macOS, and Windows; opens an issue on failure when scheduled | +| [Check URLs](/.github/workflows/check_urls.yml) | Push to main/rel-\*, monthly | Checks for broken URLs in the codebase | +| [Stale](/.github/workflows/stale.yml) | Daily | Warns and eventually closes stale issues and PRs | +| [Dependabot](/.github/dependabot.yml) | Monthly | Creates PRs for updated dependency versions | + +--- + +* **(1)** Release CIs run when: + * A PR is merged into main or a rel-\* branch + * Weekly (Monday 00:00 UTC) — publishes a Python wheel to the [onnx-weekly](https://pypi.org/project/onnx-weekly/) package on PyPI + * Any PR targeting a rel-\* branch + * Any PR labeled "run release CIs" (maintainers only) + * Manually via workflow\_dispatch + +* **(2)** Minimum supported dependency versions are listed in `[project.dependencies]` in [pyproject.toml](/pyproject.toml). + +* **(3)** The [PEP 770](https://peps.python.org/pep-0770/) SBOM (`dist-info/sboms/sbom.cdx.json`) is embedded in any wheel build where `SKBUILD_METADATA_DIR` is set (including local `pip wheel` builds). Only the cibuildwheel release pipeline patches the SBOM to reflect the actual protobuf tarball version, URL, and SHA-256 used for that specific build; other wheel builds embed the template values from `sbom.cdx.json`. diff --git a/docs/Changelog-ml.md b/docs/Changelog-ml.md new file mode 100644 index 0000000..e6bd0e5 --- /dev/null +++ b/docs/Changelog-ml.md @@ -0,0 +1,1349 @@ + +## Operator Changelog +*This file is automatically generated from the + [def files](/onnx/defs) via [this script](/onnx/defs/gen_doc.py). + Do not modify directly and instead edit operator definitions.* + +For an operator input/output's differentiability, it can be differentiable, + non-differentiable, or undefined. If a variable's differentiability + is not specified, that variable has undefined differentiability. + +# ai.onnx.ml +## Version 1 of the 'ai.onnx.ml' operator set +### **ai.onnx.ml.ArrayFeatureExtractor-1** + + Select elements of the input tensor based on the indices passed.
+ The indices are applied to the last axes of the tensor. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Inputs + +
+
X : T
+
Data to be selected
+
Y : tensor(int64)
+
The indices, based on 0 as the first index of any dimension.
+
+ +#### Outputs + +
+
Z : T
+
Selected output data as an array
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32), tensor(string)
+
The input must be a tensor of a numeric type or string. The output will be of the same tensor type.
+
+ +### **ai.onnx.ml.Binarizer-1** + + Maps the values of the input tensor to either 0 or 1, element-wise, based on the outcome of a comparison against a threshold value. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
threshold : float (default is 0.0)
+
Values greater than this are mapped to 1, others to 0.
+
+ +#### Inputs + +
+
X : T
+
Data to be binarized
+
+ +#### Outputs + +
+
Y : T
+
Binarized output data
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type. The output will be of the same tensor type.
+
+ +### **ai.onnx.ml.CastMap-1** + + Converts a map to a tensor.
The map key must be an int64 and the values will be ordered + in ascending order based on this key.
The operator supports dense packing or sparse packing. + If using sparse packing, the key cannot exceed the max_map-1 value. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
cast_to : string (default is TO_FLOAT)
+
A string indicating the desired element type of the output tensor, one of 'TO_FLOAT', 'TO_STRING', 'TO_INT64'.
+
map_form : string (default is DENSE)
+
Indicates whether to only output as many values as are in the input (dense), or position the input based on using the key of the map as the index of the output (sparse).
One of 'DENSE', 'SPARSE'.
+
max_map : int (default is 1)
+
If the value of map_form is 'SPARSE,' this attribute indicates the total length of the output tensor.
+
+ +#### Inputs + +
+
X : T1
+
The input map that is to be cast to a tensor
+
+ +#### Outputs + +
+
Y : T2
+
A tensor representing the same data as the input map, ordered by their keys
+
+ +#### Type Constraints + +
+
T1 : map(int64, string), map(int64, float)
+
The input must be an integer map to either string or float.
+
T2 : tensor(string), tensor(float), tensor(int64)
+
The output is a 1-D tensor of string, float, or integer.
+
+ +### **ai.onnx.ml.CategoryMapper-1** + + Converts strings to integers and vice versa.
+ Two sequences of equal length are used to map between integers and strings, + with strings and integers at the same index detailing the mapping.
+ Each operator converts either integers to strings or strings to integers, depending + on which default value attribute is provided. Only one default value attribute + should be defined.
+ If the string default value is set, it will convert integers to strings. + If the int default value is set, it will convert strings to integers. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
cats_int64s : list of ints
+
The integers of the map. This sequence must be the same length as the 'cats_strings' sequence.
+
cats_strings : list of strings
+
The strings of the map. This sequence must be the same length as the 'cats_int64s' sequence
+
default_int64 : int (default is -1)
+
An integer to use when an input string value is not found in the map.
One and only one of the 'default_*' attributes must be defined.
+
default_string : string (default is _Unused)
+
A string to use when an input integer value is not found in the map.
One and only one of the 'default_*' attributes must be defined.
+
+ +#### Inputs + +
+
X : T1
+
Input data
+
+ +#### Outputs + +
+
Y : T2
+
Output data. If strings are input, the output values are integers, and vice versa.
+
+ +#### Type Constraints + +
+
T1 : tensor(string), tensor(int64)
+
The input must be a tensor of strings or integers, either [N,C] or [C].
+
T2 : tensor(string), tensor(int64)
+
The output is a tensor of strings or integers. Its shape will be the same as the input shape.
+
+ +### **ai.onnx.ml.DictVectorizer-1** + + Uses an index mapping to convert a dictionary to an array.
+ Given a dictionary, each key is looked up in the vocabulary attribute corresponding to + the key type. The index into the vocabulary array at which the key is found is then + used to index the output 1-D tensor 'Y' and insert into it the value found in the dictionary 'X'.
+ The key type of the input map must correspond to the element type of the defined vocabulary attribute. + Therefore, the output array will be equal in length to the index mapping vector parameter. + All keys in the input dictionary must be present in the index mapping vector. + For each item in the input dictionary, insert its value in the output array. + Any keys not present in the input dictionary, will be zero in the output array.
+ For example: if the ``string_vocabulary`` parameter is set to ``["a", "c", "b", "z"]``, + then an input of ``{"a": 4, "c": 8}`` will produce an output of ``[4, 8, 0, 0]``. + + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
int64_vocabulary : list of ints
+
An integer vocabulary array.
One and only one of the vocabularies must be defined.
+
string_vocabulary : list of strings
+
A string vocabulary array.
One and only one of the vocabularies must be defined.
+
+ +#### Inputs + +
+
X : T1
+
A dictionary.
+
+ +#### Outputs + +
+
Y : T2
+
A 1-D tensor holding values from the input dictionary.
+
+ +#### Type Constraints + +
+
T1 : map(string, int64), map(int64, string), map(int64, float), map(int64, double), map(string, float), map(string, double)
+
The input must be a map from strings or integers to either strings or a numeric type. The key and value types cannot be the same.
+
T2 : tensor(int64), tensor(float), tensor(double), tensor(string)
+
The output will be a tensor of the value type of the input map. It's shape will be [1,C], where C is the length of the input dictionary.
+
+ +### **ai.onnx.ml.FeatureVectorizer-1** + + Concatenates input tensors into one continuous output.
+ All input shapes are 2-D and are concatenated along the second dimension. 1-D tensors are treated as [1,C]. + Inputs are copied to the output maintaining the order of the input arguments.
+ All inputs must be integers or floats, while the output will be all floating point values. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
inputdimensions : list of ints
+
The size of each input in the input list
+
+ +#### Inputs (1 - ∞) + +
+
X (variadic) : T1
+
An ordered collection of tensors, all with the same element type.
+
+ +#### Outputs + +
+
Y : tensor(float)
+
The output array, elements ordered as the inputs.
+
+ +#### Type Constraints + +
+
T1 : tensor(int32), tensor(int64), tensor(float), tensor(double)
+
The input type must be a tensor of a numeric type.
+
+ +### **ai.onnx.ml.Imputer-1** + + Replaces inputs that equal one value with another, leaving all other elements alone.
+ This operator is typically used to replace missing values in situations where they have a canonical + representation, such as -1, 0, NaN, or some extreme value.
+ One and only one of imputed_value_floats or imputed_value_int64s should be defined -- floats if the input tensor + holds floats, integers if the input tensor holds integers. The imputed values must all fit within the + width of the tensor element type. One and only one of the replaced_value_float or replaced_value_int64 should be defined, + which one depends on whether floats or integers are being processed.
+ The imputed_value attribute length can be 1 element, or it can have one element per input feature.
In other words, if the input tensor has the shape [*,F], then the length of the attribute array may be 1 or F. If it is 1, then it is broadcast along the last dimension and applied to each feature. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
imputed_value_floats : list of floats
+
Value(s) to change to
+
imputed_value_int64s : list of ints
+
Value(s) to change to.
+
replaced_value_float : float (default is 0.0)
+
A value that needs replacing.
+
replaced_value_int64 : int (default is 0)
+
A value that needs replacing.
+
+ +#### Inputs + +
+
X : T
+
Data to be processed.
+
+ +#### Outputs + +
+
Y : T
+
Imputed output data
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input type must be a tensor of a numeric type, either [N,C] or [C]. The output type will be of the same tensor type and shape.
+
+ +### **ai.onnx.ml.LabelEncoder-1** + + Converts strings to integers and vice versa.
+ If the string default value is set, it will convert integers to strings. + If the int default value is set, it will convert strings to integers.
+ Each operator converts either integers to strings or strings to integers, depending + on which default value attribute is provided. Only one default value attribute + should be defined.
+ When converting from integers to strings, the string is fetched from the + 'classes_strings' list, by simple indexing.
+ When converting from strings to integers, the string is looked up in the list + and the index at which it is found is used as the converted value. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
classes_strings : list of strings
+
A list of labels.
+
default_int64 : int (default is -1)
+
An integer to use when an input string value is not found in the map.
One and only one of the 'default_*' attributes must be defined.
+
default_string : string (default is _Unused)
+
A string to use when an input integer value is not found in the map.
One and only one of the 'default_*' attributes must be defined.
+
+ +#### Inputs + +
+
X : T1
+
Input data.
+
+ +#### Outputs + +
+
Y : T2
+
Output data. If strings are input, the output values are integers, and vice versa.
+
+ +#### Type Constraints + +
+
T1 : tensor(string), tensor(int64)
+
The input type must be a tensor of integers or strings, of any shape.
+
T2 : tensor(string), tensor(int64)
+
The output type will be a tensor of strings or integers, and will have the same shape as the input.
+
+ +### **ai.onnx.ml.LinearClassifier-1** + + Linear classifier + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
classlabels_ints : list of ints
+
Class labels when using integer labels. One and only one 'classlabels' attribute must be defined.
+
classlabels_strings : list of strings
+
Class labels when using string labels. One and only one 'classlabels' attribute must be defined.
+
coefficients : list of floats (required)
+
A collection of weights of the model(s).
+
intercepts : list of floats
+
A collection of intercepts.
+
multi_class : int (default is 0)
+
Indicates whether to do OvR or multinomial (0=OvR is the default).
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the scores vector.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT'
+
+ +#### Inputs + +
+
X : T1
+
Data to be classified.
+
+ +#### Outputs + +
+
Y : T2
+
Classification outputs (one class per example).
+
Z : tensor(float)
+
Classification scores ([N,E] - one score for each class and example
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type, and of shape [N,C] or [C]. In the latter case, it will be treated as [1,C]
+
T2 : tensor(string), tensor(int64)
+
The output will be a tensor of strings or integers.
+
+ +### **ai.onnx.ml.LinearRegressor-1** + + Generalized linear regression evaluation.
+ If targets is set to 1 (default) then univariate regression is performed.
+ If targets is set to M then M sets of coefficients must be passed in as a sequence + and M results will be output for each input n in N.
+ The coefficients array is of length n, and the coefficients for each target are contiguous. + Intercepts are optional but if provided must match the number of targets. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
coefficients : list of floats
+
Weights of the model(s).
+
intercepts : list of floats
+
Weights of the intercepts, if used.
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the regression output vector.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT'
+
targets : int (default is 1)
+
The total number of regression targets, 1 if not defined.
+
+ +#### Inputs + +
+
X : T
+
Data to be regressed.
+
+ +#### Outputs + +
+
Y : tensor(float)
+
Regression outputs (one per target, per example).
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type.
+
+ +### **ai.onnx.ml.Normalizer-1** + + Normalize the input. There are three normalization modes, which have the corresponding formulas, + defined using element-wise infix operators '/' and '^' and tensor-wide functions 'max' and 'sum':
+
+ Max: Y = X / max(X)
+ L1: Y = X / sum(X)
+ L2: Y = sqrt(X^2 / sum(X^2)}
+ In all modes, if the divisor is zero, Y == X. +
+ For batches, that is, [N,C] tensors, normalization is done along the C axis. In other words, each row + of the batch is normalized independently. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
norm : string (default is MAX)
+
One of 'MAX,' 'L1,' 'L2'
+
+ +#### Inputs + +
+
X : T
+
Data to be encoded, a tensor of shape [N,C] or [C]
+
+ +#### Outputs + +
+
Y : tensor(float)
+
Encoded output data
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type.
+
+ +### **ai.onnx.ml.OneHotEncoder-1** + + Replace each input element with an array of ones and zeros, where a single + one is placed at the index of the category that was passed in. The total category count + will determine the size of the extra dimension of the output array Y.
+ For example, if we pass a tensor with a single value of 4, and a category count of 8, + the output will be a tensor with ``[0,0,0,0,1,0,0,0]``.
+ This operator assumes every input feature is from the same set of categories.
+ If the input is a tensor of float, int32, or double, the data will be cast + to integers and the cats_int64s category list will be used for the lookups. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
cats_int64s : list of ints
+
List of categories, ints.
One and only one of the 'cats_*' attributes must be defined.
+
cats_strings : list of strings
+
List of categories, strings.
One and only one of the 'cats_*' attributes must be defined.
+
zeros : int (default is 1)
+
If true and category is not present, will return all zeros; if false and a category if not found, the operator will fail.
+
+ +#### Inputs + +
+
X : T
+
Data to be encoded.
+
+ +#### Outputs + +
+
Y : tensor(float)
+
Encoded output data, having one more dimension than X.
+
+ +#### Type Constraints + +
+
T : tensor(string), tensor(int64), tensor(int32), tensor(float), tensor(double)
+
The input must be a tensor of a numeric type.
+
+ +### **ai.onnx.ml.SVMClassifier-1** + + Support Vector Machine classifier + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
classlabels_ints : list of ints
+
Class labels if using integer labels.
One and only one of the 'classlabels_*' attributes must be defined.
+
classlabels_strings : list of strings
+
Class labels if using string labels.
One and only one of the 'classlabels_*' attributes must be defined.
+
coefficients : list of floats
+
+
kernel_params : list of floats
+
List of 3 elements containing gamma, coef0, and degree, in that order. Zero if unused for the kernel.
+
kernel_type : string (default is LINEAR)
+
The kernel type, one of 'LINEAR,' 'POLY,' 'RBF,' 'SIGMOID'.
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the score.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT'
+
prob_a : list of floats
+
First set of probability coefficients.
+
prob_b : list of floats
+
Second set of probability coefficients. This array must be same size as prob_a.
If these are provided then output Z are probability estimates, otherwise they are raw scores.
+
rho : list of floats
+
+
support_vectors : list of floats
+
+
vectors_per_class : list of ints
+
+
+ +#### Inputs + +
+
X : T1
+
Data to be classified.
+
+ +#### Outputs + +
+
Y : T2
+
Classification outputs (one class per example).
+
Z : tensor(float)
+
Class scores (one per class per example), if prob_a and prob_b are provided they are probabilities for each class, otherwise they are raw scores.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type, either [C] or [N,C].
+
T2 : tensor(string), tensor(int64)
+
The output type will be a tensor of strings or integers, depending on which of the classlabels_* attributes is used. Its size will match the batch size of the input.
+
+ +### **ai.onnx.ml.SVMRegressor-1** + + Support Vector Machine regression prediction and one-class SVM anomaly detection. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
coefficients : list of floats
+
Support vector coefficients.
+
kernel_params : list of floats
+
List of 3 elements containing gamma, coef0, and degree, in that order. Zero if unused for the kernel.
+
kernel_type : string (default is LINEAR)
+
The kernel type, one of 'LINEAR,' 'POLY,' 'RBF,' 'SIGMOID'.
+
n_supports : int (default is 0)
+
The number of support vectors.
+
one_class : int (default is 0)
+
Flag indicating whether the regression is a one-class SVM or not.
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the score.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT.'
+
rho : list of floats
+
+
support_vectors : list of floats
+
Chosen support vectors
+
+ +#### Inputs + +
+
X : T
+
Data to be regressed.
+
+ +#### Outputs + +
+
Y : tensor(float)
+
Regression outputs (one score per target per example).
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input type must be a tensor of a numeric type, either [C] or [N,C].
+
+ +### **ai.onnx.ml.Scaler-1** + + Rescale input data, for example to standardize features by removing the mean and scaling to unit variance. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
offset : list of floats
+
First, offset by this.
Can be length of features in an [N,F] tensor or length 1, in which case it applies to all features, regardless of dimension count.
+
scale : list of floats
+
Second, multiply by this.
Can be length of features in an [N,F] tensor or length 1, in which case it applies to all features, regardless of dimension count.
Must be same length as 'offset'
+
+ +#### Inputs + +
+
X : T
+
Data to be scaled.
+
+ +#### Outputs + +
+
Y : tensor(float)
+
Scaled output data.
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type.
+
+ +### **ai.onnx.ml.TreeEnsembleClassifier-1** + + Tree Ensemble classifier. Returns the top class for each of N inputs.
+ The attributes named 'nodes_X' form a sequence of tuples, associated by + index into the sequences, which must all be of equal length. These tuples + define the nodes.
+ Similarly, all fields prefixed with 'class_' are tuples of votes at the leaves. + A leaf may have multiple votes, where each vote is weighted by + the associated class_weights index.
+ One and only one of classlabels_strings or classlabels_int64s + will be defined. The class_ids are indices into this list. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
base_values : list of floats
+
Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
+
class_ids : list of ints
+
The index of the class list that each weight is for.
+
class_nodeids : list of ints
+
node id that this weight is for.
+
class_treeids : list of ints
+
The id of the tree that this node is in.
+
class_weights : list of floats
+
The weight for the class in class_id.
+
classlabels_int64s : list of ints
+
Class labels if using integer labels.
One and only one of the 'classlabels_*' attributes must be defined.
+
classlabels_strings : list of strings
+
Class labels if using string labels.
One and only one of the 'classlabels_*' attributes must be defined.
+
nodes_falsenodeids : list of ints
+
Child node if expression is false.
+
nodes_featureids : list of ints
+
Feature id for each node.
+
nodes_hitrates : list of floats
+
Popularity of each node, used for performance and may be omitted.
+
nodes_missing_value_tracks_true : list of ints
+
For each node, define what to do in the presence of a missing value: if a value is missing (NaN), use the 'true' or 'false' branch based on the value in this array.
This attribute may be left undefined, and the default value is false (0) for all nodes.
+
nodes_modes : list of strings
+
The node kind, that is, the comparison to make at the node. There is no comparison to make at a leaf node.
One of 'BRANCH_LEQ', 'BRANCH_LT', 'BRANCH_GTE', 'BRANCH_GT', 'BRANCH_EQ', 'BRANCH_NEQ', 'LEAF'
+
nodes_nodeids : list of ints
+
Node id for each node. Ids may restart at zero for each tree, but it not required to.
+
nodes_treeids : list of ints
+
Tree id for each node.
+
nodes_truenodeids : list of ints
+
Child node if expression is true.
+
nodes_values : list of floats
+
Thresholds to do the splitting on for each node.
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the score.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT.'
+
+ +#### Inputs + +
+
X : T1
+
Input of shape [N,F]
+
+ +#### Outputs + +
+
Y : T2
+
N, Top class for each point
+
Z : tensor(float)
+
The class score for each class, for each point, a tensor of shape [N,E].
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input type must be a tensor of a numeric type.
+
T2 : tensor(string), tensor(int64)
+
The output type will be a tensor of strings or integers, depending on which of the classlabels_* attributes is used.
+
+ +### **ai.onnx.ml.TreeEnsembleRegressor-1** + + Tree Ensemble regressor. Returns the regressed values for each input in N.
+ All args with nodes_ are fields of a tuple of tree nodes, and + it is assumed they are the same length, and an index i will decode the + tuple across these inputs. Each node id can appear only once + for each tree id.
+ All fields prefixed with target_ are tuples of votes at the leaves.
+ A leaf may have multiple votes, where each vote is weighted by + the associated target_weights index.
+ All trees must have their node ids start at 0 and increment by 1.
+ Mode enum is BRANCH_LEQ, BRANCH_LT, BRANCH_GTE, BRANCH_GT, BRANCH_EQ, BRANCH_NEQ, LEAF + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
aggregate_function : string (default is SUM)
+
Defines how to aggregate leaf values within a target.
One of 'AVERAGE,' 'SUM,' 'MIN,' 'MAX.'
+
base_values : list of floats
+
Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
+
n_targets : int
+
The total number of targets.
+
nodes_falsenodeids : list of ints
+
Child node if expression is false
+
nodes_featureids : list of ints
+
Feature id for each node.
+
nodes_hitrates : list of floats
+
Popularity of each node, used for performance and may be omitted.
+
nodes_missing_value_tracks_true : list of ints
+
For each node, define what to do in the presence of a NaN: use the 'true' (if the attribute value is 1) or 'false' (if the attribute value is 0) branch based on the value in this array.
This attribute may be left undefined and the default value is false (0) for all nodes.
+
nodes_modes : list of strings
+
The node kind, that is, the comparison to make at the node. There is no comparison to make at a leaf node.
One of 'BRANCH_LEQ', 'BRANCH_LT', 'BRANCH_GTE', 'BRANCH_GT', 'BRANCH_EQ', 'BRANCH_NEQ', 'LEAF'
+
nodes_nodeids : list of ints
+
Node id for each node. Node ids must restart at zero for each tree and increase sequentially.
+
nodes_treeids : list of ints
+
Tree id for each node.
+
nodes_truenodeids : list of ints
+
Child node if expression is true
+
nodes_values : list of floats
+
Thresholds to do the splitting on for each node.
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the score.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT'
+
target_ids : list of ints
+
The index of the target that each weight is for
+
target_nodeids : list of ints
+
The node id of each weight
+
target_treeids : list of ints
+
The id of the tree that each node is in.
+
target_weights : list of floats
+
The weight for each target
+
+ +#### Inputs + +
+
X : T
+
Input of shape [N,F]
+
+ +#### Outputs + +
+
Y : tensor(float)
+
N classes
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input type must be a tensor of a numeric type.
+
+ +### **ai.onnx.ml.ZipMap-1** + + Creates a map from the input and the attributes.
+ The values are provided by the input tensor, while the keys are specified by the attributes. + Must provide keys in either classlabels_strings or classlabels_int64s (but not both).
+ The columns of the tensor correspond one-by-one to the keys specified by the attributes. There must be as many columns as keys.
+ +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
classlabels_int64s : list of ints
+
The keys when using int keys.
One and only one of the 'classlabels_*' attributes must be defined.
+
classlabels_strings : list of strings
+
The keys when using string keys.
One and only one of the 'classlabels_*' attributes must be defined.
+
+ +#### Inputs + +
+
X : tensor(float)
+
The input values
+
+ +#### Outputs + +
+
Z : T
+
The output map
+
+ +#### Type Constraints + +
+
T : seq(map(string, float)), seq(map(int64, float))
+
The output will be a sequence of string or integer maps to float.
+
+ +## Version 2 of the 'ai.onnx.ml' operator set +### **ai.onnx.ml.LabelEncoder-2** + + Maps each element in the input tensor to another value.
+ The mapping is determined by the two parallel attributes, 'keys_*' and + 'values_*' attribute. The i-th value in the specified 'keys_*' attribute + would be mapped to the i-th value in the specified 'values_*' attribute. It + implies that input's element type and the element type of the specified + 'keys_*' should be identical while the output type is identical to the + specified 'values_*' attribute. If an input element can not be found in the + specified 'keys_*' attribute, the 'default_*' that matches the specified + 'values_*' attribute may be used as its output value.
+ Let's consider an example which maps a string tensor to an integer tensor. + Assume and 'keys_strings' is ["Amy", "Sally"], 'values_int64s' is [5, 6], + and 'default_int64' is '-1'. The input ["Dori", "Amy", "Amy", "Sally", + "Sally"] would be mapped to [-1, 5, 5, 6, 6].
+ Since this operator is an one-to-one mapping, its input and output shapes + are the same. Notice that only one of 'keys_*'/'values_*' can be set.
+ For key look-up, bit-wise comparison is used so even a float NaN can be + mapped to a value in 'values_*' attribute.
+ +#### Version + +This version of the operator has been available since version 2 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
default_float : float (default is -0.0)
+
A float.
+
default_int64 : int (default is -1)
+
An integer.
+
default_string : string (default is _Unused)
+
A string.
+
keys_floats : list of floats
+
A list of floats.
+
keys_int64s : list of ints
+
A list of ints.
+
keys_strings : list of strings
+
A list of strings. One and only one of 'keys_*'s should be set.
+
values_floats : list of floats
+
A list of floats.
+
values_int64s : list of ints
+
A list of ints.
+
values_strings : list of strings
+
A list of strings. One and only one of 'value_*'s should be set.
+
+ +#### Inputs + +
+
X : T1
+
Input data. It can be either tensor or scalar.
+
+ +#### Outputs + +
+
Y : T2
+
Output data.
+
+ +#### Type Constraints + +
+
T1 : tensor(string), tensor(int64), tensor(float)
+
The input type is a tensor of any shape.
+
T2 : tensor(string), tensor(int64), tensor(float)
+
Output type is determined by the specified 'values_*' attribute.
+
+ +## Version 3 of the 'ai.onnx.ml' operator set +### **ai.onnx.ml.TreeEnsembleClassifier-3** + + Tree Ensemble classifier. Returns the top class for each of N inputs.
+ The attributes named 'nodes_X' form a sequence of tuples, associated by + index into the sequences, which must all be of equal length. These tuples + define the nodes.
+ Similarly, all fields prefixed with 'class_' are tuples of votes at the leaves. + A leaf may have multiple votes, where each vote is weighted by + the associated class_weights index.
+ One and only one of classlabels_strings or classlabels_int64s + will be defined. The class_ids are indices into this list. + All fields ending with _as_tensor can be used instead of the + same parameter without the suffix if the element type is double and not float. + +#### Version + +This version of the operator has been available since version 3 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
base_values : list of floats
+
Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
+
base_values_as_tensor : tensor
+
Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
+
class_ids : list of ints
+
The index of the class list that each weight is for.
+
class_nodeids : list of ints
+
node id that this weight is for.
+
class_treeids : list of ints
+
The id of the tree that this node is in.
+
class_weights : list of floats
+
The weight for the class in class_id.
+
class_weights_as_tensor : tensor
+
The weight for the class in class_id.
+
classlabels_int64s : list of ints
+
Class labels if using integer labels.
One and only one of the 'classlabels_*' attributes must be defined.
+
classlabels_strings : list of strings
+
Class labels if using string labels.
One and only one of the 'classlabels_*' attributes must be defined.
+
nodes_falsenodeids : list of ints
+
Child node if expression is false.
+
nodes_featureids : list of ints
+
Feature id for each node.
+
nodes_hitrates : list of floats
+
Popularity of each node, used for performance and may be omitted.
+
nodes_hitrates_as_tensor : tensor
+
Popularity of each node, used for performance and may be omitted.
+
nodes_missing_value_tracks_true : list of ints
+
For each node, define what to do in the presence of a missing value: if a value is missing (NaN), use the 'true' or 'false' branch based on the value in this array.
This attribute may be left undefined, and the default value is false (0) for all nodes.
+
nodes_modes : list of strings
+
The node kind, that is, the comparison to make at the node. There is no comparison to make at a leaf node.
One of 'BRANCH_LEQ', 'BRANCH_LT', 'BRANCH_GTE', 'BRANCH_GT', 'BRANCH_EQ', 'BRANCH_NEQ', 'LEAF'
+
nodes_nodeids : list of ints
+
Node id for each node. Ids may restart at zero for each tree, but it not required to.
+
nodes_treeids : list of ints
+
Tree id for each node.
+
nodes_truenodeids : list of ints
+
Child node if expression is true.
+
nodes_values : list of floats
+
Thresholds to do the splitting on for each node.
+
nodes_values_as_tensor : tensor
+
Thresholds to do the splitting on for each node.
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the score.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT.'
+
+ +#### Inputs + +
+
X : T1
+
Input of shape [N,F]
+
+ +#### Outputs + +
+
Y : T2
+
N, Top class for each point
+
Z : tensor(float)
+
The class score for each class, for each point, a tensor of shape [N,E].
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input type must be a tensor of a numeric type.
+
T2 : tensor(string), tensor(int64)
+
The output type will be a tensor of strings or integers, depending on which of the classlabels_* attributes is used.
+
+ +### **ai.onnx.ml.TreeEnsembleRegressor-3** + + Tree Ensemble regressor. Returns the regressed values for each input in N.
+ All args with nodes_ are fields of a tuple of tree nodes, and + it is assumed they are the same length, and an index i will decode the + tuple across these inputs. Each node id can appear only once + for each tree id.
+ All fields prefixed with target_ are tuples of votes at the leaves.
+ A leaf may have multiple votes, where each vote is weighted by + the associated target_weights index.
+ All fields ending with _as_tensor can be used instead of the + same parameter without the suffix if the element type is double and not float. + All trees must have their node ids start at 0 and increment by 1.
+ Mode enum is BRANCH_LEQ, BRANCH_LT, BRANCH_GTE, BRANCH_GT, BRANCH_EQ, BRANCH_NEQ, LEAF + +#### Version + +This version of the operator has been available since version 3 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
aggregate_function : string (default is SUM)
+
Defines how to aggregate leaf values within a target.
One of 'AVERAGE,' 'SUM,' 'MIN,' 'MAX.'
+
base_values : list of floats
+
Base values for regression, added to final prediction after applying aggregate_function; the size must be the same as the classes or can be left unassigned (assumed 0)
+
base_values_as_tensor : tensor
+
Base values for regression, added to final prediction after applying aggregate_function; the size must be the same as the classes or can be left unassigned (assumed 0)
+
n_targets : int
+
The total number of targets.
+
nodes_falsenodeids : list of ints
+
Child node if expression is false
+
nodes_featureids : list of ints
+
Feature id for each node.
+
nodes_hitrates : list of floats
+
Popularity of each node, used for performance and may be omitted.
+
nodes_hitrates_as_tensor : tensor
+
Popularity of each node, used for performance and may be omitted.
+
nodes_missing_value_tracks_true : list of ints
+
For each node, define what to do in the presence of a NaN: use the 'true' (if the attribute value is 1) or 'false' (if the attribute value is 0) branch based on the value in this array.
This attribute may be left undefined and the default value is false (0) for all nodes.
+
nodes_modes : list of strings
+
The node kind, that is, the comparison to make at the node. There is no comparison to make at a leaf node.
One of 'BRANCH_LEQ', 'BRANCH_LT', 'BRANCH_GTE', 'BRANCH_GT', 'BRANCH_EQ', 'BRANCH_NEQ', 'LEAF'
+
nodes_nodeids : list of ints
+
Node id for each node. Node ids must restart at zero for each tree and increase sequentially.
+
nodes_treeids : list of ints
+
Tree id for each node.
+
nodes_truenodeids : list of ints
+
Child node if expression is true
+
nodes_values : list of floats
+
Thresholds to do the splitting on for each node.
+
nodes_values_as_tensor : tensor
+
Thresholds to do the splitting on for each node.
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the score.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT'
+
target_ids : list of ints
+
The index of the target that each weight is for
+
target_nodeids : list of ints
+
The node id of each weight
+
target_treeids : list of ints
+
The id of the tree that each node is in.
+
target_weights : list of floats
+
The weight for each target
+
target_weights_as_tensor : tensor
+
The weight for each target
+
+ +#### Inputs + +
+
X : T
+
Input of shape [N,F]
+
+ +#### Outputs + +
+
Y : tensor(float)
+
N classes
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input type must be a tensor of a numeric type.
+
+ +## Version 4 of the 'ai.onnx.ml' operator set +### **ai.onnx.ml.LabelEncoder-4** + + Maps each element in the input tensor to another value.
+ The mapping is determined by the two parallel attributes, 'keys_*' and + 'values_*' attribute. The i-th value in the specified 'keys_*' attribute + would be mapped to the i-th value in the specified 'values_*' attribute. It + implies that input's element type and the element type of the specified + 'keys_*' should be identical while the output type is identical to the + specified 'values_*' attribute. Note that the 'keys_*' and 'values_*' attributes + must have the same length. If an input element can not be found in the + specified 'keys_*' attribute, the 'default_*' that matches the specified + 'values_*' attribute may be used as its output value. The type of the 'default_*' + attribute must match the 'values_*' attribute chosen.
+ Let's consider an example which maps a string tensor to an integer tensor. + Assume and 'keys_strings' is ["Amy", "Sally"], 'values_int64s' is [5, 6], + and 'default_int64' is '-1'. The input ["Dori", "Amy", "Amy", "Sally", + "Sally"] would be mapped to [-1, 5, 5, 6, 6].
+ Since this operator is an one-to-one mapping, its input and output shapes + are the same. Notice that only one of 'keys_*'/'values_*' can be set.
+ Float keys with value 'NaN' match any input 'NaN' value regardless of bit + value. If a key is repeated, the last key takes precedence. + +#### Version + +This version of the operator has been available since version 4 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
default_float : float (default is -0.0)
+
A float.
+
default_int64 : int (default is -1)
+
An integer.
+
default_string : string (default is _Unused)
+
A string.
+
default_tensor : tensor
+
A default tensor. {"_Unused"} if values_* has string type, {-1} if values_* has integral type, and {-0.f} if values_* has float type.
+
keys_floats : list of floats
+
A list of floats.
+
keys_int64s : list of ints
+
A list of ints.
+
keys_strings : list of strings
+
A list of strings.
+
keys_tensor : tensor
+
Keys encoded as a 1D tensor. One and only one of 'keys_*'s should be set.
+
values_floats : list of floats
+
A list of floats.
+
values_int64s : list of ints
+
A list of ints.
+
values_strings : list of strings
+
A list of strings.
+
values_tensor : tensor
+
Values encoded as a 1D tensor. One and only one of 'values_*'s should be set.
+
+ +#### Inputs + +
+
X : T1
+
Input data. It must have the same element type as the keys_* attribute set.
+
+ +#### Outputs + +
+
Y : T2
+
Output data. This tensor's element type is based on the values_* attribute set.
+
+ +#### Type Constraints + +
+
T1 : tensor(string), tensor(int64), tensor(float), tensor(int32), tensor(int16), tensor(double)
+
The input type is a tensor of any shape.
+
T2 : tensor(string), tensor(int64), tensor(float), tensor(int32), tensor(int16), tensor(double)
+
Output type is determined by the specified 'values_*' attribute.
+
+ +## Version 5 of the 'ai.onnx.ml' operator set +### **ai.onnx.ml.TreeEnsemble-5** + + Tree Ensemble operator. Returns the regressed values for each input in a batch. + Inputs have dimensions `[N, F]` where `N` is the input batch size and `F` is the number of input features. + Outputs have dimensions `[N, num_targets]` where `N` is the batch size and `num_targets` is the number of targets, which is a configurable attribute. + + The encoding of this attribute is split along interior nodes and the leaves of the trees. Notably, attributes with the prefix `nodes_*` are associated with interior nodes, and attributes with the prefix `leaf_*` are associated with leaves. + The attributes `nodes_*` must all have the same length and encode a sequence of tuples, as defined by taking all the `nodes_*` fields at a given position. + + All fields prefixed with `leaf_*` represent tree leaves, and similarly define tuples of leaves and must have identical length. + + This operator can be used to implement both the previous `TreeEnsembleRegressor` and `TreeEnsembleClassifier` nodes. + The `TreeEnsembleRegressor` node maps directly to this node and requires changing how the nodes are represented. + The `TreeEnsembleClassifier` node can be implemented by adding a `ArgMax` node after this node to determine the top class. + To encode class labels, a `LabelEncoder` or `GatherND` operator may be used. + +#### Version + +This version of the operator has been available since version 5 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
aggregate_function : int (default is 1)
+
Defines how to aggregate leaf values within a target.
One of 'AVERAGE' (0) 'SUM' (1) 'MIN' (2) 'MAX (3) defaults to 'SUM' (1)
+
leaf_targetids : list of ints (required)
+
The index of the target that this leaf contributes to (this must be in range `[0, n_targets)`).
+
leaf_weights : tensor (required)
+
The weight for each leaf.
+
membership_values : tensor
+
Members to test membership of for each set membership node. List all of the members to test again in the order that the 'BRANCH_MEMBER' mode appears in `node_modes`, delimited by `NaN`s. Will have the same number of sets of values as nodes with mode 'BRANCH_MEMBER'. This may be omitted if the node doesn't contain any 'BRANCH_MEMBER' nodes.
+
n_targets : int
+
The total number of targets.
+
nodes_falseleafs : list of ints (required)
+
1 if false branch is leaf for each node and 0 if an interior node. To represent a tree that is a leaf (only has one node), one can do so by having a single `nodes_*` entry with true and false branches referencing the same `leaf_*` entry
+
nodes_falsenodeids : list of ints (required)
+
If `nodes_falseleafs` is false at an entry, this represents the position of the false branch node. This position can be used to index into a `nodes_*` entry. If `nodes_falseleafs` is false, it is an index into the leaf_* attributes.
+
nodes_featureids : list of ints (required)
+
Feature id for each node.
+
nodes_hitrates : tensor
+
Popularity of each node, used for performance and may be omitted.
+
nodes_missing_value_tracks_true : list of ints
+
For each node, define whether to follow the true branch (if attribute value is 1) or false branch (if attribute value is 0) in the presence of a NaN input feature. This attribute may be left undefined and the default value is false (0) for all nodes.
+
nodes_modes : tensor (required)
+
The comparison operation performed by the node. This is encoded as an enumeration of 0 ('BRANCH_LEQ'), 1 ('BRANCH_LT'), 2 ('BRANCH_GTE'), 3 ('BRANCH_GT'), 4 ('BRANCH_EQ'), 5 ('BRANCH_NEQ'), and 6 ('BRANCH_MEMBER'). Note this is a tensor of type uint8.
+
nodes_splits : tensor (required)
+
Thresholds to do the splitting on for each node with mode that is not 'BRANCH_MEMBER'.
+
nodes_trueleafs : list of ints (required)
+
1 if true branch is leaf for each node and 0 an interior node. To represent a tree that is a leaf (only has one node), one can do so by having a single `nodes_*` entry with true and false branches referencing the same `leaf_*` entry
+
nodes_truenodeids : list of ints (required)
+
If `nodes_trueleafs` is false at an entry, this represents the position of the true branch node. This position can be used to index into a `nodes_*` entry. If `nodes_trueleafs` is false, it is an index into the leaf_* attributes.
+
post_transform : int (default is 0)
+
Indicates the transform to apply to the score.
One of 'NONE' (0), 'SOFTMAX' (1), 'LOGISTIC' (2), 'SOFTMAX_ZERO' (3) or 'PROBIT' (4), defaults to 'NONE' (0)
+
tree_roots : list of ints (required)
+
Index into `nodes_*` for the root of each tree. The tree structure is derived from the branching of each node.
+
+ +#### Inputs + +
+
X : T
+
Input of shape [Batch Size, Number of Features]
+
+ +#### Outputs + +
+
Y : T
+
Output of shape [Batch Size, Number of targets]
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(float16)
+
The input type must be a tensor of a numeric type.
+
+ +### **ai.onnx.ml.TreeEnsembleClassifier-5** (deprecated) + + This operator is DEPRECATED. Please use TreeEnsemble with provides similar functionality. + In order to determine the top class, the ArgMax node can be applied to the output of TreeEnsemble. + To encode class labels, use a LabelEncoder operator. + Tree Ensemble classifier. Returns the top class for each of N inputs.
+ The attributes named 'nodes_X' form a sequence of tuples, associated by + index into the sequences, which must all be of equal length. These tuples + define the nodes.
+ Similarly, all fields prefixed with 'class_' are tuples of votes at the leaves. + A leaf may have multiple votes, where each vote is weighted by + the associated class_weights index.
+ One and only one of classlabels_strings or classlabels_int64s + will be defined. The class_ids are indices into this list. + All fields ending with _as_tensor can be used instead of the + same parameter without the suffix if the element type is double and not float. + +#### Version + +This version of the operator has been deprecated since version 5 of the 'ai.onnx.ml' operator set. + +### **ai.onnx.ml.TreeEnsembleRegressor-5** (deprecated) + + This operator is DEPRECATED. Please use TreeEnsemble instead which provides the same + functionality.
+ Tree Ensemble regressor. Returns the regressed values for each input in N.
+ All args with nodes_ are fields of a tuple of tree nodes, and + it is assumed they are the same length, and an index i will decode the + tuple across these inputs. Each node id can appear only once + for each tree id.
+ All fields prefixed with target_ are tuples of votes at the leaves.
+ A leaf may have multiple votes, where each vote is weighted by + the associated target_weights index.
+ All fields ending with _as_tensor can be used instead of the + same parameter without the suffix if the element type is double and not float. + All trees must have their node ids start at 0 and increment by 1.
+ Mode enum is BRANCH_LEQ, BRANCH_LT, BRANCH_GTE, BRANCH_GT, BRANCH_EQ, BRANCH_NEQ, LEAF + +#### Version + +This version of the operator has been deprecated since version 5 of the 'ai.onnx.ml' operator set. + diff --git a/docs/Changelog.md b/docs/Changelog.md new file mode 100644 index 0000000..198aa0b --- /dev/null +++ b/docs/Changelog.md @@ -0,0 +1,33668 @@ + +## Operator Changelog +*This file is automatically generated from the + [def files](/onnx/defs) via [this script](/onnx/defs/gen_doc.py). + Do not modify directly and instead edit operator definitions.* + +For an operator input/output's differentiability, it can be differentiable, + non-differentiable, or undefined. If a variable's differentiability + is not specified, that variable has undefined differentiability. + +# ai.onnx (default) +## Version 1 of the default ONNX operator set +### **Abs-1** + + Absolute takes one input data (Tensor) and produces one output data + (Tensor) where the absolute is, y = abs(x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Add-1** + + Performs element-wise binary addition (with limited broadcast support). + + If necessary the right-hand-side argument will be broadcasted to match the + shape of left-hand-side argument. When broadcasting is specified, the second + tensor can either be of element size 1 (including a scalar tensor and any + tensor with rank equal to or smaller than the first tensor), or having its + shape as a contiguous subset of the first tensor's shape. The starting of the + mutually equal shape is specified by the argument "axis", and if it is not set, + suffix matching is assumed. 1-dim expansion doesn't work yet. + + For example, the following tensor shapes are supported (with broadcast=1): + + shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor + shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor + shape(A) = (2, 3, 4, 5), shape(B) = (5,) + shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) + shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 + shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0 + + Attribute `broadcast=1` needs to be passed to enable broadcasting. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions. See doc for details.
+
broadcast : int (default is 0)
+
Pass 1 to enable broadcasting
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
A : T
+
First operand, should share the type with the second operand.
+
B : T
+
Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
+
+ +#### Outputs + +
+
C : T
+
Result, has same dimensions and type as A
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **And-1** + + Returns the tensor resulted from performing the `and` logical operation + elementwise on the input tensors `A` and `B`. + + If broadcasting is enabled, the right-hand-side argument will be broadcasted + to match the shape of left-hand-side argument. See the doc of `Add` for a + detailed description of the broadcasting rules. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions.
+
broadcast : int (default is 0)
+
Enable broadcasting
+
+ +#### Inputs + +
+
A : T
+
Left input tensor for the logical operator.
+
B : T
+
Right input tensor for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input to boolean tensor.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **ArgMax-1** + + Computes the indices of the max elements of the input tensor's element along the + provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. + If keepdims equal 0, then the resulted tensor have the reduced dimension pruned. + The type of the output tensor is integer. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
The axis in which to compute the arg indices.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : tensor(int64)
+
Reduced output tensor with integer data type.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **ArgMin-1** + + Computes the indices of the min elements of the input tensor's element along the + provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. + If keepdims equal 0, then the resulted tensor have the reduced dimension pruned. + The type of the output tensor is integer. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
The axis in which to compute the arg indices.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : tensor(int64)
+
Reduced output tensor with integer data type.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **AveragePool-1** + + AveragePool consumes an input tensor X and applies average pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + average pooling consisting of computing the average on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) + + * pad_shape[i] is sum of pads along axis i + ``` + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i] + ``` + The output of each pooling window is divided by the number of elements exclude pad. + + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **BatchNormalization-1** + + Carries out batch normalization as described in the paper + https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, + there are multiple cases for the number of outputs, which we list below: + + Output case #1: Y, mean, var, saved_mean, saved_var (training mode) + Output case #2: Y (test mode) + + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints (required)
+
legacy optimization attribute.
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero, default is 1e-5f.
+
is_test : int (default is 0)
+
If set to nonzero, run spatial batch normalization in test mode, default is 0.
+
momentum : float (default is 0.9)
+
Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum), default is 0.9f.
+
spatial : int (default is 1)
+
If true, compute the mean and variance across all spatial elements If false, compute the mean and variance across per feature.Default is 1.
+
+ +#### Inputs + +
+
X : T
+
The input 4-dimensional tensor of shape NCHW.
+
scale : T
+
The scale as a 1-dimensional tensor of size C to be applied to the output.
+
B : T
+
The bias as a 1-dimensional tensor of size C to be applied to the output.
+
mean : T
+
The running mean (training) or the estimated mean (testing) as a 1-dimensional tensor of size C.
+
var : T
+
The running variance (training) or the estimated variance (testing) as a 1-dimensional tensor of size C.
+
+ +#### Outputs (1 - 5) + +
+
Y : T
+
The output 4-dimensional tensor of the same shape as X.
+
mean (optional) : T
+
The running mean after the BatchNormalization operator. Must be in-place with the input mean. Should not be used for testing.
+
var (optional) : T
+
The running variance after the BatchNormalization operator. Must be in-place with the input var. Should not be used for testing.
+
saved_mean (optional) : T
+
Saved mean used during training to speed up gradient computation. Should not be used for testing.
+
saved_var (optional) : T
+
Saved variance used during training to speed up gradient computation. Should not be used for testing.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Cast-1** + + The operator casts the elements of a given input tensor to a data type + specified by the 'to' argument and returns an output tensor of the same size in + the converted type. The 'to' argument must be one of the data types specified + in the 'DataType' enum field in the TensorProto message. + NOTE: Casting to and from strings is not supported yet. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
to : string (required)
+
The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
+
+ +#### Inputs + +
+
input : T1
+
Input tensor to be cast.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor with the same shape as input with type specified by the 'to' argument
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool)
+
Constrain input types. Casting from strings and complex are not supported.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool)
+
Constrain output types. Casting to strings and complex are not supported.
+
+ +### **Ceil-1** + + Ceil takes one input data (Tensor) and produces one output data + (Tensor) where the ceil is, y = ceil(x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Clip-1** + + Clip operator limits the given input within an interval. The interval is + specified with arguments 'min' and 'max'. They default to + numeric_limits::lowest() and numeric_limits::max() respectively. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
max : float
+
Maximum value, above which element is replaced by max
+
min : float
+
Minimum value, under which element is replaced by min
+
+ +#### Inputs + +
+
input : T
+
Input tensor whose elements to be clipped
+
+ +#### Outputs + +
+
output : T
+
Output tensor with clipped input elements
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Concat-1** + + Concatenate a list of tensors into a single tensor + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
Which axis to concat on. Default value is 1.
+
+ +#### Inputs (1 - ∞) + +
+
inputs (variadic) : T
+
List of tensors for concatenation
+
+ +#### Outputs + +
+
concat_result : T
+
Concatenated tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ +### **Constant-1** + + A constant tensor. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
value : tensor (required)
+
The value for the elements of the output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Conv-1** + + The convolution operator consumes an input tensor and a filter, and + computes the output. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input W.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X : T
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
W : T
+
The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
+
B (optional) : T
+
Optional 1D bias to be added to the convolution, has size of M.
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **ConvTranspose-1** + + The convolution transpose operator consumes an input tensor and a filter, + and computes the output. + + If the pads parameter is provided the shape of the output is calculated via the following equation: + + output_shape[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - pads[start_i] - pads[end_i] + + output_shape can also be explicitly specified in which case pads values are auto generated using these equations: + + total_padding[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - output_shape[i] + If (auto_pads != SAME_UPPER): pads[start_i] = total_padding[i]/2; pads[end_i] = total_padding[i] - (total_padding[i]/2) + Else: pads[start_i] = total_padding[i] - (total_padding[i]/2); pads[end_i] = (total_padding[i]/2). + + + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input W.
+
output_padding : list of ints
+
The zero-padding added to one side of the output. This is also called adjs/adjustment in some frameworks.
+
output_shape : list of ints
+
The shape of the output can be explicitly set which will cause pads values to be auto generated. If output_shape is specified pads values are ignored. See doc for details for equations to generate pads
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X : T
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn)
+
W : T
+
The weight tensor that will be used in the convolutions; has size (C x M/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the weight shape will be (C x M/group x k1 x k2 x ... x kn), where (k1 x k2 x ... x kn) is the dimension of the kernel. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
+
B (optional) : T
+
Optional 1D bias to be added to the convolution, has size of M.
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, pad lengths and group count. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **DepthToSpace-1** + + DepthToSpace rearranges (permutes) data from depth into blocks of spatial data. + This is the reverse transformation of SpaceToDepth. More specifically, this op outputs a copy of + the input tensor where values from the depth dimension are moved in spatial blocks to the height + and width dimensions. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
blocksize : int (required)
+
Blocks of [blocksize, blocksize] are moved.
+
+ +#### Inputs + +
+
input : T
+
Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
+
+ +#### Outputs + +
+
output : T
+
Output tensor of [N, C/(blocksize * blocksize), H * blocksize, W * blocksize].
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Div-1** + + Performs element-wise binary division (with limited broadcast support). + + If necessary the right-hand-side argument will be broadcasted to match the + shape of left-hand-side argument. When broadcasting is specified, the second + tensor can either be of element size 1 (including a scalar tensor and any + tensor with rank equal to or smaller than the first tensor), or having its + shape as a contiguous subset of the first tensor's shape. The starting of the + mutually equal shape is specified by the argument "axis", and if it is not set, + suffix matching is assumed. 1-dim expansion doesn't work yet. + + For example, the following tensor shapes are supported (with broadcast=1): + + shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor + shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor + shape(A) = (2, 3, 4, 5), shape(B) = (5,) + shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) + shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 + shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0 + + Attribute `broadcast=1` needs to be passed to enable broadcasting. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions. See doc for details.
+
broadcast : int (default is 0)
+
Pass 1 to enable broadcasting
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
A : T
+
First operand, should share the type with the second operand.
+
B : T
+
Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
+
+ +#### Outputs + +
+
C : T
+
Result, has same dimensions and type as A
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Dropout-1** + + Dropout takes one input data (Tensor) and produces two Tensor outputs, + output (Tensor) and mask (Tensor). Depending on whether it is in + test mode or not, the output Y will either be a random dropout, or a simple + copy of the input. Note that our implementation of Dropout does scaling in + the training phase, so during testing nothing needs to be done. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
is_test : int (default is 0)
+
(int, default 0) if nonzero, run dropout in test mode where the output is simply Y = X.
+
ratio : float (default is 0.5)
+
(float, default 0.5) the ratio of random dropout
+
+ +#### Inputs + +
+
data : T
+
The input data as Tensor.
+
+ +#### Outputs (1 - 2) + +
+
output : T
+
The output.
+
mask (optional) : T
+
The output mask. If is_test is nonzero, this output is not filled.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Elu-1** + + Elu takes one input data (Tensor) and produces one output data + (Tensor) where the function `f(x) = alpha * (exp(x) - 1.) for x < + 0`, `f(x) = x for x >= 0`., is applied to the tensor elementwise. + + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Coefficient of ELU default to 1.0.
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Equal-1** + + Returns the tensor resulted from performing the `equal` logical operation + elementwise on the input tensors `A` and `B`. + + If broadcasting is enabled, the right-hand-side argument will be broadcasted + to match the shape of left-hand-side argument. See the doc of `Add` for a + detailed description of the broadcasting rules. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions.
+
broadcast : int (default is 0)
+
Enable broadcasting
+
+ +#### Inputs + +
+
A : T
+
Left input tensor for the logical operator.
+
B : T
+
Right input tensor for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool), tensor(int32), tensor(int64)
+
Constrain input to integral tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Exp-1** + + Calculates the exponential of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
input : T
+
Input tensor
+
+ +#### Outputs + +
+
output : T
+
The exponential of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Flatten-1** + + Flattens the input tensor into a 2D matrix. If input tensor has shape + (d_0, d_1, ... d_n) then the output will have shape + (d_0 X d_1 ... d_(axis-1), d_axis X d_(axis+1) ... X dn). + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [0, R], where R is the rank of the input tensor. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 ... d_n), where the shape of the input tensor is (d_0, d_1, ... d_n).
+
+ +#### Inputs + +
+
input : T
+
A tensor of rank >= axis.
+
+ +#### Outputs + +
+
output : T
+
A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Floor-1** + + Floor takes one input data (Tensor) and produces one output data + (Tensor) where the floor is, y = floor(x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **GRU-1** + + Computes an one-layer GRU. This operator is usually supported via some custom + implementation such as CuDNN. + + Notations: + + `X` - input tensor + + `z` - update gate + + `r` - reset gate + + `h` - hidden gate + + `t` - time step (t-1 means previous time step) + + `W[zrh]` - W parameter weight matrix for update, reset, and hidden gates + + `R[zrh]` - R recurrence weight matrix for update, reset, and hidden gates + + `Wb[zrh]` - W bias vectors for update, reset, and hidden gates + + `Rb[zrh]` - R bias vectors for update, reset, and hidden gates + + `WB[zrh]` - W parameter weight matrix for backward update, reset, and hidden gates + + `RB[zrh]` - R recurrence weight matrix for backward update, reset, and hidden gates + + `WBb[zrh]` - W bias vectors for backward update, reset, and hidden gates + + `RBb[zrh]` - R bias vectors for backward update, reset, and hidden gates + + `H` - Hidden state + + `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + Relu(x) - max(0, x) + + Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + + Sigmoid(x) - 1/(1 + e^{-x}) + + (NOTE: Below are optional) + + Affine(x) - alpha*x + beta + + LeakyRelu(x) - x if x >= 0 else alpha * x + + ThresholdedRelu(x) - x if x >= alpha else 0 + + ScaledTanh(x) - alpha*Tanh(beta*x) + + HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + + Elu(x) - x if x >= 0 else alpha*(e^x - 1) + + Softsign(x) - x/(1 + |x|) + + Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh): + + - zt = f(Xt*(Wz^T) + Ht-1*Rz + Wbz + Rbz) + + - rt = f(Xt*(Wr^T) + Ht-1*Rr + Wbr + Rbr) + + - ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*Rh + Rbh + Wbh) # default, when linear_before_reset = 0 + + - ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*Rh + Rbh) + Wbh) # when linear_before_reset != 0 + + - Ht = (1 - zt) (.) ht + zt (.) Ht-1 + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM.
+
activations : list of strings
+
A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is foward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
output_sequence : int (default is 0)
+
The sequence output for the hidden is optional if 0. Default 0.
+
+ +#### Inputs (3 - 6) + +
+
X : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W : T
+
The weight tensor for the gates. Concatenation of `W[zrh]` and `WB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, input_size]`.
+
R : T
+
The recurrence weight tensor. Concatenation of `R[zrh]` and `RB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, hidden_size]`.
+
B (optional) : T
+
The bias tensor for the gates. Concatenation of `[Wb[zrh], Rb[zrh]]` and `[WBb[zrh], RBb[zrh]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 6*hidden_size]`. Optional: If not specified - assumed to be 0
+
sequence_lens (optional) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs + +
+
Y (optional) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`. It is optional if `output_sequence` is 0.
+
Y_h : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **Gather-1** + + Given `data` tensor of rank r >= 1, and `indices` tensor of rank q, gather + entries of the axis dimension of `data` (by default outer-most one as axis=0) indexed by `indices`, and concatenates + them in an output tensor of rank q + (r - 1). + Example 1: + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + indices = [ + [0, 1], + [1, 2], + ] + output = [ + [ + [1.0, 1.2], + [2.3, 3.4], + ], + [ + [2.3, 3.4], + [4.5, 5.7], + ], + ] + ``` + Example 2: + ``` + data = [ + [1.0, 1.2, 1.9], + [2.3, 3.4, 3.9], + [4.5, 5.7, 5.9], + ] + indices = [ + [0, 2], + ] + axis = 1, + output = [ + [[1.0, 1.9]], + [[2.3, 3.9]], + [[4.5, 5.9]], + ] + ``` + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1]
+
+ +#### Inputs + +
+
data : T
+
Tensor of rank r >= 1.
+
indices : Tind
+
Tensor of int32/int64 indices, of any rank q. All index values are expected to be within bounds. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output : T
+
Tensor of rank q + (r - 1).
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **Gemm-1** + + General Matrix multiplication: + https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3 + Compute Y = alpha * A * B + beta * C, where input tensor A has + dimension (M X K), input tensor B has dimension (K X N), input tensor C and + output tensor Y have dimension (M X N). + If attribute broadcast is non-zero, input tensor C will be broadcasted to match + the dimension requirement. A will be transposed before doing the computation + if attribute transA is non-zero, same for B and transB. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Scalar multiplier for the product of input tensors A * B, the default value is 1.0.
+
beta : float (default is 1.0)
+
Scalar multiplier for input tensor C, the default value is 1.0.
+
broadcast : int (default is 0)
+
Whether C should be broadcasted
+
transA : int (default is 0)
+
Whether A should be transposed
+
transB : int (default is 0)
+
Whether B should be transposed
+
+ +#### Inputs + +
+
A : T
+
Input tensor A
+
B : T
+
Input tensor B
+
C : T
+
Input tensor C, can be inplace.
+
+ +#### Outputs + +
+
Y : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **GlobalAveragePool-1** + + GlobalAveragePool consumes an input tensor X and applies average pooling across + the values in the same channel. This is equivalent to AveragePool with kernel size + equal to the spatial dimension of input tensor. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **GlobalLpPool-1** + + GlobalLpPool consumes an input tensor X and applies lp pool pooling across the + the values in the same channel. This is equivalent to LpPool with kernel size + equal to the spatial dimension of input tensor. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
p : float (default is 2.0)
+
p value of the Lp norm used to pool over the input data, default is 2.0.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimension are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor from pooling across the input tensor. Dimensions will be N x C x 1 x 1
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **GlobalMaxPool-1** + + GlobalMaxPool consumes an input tensor X and applies max pooling across + the values in the same channel. This is equivalent to MaxPool with kernel size + equal to the spatial dimension of input tensor. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Greater-1** + + Returns the tensor resulted from performing the `greater` logical operation + elementwise on the input tensors `A` and `B`. + + If broadcasting is enabled, the right-hand-side argument will be broadcasted + to match the shape of left-hand-side argument. See the doc of `Add` for a + detailed description of the broadcasting rules. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions.
+
broadcast : int (default is 0)
+
Enable broadcasting
+
+ +#### Inputs + +
+
A : T
+
Left input tensor for the logical operator.
+
B : T
+
Right input tensor for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input to float tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **HardSigmoid-1** + + HardSigmoid takes one input data (Tensor) and produces one output data + (Tensor) where the HardSigmoid function, y = max(0, min(1, alpha * x + beta)), + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 0.2)
+
Value of alpha default to 0.2
+
beta : float (default is 0.5)
+
Value of beta default to 0.5
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Hardmax-1** + + The operator computes the hardmax (1 for the first maximum value, and 0 for all others) values for each layer in the batch + of the given input. The input is a 2-D tensor (Tensor) of size + (batch_size x input_feature_dimensions). The output tensor has the same shape + and contains the hardmax values of the corresponding input. + + Input does not need to explicitly be a 2D vector; rather, it will be + coerced into one. For an arbitrary n-dimensional tensor + input \in [a_0, a_1, ..., a_{k-1}, a_k, ..., a_{n-1}] and k is + the axis provided, then input will be coerced into a 2-dimensional tensor with + dimensions [a_0 * ... * a_{k-1}, a_k * ... * a_{n-1}]. For the default + case where axis=1, this means the input tensor will be coerced into a 2D tensor + of dimensions [a_0, a_1 * ... * a_{n-1}], where a_0 is often the batch size. + In this situation, we must have a_0 = N and a_1 * ... * a_{n-1} = D. + Each of these dimensions must be matched correctly, or else the operator + will throw errors. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size
+
+ +#### Inputs + +
+
input : T
+
The input tensor that's coerced into a 2D matrix of size (NxD) as described above.
+
+ +#### Outputs + +
+
output : T
+
The output values with the same shape as input tensor (the original size without coercion).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Identity-1** + + Identity operator + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
input : T
+
Input tensor
+
+ +#### Outputs + +
+
output : T
+
Tensor to copy input into.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **If-1** + + If conditional + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
else_branch : graph (required)
+
Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
+
then_branch : graph (required)
+
Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
+
+ +#### Inputs + +
+
cond : B
+
Condition for the if. The tensor must contain a single element.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : V
+
Values that are live-out to the enclosing scope. The return values in the `then_branch` and `else_branch` must be of the same shape and same data type.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
All Tensor types
+
B : tensor(bool)
+
Only bool
+
+ +### **InstanceNormalization-1** + + Carries out instance normalization as described in the paper + https://arxiv.org/abs/1607.08022. + + y = scale * (x - mean) / sqrt(variance + epsilon) + B, + where mean and variance are computed per instance per channel. + + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero, default is 1e-5f.
+
+ +#### Inputs + +
+
input : T
+
The input 4-dimensional tensor of shape NCHW.
+
scale : T
+
The input 1-dimensional scale tensor of size C.
+
B : T
+
The input 1-dimensional bias tensor of size C.
+
+ +#### Outputs + +
+
output : T
+
The output 4-dimensional tensor of the same shape as input.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **LRN-1** + + Local Response Normalization proposed in the [AlexNet paper](https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf). + It normalizes over local input regions. + The local region is defined across the channels. For an element X[n, c, d1, ..., dk] in a tensor + of shape (N x C x D1 x D2, ..., Dk), its region is + {X[n, i, d1, ..., dk] | max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2))}. + + square_sum[n, c, d1, ..., dk] = sum(X[n, i, d1, ..., dk] ^ 2), + where max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2)). + + Y[n, c, d1, ..., dk] = X[n, c, d1, ..., dk] / (bias + alpha / size * square_sum[n, c, d1, ..., dk] ) ^ beta + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 0.0001)
+
Scaling parameter.
+
beta : float (default is 0.75)
+
The exponent.
+
bias : float (default is 1.0)
+
+
size : int (required)
+
The number of channels to sum over
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y : T
+
Output tensor, which has the shape and type as input tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **LSTM-1** + + Computes an one-layer LSTM. This operator is usually supported via some + custom implementation such as CuDNN. + + Notations: + + `X` - input tensor + + `i` - input gate + + `o` - output gate + + `f` - forget gate + + `c` - cell gate + + `t` - time step (t-1 means previous time step) + + `W[iofc]` - W parameter weight matrix for input, output, forget, and cell gates + + `R[iofc]` - R recurrence weight matrix for input, output, forget, and cell gates + + `Wb[iofc]` - W bias vectors for input, output, forget, and cell gates + + `Rb[iofc]` - R bias vectors for input, output, forget, and cell gates + + `P[iof]` - P peephole weight vector for input, output, and forget gates + + `WB[iofc]` - W parameter weight matrix for backward input, output, forget, and cell gates + + `RB[iofc]` - R recurrence weight matrix for backward input, output, forget, and cell gates + + `WBb[iofc]` - W bias vectors for backward input, output, forget, and cell gates + + `RBb[iofc]` - R bias vectors for backward input, output, forget, and cell gates + + `PB[iof]` - P peephole weight vector for backward input, output, and forget gates + + `H` - Hidden state + + `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + Relu(x) - max(0, x) + + Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + + Sigmoid(x) - 1/(1 + e^{-x}) + + (NOTE: Below are optional) + + Affine(x) - alpha*x + beta + + LeakyRelu(x) - x if x >= 0 else alpha * x + + ThresholdedRelu(x) - x if x >= alpha else 0 + + ScaledTanh(x) - alpha*Tanh(beta*x) + + HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + + Elu(x) - x if x >= 0 else alpha*(e^x - 1) + + Softsign(x) - x/(1 + |x|) + + Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh, h=Tanh): + + - it = f(Xt*(Wi^T) + Ht-1*Ri + Pi (.) Ct-1 + Wbi + Rbi) + + - ft = f(Xt*(Wf^T) + Ht-1*Rf + Pf (.) Ct-1 + Wbf + Rbf) + + - ct = g(Xt*(Wc^T) + Ht-1*Rc + Wbc + Rbc) + + - Ct = ft (.) Ct-1 + it (.) ct + + - ot = f(Xt*(Wo^T) + Ht-1*Ro + Po (.) Ct + Wbo + Rbo) + + - Ht = ot (.) h(Ct) + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings
+
A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
input_forget : int (default is 0)
+
Couple the input and forget gates if 1, default 0.
+
output_sequence : int (default is 0)
+
The sequence output for the hidden is optional if 0. Default 0.
+
+ +#### Inputs (3 - 8) + +
+
X : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W : T
+
The weight tensor for the gates. Concatenation of `W[iofc]` and `WB[iofc]` (if bidirectional) along dimension 0. The tensor has shape `[num_directions, 4*hidden_size, input_size]`.
+
R : T
+
The recurrence weight tensor. Concatenation of `R[iofc]` and `RB[iofc]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 4*hidden_size, hidden_size]`.
+
B (optional) : T
+
The bias tensor for input gate. Concatenation of `[Wb[iofc], Rb[iofc]]`, and `[WBb[iofc], RBb[iofc]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 8*hidden_size]`. Optional: If not specified - assumed to be 0.
+
sequence_lens (optional) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
initial_c (optional) : T
+
Optional initial value of the cell. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
P (optional) : T
+
The weight tensor for peepholes. Concatenation of `P[iof]` and `PB[iof]` (if bidirectional) along dimension 0. It has shape `[num_directions, 3*hidde_size]`. Optional: If not specified - assumed to be 0.
+
+ +#### Outputs (0 - 3) + +
+
Y (optional) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`. It is optional if `output_sequence` is 0.
+
Y_h (optional) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
Y_c (optional) : T
+
The last output value of the cell. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **LeakyRelu-1** + + LeakyRelu takes input data (Tensor) and an argument alpha, and produces one + output data (Tensor) where the function `f(x) = alpha * x for x < 0`, + `f(x) = x for x >= 0`, is applied to the data tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 0.01)
+
Coefficient of leakage default to 0.01.
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Less-1** + + Returns the tensor resulted from performing the `less` logical operation + elementwise on the input tensors `A` and `B`. + + If broadcasting is enabled, the right-hand-side argument will be broadcasted + to match the shape of left-hand-side argument. See the doc of `Add` for a + detailed description of the broadcasting rules. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions.
+
broadcast : int (default is 0)
+
Enable broadcasting
+
+ +#### Inputs + +
+
A : T
+
Left input tensor for the logical operator.
+
B : T
+
Right input tensor for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input to float tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Log-1** + + Calculates the natural log of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
input : T
+
Input tensor
+
+ +#### Outputs + +
+
output : T
+
The natural log of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **LogSoftmax-1** + + The operator computes the logsoftmax (log of softmax) values for each layer in the batch + of the given input. The input is a 2-D tensor (Tensor) of size + (batch_size x input_feature_dimensions). The output tensor has the same shape + and contains the logsoftmax values of the corresponding input. + + Input does not need to explicitly be a 2D vector; rather, it will be + coerced into one. For an arbitrary n-dimensional tensor + input \in [a_0, a_1, ..., a_{k-1}, a_k, ..., a_{n-1}] and k is + the axis provided, then input will be coerced into a 2-dimensional tensor with + dimensions [a_0 * ... * a_{k-1}, a_k * ... * a_{n-1}]. For the default + case where axis=1, this means the input tensor will be coerced into a 2D tensor + of dimensions [a_0, a_1 * ... * a_{n-1}], where a_0 is often the batch size. + In this situation, we must have a_0 = N and a_1 * ... * a_{n-1} = D. + Each of these dimensions must be matched correctly, or else the operator + will throw errors. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size
+
+ +#### Inputs + +
+
input : T
+
The input tensor that's coerced into a 2D matrix of size (NxD) as described above.
+
+ +#### Outputs + +
+
output : T
+
The output values with the same shape as input tensor (the original size without coercion).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Loop-1** + + Generic Looping construct. This loop has multiple termination conditions: + + 1) Trip count. Iteration count specified at runtime. Set by + specifying the input M. Optional. Set to empty string to omit. + Note that a static trip count (specified at graph construction time) can be + specified by passing in a constant node for input M. + 2) Loop termination condition. This is an input to the op that determines + whether to run the first iteration and also a loop-carried dependency for + the body graph. The body graph must yield a value for the condition variable, + whether this input is provided or not. + + This table summarizes the operating modes of this operator with equivalent + C-style code: + + Operator inputs defined as (max_trip_count, condition_var). + + input ("", ""): + for (int i=0; ; ++i) { + cond = ... // Note this value is ignored, but is required in the body + } + + input ("", cond) // Note this is analogous to a while loop + bool cond = ...; + for (int i=0; cond; ++i) { + cond = ...; + } + + input ("", 1) // Note this is analogous to a do-while loop + bool cond = true + for (int i=0; cond; ++i) { + cond = ...; + } + + input (trip_count, "") // Note this is analogous to a for loop + int trip_count = ... + for (int i=0; i < trip_count; ++i) { + cond = ...; // ignored + } + + input (trip_count, cond) + int trip_count = ...; + bool cond = ...; + for (int i=0; i < trip_count && cond; ++i) { + cond = ...; + } + + + *Sample usage - cond as well as trip count* + + graph predict-net { + %a = Constant[value = ]() + %b = Constant[value = ]() + %keepgoing = Constant[value = ]() + %max_trip_count = Constant[value = ]() + %keepgoing_out, %b_out, %user_defined_vals = Loop[body = ](%max_trip_count, %keepgoing, %b) + return + } + + graph body-net ( + %i[INT32, scalar] + %keepgoing[BOOL, scalar] + %b[INT32, scalar] + ) { + %my_local = Add(%a, %b) + %b_out = Sub(%a, %b) + %keepgoing_out = Greater(%my_local, %b_out) + %user_defined_vals = Add(%b, %b) + return %keepgoing_out, %b_out, %user_defined_vals + } + + *Sample equivalent C code* + + { + /* User-defined code (enclosing scope) */ + int a = 3, b = 6; + bool keepgoing = true; // Analogous to input cond + /* End user-defined code */ + + /* Implicitly-defined code */ + const int max_trip_count = 10; // Analogous to input M + int user_defined_vals[]; // Imagine this is resizable + /* End implicitly-defined code */ + for (int i=0; i < max_trip_count && keepgoing; ++i) { + /* User-defined code (loop body) */ + int my_local = a + b; // Reading values in the enclosing scope is fine + b = a - b; // writes fine if we specify b as a loop-carried dependency + keepgoing = my_local > b; // keepgoing is a loop-carried dependency + user_defined_vals[i] = b + b; + /* End user-defined code */ + } + // my_local = 123; // Can't do this. my_local was defined in the body + + // These below values are live-out from the loop and therefore accessible + b_out; user_defined_vals; keepgoing_out; + } + + There are several things of note in this code snippet: + + 1) Values from the enclosing scope (i.e. variable a here) are in scope and can + be referenced in the inputs of the loop. + 2) Any variables which you wish to make available in the enclosing scope (i.e. + the variables b and keepgoing) must be declared as either loop-carried + dependencies (both at the op inputs and output and at the body net input and + output) or scan_outputs. + 3) Values created in the body cannot be accessed in the enclosing scope. + + Note that the semantics of this op support "diagonal" or "wavefront" execution. + (See Step 3 here for an example: + https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). + Frontends should emit multi-layer RNNs as a series of While operators (with + time being the inner looping dimension), with each successive layer consuming + the scan_outputs from the previous layer, possibly going through several + point-wise operators (e.g. dropout, residual connections, linear layer). + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies...). It has 1+N+K outputs: (condition, loop carried dependencies..., scan_outputs...). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
+
+ +#### Inputs (3 - ∞) + +
+
M (optional) : I
+
A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
+
cond (optional) : B
+
A boolean termination condition. Optional. Pass empty string to skip.
+
v_initial (variadic, heterogeneous) : V
+
The initial values of any loop-carried dependencies (values that change across loop iterations)
+
+ +#### Outputs (1 - ∞) + +
+
v_final_and_scan_outputs (variadic, heterogeneous) : V
+
Final N loop carried dependency values then K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
All Tensor types
+
I : tensor(int64)
+
tensor of int64, which should be a scalar.
+
B : tensor(bool)
+
tensor of bool, which should be a scalar.
+
+ +### **LpNormalization-1** + + Given a matrix, apply Lp-normalization along the provided axis. + The output is computed as: `output = input / Lp_norm(input, axis)`. + When the Lp norm is zero (i.e., all elements along the axis are zero), + the output is defined to be zero to avoid division by zero. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
The axis on which to apply normalization, -1 mean last axis.
+
p : int (default is 2)
+
The order of the normalization, only 1 or 2 are supported.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input matrix
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Matrix after normalization
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **LpPool-1** + + LpPool consumes an input tensor X and applies Lp pooling across the + the tensor according to kernel sizes, stride sizes, and pad lengths. + Lp pooling consisting of computing the Lp norm on all values of a subset + of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.
+
kernel_shape : list of ints
+
The size of the kernel along each axis.
+
p : float (default is 2.0)
+
p value of the Lp norm used to pool over the input data, default is 2.0.
+
pads : list of ints
+
Padding for the beginning and ending along each axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute.
+
strides : list of ints
+
Stride along each axis.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimension are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **MatMul-1** + + Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html). + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
N-dimensional matrix A
+
B : T
+
N-dimensional matrix B
+
+ +#### Outputs + +
+
Y : T
+
Matrix multiply results from A * B
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Max-1** + + Element-wise max of each of the input tensors. All inputs and outputs must + have the same shape and data type. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for Max.
+
+ +#### Outputs + +
+
max : T
+
Output tensor. Same dimension as inputs.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **MaxPool-1** + + MaxPool consumes an input tensor X and applies max pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + max pooling consisting of computing the max on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) + + * pad_shape[i] is sum of pads along axis i + ``` + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i] + ``` + The output of each pooling window is maximum number of elements exclude pad. + + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **MaxRoiPool-1** + + ROI max pool consumes an input tensor X and region of interests (RoIs) to + apply max pooling across each RoI, to produce output 4-D tensor of shape + (num_rois, channels, pooled_shape[0], pooled_shape[1]). + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
pooled_shape : list of ints (required)
+
ROI pool output shape (height, width).
+
spatial_scale : float (default is 1.0)
+
Multiplicative spatial scale factor to translate ROI coordinates from their input scale to the scale used when pooling.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
+
rois (non-differentiable) : T
+
RoIs (Regions of Interest) to pool over. Should be a 2-D tensor of shape (num_rois, 5) given as [[batch_id, x1, y1, x2, y2], ...].
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
RoI pooled output 4-D tensor of shape (num_rois, channels, pooled_shape[0], pooled_shape[1]).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Mean-1** + + Element-wise mean of each of the input tensors. All inputs and outputs must + have the same shape and data type. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for Mean.
+
+ +#### Outputs + +
+
mean : T
+
Output tensor. Same dimension as inputs.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Min-1** + + Element-wise min of each of the input tensors. All inputs and outputs must + have the same shape and data type. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for Min
+
+ +#### Outputs + +
+
min : T
+
Output tensor. Same dimension as inputs.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Mul-1** + + Performs element-wise binary multiplication (with limited broadcast support). + + If necessary the right-hand-side argument will be broadcasted to match the + shape of left-hand-side argument. When broadcasting is specified, the second + tensor can either be of element size 1 (including a scalar tensor and any + tensor with rank equal to or smaller than the first tensor), or having its + shape as a contiguous subset of the first tensor's shape. The starting of the + mutually equal shape is specified by the argument "axis", and if it is not set, + suffix matching is assumed. 1-dim expansion doesn't work yet. + + For example, the following tensor shapes are supported (with broadcast=1): + + shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor + shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor + shape(A) = (2, 3, 4, 5), shape(B) = (5,) + shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) + shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 + shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0 + + Attribute `broadcast=1` needs to be passed to enable broadcasting. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions. See doc for details.
+
broadcast : int (default is 0)
+
Pass 1 to enable broadcasting
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
A : T
+
First operand, should share the type with the second operand.
+
B : T
+
Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
+
+ +#### Outputs + +
+
C : T
+
Result, has same dimensions and type as A
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Neg-1** + + Neg takes one input data (Tensor) and produces one output data + (Tensor) where each element flipped sign, y = -x, is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Not-1** + + Returns the negation of the input tensor element-wise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input/output to boolean tensors.
+
+ +### **Or-1** + + Returns the tensor resulted from performing the `or` logical operation + elementwise on the input tensors `A` and `B`. + + If broadcasting is enabled, the right-hand-side argument will be broadcasted + to match the shape of left-hand-side argument. See the doc of `Add` for a + detailed description of the broadcasting rules. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions.
+
broadcast : int (default is 0)
+
Enable broadcasting
+
+ +#### Inputs + +
+
A : T
+
Left input tensor for the logical operator.
+
B : T
+
Right input tensor for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input to boolean tensor.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **PRelu-1** + + PRelu takes input data (Tensor) and slope tensor as input, and produces one + output data (Tensor) where the function `f(x) = slope * x for x < 0`, + `f(x) = x for x >= 0`., is applied to the data tensor elementwise. + + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
slope : T
+
Slope tensor. If `Slope` is of size 1, the value is sharedacross different channels
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Pad-1** + + Given `data` tensor, paddings, mode, and value. + Example: + Insert 0 paddings to the beginning of the second dimension. + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + paddings = [0, 0, 2, 0] + output = [ + [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ], + ] + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is constant)
+
Three modes: constant(default), reflect, edge
+
paddings : list of ints (required)
+
List of integers indicate the padding element count at the beginning and end of each axis, for 2D it is the number of pixel. `paddings` rank should be double of the input's rank. `paddings` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`.
+
value : float (default is 0.0)
+
One float, indicates the value to be filled, default is 0
+
+ +#### Inputs + +
+
data : T
+
Input tensor.
+
+ +#### Outputs + +
+
output : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Pow-1** + + Pow takes input data (Tensor) and exponent Tensor, and + produces one output data (Tensor) where the function `f(x) = x^exponent`, + is applied to the data tensor elementwise. + + If necessary the right-hand-side argument will be broadcasted to match the + shape of left-hand-side argument. When broadcasting is specified, the second + tensor can either be of element size 1 (including a scalar tensor and any + tensor with rank equal to or smaller than the first tensor), or having its + shape as a contiguous subset of the first tensor's shape. The starting of the + mutually equal shape is specified by the argument "axis", and if it is not set, + suffix matching is assumed. 1-dim expansion doesn't work yet. + + For example, the following tensor shapes are supported (with broadcast=1): + + shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor + shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor + shape(A) = (2, 3, 4, 5), shape(B) = (5,) + shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) + shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 + shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0 + + Attribute `broadcast=1` needs to be passed to enable broadcasting. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions. See doc for details.
+
broadcast : int (default is 0)
+
Pass 1 to enable broadcasting
+
+ +#### Inputs + +
+
X : T
+
Input tensor of any shape, base of the exponent.
+
Y : T
+
Input tensor of any shape broadcastable to X shape, the exponent component.
+
+ +#### Outputs + +
+
Z : T
+
Output tensor (same size as X)
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **RNN-1** + + Computes an one-layer simple RNN. This operator is usually supported + via some custom implementation such as CuDNN. + + Notations: + + `X` - input tensor + + `i` - input gate + + `t` - time step (t-1 means previous time step) + + `Wi` - W parameter weight matrix for input gate + + `Ri` - R recurrence weight matrix for input gate + + `Wbi` - W parameter bias vector for input gate + + `Rbi` - R parameter bias vector for input gate + + `WBi` - W parameter weight matrix for backward input gate + + `RBi` - R recurrence weight matrix for backward input gate + + `WBbi` - WR bias vectors for backward input gate + + `RBbi` - RR bias vectors for backward input gate + + `H` - Hidden state + + `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + Relu(x) - max(0, x) + + Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + + Sigmoid(x) - 1/(1 + e^{-x}) + + (NOTE: Below are optional) + + Affine(x) - alpha*x + beta + + LeakyRelu(x) - x if x >= 0 else alpha * x + + ThresholdedRelu(x) - x if x >= alpha else 0 + + ScaledTanh(x) - alpha*Tanh(beta*x) + + HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + + Elu(x) - x if x >= 0 else alpha*(e^x - 1) + + Softsign(x) - x/(1 + |x|) + + Softplus(x) - log(1 + e^x) + + Equations (Default: f=Tanh): + + - Ht = f(Xt*(Wi^T) + Ht-1*Ri + Wbi + Rbi) + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings (default is ['Tanh', 'Tanh'])
+
One (or two if bidirectional) activation function for input gate. The activation function must be one of the activation functions specified above. Optional: Default `Tanh` if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
output_sequence : int (default is 0)
+
The sequence output for the hidden is optional if 0. Default 0.
+
+ +#### Inputs (3 - 6) + +
+
X : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W : T
+
The weight tensor for input gate. Concatenation of `Wi` and `WBi` (if bidirectional). The tensor has shape `[num_directions, hidden_size, input_size]`.
+
R : T
+
The recurrence weight tensor. Concatenation of `Ri` and `RBi` (if bidirectional). The tensor has shape `[num_directions, hidden_size, hidden_size]`.
+
B (optional) : T
+
The bias tensor for input gate. Concatenation of `[Wbi, Rbi]` and `[WBbi, RBbi]` (if bidirectional). The tensor has shape `[num_directions, 2*hidden_size]`. Optional: If not specified - assumed to be 0.
+
sequence_lens (optional) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs (0 - 2) + +
+
Y (optional) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`. It is optional if `output_sequence` is 0.
+
Y_h (optional) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **RandomNormal-1** + + Generate a tensor with random values drawn from a normal distribution. The shape + of the tensor is specified by the `shape` argument and the parameter of the normal distribution + specified by `mean` and `scale`. + + The data type is specified by the 'dtype' argument. The 'dtype' argument must + be one of the data types specified in the 'DataType' enum field in the + TensorProto message. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int (default is 1)
+
The data type for the elements of the output tensor. Default is TensorProto::FLOAT.
+
mean : float (default is 0.0)
+
The mean of the normal distribution.
+
scale : float (default is 1.0)
+
The standard deviation of the normal distribution.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
shape : list of ints (required)
+
The shape of the output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor of random values drawn from normal distribution
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ +### **RandomNormalLike-1** + + Generate a tensor with random values drawn from a normal distribution. + The shape of the output tensor is copied from the shape of the input tensor, + and the parameters of the normal distribution are specified by `mean` and `scale`. + + The data type is specified by the 'dtype' argument, or copied from the input tensor if not provided. + The 'dtype' argument must be one of the data types specified in the 'DataType' enum field in the + TensorProto message, and be valid as an output type. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int
+
(Optional) The data type for the elements of the output tensor, if not specified, we will use the data type of the input tensor.
+
mean : float (default is 0.0)
+
The mean of the normal distribution.
+
scale : float (default is 1.0)
+
The standard deviation of the normal distribution.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
Input tensor to copy shape and optionally type information from.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of random values drawn from normal distribution
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type. If the dtype attribute is not provided this must be a valid output type.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ +### **RandomUniform-1** + + Generate a tensor with random values drawn from a uniform distribution. The shape + of the tensor is specified by the `shape` argument and the range by `low` and `high`. + + The data type is specified by the 'dtype' argument. The 'dtype' argument must + be one of the data types specified in the 'DataType' enum field in the + TensorProto message. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int (default is 1)
+
The data type for the elements of the output tensor. If not specified, default is TensorProto::FLOAT.
+
high : float (default is 1.0)
+
Upper boundary of the output values.
+
low : float (default is 0.0)
+
Lower boundary of the output values.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
shape : list of ints (required)
+
The shape of the output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor of random values drawn from uniform distribution
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ +### **RandomUniformLike-1** + + Generate a tensor with random values drawn from a uniform distribution. + The shape of the output tensor is copied from the shape of the input tensor, + and the parameters of the uniform distribution are specified by `low` and `high`. + + The data type is specified by the 'dtype' argument, or copied from the input tensor if not provided. + The 'dtype' argument must be one of the data types specified in the 'DataType' enum field in the + TensorProto message and be valid as an output type. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int
+
(Optional) The data type for the elements of the output tensor, if not specified, we will use the data type of the input tensor.
+
high : float (default is 1.0)
+
Upper boundary of the output values.
+
low : float (default is 0.0)
+
Lower boundary of the output values.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
Input tensor to copy shape and optionally type information from.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of random values drawn from uniform distribution
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type. If the dtype attribute is not provided this must be a valid output type.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ +### **Reciprocal-1** + + Reciprocal takes one input data (Tensor) and produces one output data + (Tensor) where the reciprocal is, y = 1/x, is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **ReduceL1-1** + + Computes the L1 norm of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceL2-1** + + Computes the L2 norm of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceLogSum-1** + + Computes the log sum of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or undefined otherwise. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceLogSumExp-1** + + Computes the log sum exponent of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or undefined otherwise. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceMax-1** + + Computes the max of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or the minimum value of the data type otherwise. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceMean-1** + + Computes the mean of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields undefined. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceMin-1** + + Computes the min of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields plus infinity (if supported by the datatype) or the maximum value of the data type otherwise. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceProd-1** + + Computes the product of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 1. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceSum-1** + + Computes the sum of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceSumSquare-1** + + Computes the sum square of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Relu-1** + + Relu takes one input data (Tensor) and produces one output data + (Tensor) where the rectified linear function, y = max(0, x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Reshape-1** + + Reshape the input tensor similar to numpy.reshape. + It takes a tensor as input and an argument `shape`. It outputs the reshaped tensor. + At most one dimension of the new shape can be -1. In this case, the value is + inferred from the size of the tensor and the remaining dimensions. A dimension + could also be 0, in which case the actual dimension value is unchanged (i.e. taken + from the input tensor). Shape (second input) could be an empty shape, which means converting to a scalar. + The input tensor's shape and the output tensor's shape are required to have the same number of elements. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
shape : list of ints
+
New shape
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reshaped : T
+
Reshaped data.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Selu-1** + + Selu takes one input data (Tensor) and produces one output data + (Tensor) where the scaled exponential linear unit function, + `y = gamma * (alpha * e^x - alpha) for x <= 0`, `y = gamma * x for x > 0`, + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.6732)
+
Coefficient of SELU default to 1.6732.
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
gamma : float (default is 1.0507)
+
Coefficient of SELU default to 1.0507.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Shape-1** + + Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
shape : T1
+
Shape of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor.
+
+ +### **Sigmoid-1** + + Sigmoid takes one input data (Tensor) and produces one output data + (Tensor) where the sigmoid function, y = 1 / (1 + exp(-x)), is applied to the + tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Size-1** + + Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
size : T1
+
Total number of elements of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor, which should be a scalar though.
+
+ +### **Slice-1** + + Produces a slice of the input tensor along multiple axes. Similar to numpy: + https://numpy.org/doc/stable/reference/routines.indexing.html + Slices uses `axes`, `starts` and `ends` attributes to specify the start and end + dimension for each axis in the list of axes, it uses this information to + slice the input `data` tensor. If a negative value is passed for any of the + start or end indices, it represent number of elements before the end of that + dimension. If the value passed to start or end is larger than the `n` (the + number of elements in this dimension), it represents `n`. For slicing to the + end of a dimension with unknown size, it is recommended to pass in `INT_MAX`. + If `axes` are omitted, they are set to `[0, ..., ndim-1]`. + Example 1: + data = [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + axes = [0, 1] + starts = [1, 0] + ends = [2, 3] + result = [ + [5, 6, 7], + ] + Example 2: + data = [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + starts = [0, 1] + ends = [-1, 1000] + result = [ + [2, 3, 4], + ] + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
Axes that `starts` and `ends` apply to. It's optional. If not present, will be treated as [0, 1, ..., len(`starts`) - 1].
+
ends : list of ints (required)
+
Ending indices (exclusive) of corresponding axis in axes`
+
starts : list of ints (required)
+
Starting indices of corresponding axis in `axes`
+
+ +#### Inputs + +
+
data : T
+
Tensor of data to extract slices from.
+
+ +#### Outputs + +
+
output : T
+
Sliced data tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Softmax-1** + + The operator computes the softmax (normalized exponential) values for each layer in the batch + of the given input. The input is a 2-D tensor (Tensor) of size + (batch_size x input_feature_dimensions). The output tensor has the same shape + and contains the softmax values of the corresponding input. + + Input does not need to explicitly be a 2D vector; rather, it will be + coerced into one. For an arbitrary n-dimensional tensor + input \in [a_0, a_1, ..., a_{k-1}, a_k, ..., a_{n-1}] and k is + the axis provided, then input will be coerced into a 2-dimensional tensor with + dimensions [a_0 * ... * a_{k-1}, a_k * ... * a_{n-1}]. For the default + case where axis=1, this means the input tensor will be coerced into a 2D tensor + of dimensions [a_0, a_1 * ... * a_{n-1}], where a_0 is often the batch size. + In this situation, we must have a_0 = N and a_1 * ... * a_{n-1} = D. + Each of these dimensions must be matched correctly, or else the operator + will throw errors. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size
+
+ +#### Inputs + +
+
input : T
+
The input tensor that's coerced into a 2D matrix of size (NxD) as described above.
+
+ +#### Outputs + +
+
output : T
+
The output values with the same shape as input tensor (the original size without coercion).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Softplus-1** + + Softplus takes one input data (Tensor) and produces one output data + (Tensor) where the softplus function, y = ln(exp(x) + 1), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Softsign-1** + + Calculates the softsign (x/(1+|x|)) of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The softsign (x/(1+|x|)) values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **SpaceToDepth-1** + + SpaceToDepth rearranges blocks of spatial data into depth. More specifically, + this op outputs a copy of the input tensor where values from the height and width dimensions + are moved to the depth dimension. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
blocksize : int (required)
+
Blocks of [blocksize, blocksize] are moved.
+
+ +#### Inputs + +
+
input : T
+
Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
+
+ +#### Outputs + +
+
output : T
+
Output tensor of [N, C * blocksize * blocksize, H/blocksize, W/blocksize].
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Split-1** + + Split a tensor into a list of tensors, along the specified + 'axis'. The lengths of the split can be specified using argument 'axis' or + optional second input blob to the operator. Otherwise, the tensor is split + to equal sized parts. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
Which axis to split on
+
split : list of ints
+
length of each output
+
+ +#### Inputs (1 - 2) + +
+
input : T
+
The tensor to split
+
split (optional) : T
+
Optional list of output lengths (see also arg 'split')
+
+ +#### Outputs (1 - ∞) + +
+
outputs... (variadic) : T
+
One or more outputs forming list of tensors after splitting
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
+ +### **Sqrt-1** + + Square root takes one input data (Tensor) and produces one output data + (Tensor) where the square root is, y = x^0.5, is applied to + the tensor elementwise. If x is negative, then it will return NaN. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Squeeze-1** + + Remove single-dimensional entries from the shape of a tensor. + Takes a parameter `axes` with a list of axes to squeeze. + If `axes` is not provided, all the single dimensions will be removed from + the shape. If an axis is selected with shape entry not equal to one, an error is raised. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
List of non-negative integers, indicate the dimensions to squeeze.
+
+ +#### Inputs + +
+
data : T
+
Tensors with at least max(dims) dimensions.
+
+ +#### Outputs + +
+
squeezed : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Sub-1** + + Performs element-wise binary subtraction (with limited broadcast support). + + If necessary the right-hand-side argument will be broadcasted to match the + shape of left-hand-side argument. When broadcasting is specified, the second + tensor can either be of element size 1 (including a scalar tensor and any + tensor with rank equal to or smaller than the first tensor), or having its + shape as a contiguous subset of the first tensor's shape. The starting of the + mutually equal shape is specified by the argument "axis", and if it is not set, + suffix matching is assumed. 1-dim expansion doesn't work yet. + + For example, the following tensor shapes are supported (with broadcast=1): + + shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor + shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor + shape(A) = (2, 3, 4, 5), shape(B) = (5,) + shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) + shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 + shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0 + + Attribute `broadcast=1` needs to be passed to enable broadcasting. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions. See doc for details.
+
broadcast : int (default is 0)
+
Pass 1 to enable broadcasting
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
A : T
+
First operand, should share the type with the second operand.
+
B : T
+
Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
+
+ +#### Outputs + +
+
C : T
+
Result, has same dimensions and type as A
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Sum-1** + + Element-wise sum of each of the input tensors. All inputs and outputs must + have the same shape and data type. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for Sum.
+
+ +#### Outputs + +
+
sum : T
+
Output tensor. Same dimension as inputs.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Tanh-1** + + Calculates the hyperbolic tangent of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
consumed_inputs : list of ints
+
legacy optimization attribute.
+
+ +#### Inputs + +
+
input : T
+
1-D input tensor
+
+ +#### Outputs + +
+
output : T
+
The hyperbolic tangent values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Tile-1** + + Repeat the elements of a tensor along an axis. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
input : T
+
Input tensor of any shape.
+
tiles : T
+
Number of repeated copies to make of the input tensor.
+
axis : T
+
Axis along which to repeat.
+
+ +#### Outputs + +
+
output : T
+
Output tensor of same shape and type as input.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
T1 : tensor(int64)
+
Constrain tiles and axis's type to int64 tensors.
+
+ +### **TopK-1** + + Retrieve the top-K elements along a specified axis. Given an input tensor of + shape [a_0, a_1, ..., a_{n-1}] and integer argument k, return two outputs: + -Value tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] + which contains the values of the top k elements along the specified axis + -Index tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] which + contains the indices of the top k elements (original indices from the input + tensor). + Given two equivalent values, this operator uses the indices along the axis as + a tiebreaker. That is, the element with the lower index will appear first. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
Dimension on which to do the sort.
+
k : int (required)
+
Number of top elements to retrieve
+
+ +#### Inputs + +
+
X : T
+
Tensor of shape [a_0, a_1, ..., a_{n-1}]
+
+ +#### Outputs + +
+
Values : T
+
Tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] containing top K values from the input tensor
+
Indices : I
+
Tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] containing the corresponding input tensor indices for the top K values.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **Transpose-1** + + Returns a transpose of the input tensor. (Similar to `numpy.transpose`). + The optional attribute `perm` must be a permutation of the dimensions of + the input tensor. Axis `i` of the output tensor corresponds to the axis + `perm[i]` of the input tensor. + For example, when perm=(1, 0, 2), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 1, 3). + When perm=(1, 2, 0), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 3, 1). + If the attribute `perm` is omitted, its default value is `(n-1, ..., 0)`, + where `n` is the rank of the input tensor. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
perm : list of ints
+
A list of integers. By default, reverse the dimensions, otherwise permute the axes according to the values given.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
transposed : T
+
Transposed output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Unsqueeze-1** + + Insert single-dimensional entries to the shape of a tensor. + Takes one required argument `axes`, a list of dimensions that will be inserted. + Dimension indices in `axes` are as seen in the output tensor. For example: + Given a tensor such that tensor with shape [3, 4, 5], then + Unsqueeze(tensor, axes=[0, 4]) has shape [1, 3, 4, 5, 1] + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints (required)
+
List of non-negative integers, indicate the dimensions to be inserted
+
+ +#### Inputs + +
+
data : T
+
Original tensor
+
+ +#### Outputs + +
+
expanded : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Upsample-1** + + Upsample the input tensor. + The width and height of the output tensor are: + output_width = floor(input_width * width_scale), + output_height = floor(input_height * height_scale). + Example: + Given `data` tensor, width_scale, height_scale, mode, + Upsample the input 4-D tensor in nearest mode: + data = [[[ + [1, 2], + [3, 4] + ]]] + width_scale = 2 + height_scale = 2 + mode = "nearest" + output = [[[ + [1, 1, 2, 2], + [1, 1, 2, 2], + [3, 3, 4, 4], + [3, 3, 4, 4] + ]]] + +#### Version + +No versioning maintained for experimental ops. +#### Attributes + +
+
height_scale : float (required)
+
The scale along height dimension. It takes value greater than or equal to 1.
+
mode : string (default is nearest)
+
Two interpolation modes: nearest(default), bilinear
+
width_scale : float (required)
+
The scale along width dimension. It takes value greater than or equal to 1.
+
+ +#### Inputs + +
+
X : T
+
4-D tensor, [N,C,H,W]
+
+ +#### Outputs + +
+
Y : T
+
4-D tensor after resizing, [N,C,H,W]
+
+ +#### Type Constraints + +
+
T : tensor(bool), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain output types to bool, int32, int64, float16, float, double tensors.
+
+ +### **Xor-1** + + Returns the tensor resulted from performing the `xor` logical operation + elementwise on the input tensors `A` and `B`. + + If broadcasting is enabled, the right-hand-side argument will be broadcasted + to match the shape of left-hand-side argument. See the doc of `Add` for a + detailed description of the broadcasting rules. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions.
+
broadcast : int (default is 0)
+
Enable broadcasting
+
+ +#### Inputs + +
+
A : T
+
Left input tensor for the logical operator.
+
B : T
+
Right input tensor for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input to boolean tensor.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +## Version 2 of the default ONNX operator set +### **GlobalLpPool-2** + + GlobalLpPool consumes an input tensor X and applies lp pool pooling across + the values in the same channel. This is equivalent to LpPool with kernel size + equal to the spatial dimension of input tensor. + +#### Version + +This version of the operator has been available since version 2 of the default ONNX operator set. + +#### Attributes + +
+
p : int (default is 2)
+
p value of the Lp norm used to pool over the input data.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **LpPool-2** + + LpPool consumes an input tensor X and applies Lp pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + Lp pooling consisting of computing the Lp norm on all values of a subset + of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. + +#### Version + +This version of the operator has been available since version 2 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
p : int (default is 2)
+
p value of the Lp norm used to pool over the input data.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Pad-2** + + Given `data` tensor, pads, mode, and value. + Example: + Insert 0 pads to the beginning of the second dimension. + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + pads = [0, 2, 0, 0] + output = [ + [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ], + ] + +#### Version + +This version of the operator has been available since version 2 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is constant)
+
Three modes: constant(default), reflect, edge
+
pads : list of ints (required)
+
List of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D it is the number of pixels. `pads` rank should be double of the input's rank. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`.
+
value : float (default is 0.0)
+
One float, indicates the value to be filled.
+
+ +#### Inputs + +
+
data : T
+
Input tensor.
+
+ +#### Outputs + +
+
output : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Split-2** + + Split a tensor into a list of tensors, along the specified + 'axis'. Lengths of the parts can be specified using argument 'split'. + Otherwise, the tensor is split to equal sized parts. + +#### Version + +This version of the operator has been available since version 2 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to split on.
+
split : list of ints
+
length of each output
+
+ +#### Inputs + +
+
input : T
+
The tensor to split
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic) : T
+
One or more outputs forming list of tensors after splitting
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +## Version 3 of the default ONNX operator set +### **GRU-3** + + Computes an one-layer GRU. This operator is usually supported via some custom + implementation such as CuDNN. + + Notations: + + `X` - input tensor + + `z` - update gate + + `r` - reset gate + + `h` - hidden gate + + `t` - time step (t-1 means previous time step) + + `W[zrh]` - W parameter weight matrix for update, reset, and hidden gates + + `R[zrh]` - R recurrence weight matrix for update, reset, and hidden gates + + `Wb[zrh]` - W bias vectors for update, reset, and hidden gates + + `Rb[zrh]` - R bias vectors for update, reset, and hidden gates + + `WB[zrh]` - W parameter weight matrix for backward update, reset, and hidden gates + + `RB[zrh]` - R recurrence weight matrix for backward update, reset, and hidden gates + + `WBb[zrh]` - W bias vectors for backward update, reset, and hidden gates + + `RBb[zrh]` - R bias vectors for backward update, reset, and hidden gates + + `H` - Hidden state + + `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + Relu(x) - max(0, x) + + Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + + Sigmoid(x) - 1/(1 + e^{-x}) + + (NOTE: Below are optional) + + Affine(x) - alpha*x + beta + + LeakyRelu(x) - x if x >= 0 else alpha * x + + ThresholdedRelu(x) - x if x >= alpha else 0 + + ScaledTanh(x) - alpha*Tanh(beta*x) + + HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + + Elu(x) - x if x >= 0 else alpha*(e^x - 1) + + Softsign(x) - x/(1 + |x|) + + Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh): + + - zt = f(Xt*(Wz^T) + Ht-1*Rz + Wbz + Rbz) + + - rt = f(Xt*(Wr^T) + Ht-1*Rr + Wbr + Rbr) + + - ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*Rh + Rbh + Wbh) # default, when linear_before_reset = 0 + + - ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*Rh + Rbh) + Wbh) # when linear_before_reset != 0 + + - Ht = (1 - zt) (.) ht + zt (.) Ht-1 + +#### Version + +This version of the operator has been available since version 3 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings
+
A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
linear_before_reset : int (default is 0)
+
When computing the output of the hidden gate, apply the linear transformation before multiplying by the output of the reset gate.
+
output_sequence : int (default is 0)
+
The sequence output for the hidden is optional if 0. Default 0.
+
+ +#### Inputs (3 - 6) + +
+
X : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W : T
+
The weight tensor for the gates. Concatenation of `W[zrh]` and `WB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, input_size]`.
+
R : T
+
The recurrence weight tensor. Concatenation of `R[zrh]` and `RB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, hidden_size]`.
+
B (optional) : T
+
The bias tensor for the gates. Concatenation of `[Wb[zrh], Rb[zrh]]` and `[WBb[zrh], RBb[zrh]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 6*hidden_size]`. Optional: If not specified - assumed to be 0
+
sequence_lens (optional) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs (0 - 2) + +
+
Y (optional) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`. It is optional if `output_sequence` is 0.
+
Y_h (optional) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +## Version 4 of the default ONNX operator set +### **Concat-4** + + Concatenate a list of tensors into a single tensor + +#### Version + +This version of the operator has been available since version 4 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (required)
+
Which axis to concat on
+
+ +#### Inputs (1 - ∞) + +
+
inputs (variadic) : T
+
List of tensors for concatenation
+
+ +#### Outputs + +
+
concat_result : T
+
Concatenated tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain output types to any tensor type.
+
+ +## Version 5 of the default ONNX operator set +### **Reshape-5** + + Reshape the input tensor similar to numpy.reshape. + First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. + At most one dimension of the new shape can be -1. In this case, the value is + inferred from the size of the tensor and the remaining dimensions. A dimension + could also be 0, in which case the actual dimension value is unchanged (i.e. taken + from the input tensor). Shape (second input) could be an empty shape, which means converting to a scalar. + The input tensor's shape and the output tensor's shape are required to have the same number of elements. + +#### Version + +This version of the operator has been available since version 5 of the default ONNX operator set. + +#### Inputs + +
+
data : T
+
An input tensor.
+
shape : tensor(int64)
+
Specified shape for output.
+
+ +#### Outputs + +
+
reshaped : T
+
Reshaped data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +## Version 6 of the default ONNX operator set +### **Abs-6** + + Absolute takes one input data (Tensor) and produces one output data + (Tensor) where the absolute is, y = abs(x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **Add-6** + + Performs element-wise binary addition (with limited broadcast support). + + If necessary the right-hand-side argument will be broadcasted to match the + shape of left-hand-side argument. When broadcasting is specified, the second + tensor can either be of element size 1 (including a scalar tensor and any + tensor with rank equal to or smaller than the first tensor), or having its + shape as a contiguous subset of the first tensor's shape. The starting of the + mutually equal shape is specified by the argument "axis", and if it is not set, + suffix matching is assumed. 1-dim expansion doesn't work yet. + + For example, the following tensor shapes are supported (with broadcast=1): + + shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor + shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor + shape(A) = (2, 3, 4, 5), shape(B) = (5,) + shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) + shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 + shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0 + + Attribute `broadcast=1` needs to be passed to enable broadcasting. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions. See doc for details.
+
broadcast : int (default is 0)
+
Pass 1 to enable broadcasting
+
+ +#### Inputs + +
+
A : T
+
First operand, should share the type with the second operand.
+
B : T
+
Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
+
+ +#### Outputs + +
+
C : T
+
Result, has same dimensions and type as A
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **BatchNormalization-6** + + Carries out batch normalization as described in the paper + https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, + there are multiple cases for the number of outputs, which we list below: + + Output case #1: Y, mean, var, saved_mean, saved_var (training mode) + Output case #2: Y (test mode) + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero, default is 1e-5f.
+
is_test : int (default is 0)
+
If set to nonzero, run spatial batch normalization in test mode, default is 0.
+
momentum : float (default is 0.9)
+
Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum), default is 0.9f.
+
spatial : int (default is 1)
+
If true, compute the mean and variance across all spatial elements If false, compute the mean and variance across per feature.Default is 1.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
scale : T
+
The scale as a 1-dimensional tensor of size C to be applied to the output.
+
B : T
+
The bias as a 1-dimensional tensor of size C to be applied to the output.
+
mean : T
+
The running mean (training) or the estimated mean (testing) as a 1-dimensional tensor of size C.
+
var : T
+
The running variance (training) or the estimated variance (testing) as a 1-dimensional tensor of size C.
+
+ +#### Outputs (1 - 5) + +
+
Y : T
+
The output tensor of the same shape as X.
+
mean (optional) : T
+
The running mean after the BatchNormalization operator. Must be in-place with the input mean. Should not be used for testing.
+
var (optional) : T
+
The running variance after the BatchNormalization operator. Must be in-place with the input var. Should not be used for testing.
+
saved_mean (optional) : T
+
Saved mean used during training to speed up gradient computation. Should not be used for testing.
+
saved_var (optional) : T
+
Saved variance used during training to speed up gradient computation. Should not be used for testing.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Cast-6** + + The operator casts the elements of a given input tensor to a data type + specified by the 'to' argument and returns an output tensor of the same size in + the converted type. The 'to' argument must be one of the data types specified + in the 'DataType' enum field in the TensorProto message. + NOTE: Casting to and from strings is not supported yet. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
to : int (required)
+
The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
+
+ +#### Inputs + +
+
input : T1
+
Input tensor to be cast.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor with the same shape as input with type specified by the 'to' argument
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool)
+
Constrain input types. Casting from strings and complex are not supported.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool)
+
Constrain output types. Casting to strings and complex are not supported.
+
+ +### **Ceil-6** + + Ceil takes one input data (Tensor) and produces one output data + (Tensor) where the ceil is, y = ceil(x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Clip-6** + + Clip operator limits the given input within an interval. The interval is + specified with arguments 'min' and 'max'. They default to + numeric_limits::lowest() and numeric_limits::max() respectively. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
max : float (default is (3.402823e+38))
+
Maximum value, above which element is replaced by max
+
min : float (default is (-3.402823e+38))
+
Minimum value, under which element is replaced by min
+
+ +#### Inputs + +
+
input : T
+
Input tensor whose elements to be clipped
+
+ +#### Outputs + +
+
output : T
+
Output tensor with clipped input elements
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Div-6** + + Performs element-wise binary division (with limited broadcast support). + + If necessary the right-hand-side argument will be broadcasted to match the + shape of left-hand-side argument. When broadcasting is specified, the second + tensor can either be of element size 1 (including a scalar tensor and any + tensor with rank equal to or smaller than the first tensor), or having its + shape as a contiguous subset of the first tensor's shape. The starting of the + mutually equal shape is specified by the argument "axis", and if it is not set, + suffix matching is assumed. 1-dim expansion doesn't work yet. + + For example, the following tensor shapes are supported (with broadcast=1): + + shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor + shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor + shape(A) = (2, 3, 4, 5), shape(B) = (5,) + shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) + shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 + shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0 + + Attribute `broadcast=1` needs to be passed to enable broadcasting. + + + For integer inputs, the result is computed using truncating division (rounding toward zero). + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions. See doc for details.
+
broadcast : int (default is 0)
+
Pass 1 to enable broadcasting
+
+ +#### Inputs + +
+
A : T
+
First operand, should share the type with the second operand.
+
B : T
+
Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
+
+ +#### Outputs + +
+
C : T
+
Result, has same dimensions and type as A
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Dropout-6** + + Dropout takes one input data (Tensor) and produces two Tensor outputs, + output (Tensor) and mask (Tensor). Depending on whether it is in + test mode or not, the output Y will either be a random dropout, or a simple + copy of the input. Note that our implementation of Dropout does scaling in + the training phase, so during testing nothing needs to be done. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
is_test : int (default is 0)
+
(int, default 0) if nonzero, run dropout in test mode where the output is simply Y = X.
+
ratio : float (default is 0.5)
+
(float, default 0.5) the ratio of random dropout
+
+ +#### Inputs + +
+
data : T
+
The input data as Tensor.
+
+ +#### Outputs (1 - 2) + +
+
output : T
+
The output.
+
mask (optional) : T
+
The output mask. If is_test is nonzero, this output is not filled.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Elu-6** + + Elu takes one input data (Tensor) and produces one output data + (Tensor) where the function `f(x) = alpha * (exp(x) - 1.) for x < + 0`, `f(x) = x for x >= 0`., is applied to the tensor elementwise. + + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Coefficient of ELU.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Exp-6** + + Calculates the exponential of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
input : T
+
Input tensor
+
+ +#### Outputs + +
+
output : T
+
The exponential of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Floor-6** + + Floor takes one input data (Tensor) and produces one output data + (Tensor) where the floor is, y = floor(x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Gemm-6** + + General Matrix multiplication: + https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3 + Compute Y = alpha * A * B + beta * C, where input tensor A has + dimension (M X K), input tensor B has dimension (K X N), input tensor C and + output tensor Y have dimension (M X N). + If attribute broadcast is non-zero, input tensor C will be broadcasted to match + the dimension requirement. A will be transposed before doing the computation + if attribute transA is non-zero, same for B and transB. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Scalar multiplier for the product of input tensors A * B, the default value is 1.0.
+
beta : float (default is 1.0)
+
Scalar multiplier for input tensor C, the default value is 1.0.
+
broadcast : int (default is 0)
+
Whether C should be broadcasted
+
transA : int (default is 0)
+
Whether A should be transposed
+
transB : int (default is 0)
+
Whether B should be transposed
+
+ +#### Inputs + +
+
A : T
+
Input tensor A
+
B : T
+
Input tensor B
+
C : T
+
Input tensor C
+
+ +#### Outputs + +
+
Y : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **HardSigmoid-6** + + HardSigmoid takes one input data (Tensor) and produces one output data + (Tensor) where the HardSigmoid function, y = max(0, min(1, alpha * x + beta)), + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 0.2)
+
Value of alpha.
+
beta : float (default is 0.5)
+
Value of beta.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **InstanceNormalization-6** + + Carries out instance normalization as described in the paper + https://arxiv.org/abs/1607.08022. + + y = scale * (x - mean) / sqrt(variance + epsilon) + B, + where mean and variance are computed per instance per channel. + + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
scale (differentiable) : T
+
The input 1-dimensional scale tensor of size C.
+
B (differentiable) : T
+
The input 1-dimensional bias tensor of size C.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output tensor of the same shape as input.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **LeakyRelu-6** + + LeakyRelu takes input data (Tensor) and an argument alpha, and produces one + output data (Tensor) where the function `f(x) = alpha * x for x < 0`, + `f(x) = x for x >= 0`, is applied to the data tensor elementwise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 0.01)
+
Coefficient of leakage.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Log-6** + + Calculates the natural log of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
input : T
+
Input tensor
+
+ +#### Outputs + +
+
output : T
+
The natural log of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Max-6** + + Element-wise max of each of the input tensors. All inputs and outputs must + have the same shape and data type. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for Max.
+
+ +#### Outputs + +
+
max : T
+
Output tensor. Same dimension as inputs.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Mean-6** + + Element-wise mean of each of the input tensors. All inputs and outputs must + have the same shape and data type. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for Mean.
+
+ +#### Outputs + +
+
mean : T
+
Output tensor. Same dimension as inputs.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Min-6** + + Element-wise min of each of the input tensors. All inputs and outputs must + have the same shape and data type. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for Min
+
+ +#### Outputs + +
+
min : T
+
Output tensor. Same dimension as inputs.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Mul-6** + + Performs element-wise binary multiplication (with limited broadcast support). + + If necessary the right-hand-side argument will be broadcasted to match the + shape of left-hand-side argument. When broadcasting is specified, the second + tensor can either be of element size 1 (including a scalar tensor and any + tensor with rank equal to or smaller than the first tensor), or having its + shape as a contiguous subset of the first tensor's shape. The starting of the + mutually equal shape is specified by the argument "axis", and if it is not set, + suffix matching is assumed. 1-dim expansion doesn't work yet. + + For example, the following tensor shapes are supported (with broadcast=1): + + shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor + shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor + shape(A) = (2, 3, 4, 5), shape(B) = (5,) + shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) + shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 + shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0 + + Attribute `broadcast=1` needs to be passed to enable broadcasting. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions. See doc for details.
+
broadcast : int (default is 0)
+
Pass 1 to enable broadcasting
+
+ +#### Inputs + +
+
A : T
+
First operand, should share the type with the second operand.
+
B : T
+
Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
+
+ +#### Outputs + +
+
C : T
+
Result, has same dimensions and type as A
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Neg-6** + + Neg takes one input data (Tensor) and produces one output data + (Tensor) where each element flipped sign, y = -x, is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(int32), tensor(int8), tensor(int16), tensor(int64), tensor(float16), tensor(double)
+
Constrain input and output types to signed numeric tensors.
+
+ +### **PRelu-6** + + PRelu takes input data (Tensor) and slope tensor as input, and produces one + output data (Tensor) where the function `f(x) = slope * x for x < 0`, + `f(x) = x for x >= 0`., is applied to the data tensor elementwise. + + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
Input tensor
+
slope : T
+
Slope tensor. If `Slope` is of size 1, the value is sharedacross different channels
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Reciprocal-6** + + Reciprocal takes one input data (Tensor) and produces one output data + (Tensor) where the reciprocal is, y = 1/x, is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Relu-6** + + Relu takes one input data (Tensor) and produces one output data + (Tensor) where the rectified linear function, y = max(0, x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Selu-6** + + Selu takes one input data (Tensor) and produces one output data + (Tensor) where the scaled exponential linear unit function, + `y = gamma * (alpha * e^x - alpha) for x <= 0`, `y = gamma * x for x > 0`, + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.67326)
+
Coefficient of SELU default to 1.67326319217681884765625 (i.e., float32 approximation of 1.6732632423543772848170429916717).
+
gamma : float (default is 1.0507)
+
Coefficient of SELU default to 1.05070102214813232421875 (i.e., float32 approximation of 1.0507009873554804934193349852946).
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Sigmoid-6** + + Sigmoid takes one input data (Tensor) and produces one output data + (Tensor) where the sigmoid function, y = 1 / (1 + exp(-x)), is applied to the + tensor elementwise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Sqrt-6** + + Square root takes one input data (Tensor) and produces one output data + (Tensor) where the square root is, y = x^0.5, is applied to + the tensor elementwise. If x is negative, then it will return NaN. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Sub-6** + + Performs element-wise binary subtraction (with limited broadcast support). + + If necessary the right-hand-side argument will be broadcasted to match the + shape of left-hand-side argument. When broadcasting is specified, the second + tensor can either be of element size 1 (including a scalar tensor and any + tensor with rank equal to or smaller than the first tensor), or having its + shape as a contiguous subset of the first tensor's shape. The starting of the + mutually equal shape is specified by the argument "axis", and if it is not set, + suffix matching is assumed. 1-dim expansion doesn't work yet. + + For example, the following tensor shapes are supported (with broadcast=1): + + shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor + shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor + shape(A) = (2, 3, 4, 5), shape(B) = (5,) + shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) + shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 + shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0 + + Attribute `broadcast=1` needs to be passed to enable broadcasting. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
If set, defines the broadcast dimensions. See doc for details.
+
broadcast : int (default is 0)
+
Pass 1 to enable broadcasting
+
+ +#### Inputs + +
+
A : T
+
First operand, should share the type with the second operand.
+
B : T
+
Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
+
+ +#### Outputs + +
+
C : T
+
Result, has same dimensions and type as A
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Sum-6** + + Element-wise sum of each of the input tensors. All inputs and outputs must + have the same shape and data type. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for Sum.
+
+ +#### Outputs + +
+
sum : T
+
Output tensor. Same dimension as inputs.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Tanh-6** + + Calculates the hyperbolic tangent of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
input : T
+
Input tensor
+
+ +#### Outputs + +
+
output : T
+
The hyperbolic tangent values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Tile-6** + + Constructs a tensor by tiling a given tensor. + This is the same as function `tile` in Numpy, but no broadcast. + For example A = [[1, 2], [3, 4]], B = [1, 2], tile(A, B) = [[1, 2, 1, 2], [3, 4, 3, 4]] + +#### Version + +This version of the operator has been available since version 6 of the default ONNX operator set. + +#### Inputs + +
+
input : T
+
Input tensor of any shape.
+
repeats : T1
+
1D int64 tensor of the same length as input's dimension number, includes numbers of repeated copies along input's dimensions.
+
+ +#### Outputs + +
+
output : T
+
Output tensor of the same dimensions and type as tensor input. output_dim[i] = input_dim[i] * repeats[i]
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
T1 : tensor(int64)
+
Constrain repeat's type to int64 tensors.
+
+ +## Version 7 of the default ONNX operator set +### **Acos-7** + + Calculates the arccosine (inverse of cosine) of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The arccosine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Add-7** + + Performs element-wise binary addition (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
First operand.
+
B : T
+
Second operand.
+
+ +#### Outputs + +
+
C : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **And-7** + + Returns the tensor resulted from performing the `and` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input to boolean tensor.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Asin-7** + + Calculates the arcsine (inverse of sine) of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The arcsine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Atan-7** + + Calculates the arctangent (inverse of tangent) of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The arctangent of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **AveragePool-7** + + AveragePool consumes an input tensor X and applies average pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + average pooling consisting of computing the average on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) + + * pad_shape[i] is sum of pads along axis i + ``` + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i] + ``` + The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero). + + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.
+
count_include_pad : int (default is 0)
+
Whether include pad pixels when calculating values for the edges. Default is 0, doesn't count include pad.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **BatchNormalization-7** + + Carries out batch normalization as described in the paper + https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, + there are multiple cases for the number of outputs, which we list below: + + Output case #1: Y, mean, var, saved_mean, saved_var (training mode) + Output case #2: Y (test mode) + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
momentum : float (default is 0.9)
+
Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum).
+
spatial : int (default is 1)
+
If true, compute the mean and variance across per activation. If false, compute the mean and variance across per feature over each mini-batch.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
scale : T
+
If spatial is true, the dimension of scale is (C). If spatial is false, the dimensions of scale are (C x D1 x ... x Dn)
+
B : T
+
If spatial is true, the dimension of bias is (C). If spatial is false, the dimensions of bias are (C x D1 x ... x Dn)
+
mean : T
+
If spatial is true, the dimension of the running mean (training) or the estimated mean (testing) is (C). If spatial is false, the dimensions of the running mean (training) or the estimated mean (testing) are (C x D1 x ... x Dn).
+
var : T
+
If spatial is true, the dimension of the running variance(training) or the estimated variance (testing) is (C). If spatial is false, the dimensions of the running variance(training) or the estimated variance (testing) are (C x D1 x ... x Dn).
+
+ +#### Outputs (1 - 5) + +
+
Y : T
+
The output tensor of the same shape as X
+
mean (optional) : T
+
The running mean after the BatchNormalization operator.
+
var (optional) : T
+
The running variance after the BatchNormalization operator.
+
saved_mean (optional) : T
+
Saved mean used during training to speed up gradient computation.
+
saved_var (optional) : T
+
Saved variance used during training to speed up gradient computation.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Cos-7** + + Calculates the cosine of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The cosine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Div-7** + + Performs element-wise binary division (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + + For integer inputs, the result is computed using truncating division (rounding toward zero). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
First operand.
+
B : T
+
Second operand.
+
+ +#### Outputs + +
+
C : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Dropout-7** + + Dropout takes one input data (Tensor) and produces two Tensor outputs, + output (Tensor) and mask (Tensor). Depending on whether it is in + test mode or not, the output Y will either be a random dropout, or a simple + copy of the input. Note that our implementation of Dropout does scaling in + the training phase, so during testing nothing needs to be done. + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Attributes + +
+
ratio : float (default is 0.5)
+
The ratio of random dropout
+
+ +#### Inputs + +
+
data : T
+
The input data as Tensor.
+
+ +#### Outputs (1 - 2) + +
+
output : T
+
The output.
+
mask (optional) : T
+
The output mask.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Equal-7** + + Returns the tensor resulted from performing the `equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
First input operand for the logical operator.
+
B : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool), tensor(int32), tensor(int64)
+
Constrain input to integral tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **GRU-7** + + Computes an one-layer GRU. This operator is usually supported via some custom + implementation such as CuDNN. + + Notations: + + `X` - input tensor + + `z` - update gate + + `r` - reset gate + + `h` - hidden gate + + `t` - time step (t-1 means previous time step) + + `W[zrh]` - W parameter weight matrix for update, reset, and hidden gates + + `R[zrh]` - R recurrence weight matrix for update, reset, and hidden gates + + `Wb[zrh]` - W bias vectors for update, reset, and hidden gates + + `Rb[zrh]` - R bias vectors for update, reset, and hidden gates + + `WB[zrh]` - W parameter weight matrix for backward update, reset, and hidden gates + + `RB[zrh]` - R recurrence weight matrix for backward update, reset, and hidden gates + + `WBb[zrh]` - W bias vectors for backward update, reset, and hidden gates + + `RBb[zrh]` - R bias vectors for backward update, reset, and hidden gates + + `H` - Hidden state + + `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + Relu(x) - max(0, x) + + Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + + Sigmoid(x) - 1/(1 + e^{-x}) + + (NOTE: Below are optional) + + Affine(x) - alpha*x + beta + + LeakyRelu(x) - x if x >= 0 else alpha * x + + ThresholdedRelu(x) - x if x >= alpha else 0 + + ScaledTanh(x) - alpha*Tanh(beta*x) + + HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + + Elu(x) - x if x >= 0 else alpha*(e^x - 1) + + Softsign(x) - x/(1 + |x|) + + Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh): + + - zt = f(Xt*(Wz^T) + Ht-1*(Rz^T) + Wbz + Rbz) + + - rt = f(Xt*(Wr^T) + Ht-1*(Rr^T) + Wbr + Rbr) + + - ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*(Rh^T) + Rbh + Wbh) # default, when linear_before_reset = 0 + + - ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*(Rh^T) + Rbh)) + Wbh) # when linear_before_reset != 0 + + - Ht = (1 - zt) (.) ht + zt (.) Ht-1 + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings
+
A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
linear_before_reset : int (default is 0)
+
When computing the output of the hidden gate, apply the linear transformation before multiplying by the output of the reset gate.
+
+ +#### Inputs (3 - 6) + +
+
X : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W : T
+
The weight tensor for the gates. Concatenation of `W[zrh]` and `WB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, input_size]`.
+
R : T
+
The recurrence weight tensor. Concatenation of `R[zrh]` and `RB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, hidden_size]`.
+
B (optional) : T
+
The bias tensor for the gates. Concatenation of `[Wb[zrh], Rb[zrh]]` and `[WBb[zrh], RBb[zrh]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 6*hidden_size]`. Optional: If not specified - assumed to be 0
+
sequence_lens (optional) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs (0 - 2) + +
+
Y (optional) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **Gemm-7** + + General Matrix multiplication: + https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3 + + A' = transpose(A) if transA else A + + B' = transpose(B) if transB else B + + Compute Y = alpha * A' * B' + beta * C, where input tensor A has shape (M, K) or (K, M), + input tensor B has shape (K, N) or (N, K), input tensor C is broadcastable to shape (M, N), + and output tensor Y has shape (M, N). A will be transposed before doing the + computation if attribute transA is non-zero, same for B and transB. + This operator supports **unidirectional broadcasting** (tensor C should be unidirectional broadcastable to tensor A * B); for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Scalar multiplier for the product of input tensors A * B.
+
beta : float (default is 1.0)
+
Scalar multiplier for input tensor C.
+
transA : int (default is 0)
+
Whether A should be transposed
+
transB : int (default is 0)
+
Whether B should be transposed
+
+ +#### Inputs + +
+
A : T
+
Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
+
B : T
+
Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
+
C : T
+
Input tensor C. The shape of C should be unidirectional broadcastable to (M, N).
+
+ +#### Outputs + +
+
Y : T
+
Output tensor of shape (M, N).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Greater-7** + + Returns the tensor resulted from performing the `greater` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
First input operand for the logical operator.
+
B : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input to float tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **LSTM-7** + + Computes an one-layer LSTM. This operator is usually supported via some + custom implementation such as CuDNN. + + Notations: + + `X` - input tensor + + `i` - input gate + + `o` - output gate + + `f` - forget gate + + `c` - cell gate + + `t` - time step (t-1 means previous time step) + + `W[iofc]` - W parameter weight matrix for input, output, forget, and cell gates + + `R[iofc]` - R recurrence weight matrix for input, output, forget, and cell gates + + `Wb[iofc]` - W bias vectors for input, output, forget, and cell gates + + `Rb[iofc]` - R bias vectors for input, output, forget, and cell gates + + `P[iof]` - P peephole weight vector for input, output, and forget gates + + `WB[iofc]` - W parameter weight matrix for backward input, output, forget, and cell gates + + `RB[iofc]` - R recurrence weight matrix for backward input, output, forget, and cell gates + + `WBb[iofc]` - W bias vectors for backward input, output, forget, and cell gates + + `RBb[iofc]` - R bias vectors for backward input, output, forget, and cell gates + + `PB[iof]` - P peephole weight vector for backward input, output, and forget gates + + `H` - Hidden state + + `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + Relu(x) - max(0, x) + + Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + + Sigmoid(x) - 1/(1 + e^{-x}) + + (NOTE: Below are optional) + + Affine(x) - alpha*x + beta + + LeakyRelu(x) - x if x >= 0 else alpha * x + + ThresholdedRelu(x) - x if x >= alpha else 0 + + ScaledTanh(x) - alpha*Tanh(beta*x) + + HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + + Elu(x) - x if x >= 0 else alpha*(e^x - 1) + + Softsign(x) - x/(1 + |x|) + + Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh, h=Tanh): + + - it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi) + + - ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf) + + - ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc) + + - Ct = ft (.) Ct-1 + it (.) ct + + - ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Po (.) Ct + Wbo + Rbo) + + - Ht = ot (.) h(Ct) + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings
+
A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
input_forget : int (default is 0)
+
Couple the input and forget gates if 1.
+
+ +#### Inputs (3 - 8) + +
+
X : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W : T
+
The weight tensor for the gates. Concatenation of `W[iofc]` and `WB[iofc]` (if bidirectional) along dimension 0. The tensor has shape `[num_directions, 4*hidden_size, input_size]`.
+
R : T
+
The recurrence weight tensor. Concatenation of `R[iofc]` and `RB[iofc]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 4*hidden_size, hidden_size]`.
+
B (optional) : T
+
The bias tensor for input gate. Concatenation of `[Wb[iofc], Rb[iofc]]`, and `[WBb[iofc], RBb[iofc]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 8*hidden_size]`. Optional: If not specified - assumed to be 0.
+
sequence_lens (optional) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
initial_c (optional) : T
+
Optional initial value of the cell. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
P (optional) : T
+
The weight tensor for peepholes. Concatenation of `P[iof]` and `PB[iof]` (if bidirectional) along dimension 0. It has shape `[num_directions, 3*hidde_size]`. Optional: If not specified - assumed to be 0.
+
+ +#### Outputs (0 - 3) + +
+
Y (optional) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
Y_c (optional) : T
+
The last output value of the cell. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **Less-7** + + Returns the tensor resulted from performing the `less` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
First input operand for the logical operator.
+
B : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input to float tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Mul-7** + + Performs element-wise binary multiplication (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
First operand.
+
B : T
+
Second operand.
+
+ +#### Outputs + +
+
C : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Multinomial-7** + + Generate a tensor of samples from a multinomial distribution according to the probabilities + of each of the possible outcomes. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int (default is 6)
+
(Optional) The data type for the elements of the output tensor, if not specified, we will use int32.
+
sample_size : int (default is 1)
+
Number of times to sample.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
Input tensor with shape [batch_size, class_size], where class_size is the number of all possible outcomes. Each value along the axis zero represents the unnormalized log-probability of each corresponding outcome in a batch.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor with shape [batch_size, sample_size], where sample_size is the number of times to sample. Each value along the axis zero represents the outcome of the corresponding sample in a batch.
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
T2 : tensor(int32), tensor(int64)
+
Constrain output types to integral tensors.
+
+ +### **Or-7** + + Returns the tensor resulted from performing the `or` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input to boolean tensor.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **PRelu-7** + + PRelu takes input data (Tensor) and slope tensor as input, and produces one + output data (Tensor) where the function `f(x) = slope * x for x < 0`, + `f(x) = x for x >= 0`., is applied to the data tensor elementwise. + + This operator supports **unidirectional broadcasting** (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
Input tensor
+
slope : T
+
Slope tensor. The shape of slope can be smaller than first input X; if so, its shape must be unidirectional broadcastable to X
+
+ +#### Outputs + +
+
Y : T
+
Output tensor (same size as X)
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Pow-7** + + Pow takes input data (Tensor) and exponent Tensor, and + produces one output data (Tensor) where the function `f(x) = x^exponent`, + is applied to the data tensor elementwise. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
First operand, base of the exponent.
+
Y : T
+
Second operand, power of the exponent.
+
+ +#### Outputs + +
+
Z : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **RNN-7** + + Computes an one-layer simple RNN. This operator is usually supported + via some custom implementation such as CuDNN. + + Notations: + + `X` - input tensor + + `i` - input gate + + `t` - time step (t-1 means previous time step) + + `Wi` - W parameter weight matrix for input gate + + `Ri` - R recurrence weight matrix for input gate + + `Wbi` - W parameter bias vector for input gate + + `Rbi` - R parameter bias vector for input gate + + `WBi` - W parameter weight matrix for backward input gate + + `RBi` - R recurrence weight matrix for backward input gate + + `WBbi` - WR bias vectors for backward input gate + + `RBbi` - RR bias vectors for backward input gate + + `H` - Hidden state + + `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + Relu(x) - max(0, x) + + Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + + Sigmoid(x) - 1/(1 + e^{-x}) + + (NOTE: Below are optional) + + Affine(x) - alpha*x + beta + + LeakyRelu(x) - x if x >= 0 else alpha * x + + ThresholdedRelu(x) - x if x >= alpha else 0 + + ScaledTanh(x) - alpha*Tanh(beta*x) + + HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + + Elu(x) - x if x >= 0 else alpha*(e^x - 1) + + Softsign(x) - x/(1 + |x|) + + Softplus(x) - log(1 + e^x) + + Equations (Default: f=Tanh): + + - Ht = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Wbi + Rbi) + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings (default is ['Tanh', 'Tanh'])
+
One (or two if bidirectional) activation function for input gate. The activation function must be one of the activation functions specified above. Optional: Default `Tanh` if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
+ +#### Inputs (3 - 6) + +
+
X : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W : T
+
The weight tensor for input gate. Concatenation of `Wi` and `WBi` (if bidirectional). The tensor has shape `[num_directions, hidden_size, input_size]`.
+
R : T
+
The recurrence weight tensor. Concatenation of `Ri` and `RBi` (if bidirectional). The tensor has shape `[num_directions, hidden_size, hidden_size]`.
+
B (optional) : T
+
The bias tensor for input gate. Concatenation of `[Wbi, Rbi]` and `[WBbi, RBbi]` (if bidirectional). The tensor has shape `[num_directions, 2*hidden_size]`. Optional: If not specified - assumed to be 0.
+
sequence_lens (optional) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs (0 - 2) + +
+
Y (optional) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **Sin-7** + + Calculates the sine of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The sine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Sub-7** + + Performs element-wise binary subtraction (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
First operand.
+
B : T
+
Second operand.
+
+ +#### Outputs + +
+
C : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Tan-7** + + Calculates the tangent of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The tangent of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Upsample-7** + + Upsample the input tensor. + Each dimension value of the output tensor is: + output_dimension = floor(input_dimension * scale). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is nearest)
+
Two interpolation modes: nearest (default), and linear (including bilinear, trilinear, etc)
+
scales : list of floats (required)
+
The scale array along each dimension. It takes value greater than or equal to 1. The number of elements of 'scales' should be the same as the rank of input 'X'.
+
+ +#### Inputs + +
+
X : T
+
N-D tensor
+
+ +#### Outputs + +
+
Y : T
+
N-D tensor after resizing
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Xor-7** + + Returns the tensor resulted from performing the `xor` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input to boolean tensor.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +## Version 8 of the default ONNX operator set +### **Expand-8** + + Broadcast the input tensor following the given shape and the broadcast rule. + The broadcast rule is similar to numpy.array(input) * numpy.ones(shape): + Dimensions are right alignment; + Two corresponding dimensions must have the same value, or one of them is equal to 1. + Also, this operator is similar to numpy.broadcast_to(input, shape), + but the major difference is numpy.broadcast_to() does not allow shape to be smaller than input.size(). + It is possible that the output.shape is not equal to shape, when some dimensions in shape is equal to 1, + or the shape.ndim < input.shape.ndim. + +#### Version + +This version of the operator has been available since version 8 of the default ONNX operator set. + +#### Inputs + +
+
input : T
+
Input tensor
+
shape : tensor(int64)
+
A 1-D tensor indicates the shape you want to expand to, following the broadcast rule
+
+ +#### Outputs + +
+
output : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensors.
+
+ +### **Max-8** + + Element-wise max of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 8 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for max.
+
+ +#### Outputs + +
+
max : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **MaxPool-8** + + MaxPool consumes an input tensor X and applies max pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + max pooling consisting of computing the max on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) + + * pad_shape[i] is sum of pads along axis i + ``` + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i] + ``` + The output of each pooling window is maximum number of elements exclude pad. + + +#### Version + +This version of the operator has been available since version 8 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
storage_order : int (default is 0)
+
The storage order of the tensor. 0 is row major, and 1 is column major.
+
strides : list of ints
+
Stride along each spatial axis.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs (1 - 2) + +
+
Y : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
Indices (optional) : I
+
Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x ... x Dn).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **Mean-8** + + Element-wise mean of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 8 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for mean.
+
+ +#### Outputs + +
+
mean : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Min-8** + + Element-wise min of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 8 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for min.
+
+ +#### Outputs + +
+
min : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Scan-8** + + Scan can be used to iterate over one or more scan_input tensors, + constructing zero or more scan_output tensors. It combines ideas from general recurrences, + functional programming constructs such as scan, fold, map, and zip, and is intended to enable + generalizations of RNN-like constructs for sequence-to-sequence processing. + Other tensors (referred to as state_variables here) can be used to carry a state + when iterating from one element to another (similar to hidden-state in RNNs, also referred + to as loop-carried dependences in the context of loops). All these tensors are required to + have the same shape in each iteration of the loop (a restriction imposed to enable efficient + memory allocation). Many common usages involve a single scan_input tensor (where functionality + similar to scan, fold and map can be obtained). When more than one scan_input is used, + a behavior similar to zip is obtained. + + The attribute body must be a graph, specifying the computation to be performed in + every iteration. It takes as input the current values of the state_variables and + the current iterated element of the scan_inputs. It must return the (updated) values + of the state_variables and zero or more scan_output_element tensors. The values of the + scan_output_element tensors are concatenated over all the iterations to produce the + scan_output values of the scan construct (similar to the concatenated intermediate + hidden-state values of RNN-like constructs). + + The scan operation returns the final values of the state_variables as well as the + scan_outputs. + + The operation supports batching, and the batch-axis is required to be 0. + When multiple scan_input tensors are used, they must all have the same batch-size, + and they must all have the same maximum-sequence-length (the dimensionality of the + sequence axis or scan axis). The sequence axis or scan axis is required to be 1. + + The operation has an optional sequence_lens input (of shape [BATCH_SIZE]) to + allow variable length sequences of length <= the maximum-sequence-length. If this + input is not specified, all sequences are assumed to be of length equal to + maximum-sequence-length. For variable length input sequences, the scan_outputs + will consist of a sequence of same length as the input, padded to the + maximum-sequence-length. + + The optional attribute directions can be used to scan a sequence in the reverse direction. + If this attribute is omitted, all sequences are scanned in the forward direction. + A bidirectional scan be performed by specifying the same tensor input twice in the + scan_inputs, once with a forward direction, and once with a backward direction. + + Note that because of the ONNX restriction that only the last parameter of an operator can + be variadic, the initial-states and scan-inputs are listed together as one input parameter. + Similarly, the final-states and scan-outputs are listed together as one output parameter. + The attribute num_scan_inputs indicates the number M of scan-inputs. + + The behavior of + + Scan < + num_scan_inputs = m, + body = loop-body + > (sequence_lengths, init_1, ..., init_n, scan_1, ..., scan_m) + + is equivalent to the following pseudo-code: + + // T.shape[0] denotes the batch-size of T + // The batch-size of scan_1, ..., scan_m are all required to be equal + batch_size = scan_1.shape[0]; + + // scan_i.shape[1] denotes the (max) sequence-length of scan_i + // scan_i.shape[1] is required to be equal to scan_j.shape[1] for all i,j. + max_sequence_length = scan_1.shape[1]; + + for (int batch = 0; batch < batch_size; ++batch) { + // initialize state-variables + st_1 = init_1; ... st_n = init_n; + // initialize scan-output variables: [] denotes an empty tensor + scan_out_1 = []; ...; scan_out_k = []; + // identify number of iterations: + N = (sequence_lengths specified) ? sequence_lengths[batch] : max_sequence_length; + + // execute loop + for (int t = 0; t < N; ++t) { + // generate the scan-input elements: the notation T[t] indicates the sub-tensor + // of rank one less than T obtained by indexing T at position t along axis k. + si_1 = (scan_1[batch])[t]; + ... ; + si_m = (scan_m[batch])[t]; + // execute loop-body + st_1, ..., st_n, so_1, ..., so_k = loop-body(st_1, ..., st_n, si_1, ..., si_m) + // accumulate the scan-output elements + scan_out_1 = Concat(scan_out_1, so_1); ... ; scan_out_k = Concat(scan_out_k, so_k); + } + // accumulate the outputs for this batch: + bst_1[batch] = st_1; ..., bst_n[batch] = st_n; + // Note scan-outputs will have size max_sequence_length, but only first N values will be meaningful. + // The remaining values have an undefined value. + b_scan_out_1[batch] = scan_out_1; ...; b_scan_out_k[batch] = scan_out_k; + } + return bst_1, ..., bst_n, b_scan_out_1, ..., b_scan_out_k; + + + + *Sample usage: Encoding RNN using a Scan* + + The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, + recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can + be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes + %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these + values are computed in the outer graph, they need to be passed in as extra state_variables. + + graph rnn-encoding { + %H_0 = ... + %X = ... + %Y_h, %Y = Scan[body = , num_scan_inputs=1]("", %H_0, %X) + return %Y, %Y_h + } + + graph rnn-cell-1 ( + %H_tminus1[FLOAT, tensor] + %X_t[FLOAT, tensor] + ) { + %Wi = ... + %Ri = ... + %Wbi = ... + %Rbi = ... + %t1 = X_t * (Wi^T) + %t2 = H_tminus1*(Ri^T) + %t3 = Add(%t1, %t2) + %t4 = Add(%t3, %Wbi) + %t5 = Add(%t4, %Rbi) + %Ht = Tanh(%t5) + %Accumulate = Identity(%Ht) + return %Ht, %Accumulate + } + + +#### Version + +This version of the operator has been available since version 8 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has N+M inputs: (loop state variables..., scan_input_elts...). It has N+K outputs: (loop state variables..., scan_output_elts...). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
+
directions : list of ints
+
An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
+
num_scan_inputs : int (required)
+
An attribute specifying the number of scan_inputs M.
+
+ +#### Inputs (2 - ∞) + +
+
sequence_lens (optional) : I
+
Optional tensor specifying lengths of the sequences in a batch. If this input is not specified, all sequences are assumed to be of the maximum sequence length (the dimension of the sequence axis of the scan_input tensors).
+
initial_state_and_scan_inputs (variadic, heterogeneous) : V
+
Initial values of the loop's N state variables followed by M scan_inputs
+
+ +#### Outputs (1 - ∞) + +
+
final_state_and_scan_outputs (variadic, heterogeneous) : V
+
Final values of the loop's N state variables followed by K scan_outputs
+
+ +#### Type Constraints + +
+
I : tensor(int64)
+
Int64 tensor
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
All Tensor types
+
+ +### **Sum-8** + + Element-wise sum of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 8 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for sum.
+
+ +#### Outputs + +
+
sum : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +## Version 9 of the default ONNX operator set +### **Acosh-9** + + Calculates the hyperbolic arccosine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic arccosine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Asinh-9** + + Calculates the hyperbolic arcsine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic arcsine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Atanh-9** + + Calculates the hyperbolic arctangent of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic arctangent values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **BatchNormalization-9** + + Carries out batch normalization as described in the paper + https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, + there are multiple cases for the number of outputs, which we list below: + + Output case #1: Y, mean, var, saved_mean, saved_var (training mode) + Output case #2: Y (test mode) + + For previous (depreciated) non-spatial cases, implementors are suggested + to flatten the input shape to (N x C*D1*D2 ..*Dn) before a BatchNormalization Op. + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
momentum : float (default is 0.9)
+
Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum).
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size, C is the number of channels. Statistics are computed for every channel of C over N and D1 to Dn dimensions. For image data, input dimensions become (N x C x H x W). The op also accepts single dimension input of size N in which case C is assumed to be 1
+
scale (differentiable) : T
+
Scale tensor of shape (C).
+
B (differentiable) : T
+
Bias tensor of shape (C).
+
mean (differentiable) : T
+
running (training) or estimated (testing) mean tensor of shape (C).
+
var (differentiable) : T
+
running (training) or estimated (testing) variance tensor of shape (C).
+
+ +#### Outputs (1 - 5) + +
+
Y (differentiable) : T
+
The output tensor of the same shape as X
+
mean (optional, non-differentiable) : T
+
The running mean after the BatchNormalization operator.
+
var (optional, non-differentiable) : T
+
The running variance after the BatchNormalization operator.
+
saved_mean (optional, non-differentiable) : T
+
Saved mean used during training to speed up gradient computation.
+
saved_var (optional, non-differentiable) : T
+
Saved variance used during training to speed up gradient computation.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Cast-9** + + The operator casts the elements of a given input tensor to a data type + specified by the 'to' argument and returns an output tensor of the same size in + the converted type. The 'to' argument must be one of the data types specified + in the 'DataType' enum field in the TensorProto message. + + Casting from string tensor in plain (e.g., "3.14" and "1000") and scientific numeric representations + (e.g., "1e-5" and "1E8") to float types is supported. For example, converting string "100.5" to an integer may + yield result 100. There are some string literals reserved for special floating-point values; + "+INF" (and "INF"), "-INF", and "NaN" are positive infinity, negative infinity, and not-a-number, respectively. + Any string which can exactly match "+INF" in a case-insensitive way would be mapped to positive infinite. Similarly, + this case-insensitive rule is applied to "INF" and "NaN". When casting from numeric tensors + to string tensors, plain floating-point representation (such as "314.15926") would be used. + Converting non-numerical-literal string such as "Hello World!" is an undefined behavior. Cases + of converting string representing floating-point arithmetic value, such as "2.718", to INT is an undefined behavior. + + Conversion from a numerical type to any numerical type is always allowed. + User must be aware of precision loss and value change caused by range difference between two types. + For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting + an integer 36 to Boolean may produce 1 because we truncate bits which can't be stored in the targeted type. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
to : int (required)
+
The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
+
+ +#### Inputs + +
+
input : T1
+
Input tensor to be cast.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor with the same shape as input with type specified by the 'to' argument
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **Compress-9** + + Selects slices from an input tensor along a given axis where condition evaluates to True for each axis index. + In case axis is not provided, input is flattened before elements are selected. + Compress behaves like numpy.compress: https://docs.scipy.org/doc/numpy/reference/generated/numpy.compress.html + + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
(Optional) Axis along which to take slices. If not specified, input is flattened before elements being selected.
+
+ +#### Inputs + +
+
input : T
+
Tensor of rank r >= 1.
+
condition : T1
+
Rank 1 tensor of booleans to indicate which slices or data elements to be selected. Its length can be less than the input length alone the axis or the flattened input size if axis is not specified. In such cases data slices or elements exceeding the condition length are discarded.
+
+ +#### Outputs + +
+
output : T
+
Tensor of rank r if axis is specified. Otherwise output is a Tensor of rank 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
T1 : tensor(bool)
+
Constrain to boolean tensors.
+
+ +### **Constant-9** + + A constant tensor. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
value : tensor (required)
+
The value for the elements of the output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **ConstantOfShape-9** + + Generate a tensor with given value and shape. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
value : tensor
+
(Optional) The value of the output elements.Should be a one-element tensor. If not specified, it defaults to a tensor of value 0 and datatype float32
+
+ +#### Inputs + +
+
input : T1
+
1D tensor. The shape of the expected output tensor. If empty tensor is given, the output would be a scalar. All values must be >= 0.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of shape specified by 'input'.If attribute 'value' is specified, the value and datatype of the output tensor is taken from 'value'.If attribute 'value' is not specified, the value in the output defaults to 0, and the datatype defaults to float32.
+
+ +#### Type Constraints + +
+
T1 : tensor(int64)
+
Constrain input types.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool)
+
Constrain output types to be numerics.
+
+ +### **Cosh-9** + + Calculates the hyperbolic cosine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic cosine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Erf-9** + + Computes the error function of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
input : T
+
Input tensor
+
+ +#### Outputs + +
+
output : T
+
The error function of the input tensor computed element-wise. It has the same shape and type of the input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **EyeLike-9** + + Generate a 2D tensor (matrix) with ones on the diagonal and zeros everywhere else. Only 2D + tensors are supported, i.e. input T1 must be of rank 2. The shape of the output tensor is the + same as the input tensor. The data type can be specified by the 'dtype' argument. If + 'dtype' is not specified, then the type of input tensor is used. By default, the main diagonal + is populated with ones, but attribute 'k' can be used to populate upper or lower diagonals. + The 'dtype' argument must be one of the data types specified in the 'DataType' enum field in the + TensorProto message and be valid as an output type. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int
+
(Optional) The data type for the elements of the output tensor. If not specified,the data type of the input tensor T1 is used. If input tensor T1 is also notspecified, then type defaults to 'float'.
+
k : int (default is 0)
+
(Optional) Index of the diagonal to be populated with ones. Default is 0. If T2 is the output, this op sets T2[i, i+k] = 1. k = 0 populates the main diagonal, k > 0 populates an upper diagonal, and k < 0 populates a lower diagonal.
+
+ +#### Inputs + +
+
input : T1
+
2D input tensor to copy shape, and optionally, type information from.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor, same shape as input tensor T1.
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool)
+
Constrain input types. Strings and complex are not supported.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool)
+
Constrain output types. Strings and complex are not supported.
+
+ +### **Flatten-9** + + Flattens the input tensor into a 2D matrix. If input tensor has shape + (d_0, d_1, ... d_n) then the output will have shape + (d_0 X d_1 ... d_(axis-1), d_axis X d_(axis+1) ... X dn). + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [0, R], where R is the rank of the input tensor. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 ... d_n), where the shape of the input tensor is (d_0, d_1, ... d_n).
+
+ +#### Inputs + +
+
input : T
+
A tensor of rank >= axis.
+
+ +#### Outputs + +
+
output : T
+
A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output to all tensor types.
+
+ +### **Gemm-9** + + General Matrix multiplication: + https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3 + + A' = transpose(A) if transA else A + + B' = transpose(B) if transB else B + + Compute Y = alpha * A' * B' + beta * C, where input tensor A has shape (M, K) or (K, M), + input tensor B has shape (K, N) or (N, K), input tensor C is broadcastable to shape (M, N), + and output tensor Y has shape (M, N). A will be transposed before doing the + computation if attribute transA is non-zero, same for B and transB. + This operator supports **unidirectional broadcasting** (tensor C should be unidirectional broadcastable to tensor A * B); for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Scalar multiplier for the product of input tensors A * B.
+
beta : float (default is 1.0)
+
Scalar multiplier for input tensor C.
+
transA : int (default is 0)
+
Whether A should be transposed
+
transB : int (default is 0)
+
Whether B should be transposed
+
+ +#### Inputs + +
+
A : T
+
Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
+
B : T
+
Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
+
C : T
+
Input tensor C. The shape of C should be unidirectional broadcastable to (M, N).
+
+ +#### Outputs + +
+
Y : T
+
Output tensor of shape (M, N).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64)
+
Constrain input and output types to float/int tensors.
+
+ +### **Greater-9** + + Returns the tensor resulted from performing the `greater` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
First input operand for the logical operator.
+
B : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **IsNaN-9** + + Returns which elements of the input are NaN. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
X : T1
+
input
+
+ +#### Outputs + +
+
Y : T2
+
output
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
T2 : tensor(bool)
+
Constrain output types to boolean tensors.
+
+ +### **Less-9** + + Returns the tensor resulted from performing the `less` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
First input operand for the logical operator.
+
B : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **MatMul-9** + + Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html). + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
N-dimensional matrix A
+
B : T
+
N-dimensional matrix B
+
+ +#### Outputs + +
+
Y : T
+
Matrix multiply results from A * B
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64)
+
Constrain input and output types to float/int tensors.
+
+ +### **MaxUnpool-9** + + MaxUnpool essentially computes the partial inverse of the MaxPool op. + The input information to this op is typically the output information from a MaxPool op. The first + input tensor X is the tensor that needs to be unpooled, which is typically the pooled tensor (first output) + from MaxPool. The second input tensor, I, contains the indices to the (locally maximal) elements corresponding + to the elements in the first input tensor X. Input tensor I is typically the second output of the MaxPool op. + The third (optional) input is a tensor that specifies the output size of the unpooling operation. + + MaxUnpool is intended to do 'partial' inverse of the MaxPool op. 'Partial' because all the non-maximal + values from the original input to MaxPool are set to zero in the output of the MaxUnpool op. Pooling + the result of an unpooling operation should give back the original input to the unpooling op. + + MaxUnpool can produce the same output size for several input sizes, which makes unpooling op ambiguous. + The third input argument, output_size, is meant to disambiguate the op and produce output tensor of + known/predictable size. + + In addition to the inputs, MaxUnpool takes three attributes, namely kernel_shape, strides, and pads, + which define the exact unpooling op. The attributes typically have the same values as the corresponding + pooling op that the unpooling op is trying to invert. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X : T1
+
Input data tensor that has to be unpooled. This tensor is typically the first output of the MaxPool op.Dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non-image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
I : T2
+
Input data tensor containing the indices corresponding to elements in the first input tensor X.This tensor is typically the second output of the MaxPool op.Dimensions must be the same as input tensor X. The indices are linear, i.e. computed considering the tensor as flattened 1-D tensor, assuming row-major storage. Also, the linear indices should not consider padding. So the values in indices are in the range [0, N x C x D1 x ... x Dn).
+
output_shape (optional) : T2
+
The shape of the output can be explicitly set which will cause pads values to be auto generated. If 'output_shape' is specified, 'pads' values are ignored.
+
+ +#### Outputs + +
+
output : T1
+
Output data tensor that contains the result of the unpooling.
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T2 : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **MeanVarianceNormalization-9** + + A MeanVarianceNormalization Function: Perform mean variance normalization + on the input tensor X using formula:
``` (X-EX)/sqrt(E(X-EX)^2) ``` + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints (default is ['0', '2', '3'])
+
A list of integers, along which to reduce. The default is to calculate along axes [0,2,3] for calculating mean and variance along each channel. Two variables with the same C-coordinate are associated with the same mean and variance.
+
+ +#### Inputs + +
+
X : T
+
Input tensor
+
+ +#### Outputs + +
+
Y : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **NonZero-9** + + Returns the indices of the elements that are non-zero + (in row-major order - by dimension). + NonZero behaves similar to numpy.nonzero: + https://docs.scipy.org/doc/numpy/reference/generated/numpy.nonzero.html, + but for scalar input, NonZero produces output shape (0, N) instead of (1, N), which is different from Numpy's behavior. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
input
+
+ +#### Outputs + +
+
Y : tensor(int64)
+
output
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to all tensor types.
+
+ +### **OneHot-9** + + Produces a one-hot tensor based on inputs. + The locations represented by the index values in the 'indices' input tensor will have 'on_value' + and the other locations will have 'off_value' in the output tensor, where 'on_value' and 'off_value' + are specified as part of required input argument 'values', which is a two-element tensor of format + [off_value, on_value]. The rank of the output tensor will be one greater than the rank of the + input tensor. The additional dimension is for one-hot representation. The additional dimension will + be inserted at the position specified by 'axis'. If 'axis' is not specified then then additional + dimension will be inserted as the innermost dimension, i.e. axis=-1. The size of the additional + dimension is specified by required scalar input 'depth'. The type of the output tensor is the same + as the type of the 'values' input. Any entries in the 'indices' input tensor with values outside + the range [0, depth) will result in one-hot representation with all 'off_value' values in the + output tensor. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
(Optional) Axis along which one-hot representation in added. Default: axis=-1. axis=-1 means that the additional dimension will be inserted as the innermost/last dimension in the output tensor.
+
+ +#### Inputs + +
+
indices : T1
+
Input tensor containing indices. The values must be non-negative integers. Any entries in the 'indices' input tensor with values outside the range [0, depth) will result in one-hot representation with all 'off_value' values in the output tensor.In case 'indices' is of non-integer type, the values will be casted to int64 before use.
+
depth : T2
+
Scalar or rank 1 tensor containing exactly one element, specifying the number of classes in one-hot tensor. This is also the size of the one-hot dimension (specified by 'axis' attribute) added on in the output tensor. The values in the 'indices' input tensor are expected to be in the range [0, depth). In case 'depth' is of non-integer type, it will be casted to int64 before use.
+
values : T3
+
Rank 1 tensor containing exactly two elements, in the format [off_value, on_value], where 'on_value' is the value used for filling locations specified in 'indices' input tensor, and 'off_value' is the value used for filling locations other than those specified in 'indices' input tensor.
+
+ +#### Outputs + +
+
output : T3
+
Tensor of rank one greater than input tensor 'indices', i.e. rank(output) = rank(indices) + 1. The data type for the elements of the output tensor is the same as the type of input 'values' is used.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input to only numeric types.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input to only numeric types.
+
T3 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type.
+
+ +### **PRelu-9** + + PRelu takes input data (Tensor) and slope tensor as input, and produces one + output data (Tensor) where the function `f(x) = slope * x for x < 0`, + `f(x) = x for x >= 0`., is applied to the data tensor elementwise. + + This operator supports **unidirectional broadcasting** (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
slope (differentiable) : T
+
Slope tensor. The shape of slope can be smaller than first input X; if so, its shape must be unidirectional broadcastable to X
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor (same size as X)
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64)
+
Constrain input and output types to float/int tensors.
+
+ +### **Scan-9** + + Scan can be used to iterate over one or more scan_input tensors, + constructing zero or more scan_output tensors. It combines ideas from general recurrences, + functional programming constructs such as scan, fold, map, and zip, and is intended to enable + generalizations of RNN-like constructs for sequence-to-sequence processing. + Other tensors (referred to as state_variables here) can be used to carry a state + when iterating from one element to another (similar to hidden-state in RNNs, also referred + to as loop-carried dependences in the context of loops). + Many common usages involve a single scan_input tensor (where functionality + similar to scan, fold and map can be obtained). When more than one scan_input is used, + a behavior similar to zip is obtained. + + The attribute body must be a graph, specifying the computation to be performed in + every iteration. It takes as input the current values of the state_variables and + the current iterated element of the scan_inputs. It must return the (updated) values + of the state_variables and zero or more scan_output_element tensors. The values of the + scan_output_element tensors are concatenated over all the iterations to produce the + scan_output values of the scan construct (similar to the concatenated intermediate + hidden-state values of RNN-like constructs). All the output tensors (state_variables as + well as scan_output_element tensors) are required to have the same shape in each iteration + of the loop (a restriction imposed to enable efficient memory allocation). + + Note that the iterated element passed to the body subgraph does not have a sequence + axis. It will have a rank one less than the rank of the corresponding scan_input. + + The scan operation returns the final values of the state_variables as well as the + scan_outputs. + + The optional attribute scan_input_directions specifies the direction (forward or backward) + for each scan input. If this attribute is omitted, all sequences are scanned in the forward + direction. A bidirectional scan may be performed by specifying the same tensor input twice + in the scan_inputs, once with a forward direction, and once with a backward direction. + + The scan_output of the operation is produced by concatenating the scan_output_element + values produced by the body in each iteration. The optional attribute scan_output_directions + specifies the direction in which scan_output is constructed (by appending or prepending the + scan_output_element to scan_output in each iteration) for each scan_output. If this attribute + is omitted, the scan_output_element is appended to the scan_output in each iteration. + + The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. + If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the + batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. + Note that scanning a non-zero axis may be less efficient than scanning axis zero. + + The optional attribute scan_output_axes specifies the axis along which the scan_outputs + are accumulated for each scan_output. For example, if axis 1 is the time axis (to be + scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis + value of 1. + + Note that because of the ONNX restriction that only the last parameter of an operator can + be variadic, the initial-states and scan-inputs are listed together as one input parameter. + Similarly, the final-states and scan-outputs are listed together as one output parameter. + The attribute num_scan_inputs indicates the number M of scan-inputs. + + The behavior of + + Scan < + num_scan_inputs = m, + body = loop-body, + scan_input_axes = [axis_1, ..., axis_m] + > (init_1, ..., init_n, scan_1, ..., scan_m) + + is equivalent to the following pseudo-code: + + // scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i + // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. + sequence_length = scan_1.shape[axis_1]; + + // initialize state-variables + st_1 = init_1; ... st_n = init_n; + // initialize scan-output variables: [] denotes an empty tensor + scan_out_1 = []; ...; scan_out_k = []; + // identify number of iterations: + + // execute loop + for (int t = 0; t < sequence_length; ++t) { + // generate the scan-input elements: the notation T[t] indicates the sub-tensor + // of rank one less than T obtained by indexing T at position t along axis k. + si_1 = scan_1[t]; + ... ; + si_m = scan_m[t]; + // execute loop-body + st_1, ..., st_n, so_1, ..., so_k = loop-body(st_1, ..., st_n, si_1, ..., si_m) + // accumulate the scan-output elements + scan_out_1 = Concat(scan_out_1, so_1); ... ; scan_out_k = Concat(scan_out_k, so_k); + } + + return st_1, ..., st_n, scan_out_1, ..., scan_out_k; + + *Sample usage: Encoding RNN using a Scan* + + The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, + recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can + be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes + %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these + values are computed in the outer graph, they need to be passed in as extra state_variables. + + graph rnn-encoding { + %H_0 = ... + %X = ... + %Y_h, %Y = Scan[body = , num_scan_inputs=1](%H_0, %X) + return %Y, %Y_h + } + + graph rnn-cell-1 ( + %H_tminus1[FLOAT, tensor] + %X_t[FLOAT, tensor] + ) { + %Wi = ... + %Ri = ... + %Wbi = ... + %Rbi = ... + %t1 = X_t * (Wi^T) + %t2 = H_tminus1*(Ri^T) + %t3 = Add(%t1, %t2) + %t4 = Add(%t3, %Wbi) + %t5 = Add(%t4, %Rbi) + %Ht = Tanh(%t5) + %Accumulate = Identity(%Ht) + return %Ht, %Accumulate + } + + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has N+M inputs: (loop state variables..., scan_input_elts...). It has N+K outputs: (loop state variables..., scan_output_elts...). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
+
num_scan_inputs : int (required)
+
An attribute specifying the number of scan_inputs M.
+
scan_input_axes : list of ints
+
An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input.
+
scan_input_directions : list of ints
+
An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
+
scan_output_axes : list of ints
+
An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output.
+
scan_output_directions : list of ints
+
An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
+
+ +#### Inputs (1 - ∞) + +
+
initial_state_and_scan_inputs (variadic, heterogeneous) : V
+
Initial values of the loop's N state variables followed by M scan_inputs
+
+ +#### Outputs (1 - ∞) + +
+
final_state_and_scan_outputs (variadic, heterogeneous) : V
+
Final values of the loop's N state variables followed by K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
All Tensor types
+
+ +### **Scatter-9** + + Given `data`, `updates` and `indices` input tensors of rank r >= 1, write the values provided by `updates` + into the first input, `data`, along `axis` dimension of `data` (by default outer-most one as axis=0) at corresponding `indices`. + For each entry in `updates`, the target index in `data` is specified by corresponding entry in `indices` + for dimension = axis, and index in source for dimension != axis. For instance, in a 2-D tensor case, + data[indices[i][j]][j] = updates[i][j] if axis = 0, or data[i][indices[i][j]] = updates[i][j] if axis = 1, + where i and j are loop counters from 0 up to the respective size in `updates` - 1. + Example 1: + data = [ + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + ] + indices = [ + [1, 0, 2], + [0, 2, 1], + ] + updates = [ + [1.0, 1.1, 1.2], + [2.0, 2.1, 2.2], + ] + output = [ + [2.0, 1.1, 0.0] + [1.0, 0.0, 2.2] + [0.0, 2.1, 1.2] + ] + Example 2: + data = [[1.0, 2.0, 3.0, 4.0, 5.0]] + indices = [[1, 3]] + updates = [[1.1, 2.1]] + axis = 1 + output = [[1.0, 1.1, 3.0, 2.1, 5.0]] + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1]
+
+ +#### Inputs + +
+
data : T
+
Tensor of rank r >= 1.
+
indices : Tind
+
Tensor of int32/int64 indices, of r >= 1 (same rank as input).
+
updates : T
+
Tensor of rank r >=1 (same rank and shape as indices)
+
+ +#### Outputs + +
+
output : T
+
Tensor of rank r >= 1 (same rank as input).
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input and output types can be of any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **Shrink-9** + + Shrink takes one input data (Tensor) and produces one Tensor output, + having same datatype and shape with input. It has two attributes, lambd and + bias. The formula of this operator is: If x < -lambd, y = x + bias; + If x > lambd, y = x - bias; Otherwise, y = 0. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
bias : float (default is 0.0)
+
The bias value added to output. Default is 0.
+
lambd : float (default is 0.5)
+
The lambd value for the Shrink formulation. Default is 0.5.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
The input data as Tensor.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input to only numeric types.
+
+ +### **Sign-9** + + Calculate the sign of the given input tensor element-wise. + If input > 0, output 1. if input < 0, output -1. if input == 0, output 0. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
input : T
+
Input tensor
+
+ +#### Outputs + +
+
output : T
+
The sign of the input tensor computed element-wise. It has the same shape and type of the input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **Sinh-9** + + Calculates the hyperbolic sine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic sine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **TfIdfVectorizer-9** + + This transform extracts n-grams from the input sequence and save them as a vector. Input can + be either a 1-D or 2-D tensor. For 1-D input, output is the n-gram representation of that input. + For 2-D input, the output is also a 2-D tensor whose i-th row is the n-gram representation of the i-th input row. + More specifically, if input shape is [C], the corresponding output shape would be [max(ngram_indexes) + 1]. + If input shape is [N, C], this operator produces a [N, max(ngram_indexes) + 1]-tensor. + + In contrast to standard n-gram extraction, here, the indexes of extracting an n-gram from the original + sequence are not necessarily consecutive numbers. The discontinuity between indexes are controlled by the number of skips. + If the number of skips is 2, we should skip two tokens when scanning through the original sequence. + Let's consider an example. Assume that input sequence is [94, 17, 36, 12, 28] and the number of skips is 2. + The associated 2-grams are [94, 12] and [17, 28] respectively indexed by [0, 3] and [1, 4]. + If the number of skips becomes 0, the 2-grams generated are [94, 17], [17, 36], [36, 12], [12, 28] + indexed by [0, 1], [1, 2], [2, 3], [3, 4], respectively. + + The output vector (denoted by Y) stores the count of each n-gram; + Y[ngram_indexes[i]] indicates the times that the i-th n-gram is found. The attribute ngram_indexes is used to determine the mapping + between index i and the corresponding n-gram's output coordinate. If pool_int64s is [94, 17, 17, 36], ngram_indexes is [1, 0], + ngram_counts=[0, 0], then the Y[0] (first element in Y) and Y[1] (second element in Y) are the counts of [17, 36] and [94, 17], + respectively. An n-gram which cannot be found in pool_strings/pool_int64s should be ignored and has no effect on the output. + Note that we may consider all skips up to S when generating the n-grams. + + The examples used above are true if mode is "TF". If mode is "IDF", all the counts larger than 1 would be truncated to 1 and + the i-th element in weights would be used to scale (by multiplication) the count of the i-th n-gram in pool. If mode is "TFIDF", + this operator first computes the counts of all n-grams and then scale them by the associated values in the weights attribute. + + Only one of pool_strings and pool_int64s can be set. If pool_int64s is set, the input should be an integer tensor. + If pool_strings is set, the input must be a string tensor. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
max_gram_length : int (required)
+
Maximum n-gram length. If this value is 3, 3-grams will be used to generate the output.
+
max_skip_count : int (required)
+
Maximum number of items (integers/strings) to be skipped when constructing an n-gram from X. If max_skip_count=1, min_gram_length=2, max_gram_length=3, this operator may generate 2-grams with skip_count=0 and skip_count=1, and 3-grams with skip_count=0 and skip_count=1
+
min_gram_length : int (required)
+
Minimum n-gram length. If this value is 2 and max_gram_length is 3, output may contain counts of 2-grams and 3-grams.
+
mode : string (required)
+
The weighting criteria. It can be one of "TF" (term frequency), "IDF" (inverse document frequency), and "TFIDF" (the combination of TF and IDF)
+
ngram_counts : list of ints (required)
+
The starting indexes of 1-grams, 2-grams, and so on in pool. It is useful when determining the boundary between two consecutive collections of n-grams. For example, if ngram_counts is [0, 17, 36], the first index (zero-based) of 1-gram/2-gram/3-gram in pool are 0/17/36. This format is essentially identical to CSR (or CSC) sparse matrix format, and we choose to use this due to its popularity.
+
ngram_indexes : list of ints (required)
+
list of int64s (type: AttributeProto::INTS). This list is parallel to the specified 'pool_*' attribute. The i-th element in ngram_indexes indicate the coordinate of the i-th n-gram in the output tensor.
+
pool_int64s : list of ints
+
List of int64 n-grams learned from the training set. Either this or pool_strings attributes must be present but not both. It's an 1-D tensor starting with the collections of all 1-grams and ending with the collections of n-grams. The i-th element in pool stores the n-gram that should be mapped to coordinate ngram_indexes[i] in the output vector.
+
pool_strings : list of strings
+
List of strings n-grams learned from the training set. Either this or pool_int64s attributes must be present but not both. It's an 1-D tensor starting with the collections of all 1-grams and ending with the collections of n-grams. The i-th element in pool stores the n-gram that should be mapped to coordinate ngram_indexes[i] in the output vector.
+
weights : list of floats
+
list of floats. This attribute stores the weight of each n-gram in pool. The i-th element in weights is the weight of the i-th n-gram in pool. Its length equals to the size of ngram_indexes. By default, weights is an all-one tensor.This attribute is used when mode is "IDF" or "TFIDF" to scale the associated word counts.
+
+ +#### Inputs + +
+
X (non-differentiable) : T
+
Input for n-gram extraction
+
+ +#### Outputs + +
+
Y (non-differentiable) : T1
+
Ngram results
+
+ +#### Type Constraints + +
+
T : tensor(string), tensor(int32), tensor(int64)
+
Input is either string UTF-8 or int32/int64
+
T1 : tensor(float)
+
1-D tensor of floats
+
+ +### **Upsample-9** + + Upsample the input tensor. + Each dimension value of the output tensor is: + output_dimension = floor(input_dimension * scale). + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is nearest)
+
Two interpolation modes: nearest (default), and linear (including bilinear, trilinear, etc)
+
+ +#### Inputs + +
+
X : T
+
N-D tensor
+
scales : tensor(float)
+
The scale array along each dimension. It takes value greater than or equal to 1. The number of elements of 'scales' should be the same as the rank of input 'X'.
+
+ +#### Outputs + +
+
Y : T
+
N-D tensor after resizing
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input 'X' and output 'Y' to all tensor types.
+
+ +### **Where-9** + + Return elements, either from X or Y, depending on condition. + Where behaves like + [numpy.where](https://docs.scipy.org/doc/numpy/reference/generated/numpy.where.html) + with three parameters. + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Inputs + +
+
condition (non-differentiable) : B
+
When True (nonzero), yield X, otherwise yield Y
+
X (differentiable) : T
+
values selected at indices where condition is True
+
Y (differentiable) : T
+
values selected at indices where condition is False
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of shape equal to the broadcasted shape of condition, X, and Y.
+
+ +#### Type Constraints + +
+
B : tensor(bool)
+
Constrain to boolean tensors.
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +## Version 10 of the default ONNX operator set +### **AveragePool-10** + + AveragePool consumes an input tensor X and applies average pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + average pooling consisting of computing the average on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled + + ``` + * pad_shape[i] is sum of pads along axis i + ``` + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i] + ``` + The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero). + + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
count_include_pad : int (default is 0)
+
Whether include pad pixels when calculating values for the edges. Default is 0, doesn't count include pad.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **ConvInteger-10** + + The integer convolution operator consumes an input tensor, its zero-point, a filter, and its zero-point, + and computes the output. The production MUST never overflow. The accumulation may overflow if and only if in 32 bits. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each axis.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into. default is 1.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input 'w'.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0.The value represent the number of pixels added to the beginning and end part of the corresponding axis.`pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number ofpixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`.This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaultsto 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each axis.
+
+ +#### Inputs (2 - 4) + +
+
x : T1
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
w : T2
+
The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
+
x_zero_point (optional) : T1
+
Zero point tensor for input 'x'. It's optional and default value is 0. It's a scalar, which means a per-tensor/layer quantization.
+
w_zero_point (optional) : T2
+
Zero point tensor for input 'w'. It's optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it's a 1-D tensor, its number of elements should be equal to the number of output channels (M)
+
+ +#### Outputs + +
+
y : T3
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8)
+
Constrain input x and its zero point data type to 8-bit integer tensor.
+
T2 : tensor(int8), tensor(uint8)
+
Constrain input w and its zero point data type to 8-bit integer tensor.
+
T3 : tensor(int32)
+
Constrain output y data type to 32-bit integer tensor.
+
+ +### **DequantizeLinear-10** + + The linear dequantization operator. It consumes a quantized tensor, a scale, a zero point to compute the full precision tensor. + The dequantization formula is y = (x - x_zero_point) * x_scale. 'x_scale' and 'x_zero_point' are both scalars. + 'x_zero_point' and 'x' must have same type. 'x' and 'y' must have same shape. In the case of dequantizing int32, + there's no zero point (zero point is supposed to be 0). + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Inputs (2 - 3) + +
+
x : T
+
N-D quantized input tensor to be de-quantized.
+
x_scale : tensor(float)
+
Scale for input 'x'. It's a scalar, which means a per-tensor/layer quantization.
+
x_zero_point (optional) : T
+
Zero point for input 'x'. It's a scalar, which means a per-tensor/layer quantization. It's optional. 0 is the default value when it's not specified.
+
+ +#### Outputs + +
+
y : tensor(float)
+
N-D full precision output tensor. It has same shape as input 'x'.
+
+ +#### Type Constraints + +
+
T : tensor(int8), tensor(uint8), tensor(int32)
+
Constrain 'x_zero_point' and 'x' to 8-bit/32-bit integer tensor.
+
+ +### **Dropout-10** + + Dropout takes one input floating tensor and produces two tensor outputs, + output (floating tensor) and mask (`Tensor`). Depending on whether it is + in test mode or not, the output Y will either be a random dropout, or a simple + copy of the input. Note that our implementation of Dropout does scaling in + the training phase, so during testing nothing needs to be done. + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
ratio : float (default is 0.5)
+
The ratio of random dropout
+
+ +#### Inputs + +
+
data : T
+
The input data as Tensor.
+
+ +#### Outputs (1 - 2) + +
+
output : T
+
The output.
+
mask (optional) : T1
+
The output mask.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(bool)
+
Constrain output mask types to boolean tensors.
+
+ +### **IsInf-10** + + Map infinity to true and other values to false. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
detect_negative : int (default is 1)
+
(Optional) Whether map negative infinity to true. Default to 1 so that negative infinity induces true. Set this attribute to 0 if negative infinity should be mapped to false.
+
detect_positive : int (default is 1)
+
(Optional) Whether map positive infinity to true. Default to 1 so that positive infinity induces true. Set this attribute to 0 if positive infinity should be mapped to false.
+
+ +#### Inputs + +
+
X (non-differentiable) : T1
+
input
+
+ +#### Outputs + +
+
Y (non-differentiable) : T2
+
output
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
T2 : tensor(bool)
+
Constrain output types to boolean tensors.
+
+ +### **MatMulInteger-10** + + Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html). + The production MUST never overflow. The accumulation may overflow if and only if in 32 bits. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Inputs (2 - 4) + +
+
A (non-differentiable) : T1
+
N-dimensional matrix A
+
B (non-differentiable) : T2
+
N-dimensional matrix B
+
a_zero_point (optional, non-differentiable) : T1
+
Zero point tensor for input 'A'. It's optional and default value is 0. It could be a scalar or N-D tensor. Scalar refers to per tensor quantization whereas N-D refers to per row quantization. If the input is 2D of shape [M, K] then zero point tensor may be an M element vector [zp_1, zp_2, ..., zp_M]. If the input is N-D tensor with shape [D1, D2, M, K] then zero point tensor may have shape [D1, D2, M, 1].
+
b_zero_point (optional, non-differentiable) : T2
+
Zero point tensor for input 'B'. It's optional and default value is 0. It could be a scalar or a N-D tensor, Scalar refers to per tensor quantization whereas N-D refers to per col quantization. If the input is 2D of shape [K, N] then zero point tensor may be an N element vector [zp_1, zp_2, ..., zp_N]. If the input is N-D tensor with shape [D1, D2, K, N] then zero point tensor may have shape [D1, D2, 1, N].
+
+ +#### Outputs + +
+
Y (non-differentiable) : T3
+
Matrix multiply results from A * B
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8)
+
Constrain input A data type to 8-bit integer tensor.
+
T2 : tensor(int8), tensor(uint8)
+
Constrain input B data type to 8-bit integer tensor.
+
T3 : tensor(int32)
+
Constrain output Y data type as 32-bit integer tensor.
+
+ +### **MaxPool-10** + + MaxPool consumes an input tensor X and applies max pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + max pooling consisting of computing the max on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled + + ``` + * pad_shape[i] is sum of pads along axis i + ``` + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i] + ``` + The output of each pooling window is maximum number of elements exclude pad. + + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
dilations : list of ints
+
Dilation value along each spatial axis of filter.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
storage_order : int (default is 0)
+
The storage order of the tensor. 0 is row major, and 1 is column major.
+
strides : list of ints
+
Stride along each spatial axis.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs (1 - 2) + +
+
Y : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
Indices (optional) : I
+
Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x ... x Dn).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **Mod-10** + + Performs element-wise binary modulus (with Numpy-style broadcasting support). + The sign of the remainder is the same as that of the Divisor. + + Mod operator can also behave like C fmod() or numpy.fmod. In this case, the sign of the remainder however, will be the same as the Dividend + (in contrast to integer mod). To force a behavior like numpy.fmod() an 'fmod' Attribute is provided. + This attribute is set to 0 by default causing the behavior to be like integer mod. + Setting this attribute to 1 causes the remainder to be calculated similar to that of numpy.fmod(). + + If the input type is floating point, then `fmod` attribute must be set to 1. + + In case of dividend being zero, the results will be platform dependent. + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
fmod : int (default is 0)
+
Whether the operator should behave like fmod (default=0 meaning it will do integer mods); Set this to 1 to force fmod treatment
+
+ +#### Inputs + +
+
A : T
+
Dividend tensor
+
B : T
+
Divisor tensor
+
+ +#### Outputs + +
+
C : T
+
Remainder tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **NonMaxSuppression-10** + + Filter out boxes that have high intersection-over-union (IOU) overlap with previously selected boxes. + Bounding boxes with score less than score_threshold are removed. Bounding box format is indicated by attribute center_point_box. + Boxes are suppressed if their IOU with a previously selected box is strictly greater than iou_threshold (i.e., boxes with IOU exactly equal to the threshold are kept). + Note that this algorithm is agnostic to where the origin is in the coordinate system and more generally is invariant to + orthogonal transformations and translations of the coordinate system; thus translating or reflections of the coordinate system + result in the same boxes being selected by the algorithm. + The selected_indices output is a set of integers indexing into the input collection of bounding boxes representing the selected boxes. + The bounding box coordinates corresponding to the selected indices can then be obtained using the Gather or GatherND operation. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
center_point_box : int (default is 0)
+
Integer indicate the format of the box data. The default is 0. 0 - the box data is supplied as [y1, x1, y2, x2] where (y1, x1) and (y2, x2) are the coordinates of any diagonal pair of box corners and the coordinates can be provided as normalized (i.e., lying in the interval [0, 1]) or absolute. Mostly used for TF models. 1 - the box data is supplied as [x_center, y_center, width, height]. Mostly used for Pytorch models.
+
+ +#### Inputs (2 - 5) + +
+
boxes : tensor(float)
+
An input tensor with shape [num_batches, spatial_dimension, 4]. The single box data format is indicated by center_point_box.
+
scores : tensor(float)
+
An input tensor with shape [num_batches, num_classes, spatial_dimension]
+
max_output_boxes_per_class (optional) : tensor(int64)
+
Integer representing the maximum number of boxes to be selected per batch per class. It is a scalar. Default to 0, which means no output.
+
iou_threshold (optional) : tensor(float)
+
Float representing the threshold for deciding whether boxes overlap too much with respect to IOU. It is scalar. Value range [0, 1]. Default to 0.
+
score_threshold (optional) : tensor(float)
+
Float representing the threshold for deciding when to remove boxes based on score. It is a scalar.
+
+ +#### Outputs + +
+
selected_indices : tensor(int64)
+
selected indices from the boxes tensor. [num_selected_indices, 3], the selected index format is [batch_index, class_index, box_index].
+
+ +#### Type Constraints + + +### **QLinearConv-10** + + The convolution operator consumes a quantized input tensor, its scale and zero point, + a quantized filter, its scale and zero point, and output's scale and zero point, + and computes the quantized output. Each scale and zero-point pair must have same shape. + It means they must be either scalars (per tensor) or 1-D tensors (per output channel). + Each input or output and its related zero point must have same type. + When bias is present it must be quantized using scale = input scale * weight scale and + zero point as 0. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each spatial axis.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into. default is 1.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input 'w'.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0.The value represent the number of pixels added to the beginning and end part of the corresponding axis.`pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number ofpixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`.This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaultsto 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs (8 - 9) + +
+
x : T1
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
x_scale : tensor(float)
+
Scale tensor for input 'x'. It's a scalar, which means a per-tensor/layer quantization.
+
x_zero_point : T1
+
Zero point tensor for input 'x'. It's a scalar, which means a per-tensor/layer quantization.
+
w : T2
+
The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
+
w_scale : tensor(float)
+
Scale tensor for input 'w'. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it's a 1-D tensor, its number of elements should be equal to the number of output channels (M).
+
w_zero_point : T2
+
Zero point tensor for input 'w'. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it's a 1-D tensor, its number of elements should be equal to the number of output channels (M).
+
y_scale : tensor(float)
+
Scale tensor for output 'y'. It's a scalar, which means a per-tensor/layer quantization.
+
y_zero_point : T3
+
Zero point tensor for output 'y'. It's a scalar, which means a per-tensor/layer quantization.
+
B (optional) : T4
+
Optional 1D bias to be added to the convolution, has size of M. Bias must be quantized using scale = x_scale * w_scale and zero_point = 0
+
+ +#### Outputs + +
+
y : T3
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8)
+
Constrain input type to 8-bit integer tensor.
+
T2 : tensor(int8), tensor(uint8)
+
Constrain filter type to 8-bit integer tensor.
+
T3 : tensor(int8), tensor(uint8)
+
Constrain output type to 8-bit integer tensor.
+
T4 : tensor(int32)
+
Constrain bias type to 32-bit integer tensor.
+
+ +### **QLinearMatMul-10** + + Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html). + It consumes two quantized input tensors, their scales and zero points, scale and zero point of output, + and computes the quantized output. The quantization formula is y = saturate((x / y_scale) + y_zero_point). + For (x / y_scale), it is rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details. + Scale and zero point must have same shape. They must be either scalar (per tensor) or N-D tensor + (per row for 'a' and per column for 'b'). Scalar refers to per tensor quantization whereas N-D refers to per row + or per column quantization. If the input is 2D of shape [M, K] then zero point and scale tensor may be + an M element vector [v_1, v_2, ..., v_M] for per row quantization and K element vector of shape [v_1, v_2, ..., v_K] + for per column quantization. If the input is N-D tensor with shape [D1, D2, M, K] then zero point and scale tensor may + have shape [D1, D2, M, 1] for per row quantization and shape [D1, D2, 1, K] for per column quantization. + Production must never overflow, and accumulation may overflow if and only if in 32 bits. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Inputs + +
+
a (non-differentiable) : T1
+
N-dimensional quantized matrix a
+
a_scale (non-differentiable) : tensor(float)
+
scale of quantized input a
+
a_zero_point (non-differentiable) : T1
+
zero point of quantized input a
+
b (non-differentiable) : T2
+
N-dimensional quantized matrix b
+
b_scale (non-differentiable) : tensor(float)
+
scale of quantized input b
+
b_zero_point (non-differentiable) : T2
+
zero point of quantized input b
+
y_scale (non-differentiable) : tensor(float)
+
scale of quantized output y
+
y_zero_point (non-differentiable) : T3
+
zero point of quantized output y
+
+ +#### Outputs + +
+
y (non-differentiable) : T3
+
Quantized matrix multiply results from a * b
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8)
+
Constrain input a and its zero point data type to 8-bit integer tensor.
+
T2 : tensor(int8), tensor(uint8)
+
Constrain input b and its zero point data type to 8-bit integer tensor.
+
T3 : tensor(int8), tensor(uint8)
+
Constrain output y and its zero point data type to 8-bit integer tensor.
+
+ +### **QuantizeLinear-10** + + The linear per-tensor/layer quantization operator. It consumes a high precision tensor, a scale, a zero point to compute the low precision / quantized tensor. + The quantization formula is y = saturate ((x / y_scale) + y_zero_point). For saturation, it saturates to [0, 255] if it's uint8, or [-128, 127] if it's int8. + For (x / y_scale), it's rounding to the nearest even. Refer to https://en.wikipedia.org/wiki/Rounding for details. 'y_zero_point' and 'y' must have same type. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Inputs (2 - 3) + +
+
x : T1
+
N-D full precision Input tensor to be quantized.
+
y_scale : tensor(float)
+
Scale for doing quantization to get 'y'. It's a scalar, which means a per-tensor/layer quantization.
+
y_zero_point (optional) : T2
+
Zero point for doing quantization to get 'y'. It's a scalar, which means a per-tensor/layer quantization. Default value is uint8 typed 0 if it's not specified.
+
+ +#### Outputs + +
+
y : T2
+
N-D quantized output tensor. It has same shape as input 'x'.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(int32)
+
Constrain 'x' to float or int32 tensor.
+
T2 : tensor(int8), tensor(uint8)
+
Constrain 'y_zero_point' and 'y' to 8-bit integer tensor.
+
+ +### **Resize-10** + + Resize the input tensor. + Each dimension value of the output tensor is: + output_dimension = floor(input_dimension * scale). + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is nearest)
+
Two interpolation modes: nearest (default), and linear (including bilinear, trilinear, etc)
+
+ +#### Inputs + +
+
X : T
+
N-D tensor
+
scales : tensor(float)
+
The scale array along each dimension. It takes value greater than 0. If it's less than 1, it's sampling down, otherwise, it's upsampling. The number of elements of 'scales' should be the same as the rank of input 'X'.
+
+ +#### Outputs + +
+
Y : T
+
N-D tensor after resizing
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input 'X' and output 'Y' to all tensor types.
+
+ +### **ReverseSequence-10** + + Reverse batch of sequences having different lengths specified by `sequence_lens`. + + For each slice i iterating on batch axis, the operator reverses the first sequence_lens[i] elements on time axis, + and copies elements whose index's beyond sequence_lens[i] to the output. So the output slice i contains reversed + sequences on the first sequence_lens[i] elements, then have original values copied for the other elements. + + Example 1: + input = [[0.0, 4.0, 8.0, 12.0], + [1.0, 5.0, 9.0, 13.0], + [2.0, 6.0, 10.0, 14.0], + [3.0, 7.0, 11.0, 15.0]] + sequence_lens = [4, 3, 2, 1] + time_axis = 0 + batch_axis = 1 + + output = [[3.0, 6.0, 9.0, 12.0], + [2.0, 5.0, 8.0, 13.0], + [1.0, 4.0, 10.0, 14.0], + [0.0, 7.0, 11.0, 15.0]] + + Example 2: + input = [[0.0, 1.0, 2.0, 3.0 ], + [4.0, 5.0, 6.0, 7.0 ], + [8.0, 9.0, 10.0, 11.0], + [12.0, 13.0, 14.0, 15.0]] + sequence_lens = [1, 2, 3, 4] + time_axis = 1 + batch_axis = 0 + + output = [[0.0, 1.0, 2.0, 3.0 ], + [5.0, 4.0, 6.0, 7.0 ], + [10.0, 9.0, 8.0, 11.0], + [15.0, 14.0, 13.0, 12.0]] + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
batch_axis : int (default is 1)
+
(Optional) Specify which axis is batch axis. Must be one of 1 (default), or 0.
+
time_axis : int (default is 0)
+
(Optional) Specify which axis is time axis. Must be one of 0 (default), or 1.
+
+ +#### Inputs + +
+
input : T
+
Tensor of rank r >= 2.
+
sequence_lens : tensor(int64)
+
Tensor specifying lengths of the sequences in a batch. It has shape `[batch_size]`.
+
+ +#### Outputs + +
+
Y : T
+
Tensor with same shape of input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input and output types can be of any tensor type.
+
+ +### **RoiAlign-10** + + Region of Interest (RoI) align operation described in the + [Mask R-CNN paper](https://arxiv.org/abs/1703.06870). + RoiAlign consumes an input tensor X and region of interests (rois) + to apply pooling across each RoI; it produces a 4-D tensor of shape + (num_rois, C, output_height, output_width). + + RoiAlign is proposed to avoid the misalignment by removing + quantizations while converting from original image into feature + map and from feature map into RoI feature; in each ROI bin, + the value of the sampled locations are computed directly + through bilinear interpolation. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is avg)
+
The pooling method. Two modes are supported: 'avg' and 'max'. Default is 'avg'.
+
output_height : int (default is 1)
+
default 1; Pooled output Y's height.
+
output_width : int (default is 1)
+
default 1; Pooled output Y's width.
+
sampling_ratio : int (default is 0)
+
Number of sampling points in the interpolation grid used to compute the output value of each pooled output bin. If > 0, then exactly sampling_ratio x sampling_ratio grid points are used. If == 0, then an adaptive number of grid points are used (computed as ceil(roi_width / output_width), and likewise for height). Default is 0.
+
spatial_scale : float (default is 1.0)
+
Multiplicative spatial scale factor to translate ROI coordinates from their input spatial scale to the scale used when pooling, i.e., spatial scale of the input feature map X relative to the input image. E.g.; default is 1.0f.
+
+ +#### Inputs + +
+
X : T1
+
Input data tensor from the previous operator; 4-D feature map of shape (N, C, H, W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
+
rois : T1
+
RoIs (Regions of Interest) to pool over; rois is 2-D input of shape (num_rois, 4) given as [[x1, y1, x2, y2], ...]. The RoIs' coordinates are in the coordinate system of the input image. Each coordinate set has a 1:1 correspondence with the 'batch_indices' input.
+
batch_indices : T2
+
1-D tensor of shape (num_rois,) with each element denoting the index of the corresponding image in the batch.
+
+ +#### Outputs + +
+
Y : T1
+
RoI pooled output, 4-D tensor of shape (num_rois, C, output_height, output_width). The r-th batch element Y[r-1] is a pooled feature map corresponding to the r-th RoI X[r-1].
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double)
+
Constrain types to float tensors.
+
T2 : tensor(int64)
+
Constrain types to int tensors.
+
+ +### **Slice-10** + + Produces a slice of the input tensor along multiple axes. Similar to numpy: + https://numpy.org/doc/stable/reference/routines.indexing.html + Slices uses `starts`, `ends`, `axes` and `steps` inputs to specify the start and end + dimension and step for each axis in the list of axes, it uses this information to + slice the input `data` tensor. If a negative value is passed for any of the + start or end indices, it represent number of elements before the end of that + dimension. If the value passed to start or end is larger than the `n` (the + number of elements in this dimension), it represents `n`. For slicing to the + end of a dimension with unknown size, it is recommended to pass in `INT_MAX`. + If a negative value is passed for step, it represents slicing backward. + If `axes` are omitted, they are set to `[0, ..., ndim-1]`. + If `steps` are omitted, they are set to `[1, ..., 1]` of length `len(starts)` + Example 1: + data = [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + axes = [0, 1] + starts = [1, 0] + ends = [2, 3] + steps = [1, 2] + result = [ + [5, 7], + ] + Example 2: + data = [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + starts = [0, 1] + ends = [-1, 1000] + result = [ + [2, 3, 4], + ] + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Inputs (3 - 5) + +
+
data : T
+
Tensor of data to extract slices from.
+
starts : Tind
+
1-D tensor of starting indices of corresponding axis in `axes`
+
ends : Tind
+
1-D tensor of ending indices (exclusive) of corresponding axis in `axes`
+
axes (optional) : Tind
+
1-D tensor of axes that `starts` and `ends` apply to.
+
steps (optional) : Tind
+
1-D tensor of slice step of corresponding axis in `axes`. Default to 1.
+
+ +#### Outputs + +
+
output : T
+
Sliced data tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **StringNormalizer-10** + + StringNormalization performs string operations for basic cleaning. + This operator has only one input (denoted by X) and only one output + (denoted by Y). This operator first examines the elements in the X, + and removes elements specified in "stopwords" attribute. + After removing stop words, the intermediate result can be further lowercased, + uppercased, or just returned depending the "case_change_action" attribute. + This operator only accepts [C]- and [1, C]-tensor. + If all elements in X are dropped, the output will be the empty value of string tensor with shape [1] + if input shape is [C] and shape [1, 1] if input shape is [1, C]. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
case_change_action : string (default is NONE)
+
string enum that cases output to be lowercased/uppercases/unchanged. Valid values are "LOWER", "UPPER", "NONE". Default is "NONE"
+
is_case_sensitive : int (default is 0)
+
Boolean. Whether the identification of stop words in X is case-sensitive. Default is false
+
locale : string
+
Environment dependent string that denotes the locale according to which output strings needs to be upper/lowercased.Default en_US or platform specific equivalent as decided by the implementation.
+
stopwords : list of strings
+
List of stop words. If not set, no word would be removed from X.
+
+ +#### Inputs + +
+
X : tensor(string)
+
UTF-8 strings to normalize
+
+ +#### Outputs + +
+
Y : tensor(string)
+
UTF-8 Normalized strings
+
+ +#### Type Constraints + + +### **ThresholdedRelu-10** + + ThresholdedRelu takes one input data (Tensor) and produces one output data + (Tensor) where the rectified linear function, y = x for x > alpha, y = 0 otherwise, + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Threshold value
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **TopK-10** + + Retrieve the top-K elements along a specified axis. Given an input tensor of + shape [a_0, a_1, ..., a_{n-1}] and integer argument k, return two outputs: + -Value tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] + which contains the values of the top k elements along the specified axis + -Index tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] which + contains the indices of the top k elements (original indices from the input + tensor). + + Given two equivalent values, this operator uses the indices along the axis as + a tiebreaker. That is, the element with the lower index will appear first. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
Dimension on which to do the sort.
+
+ +#### Inputs + +
+
X : T
+
Tensor of shape [a_0, a_1, ..., a_{n-1}]
+
K : tensor(int64)
+
A 1-D tensor containing a single positive value corresponding to the number of top elements to retrieve
+
+ +#### Outputs + +
+
Values : T
+
Tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] containing top K values from the input tensor
+
Indices : I
+
Tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] containing the corresponding input tensor indices for the top K values.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **Upsample-10** (deprecated) + + Upsample the input tensor. + Each dimension value of the output tensor is: + output_dimension = floor(input_dimension * scale). + +#### Version + +This version of the operator has been deprecated since version 10 of the default ONNX operator set. + +## Version 11 of the default ONNX operator set +### **ArgMax-11** + + Computes the indices of the max elements of the input tensor's element along the + provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. + If keepdims equal 0, then the resulting tensor has the reduced dimension pruned. + The input tensor must not be empty. + The type of the output tensor is integer. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : tensor(int64)
+
Reduced output tensor with integer data type.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **ArgMin-11** + + Computes the indices of the min elements of the input tensor's element along the + provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. + If keepdims equal 0, then the resulting tensor has the reduced dimension pruned. + The input tensor must not be empty. + The type of the output tensor is integer. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : tensor(int64)
+
Reduced output tensor with integer data type.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **AveragePool-11** + + AveragePool consumes an input tensor X and applies average pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + average pooling consisting of computing the average on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled + + ``` + * pad_shape[i] is sum of pads along axis i + ``` + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following when ceil_mode is enabled: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + or when ceil_mode is disabled: + ``` + VALID: output_spatial_shape[i] = floor((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = floor(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i] + ``` + The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero). + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
count_include_pad : int (default is 0)
+
Whether include pad pixels when calculating values for the edges. Default is 0, doesn't count include pad.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **BitShift-11** + + Bitwise shift operator performs element-wise operation. For each input element, if the + attribute "direction" is "RIGHT", this operator moves its binary representation toward + the right side so that the input value is effectively decreased. If the attribute "direction" + is "LEFT", bits of binary representation moves toward the left side, which results the + increase of its actual value. The input X is the tensor to be shifted and another input + Y specifies the amounts of shifting. For example, if "direction" is "Right", X is [1, 4], + and S is [1, 1], the corresponding output Z would be [0, 2]. If "direction" is "LEFT" with + X=[1, 2] and S=[1, 2], the corresponding output Y would be [2, 8]. + + Because this operator supports Numpy-style broadcasting, X's and Y's shapes are + not necessarily identical. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
direction : string (required)
+
Direction of moving bits. It can be either "RIGHT" (for right shift) or "LEFT" (for left shift).
+
+ +#### Inputs + +
+
X (non-differentiable) : T
+
First operand, input to be shifted.
+
Y (non-differentiable) : T
+
Second operand, amounts of shift.
+
+ +#### Outputs + +
+
Z (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64)
+
Constrain input and output types to integer tensors.
+
+ +### **Clip-11** + + Clip operator limits the given input within an interval. The interval is + specified by the inputs 'min' and 'max'. They default to + numeric_limits::lowest() and numeric_limits::max(), respectively. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs (1 - 3) + +
+
input : T
+
Input tensor whose elements to be clipped
+
min (optional) : T
+
Minimum value, under which element is replaced by min. It must be a scalar(tensor of empty shape).
+
max (optional) : T
+
Maximum value, above which element is replaced by max. It must be a scalar(tensor of empty shape).
+
+ +#### Outputs + +
+
output : T
+
Output tensor with clipped input elements
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Compress-11** + + Selects slices from an input tensor along a given axis where condition evaluates to True for each axis index. + In case axis is not provided, input is flattened before elements are selected. + Compress behaves like numpy.compress: https://docs.scipy.org/doc/numpy/reference/generated/numpy.compress.html + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
(Optional) Axis along which to take slices. If not specified, input is flattened before elements being selected. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Tensor of rank r >= 1.
+
condition (non-differentiable) : T1
+
Rank 1 tensor of booleans to indicate which slices or data elements to be selected. Its length can be less than the input length along the axis or the flattened input size if axis is not specified. In such cases data slices or elements exceeding the condition length are discarded.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank r if axis is specified. Otherwise output is a Tensor of rank 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
T1 : tensor(bool)
+
Constrain to boolean tensors.
+
+ +### **Concat-11** + + Concatenate a list of tensors into a single tensor. All input tensors must have the same shape, except for the dimension size of the axis to concatenate on. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (required)
+
Which axis to concat on. A negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(inputs)..
+
+ +#### Inputs (1 - ∞) + +
+
inputs (variadic) : T
+
List of tensors for concatenation
+
+ +#### Outputs + +
+
concat_result : T
+
Concatenated tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain output types to any tensor type.
+
+ +### **ConcatFromSequence-11** + + Concatenate a sequence of tensors into a single tensor. + All input tensors must have the same shape, except for the dimension size of the axis to concatenate on. + By default 'new_axis' is 0, the behavior is similar to numpy.concatenate. + When 'new_axis' is 1, the behavior is similar to numpy.stack. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (required)
+
Which axis to concat on. Accepted range in `[-r, r - 1]`, where `r` is the rank of input tensors. When `new_axis` is 1, accepted range is `[-r - 1, r]`.
+
new_axis : int (default is 0)
+
Insert and concatenate on a new axis or not, default 0 means do not insert new axis.
+
+ +#### Inputs + +
+
input_sequence : S
+
Sequence of tensors for concatenation
+
+ +#### Outputs + +
+
concat_result : T
+
Concatenated tensor
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input types to any tensor type.
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain output types to any tensor type.
+
+ +### **Constant-11** + + A constant tensor. Exactly one of the two attributes, either value or sparse_value, + must be specified. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
sparse_value : sparse_tensor
+
The value for the elements of the output tensor in sparse format.
+
value : tensor
+
The value for the elements of the output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Conv-11** + + The convolution operator consumes an input tensor and a filter, and + computes the output. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input W.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults is 1 along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X (differentiable) : T
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
W (differentiable) : T
+
The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.
+
B (optional, differentiable) : T
+
Optional 1D bias to be added to the convolution, has size of M.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **ConvTranspose-11** + + The convolution transpose operator consumes an input tensor and a filter, + and computes the output. + + If the pads parameter is provided the shape of the output is calculated via the following equation: + + output_shape[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - pads[start_i] - pads[end_i] + + output_shape can also be explicitly specified in which case pads values are auto generated using these equations: + + total_padding[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - output_shape[i] + If (auto_pads == SAME_UPPER): pads[start_i] = total_padding[i]/2; pads[end_i] = total_padding[i] - (total_padding[i]/2) + Else: pads[start_i] = total_padding[i] - (total_padding[i]/2); pads[end_i] = (total_padding[i]/2). + + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = input_shape[i] * strides[i]` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each spatial axis.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input W.
+
output_padding : list of ints
+
Additional elements added to the side with higher coordinate indices in the output. Each padding value in "output_padding" must be less than the corresponding stride/dilation dimension. By default, this attribute is a zero vector. Note that this attribute doesn't directly affect the computed output values. It only controls the selection of the computed values, so changing this attribute only adds or removes output elements. If "output_shape" is explicitly provided, "output_padding" does not contribute additional size to "output_shape" but participates in the computation of the needed padding amount. This is also called adjs or adjustment in some frameworks.
+
output_shape : list of ints
+
The shape of the output can be explicitly set which will cause pads values to be auto generated. If output_shape is specified pads values are ignored. See doc for details for equations to generate pads. Note that the output_shape attribute value should not include dimensions for batch size and channels, which are automatically inferred.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X (differentiable) : T
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn)
+
W (differentiable) : T
+
The weight tensor that will be used in the convolutions; has size (C x M/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the weight shape will be (C x M/group x k1 x k2 x ... x kn), where (k1 x k2 x ... x kn) is the dimension of the kernel. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
+
B (optional, differentiable) : T
+
Optional 1D bias to be added to the convolution, has size of M.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, pad lengths and group count. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **CumSum-11** + + Performs cumulative sum of the input elements along the given axis. + By default, it will do the sum inclusively meaning the first element is copied as is. + Through an `exclusive` attribute, this behavior can change to exclude the first element. + It can also perform summation in the opposite direction of the axis. For that, set `reverse` attribute to 1. + + Example: + ``` + input_x = [1, 2, 3] + axis=0 + output = [1, 3, 6] + exclusive=1 + output = [0, 1, 3] + exclusive=0 + reverse=1 + output = [6, 5, 3] + exclusive=1 + reverse=1 + output = [5, 3, 0] + ``` + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
exclusive : int (default is 0)
+
If set to 1 will return exclusive sum in which the top element is not included. In other terms, if set to 1, the j-th output element would be the sum of the first (j-1) elements. Otherwise, it would be the sum of the first j elements.
+
reverse : int (default is 0)
+
If set to 1 will perform the sums in reverse direction.
+
+ +#### Inputs + +
+
x (differentiable) : T
+
An input tensor that is to be processed.
+
axis (non-differentiable) : T2
+
A 0-D tensor. Must be in the range [-rank(x), rank(x)-1]. Negative value means counting dimensions from the back.
+
+ +#### Outputs + +
+
y (differentiable) : T
+
Output tensor of the same type as 'x' with cumulative sums of the x's elements
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float), tensor(double)
+
Input can be of any tensor type.
+
T2 : tensor(int32), tensor(int64)
+
axis tensor can be int32 or int64 only
+
+ +### **DepthToSpace-11** + + DepthToSpace rearranges (permutes) data from depth into blocks of spatial data. + This is the reverse transformation of SpaceToDepth. More specifically, this op outputs a copy of + the input tensor where values from the depth dimension are moved in spatial blocks to the height + and width dimensions. By default, `mode` = `DCR`. + In the DCR mode, elements along the depth dimension from the input tensor are rearranged in the + following order: depth, column, and then row. The output y is computed from the input x as below: + + b, c, h, w = x.shape + + tmp = np.reshape(x, [b, blocksize, blocksize, c // (blocksize**2), h, w]) + + tmp = np.transpose(tmp, [0, 3, 4, 1, 5, 2]) + + y = np.reshape(tmp, [b, c // (blocksize**2), h * blocksize, w * blocksize]) + + + In the CRD mode, elements along the depth dimension from the input tensor are rearranged in the + following order: column, row, and the depth. The output y is computed from the input x as below: + + b, c, h, w = x.shape + + tmp = np.reshape(x, [b, c // (blocksize ** 2), blocksize, blocksize, h, w]) + + tmp = np.transpose(tmp, [0, 1, 4, 2, 5, 3]) + + y = np.reshape(tmp, [b, c // (blocksize ** 2), h * blocksize, w * blocksize]) + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
blocksize : int (required)
+
Blocks of [blocksize, blocksize] are moved.
+
mode : string (default is DCR)
+
DCR (default) for depth-column-row order re-arrangement. Use CRD for column-row-depth order.
+
+ +#### Inputs + +
+
input : T
+
Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
+
+ +#### Outputs + +
+
output : T
+
Output tensor of [N, C/(blocksize * blocksize), H * blocksize, W * blocksize].
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Det-11** + + Det calculates determinant of a square matrix or batches of square matrices. + Det takes one input tensor of shape `[*, M, M]`, where `*` is zero or more batch dimensions, + and the inner-most 2 dimensions form square matrices. + The output is a tensor of shape `[*]`, containing the determinants of all input submatrices. + e.g., When the input is 2-D, the output is a scalar(shape is empty: `[]`). + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to floating-point tensors.
+
+ +### **DynamicQuantizeLinear-11** + + A Function to fuse calculation for Scale, Zero Point and FP32->8Bit conversion of FP32 Input data. + Outputs Scale, ZeroPoint and Quantized Input for a given FP32 Input. + Scale is calculated as: + ``` + y_scale = (maximum(0, max(x)) - minimum(0, min(x))) / (qmax - qmin) + ``` + + * where qmax and qmin are max and min values for quantization range i.e. [0, 255] in case of uint8 + * data range is adjusted to include 0. + + Zero point is calculated as: + ``` + intermediate_zero_point = qmin - min(x)/y_scale + y_zero_point = cast(round(saturate(intermediate_zero_point))) + ``` + + * where qmax and qmin are max and min values for quantization range .i.e [0, 255] in case of uint8 + * for saturation, it saturates to [0, 255] if it's uint8, or [-127, 127] if it's int8. Right now only uint8 is supported. + * rounding to nearest ties to even. + + Data quantization formula is: + ``` + y = saturate (round (x / y_scale) + y_zero_point) + ``` + + * for saturation, it saturates to [0, 255] if it's uint8, or [-127, 127] if it's int8. Right now only uint8 is supported. + * rounding to nearest ties to even. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
x : T1
+
Input tensor
+
+ +#### Outputs + +
+
y : T2
+
Quantized output tensor
+
y_scale : tensor(float)
+
Output scale. It's a scalar, which means a per-tensor/layer quantization.
+
y_zero_point : T2
+
Output zero point. It's a scalar, which means a per-tensor/layer quantization.
+
+ +#### Type Constraints + +
+
T1 : tensor(float)
+
Constrain 'x' to float tensor.
+
T2 : tensor(uint8)
+
Constrain 'y_zero_point' and 'y' to 8-bit unsigned integer tensor.
+
+ +### **Equal-11** + + Returns the tensor resulted from performing the `equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
A : T
+
First input operand for the logical operator.
+
B : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Flatten-11** + + Flattens the input tensor into a 2D matrix. If input tensor has shape + (d_0, d_1, ... d_n) then the output will have shape + (d_0 X d_1 ... d_(axis-1), d_axis X d_(axis+1) ... X dn). + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 ... d_n), where the shape of the input tensor is (d_0, d_1, ... d_n).
+
+ +#### Inputs + +
+
input : T
+
A tensor of rank >= axis.
+
+ +#### Outputs + +
+
output : T
+
A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output to all tensor types.
+
+ +### **Gather-11** + + Given `data` tensor of rank r >= 1, and `indices` tensor of rank q, gather + entries of the axis dimension of `data` (by default outer-most one as axis=0) indexed by `indices`, and concatenates + them in an output tensor of rank q + (r - 1). + + axis = 0 : + + Let + k = indices[i_{0}, ..., i_{q-1}] + Then + output[i_{0}, ..., i_{q-1}, j_{0}, ..., j_{r-2}] = input[k , j_{0}, ..., j_{r-2}] + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + indices = [ + [0, 1], + [1, 2], + ] + output = [ + [ + [1.0, 1.2], + [2.3, 3.4], + ], + [ + [2.3, 3.4], + [4.5, 5.7], + ], + ] + ``` + axis = 1 : + + Let + k = indices[i_{0}, ..., i_{q-1}] + Then + output[j_{0}, i_{0}, ..., i_{q-1}, j_{1}, ..., j_{r-2}] = input[j_{0}, k, j_{1}, ..., j_{r-2}] + + ``` + data = [ + [1.0, 1.2, 1.9], + [2.3, 3.4, 3.9], + [4.5, 5.7, 5.9], + ] + indices = [ + [0, 2], + ] + axis = 1, + output = [ + [[1.0, 1.9]], + [[2.3, 3.9]], + [[4.5, 5.9]], + ] + ``` + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Inputs + +
+
data : T
+
Tensor of rank r >= 1.
+
indices : Tind
+
Tensor of int32/int64 indices, of any rank q. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output : T
+
Tensor of rank q + (r - 1).
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **GatherElements-11** + + GatherElements takes two inputs `data` and `indices` of the same rank r >= 1 + and an optional attribute `axis` that identifies an axis of `data` + (by default, the outer-most axis, that is axis 0). It is an indexing operation + that produces its output by indexing into the input data tensor at index + positions determined by elements of the `indices` tensor. + Its output shape is the same as the shape of `indices` and consists of one value + (gathered from the `data`) for each element in `indices`. + + For instance, in the 3-D case (r = 3), the output produced is determined + by the following equations: + ``` + out[i][j][k] = input[index[i][j][k]][j][k] if axis = 0, + out[i][j][k] = input[i][index[i][j][k]][k] if axis = 1, + out[i][j][k] = input[i][j][index[i][j][k]] if axis = 2, + ``` + + This operator is also the inverse of ScatterElements. It is similar to Torch's gather operation. + + Example 1: + ``` + data = [ + [1, 2], + [3, 4], + ] + indices = [ + [0, 0], + [1, 0], + ] + axis = 1 + output = [ + [ + [1, 1], + [4, 3], + ], + ] + ``` + Example 2: + ``` + data = [ + [1, 2, 3], + [4, 5, 6], + [7, 8, 9], + ] + indices = [ + [1, 2, 0], + [2, 0, 0], + ] + axis = 0 + output = [ + [ + [4, 8, 3], + [7, 2, 3], + ], + ] + ``` + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Inputs + +
+
data : T
+
Tensor of rank r >= 1.
+
indices : Tind
+
Tensor of int32/int64 indices, with the same rank r as the input. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output : T
+
Tensor of the same shape as indices.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **GatherND-11** + + Given `data` tensor of rank `r` >= 1, and `indices` tensor of rank `q` >= 1, this operator gathers + slices of `data` into an output tensor of rank `q + r - indices_shape[-1] - 1`. + + `indices` is an q-dimensional integer tensor, best thought of as a `(q-1)`-dimensional tensor of index-tuples into `data`, + where each element defines a slice of `data` + + Some salient points about the inputs' rank and shape: + + 1) r >= 1 and q >= 1 are to be honored. There is no dependency condition to be met between ranks `r` and `q` + + 2) The `indices_shape[-1]` should have a value between 1 (inclusive) and rank `r` (inclusive) + + 3) All values in `indices` are expected to be within bounds [-s, s-1] along axis of size `s` (i.e.) `-data_shape[i] <= indices[...,i] <= data_shape[i] - 1`. + It is an error if any of the index values are out of bounds. + + The output is computed as follows: + + The output tensor is obtained by mapping each index-tuple in the `indices` tensor to the corresponding slice of the input `data`. + + 1) If `indices_shape[-1] > r` => error condition + + 2) If `indices_shape[-1] == r`, since the rank of `indices` is `q`, `indices` can be thought of as a `(q-1)`-dimensional tensor + containing 1-D tensors of dimension `r`. Let us think of each such `r` ranked tensor as `indices_slice`. + Each *scalar value* corresponding to `data[indices_slice]` is filled into the corresponding location of the `(q-1)`-dimensional tensor + to form the `output` tensor (Example 1 below) + + 3) If `indices_shape[-1] < r`, since the rank of `indices` is `q`, `indices` can be thought of as a `(q-1)`-dimensional tensor + containing 1-D tensors of dimension `< r`. Let us think of each such tensors as `indices_slice`. + Each *tensor slice* corresponding to `data[indices_slice , :]` is filled into the corresponding location of the `(q-1)`-dimensional tensor + to form the `output` tensor (Examples 2, 3, and 4 below) + + This operator is the inverse of `ScatterND`. + + `Example 1` + + data = [[0,1],[2,3]] # data_shape = [2, 2] + + indices = [[0,0],[1,1]] # indices_shape = [2, 2] + + output = [0,3] # output_shape = [2] + + `Example 2` + + data = [[0,1],[2,3]] # data_shape = [2, 2] + + indices = [[1],[0]] # indices_shape = [2, 1] + + output = [[2,3],[0,1]] # output_shape = [2, 2] + + `Example 3` + + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + + indices = [[0,1],[1,0]] # indices_shape = [2, 2] + + output = [[2,3],[4,5]] # output_shape = [2, 2] + + `Example 4` + + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + + indices = [[[0,1]],[[1,0]]] # indices_shape = [2, 1, 2] + + output = [[[2,3]],[[4,5]]] # output_shape = [2, 1, 2] + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
data : T
+
Tensor of rank r >= 1.
+
indices : tensor(int64)
+
Tensor of rank q >= 1. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output : T
+
Tensor of rank q + r - indices_shape[-1] - 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
+ +### **Gemm-11** + + General Matrix multiplication: + https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3 + + A' = transpose(A) if transA else A + + B' = transpose(B) if transB else B + + Compute Y = alpha * A' * B' + beta * C, where input tensor A has shape (M, K) or (K, M), + input tensor B has shape (K, N) or (N, K), input tensor C is broadcastable to shape (M, N), + and output tensor Y has shape (M, N). A will be transposed before doing the + computation if attribute transA is non-zero, same for B and transB. + This operator supports **unidirectional broadcasting** (tensor C should be unidirectional broadcastable to tensor A * B); for more details please check [the doc](Broadcasting.md). + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Scalar multiplier for the product of input tensors A * B.
+
beta : float (default is 1.0)
+
Scalar multiplier for input tensor C.
+
transA : int (default is 0)
+
Whether A should be transposed
+
transB : int (default is 0)
+
Whether B should be transposed
+
+ +#### Inputs (2 - 3) + +
+
A : T
+
Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
+
B : T
+
Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
+
C (optional) : T
+
Optional input tensor C. If not specified, the computation is done as if C is a scalar 0. The shape of C should be unidirectional broadcastable to (M, N).
+
+ +#### Outputs + +
+
Y : T
+
Output tensor of shape (M, N).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64)
+
Constrain input and output types to float/int tensors.
+
+ +### **Hardmax-11** + + The operator computes the hardmax (1 for the first maximum value, and 0 for all others) values for each layer in the batch + of the given input. + + The input does not need to explicitly be a 2D vector; rather, it will be + coerced into one. For an arbitrary n-dimensional tensor + input \in [a_0, a_1, ..., a_{k-1}, a_k, ..., a_{n-1}] and k is + the axis provided, then input will be coerced into a 2-dimensional tensor with + dimensions [a_0 * ... * a_{k-1}, a_k * ... * a_{n-1}]. For the default + case where axis=1, this means the input tensor will be coerced into a 2D tensor + of dimensions [a_0, a_1 * ... * a_{n-1}], where a_0 is often the batch size. + In this situation, we must have a_0 = N and a_1 * ... * a_{n-1} = D. + Each of these dimensions must be matched correctly, or else the operator + will throw errors. The output tensor has the same shape + and contains the hardmax values of the corresponding input. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
+ +#### Inputs + +
+
input : T
+
The input tensor that's coerced into a 2D matrix of size (NxD) as described above.
+
+ +#### Outputs + +
+
output : T
+
The output values with the same shape as input tensor (the original size without coercion).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **If-11** + + If conditional + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
else_branch : graph (required)
+
Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
+
then_branch : graph (required)
+
Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
+
+ +#### Inputs + +
+
cond : B
+
Condition for the if. The tensor must contain a single element.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : V
+
Values that are live-out to the enclosing scope. The return values in the `then_branch` and `else_branch` must be of the same data type. The `then_branch` and `else_branch` may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if-node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of `then_branch` is typed float tensor with shape [2] and the first output of `else_branch` is another float tensor with shape [3], If's first output should have (a) no shape set, or (b) a shape of rank 1 with neither `dim_value` nor `dim_param` set, or (c) a shape of rank 1 with a unique `dim_param`. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
All Tensor types
+
B : tensor(bool)
+
Only bool
+
+ +### **LogSoftmax-11** + + The operator computes the logsoftmax (log of softmax) values for each layer in the batch + of the given input. + + The input does not need to explicitly be a 2D vector; rather, it will be + coerced into one. For an arbitrary n-dimensional tensor + input \in [a_0, a_1, ..., a_{k-1}, a_k, ..., a_{n-1}] and k is + the axis provided, then input will be coerced into a 2-dimensional tensor with + dimensions [a_0 * ... * a_{k-1}, a_k * ... * a_{n-1}]. For the default + case where axis=1, this means the input tensor will be coerced into a 2D tensor + of dimensions [a_0, a_1 * ... * a_{n-1}], where a_0 is often the batch size. + In this situation, we must have a_0 = N and a_1 * ... * a_{n-1} = D. + Each of these dimensions must be matched correctly, or else the operator + will throw errors. The output tensor has the same shape + and contains the logsoftmax values of the corresponding input. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
+ +#### Inputs + +
+
input : T
+
The input tensor that's coerced into a 2D matrix of size (NxD) as described above.
+
+ +#### Outputs + +
+
output : T
+
The output values with the same shape as input tensor (the original size without coercion).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Loop-11** + + Generic Looping construct. This loop has multiple termination conditions: + + 1) Trip count. Iteration count specified at runtime. Set by + specifying the input M. Optional. Set to empty string to omit. + Note that a static trip count (specified at graph construction time) can be + specified by passing in a constant node for input M. + 2) Loop termination condition. This is an input to the op that determines + whether to run the first iteration and also a loop-carried dependency for + the body graph. The body graph must yield a value for the condition variable, + whether this input is provided or not. + + This table summarizes the operating modes of this operator with equivalent + C-style code: + + Operator inputs defined as (max_trip_count, condition_var). + + input ("", ""): + for (int i=0; ; ++i) { + cond = ... // Note this value is ignored, but is required in the body + } + + input ("", cond) // Note this is analogous to a while loop + bool cond = ...; + for (int i=0; cond; ++i) { + cond = ...; + } + + input ("", 1) // Note this is analogous to a do-while loop + bool cond = true + for (int i=0; cond; ++i) { + cond = ...; + } + + input (trip_count, "") // Note this is analogous to a for loop + int trip_count = ... + for (int i=0; i < trip_count; ++i) { + cond = ...; // ignored + } + + input (trip_count, cond) + int trip_count = ...; + bool cond = ...; + for (int i=0; i < trip_count && cond; ++i) { + cond = ...; + } + + + *Sample usage - cond as well as trip count* + + graph predict-net { + %a = Constant[value = ]() + %b = Constant[value = ]() + %keepgoing = Constant[value = ]() + %max_trip_count = Constant[value = ]() + %keepgoing_out, %b_out, %user_defined_vals = Loop[body = ](%max_trip_count, %keepgoing, %b) + return + } + + graph body-net ( + %i[INT32, scalar] // iteration number + %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used + %b_in[INT32, scalar] // incoming value of loop-carried-dependency b + ) { + %my_local = Add(%a, %b_in) + %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b + %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition + %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated + return %keepgoing_out, %b_out, %user_defined_val + } + + *Sample equivalent C code* + + { + /* User-defined code (enclosing scope) */ + int a = 3, b = 6; + bool keepgoing = true; // Analogous to input cond + /* End user-defined code */ + + /* Implicitly-defined code */ + const int max_trip_count = 10; // Analogous to input M + int user_defined_vals[]; // Imagine this is resizable + /* End implicitly-defined code */ + /* initialize loop-carried variables and scan-output variables */ + bool keepgoing_out = keepgoing + int b_out = b + + for (int i=0; i < max_trip_count && keepgoing_out; ++i) { + /* Implicitly-defined code: bind actual parameter values + to formal parameter variables of loop-body */ + bool keepgoing_in = keepgoing_out; + bool b_in = b_out; + + /* User-defined code (loop body) */ + int my_local = a + b_in; // Reading value "a" from the enclosing scope is fine + b_out = a - b_in; + keepgoing_out = my_local > b_out; + user_defined_val = b_in + b_in; // b_in and b_out are different variables + /* End user-defined code */ + + /* Implicitly defined-code */ + user_defined_vals[i] = user_defined_val // accumulate scan-output values + } + // int t = my_local; // Can't do this. my_local is not accessible here. + + // The values below are bound to the output variables of the loop and therefore accessible + // b_out; user_defined_vals; keepgoing_out; + } + + There are several things of note in this code snippet: + + 1) Values from the enclosing scope (i.e. variable "a" here) are in scope and can + be referenced in the inputs of the loop. + 2) Any values computed in the loop body that needs to be used in a subsequent + iteration or after the loop are modeled using a pair of variables in the loop-body, + consisting of an input variable (eg., b_in) and an output variable (eg., b_out). + These are referred to as loop-carried dependences. The loop operation node + supplies the input value of the input variable for the first iteration, and + returns the output value of the output variable produced by the final + iteration. + 3) Scan_output variables are used to implicitly concatenate values computed across + all the iterations. In the above example, the value of user_defined_val computed + over all iterations are concatenated and returned as the value of user_defined_vals + after the loop. + 4) Values created in the body cannot be accessed in the enclosing scope, + except using the mechanism described above. + + Note that the semantics of this op support "diagonal" or "wavefront" execution. + (See Step 3 here for an example: + https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). + Frontends should emit multi-layer RNNs as a series of While operators (with + time being the inner looping dimension), with each successive layer consuming + the scan_outputs from the previous layer, possibly going through several + point-wise operators (e.g. dropout, residual connections, linear layer). + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies...). It has 1+N+K outputs: (condition, loop carried dependencies..., scan_outputs...). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
+
+ +#### Inputs (2 - ∞) + +
+
M (optional) : I
+
A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
+
cond (optional) : B
+
A boolean termination condition. Optional. Pass empty string to skip.
+
v_initial (variadic, heterogeneous) : V
+
The initial values of any loop-carried dependencies (values that change across loop iterations)
+
+ +#### Outputs (1 - ∞) + +
+
v_final_and_scan_outputs (variadic, heterogeneous) : V
+
Final N loop carried dependency values then K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
All Tensor types
+
I : tensor(int64)
+
tensor of int64, which should be a scalar.
+
B : tensor(bool)
+
tensor of bool, which should be a scalar.
+
+ +### **LpPool-11** + + LpPool consumes an input tensor X and applies Lp pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + Lp pooling consisting of computing the Lp norm on all values of a subset + of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
p : int (default is 2)
+
p value of the Lp norm used to pool over the input data.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **MaxPool-11** + + MaxPool consumes an input tensor X and applies max pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + max pooling consisting of computing the max on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled + + ``` + * pad_shape[i] is sum of pads along axis i + ``` + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i] + ``` + The output of each pooling window is maximum number of elements exclude pad. + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
dilations : list of ints
+
Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
storage_order : int (default is 0)
+
The storage order of the tensor. 0 is row major, and 1 is column major.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs (1 - 2) + +
+
Y : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
Indices (optional) : I
+
Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x ... x Dn).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **MaxUnpool-11** + + MaxUnpool essentially computes the partial inverse of the MaxPool op. + The input information to this op is typically the output information from a MaxPool op. The first + input tensor X is the tensor that needs to be unpooled, which is typically the pooled tensor (first output) + from MaxPool. The second input tensor, I, contains the indices to the (locally maximal) elements corresponding + to the elements in the first input tensor X. Input tensor I is typically the second output of the MaxPool op. + The third (optional) input is a tensor that specifies the output size of the unpooling operation. + + MaxUnpool is intended to do 'partial' inverse of the MaxPool op. 'Partial' because all the non-maximal + values from the original input to MaxPool are set to zero in the output of the MaxUnpool op. Pooling + the result of an unpooling operation should give back the original input to the unpooling op. + + MaxUnpool can produce the same output size for several input sizes, which makes unpooling op ambiguous. + The third input argument, output_size, is meant to disambiguate the op and produce output tensor of + known/predictable size. + + In addition to the inputs, MaxUnpool takes three attributes, namely kernel_shape, strides, and pads, + which define the exact unpooling op. The attributes typically have the same values as the corresponding + pooling op that the unpooling op is trying to invert. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X (differentiable) : T1
+
Input data tensor that has to be unpooled. This tensor is typically the first output of the MaxPool op.Dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non-image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
I (non-differentiable) : T2
+
Input data tensor containing the indices corresponding to elements in the first input tensor X.This tensor is typically the second output of the MaxPool op.Dimensions must be the same as input tensor X. The indices are linear, i.e. computed considering the tensor as flattened 1-D tensor, assuming row-major storage. Also, the linear indices should not consider padding. So the values in indices are in the range [0, N x C x D1 x ... x Dn).
+
output_shape (optional, non-differentiable) : T2
+
The shape of the output can be explicitly set which will cause pads values to be auto generated. If 'output_shape' is specified, 'pads' values are ignored.
+
+ +#### Outputs + +
+
output (differentiable) : T1
+
Output data tensor that contains the result of the unpooling.
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T2 : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **NonMaxSuppression-11** + + Filter out boxes that have high intersection-over-union (IOU) overlap with previously selected boxes. + Bounding boxes with score less than score_threshold are removed. Bounding box format is indicated by attribute center_point_box. + Boxes are suppressed if their IOU with a previously selected box is strictly greater than iou_threshold (i.e., boxes with IOU exactly equal to the threshold are kept). + Note that this algorithm is agnostic to where the origin is in the coordinate system and more generally is invariant to + orthogonal transformations and translations of the coordinate system; thus translating or reflections of the coordinate system + result in the same boxes being selected by the algorithm. + The selected_indices output is a set of integers indexing into the input collection of bounding boxes representing the selected boxes. + The bounding box coordinates corresponding to the selected indices can then be obtained using the Gather or GatherND operation. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
center_point_box : int (default is 0)
+
Integer indicate the format of the box data. The default is 0. 0 - the box data is supplied as [y1, x1, y2, x2] where (y1, x1) and (y2, x2) are the coordinates of any diagonal pair of box corners and the coordinates can be provided as normalized (i.e., lying in the interval [0, 1]) or absolute. Mostly used for TF models. 1 - the box data is supplied as [x_center, y_center, width, height]. Mostly used for Pytorch models.
+
+ +#### Inputs (2 - 5) + +
+
boxes : tensor(float)
+
An input tensor with shape [num_batches, spatial_dimension, 4]. The single box data format is indicated by center_point_box.
+
scores : tensor(float)
+
An input tensor with shape [num_batches, num_classes, spatial_dimension]
+
max_output_boxes_per_class (optional) : tensor(int64)
+
Integer representing the maximum number of boxes to be selected per batch per class. It is a scalar. Default to 0, which means no output.
+
iou_threshold (optional) : tensor(float)
+
Float representing the threshold for deciding whether boxes overlap too much with respect to IOU. Boxes with IoU strictly greater than this threshold are suppressed. It is scalar. Value range [0, 1]. Default to 0.
+
score_threshold (optional) : tensor(float)
+
Float representing the threshold for deciding when to remove boxes based on score. It is a scalar.
+
+ +#### Outputs + +
+
selected_indices : tensor(int64)
+
selected indices from the boxes tensor. [num_selected_indices, 3], the selected index format is [batch_index, class_index, box_index].
+
+ +#### Type Constraints + + +### **OneHot-11** + + Produces a one-hot tensor based on inputs. + The locations represented by the index values in the 'indices' input tensor will have 'on_value' + and the other locations will have 'off_value' in the output tensor, where 'on_value' and 'off_value' + are specified as part of required input argument 'values', which is a two-element tensor of format + [off_value, on_value]. The rank of the output tensor will be one greater than the rank of the + input tensor. The additional dimension is for one-hot representation. The additional dimension will + be inserted at the position specified by 'axis'. If 'axis' is not specified then then additional + dimension will be inserted as the innermost dimension, i.e. axis=-1. The size of the additional + dimension is specified by required scalar input 'depth'. The type of the output tensor is the same + as the type of the 'values' input. Any entries in the 'indices' input tensor with values outside + the range [-depth, depth-1] will result in one-hot representation with all 'off_value' values in the + output tensor. + + when axis = 0: + output[input[i, j, k], i, j, k] = 1 for all i, j, k and 0 otherwise. + + when axis = -1: + output[i, j, k, input[i, j, k]] = 1 for all i, j, k and 0 otherwise. + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
(Optional) Axis along which one-hot representation in added. Default: axis=-1. axis=-1 means that the additional dimension will be inserted as the innermost/last dimension in the output tensor. Negative value means counting dimensions from the back. Accepted range is [-r-1, r] where r = rank(indices).
+
+ +#### Inputs + +
+
indices (non-differentiable) : T1
+
Input tensor containing indices. Any entries in the 'indices' input tensor with values outside the range [-depth, depth-1] will result in one-hot representation with all 'off_value' values in the output tensor.In case 'indices' is of non-integer type, the values will be casted to int64 before use.
+
depth (non-differentiable) : T2
+
Scalar or Rank 1 tensor containing exactly one element, specifying the number of classes in one-hot tensor. This is also the size of the one-hot dimension (specified by 'axis' attribute) added on in the output tensor. The values in the 'indices' input tensor are expected to be in the range [-depth, depth-1]. In case 'depth' is of non-integer type, it will be casted to int64 before use.
+
values (non-differentiable) : T3
+
Rank 1 tensor containing exactly two elements, in the format [off_value, on_value], where 'on_value' is the value used for filling locations specified in 'indices' input tensor, and 'off_value' is the value used for filling locations other than those specified in 'indices' input tensor.
+
+ +#### Outputs + +
+
output (non-differentiable) : T3
+
Tensor of rank one greater than input tensor 'indices', i.e. rank(output) = rank(indices) + 1. The data type for the elements of the output tensor is the same as the type of input 'values' is used.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input to only numeric types.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input to only numeric types.
+
T3 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type.
+
+ +### **Pad-11** + + Given a tensor containing the data to be padded (`data`), a tensor containing the number of start and end pad values for axis (`pads`), (optionally) a `mode`, and (optionally) `constant_value`, + a padded tensor (`output`) is generated. + + The three supported `modes` are (similar to corresponding modes supported by `numpy.pad`): + + 1) `constant`(default) - pads with a given constant value as specified by `constant_value` (which defaults to 0) + + 2) `reflect` - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis + + 3) `edge` - pads with the edge values of array + + + Example 1 (`constant` mode): + Insert 0 pads to the beginning of the second dimension. + + data = + [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'constant' + + constant_value = 0.0 + + output = + [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ] + + + Example 2 (`reflect` mode): + data = + [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'reflect' + + output = + [ + [1.0, 1.2, 1.0, 1.2], + [2.3, 3.4, 2.3, 3.4], + [4.5, 5.7, 4.5, 5.7], + ] + + + Example 3 (`edge` mode): + data = + [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'edge' + + output = + [ + [1.0, 1.0, 1.0, 1.2], + [2.3, 2.3, 2.3, 3.4], + [4.5, 4.5, 4.5, 5.7], + ] + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is constant)
+
Supported modes: `constant`(default), `reflect`, `edge`
+
+ +#### Inputs (2 - 3) + +
+
data : T
+
Input tensor.
+
pads : tensor(int64)
+
Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. `pads` should be a 1D tensor of shape [2 * input_rank]. `pads` format should be: [x1_begin, x2_begin,...,x1_end, x2_end,...], where xi_begin is the number of pad values added at the beginning of axis `i` and xi_end, the number of pad values added at the end of axis `i`.
+
constant_value (optional) : T
+
(Optional) A scalar value to be used if the mode chosen is `constant` (by default it is 0).
+
+ +#### Outputs + +
+
output : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output to only numeric types.
+
+ +### **Range-11** + + Generate a tensor containing a sequence of numbers that begin at `start` and extends by increments of `delta` + up to `limit` (exclusive). + + The number of elements in the output of range is computed as below: + ``` + number_of_elements = max( ceil( (limit - start) / delta ) , 0 ) + ``` + The pseudocode determining the contents of the output is shown below: + ``` + for(int i=0; i +
start : T
+
Scalar. First entry for the range of output values.
+
limit : T
+
Scalar. Exclusive upper limit for the range of output values.
+
delta : T
+
Scalar. Value to step by.
+ + +#### Outputs + +
+
output : T
+
A 1-D tensor with same type as the inputs containing generated range of values.
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int16), tensor(int32), tensor(int64)
+
Constrain input types to common numeric type tensors.
+
+ +### **ReduceL1-11** + + Computes the L1 norm of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceL2-11** + + Computes the L2 norm of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceLogSum-11** + + Computes the log sum of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceLogSumExp-11** + + Computes the log sum exponent of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceMax-11** + + Computes the max of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or the minimum value of the data type otherwise. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceMean-11** + + Computes the mean of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceMin-11** + + Computes the min of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields plus infinity (if supported by the datatype) or the maximum value of the data type otherwise. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceProd-11** + + Computes the product of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceSum-11** + + Computes the sum of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ReduceSumSquare-11** + + Computes the sum square of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Resize-11** + + Resize the input tensor. In general, it calculates every value in the output tensor as a weighted average of neighborhood (a.k.a. sampling locations) in the input tensor. + Each dimension value of the output tensor is: + output_dimension = floor(input_dimension * (roi_end - roi_start) * scale) if input \"sizes\" is not specified. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
coordinate_transformation_mode : string (default is half_pixel)
+
+This attribute describes how to transform the coordinate in the resized tensor to the coordinate in the original tensor.
+ +The coordinate of each dimension is transformed individually. Let's describe a case using axis x as an example. +Denote x_resized as the coordinate of axis x in the resized tensor, x_original as the coordinate of axis x in the original tensor, length_original as the length of the original tensor in axis x, length_resized as the length of the resized tensor in axis x, roi_x = (start_x, end_x) of the axis x in input "roi", scale = length_resized / length_original,
+ +if coordinate_transformation_mode is "half_pixel",
+x_original = (x_resized + 0.5) / scale - 0.5,
+ +if coordinate_transformation_mode is "pytorch_half_pixel",
+x_original = length_resized > 1 ? (x_resized + 0.5) / scale - 0.5 : 0,
+ +if coordinate_transformation_mode is "align_corners",
+x_original = x_resized * (length_original - 1) / (length_resized - 1),
+ +if coordinate_transformation_mode is "asymmetric",
+x_original = x_resized / scale,
+ +if coordinate_transformation_mode is "tf_half_pixel_for_nn",
+x_original = (x_resized + 0.5) / scale,
+ +if coordinate_transformation_mode is "tf_crop_and_resize",
+x_original = length_resized > 1 ? start_x * (length_original - 1) + x_resized * (end_x - start_x) * (length_original - 1) / (length_resized - 1) : 0.5 * (start_x + end_x) * (length_original - 1).
+
cubic_coeff_a : float (default is -0.75)
+
The coefficient 'a' used in cubic interpolation. Two common choice are -0.5 (in some cases of TensorFlow) and -0.75 (in PyTorch). Check out Equation (4) in https://ieeexplore.ieee.org/document/1163711 for the details. This attribute is valid only if "mode" is "cubic".
+
exclude_outside : int (default is 0)
+
If set to 1, the weight of sampling locations outside the tensor will be set to 0 and the weight will be renormalized so that their sum is 1.0. The default value is 0.
+
extrapolation_value : float (default is 0.0)
+
When coordinate_transformation_mode is "tf_crop_and_resize" and x_original is outside the range [0, length_original - 1], this value is used as the corresponding output value. Default is 0.0f.
+
mode : string (default is nearest)
+
Three interpolation modes: nearest (default), linear and cubic. The "linear" mode includes linear interpolation for 1D tensor and N-linear interpolation for N-D tensor (for example, bilinear interpolation for 2D tensor). The "cubic" mode includes cubic interpolation for 1D tensor and N-cubic interpolation for N-D tensor (for example, bicubic interpolation for 2D tensor).
+
nearest_mode : string (default is round_prefer_floor)
+
Four modes: round_prefer_floor (default, as known as round half down), round_prefer_ceil (as known as round half up), floor, ceil. Only used by nearest interpolation. It indicates how to get "nearest" pixel in input tensor from x_original, so this attribute is valid only if "mode" is "nearest".
+
+ +#### Inputs (3 - 4) + +
+
X : T1
+
N-D tensor
+
roi : T2
+
1-D tensor given as [start1, ..., startN, end1, ..., endN], where N is the rank of X. The RoIs' coordinates are normalized in the coordinate system of the input image. It only takes effect when coordinate_transformation_mode is "tf_crop_and_resize"
+
scales : tensor(float)
+
The scale array along each dimension. It takes value greater than 0. If it's less than 1, it's sampling down, otherwise, it's upsampling. The number of elements of 'scales' should be the same as the rank of input 'X'. If 'size' is needed, the user must set 'scales' to an empty tensor.
+
sizes (optional) : tensor(int64)
+
The size of the output tensor. The number of elements of 'sizes' should be the same as the rank of input 'X'. May only be set if 'scales' is set to an empty tensor.
+
+ +#### Outputs + +
+
Y : T1
+
N-D tensor after resizing
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input 'X' and output 'Y' to all tensor types.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Constrain roi type to float or double.
+
+ +### **Round-11** + + Round takes one input Tensor and rounds the values, element-wise, meaning + it finds the nearest integer for each value. + In case of halves, the rule is to round them to the nearest even integer. + If input x is integral, +0, -0, NaN, or infinite, x itself is returned. + The output tensor has the same shape and type as the input. + + Examples: + ``` + round([0.9]) = [1.0] + round([2.5]) = [2.0] + round([2.3]) = [2.0] + round([1.5]) = [2.0] + round([-4.5]) = [-4.0] + ``` + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Scan-11** + + Scan can be used to iterate over one or more scan_input tensors, + constructing zero or more scan_output tensors. It combines ideas from general recurrences, + functional programming constructs such as scan, fold, map, and zip, and is intended to enable + generalizations of RNN-like constructs for sequence-to-sequence processing. + Other tensors (referred to as state_variables here) can be used to carry a state + when iterating from one element to another (similar to hidden-state in RNNs, also referred + to as loop-carried dependences in the context of loops). + Many common usages involve a single scan_input tensor (where functionality + similar to scan, fold and map can be obtained). When more than one scan_input is used, + a behavior similar to zip is obtained. + + The attribute body must be a graph, specifying the computation to be performed in + every iteration. It takes as input the current values of the state_variables and + the current iterated element of the scan_inputs. It must return the (updated) values + of the state_variables and zero or more scan_output_element tensors. The values of the + scan_output_element tensors are concatenated over all the iterations to produce the + scan_output values of the scan construct (similar to the concatenated intermediate + hidden-state values of RNN-like constructs). All the output tensors (state_variables as + well as scan_output_element tensors) are required to have the same shape in each iteration + of the loop (a restriction imposed to enable efficient memory allocation). + + Note that the iterated element passed to the body subgraph does not have a sequence + axis. It will have a rank one less than the rank of the corresponding scan_input. + + The scan operation returns the final values of the state_variables as well as the + scan_outputs. + + The optional attribute scan_input_directions specifies the direction (forward or backward) + for each scan input. If this attribute is omitted, all sequences are scanned in the forward + direction. A bidirectional scan may be performed by specifying the same tensor input twice + in the scan_inputs, once with a forward direction, and once with a backward direction. + + The scan_output of the operation is produced by concatenating the scan_output_element + values produced by the body in each iteration. The optional attribute scan_output_directions + specifies the direction in which scan_output is constructed (by appending or prepending the + scan_output_element to scan_output in each iteration) for each scan_output. If this attribute + is omitted, the scan_output_element is appended to the scan_output in each iteration. + + The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. + If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the + batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. + Note that scanning a non-zero axis may be less efficient than scanning axis zero. + + The optional attribute scan_output_axes specifies the axis along which the scan_outputs + are accumulated for each scan_output. For example, if axis 1 is the time axis (to be + scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis + value of 1. + + Note that because of the ONNX restriction that only the last parameter of an operator can + be variadic, the initial-states and scan-inputs are listed together as one input parameter. + Similarly, the final-states and scan-outputs are listed together as one output parameter. + The attribute num_scan_inputs indicates the number M of scan-inputs. + + The behavior of + + Scan < + num_scan_inputs = m, + body = loop-body, + scan_input_axes = [axis_1, ..., axis_m] + > (init_1, ..., init_n, scan_1, ..., scan_m) + + is equivalent to the following pseudo-code: + + // scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i + // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. + sequence_length = scan_1.shape[axis_1]; + + // initialize state-variables + st_1 = init_1; ... st_n = init_n; + // initialize scan-output variables: [] denotes an empty tensor + scan_out_1 = []; ...; scan_out_k = []; + // identify number of iterations: + + // execute loop + for (int t = 0; t < sequence_length; ++t) { + // generate the scan-input elements: the notation T[t] indicates the sub-tensor + // of rank one less than T obtained by indexing T at position t along axis k. + si_1 = scan_1[t]; + ... ; + si_m = scan_m[t]; + // execute loop-body + st_1, ..., st_n, so_1, ..., so_k = loop-body(st_1, ..., st_n, si_1, ..., si_m) + // accumulate the scan-output elements + scan_out_1 = Concat(scan_out_1, so_1); ... ; scan_out_k = Concat(scan_out_k, so_k); + } + + return st_1, ..., st_n, scan_out_1, ..., scan_out_k; + + *Sample usage: Encoding RNN using a Scan* + + The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, + recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can + be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes + %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these + values are computed in the outer graph, they need to be passed in as extra state_variables. + + graph rnn-encoding { + %H_0 = ... + %X = ... + %Y_h, %Y = Scan[body = , num_scan_inputs=1](%H_0, %X) + return %Y, %Y_h + } + + graph rnn-cell-1 ( + %H_tminus1[FLOAT, tensor] + %X_t[FLOAT, tensor] + ) { + %Wi = ... + %Ri = ... + %Wbi = ... + %Rbi = ... + %t1 = X_t * (Wi^T) + %t2 = H_tminus1*(Ri^T) + %t3 = Add(%t1, %t2) + %t4 = Add(%t3, %Wbi) + %t5 = Add(%t4, %Rbi) + %Ht = Tanh(%t5) + %Accumulate = Identity(%Ht) + return %Ht, %Accumulate + } + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has N+M inputs: (loop state variables..., scan_input_elts...). It has N+K outputs: (loop state variables..., scan_output_elts...). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
+
num_scan_inputs : int (required)
+
An attribute specifying the number of scan_inputs M.
+
scan_input_axes : list of ints
+
An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
scan_input_directions : list of ints
+
An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
+
scan_output_axes : list of ints
+
An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
+
scan_output_directions : list of ints
+
An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
+
+ +#### Inputs (1 - ∞) + +
+
initial_state_and_scan_inputs (variadic, heterogeneous) : V
+
Initial values of the loop's N state variables followed by M scan_inputs
+
+ +#### Outputs (1 - ∞) + +
+
final_state_and_scan_outputs (variadic, heterogeneous) : V
+
Final values of the loop's N state variables followed by K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
All Tensor types
+
+ +### **Scatter-11** (deprecated) + + This operator is deprecated. Please use ScatterElements, which provides the same functionality. + + Scatter takes three inputs `data`, `updates`, and `indices` of the same + rank r >= 1 and an optional attribute axis that identifies an axis of `data` + (by default, the outer-most axis, that is axis 0). The output of the operation + is produced by creating a copy of the input `data`, and then updating its value + to values specified by `updates` at specific index positions specified by + `indices`. Its output shape is the same as the shape of `data`. + + For each entry in `updates`, the target index in `data` is obtained by combining + the corresponding entry in `indices` with the index of the entry itself: the + index-value for dimension = axis is obtained from the value of the corresponding + entry in `indices` and the index-value for dimension != axis is obtained from the + index of the entry itself. + + For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry + is performed as below: + ``` + output[indices[i][j]][j] = updates[i][j] if axis = 0, + output[i][indices[i][j]] = updates[i][j] if axis = 1, + ``` + + This operator is the inverse of GatherElements. It is similar to Torch's Scatter operation. + + Example 1: + ``` + data = [ + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + ] + indices = [ + [1, 0, 2], + [0, 2, 1], + ] + updates = [ + [1.0, 1.1, 1.2], + [2.0, 2.1, 2.2], + ] + output = [ + [2.0, 1.1, 0.0] + [1.0, 0.0, 2.2] + [0.0, 2.1, 1.2] + ] + ``` + Example 2: + ``` + data = [[1.0, 2.0, 3.0, 4.0, 5.0]] + indices = [[1, 3]] + updates = [[1.1, 2.1]] + axis = 1 + output = [[1.0, 1.1, 3.0, 2.1, 5.0]] + ``` + +#### Version + +This version of the operator has been deprecated since version 11 of the default ONNX operator set. + +### **ScatterElements-11** + + ScatterElements takes three inputs `data`, `updates`, and `indices` of the same + rank r >= 1 and an optional attribute axis that identifies an axis of `data` + (by default, the outer-most axis, that is axis 0). The output of the operation + is produced by creating a copy of the input `data`, and then updating its value + to values specified by `updates` at specific index positions specified by + `indices`. Its output shape is the same as the shape of `data`. + + For each entry in `updates`, the target index in `data` is obtained by combining + the corresponding entry in `indices` with the index of the entry itself: the + index-value for dimension = axis is obtained from the value of the corresponding + entry in `indices` and the index-value for dimension != axis is obtained from the + index of the entry itself. + + For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry + is performed as below: + ``` + output[indices[i][j]][j] = updates[i][j] if axis = 0, + output[i][indices[i][j]] = updates[i][j] if axis = 1, + ``` + + This operator is the inverse of GatherElements. It is similar to Torch's Scatter operation. + + Example 1: + ``` + data = [ + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + ] + indices = [ + [1, 0, 2], + [0, 2, 1], + ] + updates = [ + [1.0, 1.1, 1.2], + [2.0, 2.1, 2.2], + ] + output = [ + [2.0, 1.1, 0.0] + [1.0, 0.0, 2.2] + [0.0, 2.1, 1.2] + ] + ``` + Example 2: + ``` + data = [[1.0, 2.0, 3.0, 4.0, 5.0]] + indices = [[1, 3]] + updates = [[1.1, 2.1]] + axis = 1 + output = [[1.0, 1.1, 3.0, 2.1, 5.0]] + ``` + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Inputs + +
+
data : T
+
Tensor of rank r >= 1.
+
indices : Tind
+
Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
updates : T
+
Tensor of rank r >=1 (same rank and shape as indices)
+
+ +#### Outputs + +
+
output : T
+
Tensor of rank r >= 1 (same rank as input).
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input and output types can be of any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **ScatterND-11** + + ScatterND takes three inputs `data` tensor of rank r >= 1, `indices` tensor of rank q >= 1, + and `updates` tensor of rank q + r - indices.shape[-1] - 1. The output of the operation + is produced by creating a copy of the input `data`, and then updating its value to values + specified by `updates` at specific index positions specified by `indices`. Its output shape + is the same as the shape of `data`. Note that `indices` should not have duplicate entries. + That is, two or more `updates` for the same index-location is not supported. + + `indices` is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of `indices`. + `indices` is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into `data`. + Hence, k can be a value at most the rank of `data`. When k equals rank(data), each update entry specifies an + update to a single element of the tensor. When k is less than rank(data) each update entry specifies an + update to a slice of the tensor. Index values are allowed to be negative, as per the usual + convention for counting backwards from the end, but are expected in the valid range. + + `updates` is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the + first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. + The remaining dimensions of `updates` correspond to the dimensions of the + replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, + corresponding to the trailing (r-k) dimensions of `data`. Thus, the shape of `updates` + must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation + of shapes. + + The `output` is calculated via the following equation: + + output = np.copy(data) + update_indices = indices.shape[:-1] + for idx in np.ndindex(update_indices): + output[tuple(indices[idx])] = updates[idx] + + The order of iteration in the above loop is not specified. + In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. + This ensures that the output value does not depend on the iteration order. + + This operator is the inverse of GatherND. + + Example 1: + ``` + data = [1, 2, 3, 4, 5, 6, 7, 8] + indices = [[4], [3], [1], [7]] + updates = [9, 10, 11, 12] + output = [1, 11, 3, 10, 9, 6, 7, 12] + ``` + + Example 2: + ``` + data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] + indices = [[0], [2]] + updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]] + output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] + ``` + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
data : T
+
Tensor of rank r >= 1.
+
indices : tensor(int64)
+
Tensor of rank q >= 1.
+
updates : T
+
Tensor of rank q + r - indices_shape[-1] - 1.
+
+ +#### Outputs + +
+
output : T
+
Tensor of rank r >= 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
+ +### **SequenceAt-11** + + Outputs a tensor copy from the tensor at 'position' in 'input_sequence'. + Accepted range for 'position' is in `[-n, n - 1]`, where `n` is the number of tensors in 'input_sequence'. + Negative value means counting positions from the back. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
input_sequence : S
+
Input sequence.
+
position : I
+
Position of the tensor in the sequence. Negative value means counting positions from the back. Accepted range in `[-n, n - 1]`, where `n` is the number of tensors in 'input_sequence'. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
+
+ +#### Outputs + +
+
tensor : T
+
Output tensor at the specified position in the input sequence.
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain to any tensor type.
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type.
+
I : tensor(int32), tensor(int64)
+
Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
+
+ +### **SequenceConstruct-11** + + Construct a tensor sequence containing 'inputs' tensors. + All tensors in 'inputs' must have the same data type. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
inputs (variadic) : T
+
Tensors.
+
+ +#### Outputs + +
+
output_sequence : S
+
Sequence enclosing the input tensors.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input types to any tensor type.
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain output types to any tensor type.
+
+ +### **SequenceEmpty-11** + + Construct an empty tensor sequence, with given data type. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int
+
(Optional) The data type of the tensors in the output sequence. The default type is 'float'.
+
+ +#### Inputs + + +#### Outputs + +
+
output : S
+
Empty sequence.
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain output types to any tensor type.
+
+ +### **SequenceErase-11** + + Outputs a tensor sequence that removes the tensor at 'position' from 'input_sequence'. + Accepted range for 'position' is in `[-n, n - 1]`, where `n` is the number of tensors in 'input_sequence'. + Negative value means counting positions from the back. + 'position' is optional, by default it erases the last tensor from 'input_sequence'. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs (1 - 2) + +
+
input_sequence : S
+
Input sequence.
+
position (optional) : I
+
Position of the tensor in the sequence. Negative value means counting positions from the back. Accepted range in `[-n, n - 1]`, where `n` is the number of tensors in 'input_sequence'. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
+
+ +#### Outputs + +
+
output_sequence : S
+
Output sequence that has the tensor at the specified position removed.
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain to any tensor type.
+
I : tensor(int32), tensor(int64)
+
Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
+
+ +### **SequenceInsert-11** + + Outputs a tensor sequence that inserts 'tensor' into 'input_sequence' at 'position'. + 'tensor' must have the same data type as 'input_sequence'. + Accepted range for 'position' is in `[-n, n]`, where `n` is the number of tensors in 'input_sequence'. + Negative value means counting positions from the back. + 'position' is optional, by default it inserts 'tensor' to the back of 'input_sequence'. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs (2 - 3) + +
+
input_sequence : S
+
Input sequence.
+
tensor : T
+
Input tensor to be inserted into the input sequence.
+
position (optional) : I
+
Position in the sequence where the new tensor is inserted. It is optional and default is to insert to the back of the sequence. Negative value means counting positions from the back. Accepted range in `[-n, n]`, where `n` is the number of tensors in 'input_sequence'. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
+
+ +#### Outputs + +
+
output_sequence : S
+
Output sequence that contains the inserted tensor at given position.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type.
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain to any tensor type.
+
I : tensor(int32), tensor(int64)
+
Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
+
+ +### **SequenceLength-11** + + Produces a scalar(tensor of empty shape) containing the number of tensors in 'input_sequence'. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
input_sequence : S
+
Input sequence.
+
+ +#### Outputs + +
+
length : I
+
Length of input sequence. It must be a scalar(tensor of empty shape).
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain to any tensor type.
+
I : tensor(int64)
+
Constrain output to integral tensor. It must be a scalar(tensor of empty shape).
+
+ +### **Slice-11** + + Produces a slice of the input tensor along multiple axes. Similar to numpy: + https://numpy.org/doc/stable/reference/routines.indexing.html + Slices uses `starts`, `ends`, `axes` and `steps` inputs to specify the start and end + dimension and step for each axis in the list of axes, it uses this information to + slice the input `data` tensor. If a negative value is passed for any of the + start or end indices, it represents number of elements before the end of that + dimension. If the value passed to start or end is larger than the `n` (the + number of elements in this dimension), it represents `n`. For slicing to the + end of a dimension with unknown size, it is recommended to pass in `INT_MAX` + when slicing forward and 'INT_MIN' when slicing backward. + If a negative value is passed for step, it represents slicing backward. + However step value cannot be 0. + If `axes` are omitted, they are set to `[0, ..., ndim-1]`. + If `steps` are omitted, they are set to `[1, ..., 1]` of length `len(starts)` + Example 1: + data = [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + axes = [0, 1] + starts = [1, 0] + ends = [2, 3] + steps = [1, 2] + result = [ + [5, 7], + ] + Example 2: + data = [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + starts = [0, 1] + ends = [-1, 1000] + result = [ + [2, 3, 4], + ] + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs (3 - 5) + +
+
data : T
+
Tensor of data to extract slices from.
+
starts : Tind
+
1-D tensor of starting indices of corresponding axis in `axes`
+
ends : Tind
+
1-D tensor of ending indices (exclusive) of corresponding axis in `axes`
+
axes (optional) : Tind
+
1-D tensor of axes that `starts` and `ends` apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
steps (optional) : Tind
+
1-D tensor of slice step of corresponding axis in `axes`. Negative value means slicing backward. 'steps' cannot be 0. Defaults to 1.
+
+ +#### Outputs + +
+
output : T
+
Sliced data tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **Softmax-11** + + The operator computes the softmax (normalized exponential) values for each layer in the batch + of the given input. + + The input does not need to explicitly be a 2D vector; rather, it will be + coerced into one. For an arbitrary n-dimensional tensor + input \in [a_0, a_1, ..., a_{k-1}, a_k, ..., a_{n-1}] and k is + the axis provided, then input will be coerced into a 2-dimensional tensor with + dimensions [a_0 * ... * a_{k-1}, a_k * ... * a_{n-1}]. For the default + case where axis=1, this means the input tensor will be coerced into a 2D tensor + of dimensions [a_0, a_1 * ... * a_{n-1}], where a_0 is often the batch size. + In this situation, we must have a_0 = N and a_1 * ... * a_{n-1} = D. + Each of these dimensions must be matched correctly, or else the operator + will throw errors. The output tensor has the same shape + and contains the softmax values of the corresponding input. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
+ +#### Inputs + +
+
input : T
+
The input tensor that's coerced into a 2D matrix of size (NxD) as described above.
+
+ +#### Outputs + +
+
output : T
+
The output values with the same shape as input tensor (the original size without coercion).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Split-11** + + Split a tensor into a list of tensors, along the specified + 'axis'. Lengths of the parts can be specified using argument 'split'. + Otherwise, the tensor is split to equal sized parts. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1] where r = rank(input).
+
split : list of ints
+
length of each output. Values should be >= 0.
+
+ +#### Inputs + +
+
input : T
+
The tensor to split
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic) : T
+
One or more outputs forming list of tensors after splitting
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **SplitToSequence-11** + + Split a tensor into a sequence of tensors, along the specified 'axis'. + Lengths of the parts can be specified using the optional argument 'split'. + If the argument `split' is not specified, a default scalar value of 1 + is used as the value of `split'. + 'split' must contain only positive numbers. + 'split' is either a scalar (tensor of empty shape), or a 1-D tensor. + If 'split' is a scalar, then 'input' will be split into chunks all of size 'split' + if possible. The last chunk alone may be smaller than 'split' if the 'input' size + along the given axis 'axis' is not divisible by 'split'. + If 'split' is a 1-dimensional tensor, the input tensor is split into 'size(split)' chunks, + with lengths of the parts on 'axis' specified in 'split'. In this scenario, the sum of entries + in 'split' must be equal to the dimension size of input tensor on 'axis'. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1].
+
keepdims : int (default is 1)
+
Keep the split dimension or not. Default 1, which means we keep split dimension. If input 'split' is specified, this attribute is ignored.
+
+ +#### Inputs (1 - 2) + +
+
input : T
+
The tensor to split
+
split (optional) : I
+
Length of each output. It can be either a scalar(tensor of empty shape), or a 1-D tensor. All values must be >= 0.
+
+ +#### Outputs + +
+
output_sequence : S
+
One or more outputs forming a sequence of tensors after splitting
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input types to all tensor types.
+
I : tensor(int32), tensor(int64)
+
Constrain split size to integral tensor.
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain output types to all tensor types.
+
+ +### **Squeeze-11** + + Remove single-dimensional entries from the shape of a tensor. + Takes a parameter `axes` with a list of axes to squeeze. + If `axes` is not provided, all the single dimensions will be removed from + the shape. If an axis is selected with shape entry not equal to one, an error is raised. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
List of integers indicating the dimensions to squeeze. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Inputs + +
+
data : T
+
Tensors with at least max(dims) dimensions.
+
+ +#### Outputs + +
+
squeezed : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **TopK-11** + + Retrieve the top-K largest or smallest elements along a specified axis. Given an input tensor of + shape [a_0, a_1, ..., a_{n-1}] and integer argument k, return two outputs: + + * Value tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] + which contains the values of the top k elements along the specified axis + * Index tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] which + contains the indices of the top k elements (original indices from the input + tensor). + + * If "largest" is 1 (the default value) then the k largest elements are returned. + * If "sorted" is 1 (the default value) then the resulting k elements will be sorted. + * If "sorted" is 0, order of returned 'Values' and 'Indices' are undefined. + + Given two equivalent values, this operator uses the indices along the axis as + a tiebreaker. That is, the element with the lower index will appear first. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
Dimension on which to do the sort. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
largest : int (default is 1)
+
Whether to return the top-K largest or smallest elements.
+
sorted : int (default is 1)
+
Whether to return the elements in sorted order.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Tensor of shape [a_0, a_1, ..., a_{n-1}]
+
K (non-differentiable) : tensor(int64)
+
A 1-D tensor containing a single positive value corresponding to the number of top elements to retrieve
+
+ +#### Outputs + +
+
Values (differentiable) : T
+
Tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] containing top K values from the input tensor
+
Indices (non-differentiable) : I
+
Tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] containing the corresponding input tensor indices for the top K values.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to numeric tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **Unique-11** + + Find the unique elements of a tensor. When an optional attribute 'axis' is provided, unique subtensors sliced along the 'axis' are returned. + Otherwise the input tensor is flattened and unique values of the flattened tensor are returned. + + This operator returns the unique values or sliced unique subtensors of the input tensor and three optional outputs. + The first output tensor 'Y' contains all unique values or subtensors of the input. + The second optional output tensor 'indices' contains indices of 'Y' elements' first occurrence in 'X'. + The third optional output tensor 'inverse_indices' contains, for elements of 'X', its corresponding indices in 'Y'. + The fourth optional output tensor 'counts' contains the count of each element of 'Y' in the input. + + Outputs are either sorted in ascending order or optionally in the order of the first occurrence of the values in the input. + + https://docs.scipy.org/doc/numpy/reference/generated/numpy.unique.html + + Example 1: + ``` + input_X = [2, 1, 1, 3, 4, 3] + attribute_sorted = 0 + attribute_axis = None + output_Y = [2, 1, 3, 4] + output_indices = [0, 1, 3, 4] + output_inverse_indices = [0, 1, 1, 2, 3, 2] + output_counts = [1, 2, 2, 1] + ``` + + Example 2: + ``` + input_X = [[1, 3], [2, 3]] + attribute_sorted = 1 + attribute_axis = None + output_Y = [1, 2, 3] + output_indices = [0, 2, 1] + output_inverse_indices = [0, 2, 1, 2] + output_counts = [1, 1, 2] + ``` + + Example 3: + ``` + input_X = [[1, 0, 0], [1, 0, 0], [2, 3, 4]] + attribute_sorted = 1 + attribute_axis = 0 + output_Y = [[1, 0, 0], [2, 3, 4]] + output_indices = [0, 2] + output_inverse_indices = [0, 0, 1] + output_counts = [2, 1] + ``` + + Example 4: + ``` + input_x = [[[1., 1.], [0., 1.], [2., 1.], [0., 1.]], + [[1., 1.], [0., 1.], [2., 1.], [0., 1.]]] + attribute_sorted = 1 + attribute_axis = 1 + ``` + + intermediate data are presented below for better understanding: + there are 4 subtensors sliced along axis 1 of input_x (shape = (2, 4, 2)): + ``` + A: [[1, 1], [1, 1]], + [[0, 1], [0, 1]], + [[2, 1], [2, 1]], + [[0, 1], [0, 1]]. + ``` + + there are 3 unique subtensors: + ``` + [[1, 1], [1, 1]], + [[0, 1], [0, 1]], + [[2, 1], [2, 1]]. + ``` + + sorted unique subtensors: + ``` + B: [[0, 1], [0, 1]], + [[1, 1], [1, 1]], + [[2, 1], [2, 1]]. + ``` + + output_Y is constructed from B: + ``` + [[[0. 1.], [1. 1.], [2. 1.]], + [[0. 1.], [1. 1.], [2. 1.]]] + ``` + + output_indices is to map from B to A: + ``` + [1, 0, 2] + ``` + + output_inverse_indices is to map from A to B: + ``` + [1, 0, 2, 0] + ``` + + output_counts: + ``` + [2, 1, 1] + ``` + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
(Optional) The dimension to apply unique. If not specified, the unique elements of the flattened input are returned. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
sorted : int (default is 1)
+
(Optional) Whether to sort the unique elements in ascending order before returning as output. Must be one of 0, or 1 (default).
+
+ +#### Inputs + +
+
X (non-differentiable) : T
+
A N-D input tensor that is to be processed.
+
+ +#### Outputs (1 - 4) + +
+
Y (non-differentiable) : T
+
A tensor of the same type as 'X' containing all the unique values or subtensors sliced along a provided 'axis' in 'X', either sorted or maintained in the same order they occur in input 'X'
+
indices (optional, non-differentiable) : tensor(int64)
+
A 1-D INT64 tensor containing indices of 'Y' elements' first occurrence in 'X'. When 'axis' is provided, it contains indices to subtensors in input 'X' on the 'axis'. When 'axis' is not provided, it contains indices to values in the flattened input tensor.
+
inverse_indices (optional, non-differentiable) : tensor(int64)
+
A 1-D INT64 tensor containing, for elements of 'X', its corresponding indices in 'Y'. When 'axis' is provided, it contains indices to subtensors in output 'Y' on the 'axis'. When 'axis' is not provided, it contains indices to values in output 'Y'.
+
counts (optional, non-differentiable) : tensor(int64)
+
A 1-D INT64 tensor containing the count of each element of 'Y' in input 'X'
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input can be of any tensor type.
+
+ +### **Unsqueeze-11** + + Insert single-dimensional entries to the shape of an input tensor (`data`). + Takes one required argument `axes` - which contains a list of dimension indices and this operator will insert a dimension of value `1` into the corresponding index of the output tensor (`expanded`). + + For example: + Given an input tensor (`data`) of shape [3, 4, 5], then + Unsqueeze(data, axes=[0, 4]) outputs a tensor (`expanded`) containing same data as `data` but with shape [1, 3, 4, 5, 1]. + + The attribute `axes` should not contain any duplicate entries. It is an error if it contains duplicates. + The rank of the output tensor (`output_rank`) is the rank of the input tensor (`data`) plus the number of values in `axes`. + Each value in `axes` should be within the (inclusive) range [-output_rank , output_rank - 1]. + The order of values in `axes` does not matter and can come in any order. + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints (required)
+
List of integers indicating the dimensions to be inserted. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(expanded).
+
+ +#### Inputs + +
+
data : T
+
Original tensor
+
+ +#### Outputs + +
+
expanded : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +## Version 12 of the default ONNX operator set +### **ArgMax-12** + + Computes the indices of the max elements of the input tensor's element along the + provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. + If keepdims equal 0, then the resulting tensor has the reduced dimension pruned. + If select_last_index is True (default False), the index of the last occurrence of the max + is selected if the max appears more than once in the input. Otherwise the index of the + first occurrence is selected. + The type of the output tensor is integer. + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
select_last_index : int (default is 0)
+
Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index).
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : tensor(int64)
+
Reduced output tensor with integer data type.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **ArgMin-12** + + Computes the indices of the min elements of the input tensor's element along the + provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. + If keepdims equal 0, then the resulting tensor has the reduced dimension pruned. + If select_last_index is True (default False), the index of the last occurrence of the min + is selected if the min appears more than once in the input. Otherwise the index of the + first occurrence is selected. + The type of the output tensor is integer. + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
select_last_index : int (default is 0)
+
Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index).
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : tensor(int64)
+
Reduced output tensor with integer data type.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **Celu-12** + + Continuously Differentiable Exponential Linear Units: + Perform the linear unit element-wise on the input tensor X + using formula: + + ``` + max(0,x) + min(0,alpha*(exp(x/alpha)-1)) + ``` + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
The Alpha value in Celu formula which control the shape of the unit. The default value is 1.0.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float)
+
Constrain input and output types to float32 tensors.
+
+ +### **Clip-12** + + Clip operator limits the given input within an interval. The interval is + specified by the inputs 'min' and 'max'. They default to + numeric_limits::lowest() and numeric_limits::max(), respectively. + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Inputs (1 - 3) + +
+
input : T
+
Input tensor whose elements to be clipped
+
min (optional) : T
+
Minimum value, under which element is replaced by min. It must be a scalar(tensor of empty shape).
+
max (optional) : T
+
Maximum value, above which element is replaced by max. It must be a scalar(tensor of empty shape).
+
+ +#### Outputs + +
+
output : T
+
Output tensor with clipped input elements
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numeric tensors.
+
+ +### **Constant-12** + + This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, + or value_* must be specified. + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
sparse_value : sparse_tensor
+
The value for the elements of the output tensor in sparse format.
+
value : tensor
+
The value for the elements of the output tensor.
+
value_float : float
+
The value for the sole element for the scalar, float32, output tensor.
+
value_floats : list of floats
+
The values for the elements for the 1D, float32, output tensor.
+
value_int : int
+
The value for the sole element for the scalar, int64, output tensor.
+
value_ints : list of ints
+
The values for the elements for the 1D, int64, output tensor.
+
value_string : string
+
The value for the sole element for the scalar, UTF-8 string, output tensor.
+
value_strings : list of strings
+
The values for the elements for the 1D, UTF-8 string, output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Dropout-12** + + Dropout takes an input floating-point tensor, an optional input ratio (floating-point scalar) and an optional input training_mode (boolean scalar). It produces two tensor outputs, + output (floating-point tensor) and mask (optional `Tensor`). If `training_mode` is true then the output Y will be a random dropout; + Note that this Dropout scales the masked input data by the following equation, so to convert the trained model into inference mode, + the user can simply not pass `training_mode` input or set it to false. + ``` + output = scale * data * mask, + ``` + where + ``` + scale = 1. / (1. - ratio). + ``` + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
seed : int
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs (1 - 3) + +
+
data : T
+
The input data as Tensor.
+
ratio (optional) : T1
+
The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it's non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
+
training_mode (optional) : T2
+
If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
+
+ +#### Outputs (1 - 2) + +
+
output : T
+
The output.
+
mask (optional) : T2
+
The output mask.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(float16), tensor(float), tensor(double)
+
Constrain input 'ratio' types to float tensors.
+
T2 : tensor(bool)
+
Constrain output 'mask' types to boolean tensors.
+
+ +### **Einsum-12** + + An einsum of the form `term1, term2 -> output-term` produces an output tensor using the following equation + + ``` + output[output-term] = reduce-sum( input1[term1] * input2[term2] ) + ``` + + where the reduce-sum performs a summation over all the indices occurring in the input terms (term1, term2) + that do not occur in the output-term. + + The Einsum operator evaluates algebraic tensor operations on a sequence of tensors, using the Einstein summation + convention. The equation string contains a comma-separated sequence of lower case letters. Each term corresponds to + an operand tensor, and the characters within the terms correspond to operands dimensions. + + This sequence may be followed by "->" to separate the left and right hand side of the equation. + If the equation contains "->" followed by the right-hand side, the explicit (not classical) form of the Einstein + summation is performed, and the right-hand side indices indicate output tensor dimensions. In other cases, + output indices are (implicitly) set to the alphabetically sorted sequence of indices appearing exactly once in the + equation. + + When a dimension character is repeated in the left-hand side, it represents summation along the dimension. + + The equation may contain ellipsis ("...") to enable broadcasting. Ellipsis must indicate a fixed number of dimensions. + Specifically, every occurrence of ellipsis in the equation must represent the same number of dimensions. + The right-hand side may contain exactly one ellipsis. In implicit mode, the ellipsis dimensions are set to the + beginning of the output. The equation string may contain space (U+0020) character. + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
equation : string (required)
+
Einsum expression string.
+
+ +#### Inputs (1 - ∞) + +
+
Inputs (variadic, differentiable) : T
+
Operands
+
+ +#### Outputs + +
+
Output (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numerical tensor types.
+
+ +### **GatherND-12** + + Given `data` tensor of rank `r` >= 1, `indices` tensor of rank `q` >= 1, and `batch_dims` integer `b`, this operator gathers + slices of `data` into an output tensor of rank `q + r - indices_shape[-1] - 1 - b`. + + `indices` is an q-dimensional integer tensor, best thought of as a `(q-1)`-dimensional tensor of index-tuples into `data`, + where each element defines a slice of `data` + + `batch_dims` (denoted as `b`) is an integer indicating the number of batch dimensions, i.e the leading `b` number of dimensions of + `data` tensor and `indices` are representing the batches, and the gather starts from the `b+1` dimension. + + Some salient points about the inputs' rank and shape: + + 1) r >= 1 and q >= 1 are to be honored. There is no dependency condition to be met between ranks `r` and `q` + + 2) The first `b` dimensions of the shape of `indices` tensor and `data` tensor must be equal. + + 3) b < min(q, r) is to be honored. + + 4) The `indices_shape[-1]` should have a value between 1 (inclusive) and rank `r-b` (inclusive) + + 5) All values in `indices` are expected to be within bounds [-s, s-1] along axis of size `s` (i.e.) `-data_shape[i] <= indices[...,i] <= data_shape[i] - 1`. + It is an error if any of the index values are out of bounds. + + The output is computed as follows: + + The output tensor is obtained by mapping each index-tuple in the `indices` tensor to the corresponding slice of the input `data`. + + 1) If `indices_shape[-1] > r-b` => error condition + + 2) If `indices_shape[-1] == r-b`, since the rank of `indices` is `q`, `indices` can be thought of as `N` `(q-b-1)`-dimensional tensors + containing 1-D tensors of dimension `r-b`, where `N` is an integer equals to the product of 1 and all the elements in the batch dimensions + of the indices_shape. Let us think of each such `r-b` ranked tensor as `indices_slice`. Each *scalar value* corresponding to `data[0:b-1,indices_slice]` + is filled into the corresponding location of the `(q-b-1)`-dimensional tensor to form the `output` tensor (Example 1 below) + + 3) If `indices_shape[-1] < r-b`, since the rank of `indices` is `q`, `indices` can be thought of as `N` `(q-b-1)`-dimensional tensor + containing 1-D tensors of dimension `< r-b`. Let us think of each such tensors as `indices_slice`. Each *tensor slice* corresponding + to `data[0:b-1, indices_slice , :]` is filled into the corresponding location of the `(q-b-1)`-dimensional tensor + to form the `output` tensor (Examples 2, 3, 4 and 5 below) + + This operator is the inverse of `ScatterND`. + + `Example 1` + + batch_dims = 0 + + data = [[0,1],[2,3]] # data_shape = [2, 2] + + indices = [[0,0],[1,1]] # indices_shape = [2, 2] + + output = [0,3] # output_shape = [2] + + `Example 2` + + batch_dims = 0 + + data = [[0,1],[2,3]] # data_shape = [2, 2] + + indices = [[1],[0]] # indices_shape = [2, 1] + + output = [[2,3],[0,1]] # output_shape = [2, 2] + + `Example 3` + + batch_dims = 0 + + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + + indices = [[0,1],[1,0]] # indices_shape = [2, 2] + + output = [[2,3],[4,5]] # output_shape = [2, 2] + + `Example 4` + + batch_dims = 0 + + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + + indices = [[[0,1]],[[1,0]]] # indices_shape = [2, 1, 2] + + output = [[[2,3]],[[4,5]]] # output_shape = [2, 1, 2] + + `Example 5` + + batch_dims = 1 + + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + + indices = [[1],[0]] # indices_shape = [2, 1] + + output = [[2,3],[4,5]] # output_shape = [2, 2] + + + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
batch_dims : int (default is 0)
+
The number of batch dimensions. The gather of indexing starts from dimension of data[batch_dims:]
+
+ +#### Inputs + +
+
data : T
+
Tensor of rank r >= 1.
+
indices : tensor(int64)
+
Tensor of rank q >= 1. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output : T
+
Tensor of rank q + r - indices_shape[-1] - 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
+ +### **GreaterOrEqual-12** + + Returns the tensor resulted from performing the `greater_equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **LessOrEqual-12** + + Returns the tensor resulted from performing the `less_equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Max-12** + + Element-wise max of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for max.
+
+ +#### Outputs + +
+
max : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to numeric tensors.
+
+ +### **MaxPool-12** + + MaxPool consumes an input tensor X and applies max pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + max pooling consisting of computing the max on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape is calculated differently + depending on whether explicit padding is used, where pads is employed, or auto padding is used, where auto_pad is utilized. + With explicit padding (https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html?highlight=maxpool#torch.nn.MaxPool2d): + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled. `pad_shape[i]` is the sum of pads along axis `i`. + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following when ceil_mode is enabled: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + or when ceil_mode is disabled (https://www.tensorflow.org/api_docs/python/tf/keras/layers/AveragePooling2D): + ``` + VALID: output_spatial_shape[i] = floor((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i]) + 1 + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = floor((input_spatial_shape[i] - 1) / strides_spatial_shape[i]) + 1 + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i] + ``` + The output of each pooling window is maximum number of elements exclude pad. + + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
dilations : list of ints
+
Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
storage_order : int (default is 0)
+
The storage order of the tensor. 0 is row major, and 1 is column major. This attribute is used only to convert an n-tuple index value into a single integer value for producing the second output.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs (1 - 2) + +
+
Y (differentiable) : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
Indices (optional, non-differentiable) : I
+
Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x ... x Dn).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(uint8)
+
Constrain input and output types to float and 8 bit tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **Min-12** + + Element-wise min of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic) : T
+
List of tensors for min.
+
+ +#### Outputs + +
+
min : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to numeric tensors.
+
+ +### **NegativeLogLikelihoodLoss-12** + + A NegativeLogLikelihoodLoss operator computes (weighted) negative log likelihood loss. + Its "input" tensor has the shape of (N, C, d1, d2, ..., dk) where k >= 0. + The "input" tensor contains log-probabilities for input[n, :, d_1, d_2,..., d_k] being in a class of [0, C). + The operator's "target" input tensor has the shape of (N, d1, d2, ..., dk). It encodes class labels (one of C classes) + or it may contain a special value (indicated by an attribute ignore_index) for N x d1 x d2 x ... x dk samples. + The loss value for input[n, :, d_1, d_2,...d_k] being classified as class c = target[n][d_1][d_2]...[d_k] is computed as: + loss[n][d_1][d_2]...[d_k] = -input[n][c][d_1][d_2]...[d_k]. + When an optional "weight" is provided, the sample loss is calculated as: + loss[n][d_1][d_2]...[d_k] = -input[n][c][d_1][d_2]...[d_k] * weight[c]. + loss is zero for the case when target-value equals ignore_index. + + loss[n][d_1][d_2]...[d_k] = 0, when target[n][d_1][d_2]...[d_k] = ignore_index + If "reduction" attribute is set to "none", the operator's output will be the above loss with shape (N, d1, d2, ..., dk). + If "reduction" attribute is set to "mean" (the default attribute value), the output loss is (weight) averaged: + mean(loss), if "weight" is not provided, + or if weight is provided, + sum(loss) / sum(weight[target[n][d_1][d_2]...[d_k]]]), for all samples. + If "reduction" attribute is set to "sum", the output is a scalar: + sum(loss). + See also https://pytorch.org/docs/stable/nn.html#torch.nn.NLLLoss. + Example 1: + // negative log likelihood loss, "none" reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + loss = np.zeros((N, d1)) + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] + // print(loss) + // [[-3. -2.] + // [-0. -2.]] + Example 2: + // weighted negative log likelihood loss, sum reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + weight = [0.2, 0.3, 0.1] + loss = np.zeros((N, d1)) + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] * weight[c] + loss = np.sum(loss) + // print(loss) + // -1.1 + Example 3: + // weighted negative log likelihood loss, mean reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + weight = [0.2, 0.3, 0.1] + loss = np.zeros((N, d1)) + weight_total = 0 + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] * weight[c] + weight_total = weight_total + weight[c] + loss = np.sum(loss) / weight_total + // print(loss) + // -1.57 + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
ignore_index : int
+
Specifies a target value that is ignored and does not contribute to the input gradient. It's an optional value.
+
reduction : string (default is mean)
+
Type of reduction to apply to loss: none, sum, mean (default). 'none': the output is the loss for each sample. 'sum': the output will be summed. 'mean': the sum of the output will be divided by the sum of applied weights.
+
+ +#### Inputs (2 - 3) + +
+
input : T
+
Input tensor of shape (N, C) or (N, C, d1, d2, ..., dk).
+
target : Tind
+
Target tensor of shape (N) or (N, d1, d2, ..., dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
+
weight (optional) : T
+
Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
+
+ +#### Outputs + +
+
loss : T
+
The negative log likelihood loss
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input, weight, and output types to floating-point tensors.
+
Tind : tensor(int32), tensor(int64)
+
Constrain target to integer types
+
+ +### **Pow-12** + + Pow takes input data (Tensor) and exponent Tensor, and + produces one output data (Tensor) where the function `f(x) = x^exponent`, + is applied to the data tensor elementwise. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Inputs + +
+
X : T
+
First operand, base of the exponent.
+
Y : T1
+
Second operand, power of the exponent.
+
+ +#### Outputs + +
+
Z : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input X and output types to float/int tensors.
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input Y types to float/int tensors.
+
+ +### **ReduceMax-12** + + Computes the max of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(uint8), tensor(int8)
+
Constrain input and output types to high-precision and 8 bit numeric tensors.
+
+ +### **ReduceMin-12** + + Computes the min of the input tensor's element along the provided axes. The resulting + tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then + the resulted tensor have the reduced dimension pruned. + + The above behavior is similar to numpy, with the exception that numpy defaults keepdims to + False instead of True. + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(uint8), tensor(int8)
+
Constrain input and output types to high-precision and 8 bit numeric tensors.
+
+ +### **SoftmaxCrossEntropyLoss-12** + + Loss function that measures the softmax cross entropy + between 'scores' and 'labels'. + This operator first computes a loss tensor whose shape is identical to the labels input. + If the input is 2-D with shape (N, C), the loss tensor may be a N-element vector L = (l_1, l_2, ..., l_N). + If the input is N-D tensor with shape (N, C, D1, D2, ..., Dk), + the loss tensor L may have (N, D1, D2, ..., Dk) as its shape and L[i,][j_1][j_2]...[j_k] denotes a scalar element in L. + After L is available, this operator can optionally do a reduction operator. + + shape(scores): (N, C) where C is the number of classes, or (N, C, D1, D2,..., Dk), + with K >= 1 in case of K-dimensional loss. + shape(labels): (N) where each value is 0 <= labels[i] <= C-1, or (N, D1, D2,..., Dk), + with K >= 1 in case of K-dimensional loss. + + The loss for one sample, l_i, can calculated as follows: + l[i][d1][d2]...[dk] = -y[i][c][d1][d2]..[dk], where i is the index of classes. + or + l[i][d1][d2]...[dk] = -y[i][c][d1][d2]..[dk] * weights[c], if 'weights' is provided. + + loss is zero for the case when label-value equals ignore_index. + l[i][d1][d2]...[dk] = 0, when labels[n][d1][d2]...[dk] = ignore_index + + where: + p = Softmax(scores) + y = Log(p) + c = labels[i][d1][d2]...[dk] + + Finally, L is optionally reduced: + If reduction = 'none', the output is L with shape (N, D1, D2, ..., Dk). + If reduction = 'sum', the output is scalar: Sum(L). + If reduction = 'mean', the output is scalar: ReduceMean(L), or if weight is provided: ReduceSum(L) / ReduceSum(W), + where tensor W is of shape (N, D1, D2, ..., Dk) and W[n][d1][d2]...[dk] = weights[labels[i][d1][d2]...[dk]]. + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
ignore_index : int
+
Specifies a target value that is ignored and does not contribute to the input gradient. It's an optional value.
+
reduction : string (default is mean)
+
Type of reduction to apply to loss: none, sum, mean(default). 'none': no reduction will be applied, 'sum': the output will be summed. 'mean': the sum of the output will be divided by the number of elements in the output.
+
+ +#### Inputs (2 - 3) + +
+
scores : T
+
The predicted outputs with shape [batch_size, class_size], or [batch_size, class_size, D1, D2 , ..., Dk], where K is the number of dimensions.
+
labels : Tind
+
The ground truth output tensor, with shape [batch_size], or [batch_size, D1, D2, ..., Dk], where K is the number of dimensions. Labels element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the label values should either be in the range [0, C) or have the value ignore_index.
+
weights (optional) : T
+
A manual rescaling weight given to each class. If given, it has to be a 1D Tensor assigning weight to each of the classes. Otherwise, it is treated as if having all ones.
+
+ +#### Outputs (1 - 2) + +
+
output : T
+
Weighted loss float Tensor. If reduction is 'none', this has the shape of [batch_size], or [batch_size, D1, D2, ..., Dk] in case of K-dimensional loss. Otherwise, it is a scalar.
+
log_prob (optional) : T
+
Log probability tensor. If the output of softmax is prob, its value is log(prob).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
Tind : tensor(int32), tensor(int64)
+
Constrain target to integer types
+
+ +## Version 13 of the default ONNX operator set +### **Abs-13** + + Absolute takes one input data (Tensor) and produces one output data + (Tensor) where absolute value, y = abs(x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ +### **Add-13** + + Performs element-wise binary addition (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **ArgMax-13** + + Computes the indices of the max elements of the input tensor's element along the + provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. + If keepdims equals 0, then the resulting tensor has the reduced dimension pruned. + If select_last_index is True (default False), the index of the last occurrence of the max + is selected if the max appears more than once in the input. Otherwise the index of the + first occurrence is selected. + The type of the output tensor is integer. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
select_last_index : int (default is 0)
+
Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index).
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (non-differentiable) : tensor(int64)
+
Reduced output tensor with integer data type.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ +### **ArgMin-13** + + Computes the indices of the min elements of the input tensor's element along the + provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. + If keepdims equals 0, then the resulting tensor has the reduced dimension pruned. + If select_last_index is True (default False), the index of the last occurrence of the min + is selected if the min appears more than once in the input. Otherwise the index of the + first occurrence is selected. + The type of the output tensor is integer. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
select_last_index : int (default is 0)
+
Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index).
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (non-differentiable) : tensor(int64)
+
Reduced output tensor with integer data type.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ +### **Cast-13** + + The operator casts the elements of a given input tensor to a data type + specified by the 'to' argument and returns an output tensor of the same size in + the converted type. The 'to' argument must be one of the data types specified + in the 'DataType' enum field in the TensorProto message. + + Casting from string tensor in plain (e.g., "3.14" and "1000") and scientific numeric representations + (e.g., "1e-5" and "1E8") to float types is supported. For example, converting string "100.5" to an integer may + yield result 100. There are some string literals reserved for special floating-point values; + "+INF" (and "INF"), "-INF", and "NaN" are positive infinity, negative infinity, and not-a-number, respectively. + Any string which can exactly match "+INF" in a case-insensitive way would be mapped to positive infinite. Similarly, + this case-insensitive rule is applied to "INF" and "NaN". When casting from numeric tensors + to string tensors, plain floating-point representation (such as "314.15926") would be used. + Converting non-numerical-literal string such as "Hello World!" is an undefined behavior. Cases + of converting string representing floating-point arithmetic value, such as "2.718", to INT is an undefined behavior. + + Conversion from a numerical type to any numerical type is always allowed. + User must be aware of precision loss and value change caused by range difference between two types. + For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting + an integer 36 to Boolean may produce 1 because we truncate bits which can't be stored in the targeted type. + + In more detail, the conversion among numerical types should follow these rules: + + * Casting from floating point to: + * floating point: +/- infinity if OOR (out of range). + * fixed point: undefined if OOR. + * bool: +/- 0.0 to False; all else to True. + * Casting from fixed point to: + * floating point: +/- infinity if OOR. (+ infinity in the case of uint) + * fixed point: when OOR, discard higher bits and reinterpret (with respect to two's complement representation for + signed types). For example, 200 (int16) -> -56 (int8). + * bool: zero to False; nonzero to True. + * Casting from bool to: + * floating point: `{1.0, 0.0}`. + * fixed point: `{1, 0}`. + * bool: no change. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
to : int (required)
+
The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor with the same shape as input with type specified by the 'to' argument
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string), tensor(bfloat16)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string), tensor(bfloat16)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **Ceil-13** + + Ceil takes one input data (Tensor) and produces one output data + (Tensor) where the ceil is, y = ceil(x), is applied to + the tensor elementwise. If x is integral, +0, -0, NaN, or infinite, x itself is returned. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **Clip-13** + + Clip operator limits the given input within an interval. The interval is + specified by the inputs 'min' and 'max'. They default to + numeric_limits::lowest() and numeric_limits::max(), respectively. + When 'min' is greater than 'max', the clip operator sets all the 'input' values to + the value of 'max'. Thus, this is equivalent to 'Min(max, Max(input, min))'. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs (1 - 3) + +
+
input (differentiable) : T
+
Input tensor whose elements to be clipped
+
min (optional, non-differentiable) : T
+
Minimum value, under which element is replaced by min. It must be a scalar(tensor of empty shape).
+
max (optional, non-differentiable) : T
+
Maximum value, above which element is replaced by max. It must be a scalar(tensor of empty shape).
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor with clipped input elements
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ +### **Concat-13** + + Concatenate a list of tensors into a single tensor. All input tensors must have the same shape, except for the dimension size of the axis to concatenate on. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (required)
+
Which axis to concat on. A negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(inputs)..
+
+ +#### Inputs (1 - ∞) + +
+
inputs (variadic, differentiable) : T
+
List of tensors for concatenation
+
+ +#### Outputs + +
+
concat_result (differentiable) : T
+
Concatenated tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain output types to any tensor type.
+
+ +### **Constant-13** + + This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, + or value_* must be specified. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
sparse_value : sparse_tensor
+
The value for the elements of the output tensor in sparse format.
+
value : tensor
+
The value for the elements of the output tensor.
+
value_float : float
+
The value for the sole element for the scalar, float32, output tensor.
+
value_floats : list of floats
+
The values for the elements for the 1D, float32, output tensor.
+
value_int : int
+
The value for the sole element for the scalar, int64, output tensor.
+
value_ints : list of ints
+
The values for the elements for the 1D, int64, output tensor.
+
value_string : string
+
The value for the sole element for the scalar, UTF-8 string, output tensor.
+
value_strings : list of strings
+
The values for the elements for the 1D, UTF-8 string, output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **DepthToSpace-13** + + DepthToSpace rearranges (permutes) data from depth into blocks of spatial data. + This is the reverse transformation of SpaceToDepth. More specifically, this op outputs a copy of + the input tensor where values from the depth dimension are moved in spatial blocks to the height + and width dimensions. By default, `mode` = `DCR`. + In the DCR mode, elements along the depth dimension from the input tensor are rearranged in the + following order: depth, column, and then row. The output y is computed from the input x as below: + + ``` + b, c, h, w = x.shape + tmp = np.reshape(x, [b, blocksize, blocksize, c // (blocksize**2), h, w]) + tmp = np.transpose(tmp, [0, 3, 4, 1, 5, 2]) + y = np.reshape(tmp, [b, c // (blocksize**2), h * blocksize, w * blocksize]) + ``` + + In the CRD mode, elements along the depth dimension from the input tensor are rearranged in the + following order: column, row, and the depth. The output y is computed from the input x as below: + + ``` + b, c, h, w = x.shape + tmp = np.reshape(x, [b, c // (blocksize ** 2), blocksize, blocksize, h, w]) + tmp = np.transpose(tmp, [0, 1, 4, 2, 5, 3]) + y = np.reshape(tmp, [b, c // (blocksize ** 2), h * blocksize, w * blocksize]) + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
blocksize : int (required)
+
Blocks of [blocksize, blocksize] are moved.
+
mode : string (default is DCR)
+
DCR (default) for depth-column-row order re-arrangement. Use CRD for column-row-depth order.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor of [N, C/(blocksize * blocksize), H * blocksize, W * blocksize].
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **DequantizeLinear-13** + + The linear dequantization operator. It consumes a quantized tensor, a scale, and a zero point to compute the full precision tensor. + The dequantization formula is `y = (x - x_zero_point) * x_scale`. `x_scale` and `x_zero_point` must have same shape, and can be either a scalar + for per-tensor / per layer quantization, or a 1-D tensor for per-axis quantization. + `x_zero_point` and `x` must have same type. `x` and `y` must have same shape. In the case of dequantizing int32, + there's no zero point (zero point is supposed to be 0). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Ignored for per-tensor quantization. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
+ +#### Inputs (2 - 3) + +
+
x : T
+
N-D quantized input tensor to be de-quantized.
+
x_scale : tensor(float)
+
Scale for input 'x'. It can be a scalar, which means a per-tensor/layer dequantization, or a 1-D tensor for per-axis dequantization.
+
x_zero_point (optional) : T
+
Zero point for input 'x'. Shape must match x_scale. It's optional. Zero point is 0 when it's not specified.
+
+ +#### Outputs + +
+
y : tensor(float)
+
N-D full precision output tensor. It has same shape as input 'x'.
+
+ +#### Type Constraints + +
+
T : tensor(int8), tensor(uint8), tensor(int32)
+
Constrain 'x_zero_point' and 'x' to 8-bit/32-bit integer tensor.
+
+ +### **Div-13** + + Performs element-wise binary division (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + + For integer inputs, the result is computed using truncating division (rounding toward zero). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Dropout-13** + + Dropout takes an input floating-point tensor, an optional input ratio (floating-point scalar) and an optional input training_mode (boolean scalar). It produces two tensor outputs, + output (floating-point tensor) and mask (optional `Tensor`). If `training_mode` is true then the output Y will be a random dropout; + Note that this Dropout scales the masked input data by the following equation, so to convert the trained model into inference mode, + the user can simply not pass `training_mode` input or set it to false. + ``` + output = scale * data * mask, + ``` + where + ``` + scale = 1. / (1. - ratio). + ``` + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
seed : int
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs (1 - 3) + +
+
data (differentiable) : T
+
The input data as Tensor.
+
ratio (optional, non-differentiable) : T1
+
The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it's non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
+
training_mode (optional, non-differentiable) : T2
+
If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
+
+ +#### Outputs (1 - 2) + +
+
output (differentiable) : T
+
The output.
+
mask (optional, non-differentiable) : T2
+
The output mask.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
T1 : tensor(float16), tensor(float), tensor(double)
+
Constrain input 'ratio' types to float tensors.
+
T2 : tensor(bool)
+
Constrain output 'mask' types to boolean tensors.
+
+ +### **Equal-13** + + Returns the tensor resulted from performing the `equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Erf-13** + + Computes the error function of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The error function of the input tensor computed element-wise. It has the same shape and type of the input.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Exp-13** + + Calculates the exponential of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The exponential of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Expand-13** + + Broadcast the input tensor following the given shape and the broadcast rule. + The broadcast rule is similar to numpy.array(input) * numpy.ones(shape): + Dimensions are right alignment; + Two corresponding dimensions must have the same value, or one of them is equal to 1. + Also, this operator is similar to numpy.broadcast_to(input, shape), + but the major difference is numpy.broadcast_to() does not allow shape to be smaller than input.size(). + It is possible that the output.shape is not equal to shape, when some dimensions in shape is equal to 1, + or the shape.ndim < input.shape.ndim. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
shape (non-differentiable) : tensor(int64)
+
A 1-D tensor indicates the shape you want to expand to, following the broadcast rule
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensors.
+
+ +### **Flatten-13** + + Flattens the input tensor into a 2D matrix. If input tensor has shape + (d_0, d_1, ... d_n) then the output will have shape + (d_0 X d_1 ... d_(axis-1), d_axis X d_(axis+1) ... X dn). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 ... d_n), where the shape of the input tensor is (d_0, d_1, ... d_n).
+
+ +#### Inputs + +
+
input (differentiable) : T
+
A tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output to all tensor types.
+
+ +### **Floor-13** + + Floor takes one input data (Tensor) and produces one output data + (Tensor) where the floor is, y = floor(x), is applied to + the tensor elementwise. If x is integral, +0, -0, NaN, or infinite, x itself is returned. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **Gather-13** + + Given `data` tensor of rank r >= 1, and `indices` tensor of rank q, gather + entries of the axis dimension of `data` (by default outer-most one as axis=0) indexed by `indices`, and concatenates + them in an output tensor of rank q + (r - 1). + + It is an indexing operation that indexes into the input `data` along a single (specified) axis. + Each entry in `indices` produces a `r-1` dimensional slice of the input tensor. + The entire operation produces, conceptually, a `q`-dimensional tensor of `r-1` dimensional slices, + which is arranged into a `q + (r-1)`-dimensional tensor, with the `q` dimensions taking the + place of the original `axis` that is being indexed into. + + The following few examples illustrate how `Gather` works for specific shapes of `data`, + `indices`, and given value of `axis`: + | data shape | indices shape | axis | output shape | output equation | + | --- | --- | --- | --- | --- | + | (P, Q) | ( ) (a scalar) | 0 | (Q) | output[q] = data[indices, q] | + | (P, Q, R) | ( ) (a scalar) | 1 | (P, R) | output[p, r] = data[p, indices, r] | + | (P, Q) | (R, S) | 0 | (R, S, Q) | output[r, s, q] = data[ [indices[r, s], q] | + | (P, Q) | (R, S) | 1 | (P, R, S) | output[p, r, s] = data[ p, indices[r, s]] | + + More generally, if `axis = 0`, let `k = indices[i_{0}, ..., i_{q-1}]` + then `output[i_{0}, ..., i_{q-1}, j_{0}, ..., j_{r-2}] = input[k , j_{0}, ..., j_{r-2}]`: + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + indices = [ + [0, 1], + [1, 2], + ] + output = [ + [ + [1.0, 1.2], + [2.3, 3.4], + ], + [ + [2.3, 3.4], + [4.5, 5.7], + ], + ] + ``` + + If `axis = 1`, let `k = indices[i_{0}, ..., i_{q-1}]` + then `output[j_{0}, i_{0}, ..., i_{q-1}, j_{1}, ..., j_{r-2}] = input[j_{0}, k, j_{1}, ..., j_{r-2}]`: + + ``` + data = [ + [1.0, 1.2, 1.9], + [2.3, 3.4, 3.9], + [4.5, 5.7, 5.9], + ] + indices = [ + [0, 2], + ] + axis = 1, + output = [ + [[1.0, 1.9]], + [[2.3, 3.9]], + [[4.5, 5.9]], + ] + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : Tind
+
Tensor of int32/int64 indices, of any rank q. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank q + (r - 1).
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **GatherElements-13** + + GatherElements takes two inputs `data` and `indices` of the same rank r >= 1 + and an optional attribute `axis` that identifies an axis of `data` + (by default, the outer-most axis, that is axis 0). It is an indexing operation + that produces its output by indexing into the input data tensor at index + positions determined by elements of the `indices` tensor. + Its output shape is the same as the shape of `indices` and consists of one value + (gathered from the `data`) for each element in `indices`. + + For instance, in the 3-D case (r = 3), the output produced is determined + by the following equations: + ``` + out[i][j][k] = input[index[i][j][k]][j][k] if axis = 0, + out[i][j][k] = input[i][index[i][j][k]][k] if axis = 1, + out[i][j][k] = input[i][j][index[i][j][k]] if axis = 2, + ``` + + This operator is also the inverse of ScatterElements. It is similar to Torch's gather operation. + + Example 1: + ``` + data = [ + [1, 2], + [3, 4], + ] + indices = [ + [0, 0], + [1, 0], + ] + axis = 1 + output = [ + [1, 1], + [4, 3], + ] + ``` + Example 2: + ``` + data = [ + [1, 2, 3], + [4, 5, 6], + [7, 8, 9], + ] + indices = [ + [1, 2, 0], + [2, 0, 0], + ] + axis = 0 + output = [ + [4, 8, 3], + [7, 2, 3], + ] + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : Tind
+
Tensor of int32/int64 indices, with the same rank r as the input. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of the same shape as indices.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **GatherND-13** + + Given `data` tensor of rank `r` >= 1, `indices` tensor of rank `q` >= 1, and `batch_dims` integer `b`, this operator gathers + slices of `data` into an output tensor of rank `q + r - indices_shape[-1] - 1 - b`. + + `indices` is an q-dimensional integer tensor, best thought of as a `(q-1)`-dimensional tensor of index-tuples into `data`, + where each element defines a slice of `data` + + `batch_dims` (denoted as `b`) is an integer indicating the number of batch dimensions, i.e the leading `b` number of dimensions of + `data` tensor and `indices` are representing the batches, and the gather starts from the `b+1` dimension. + + Some salient points about the inputs' rank and shape: + + 1) r >= 1 and q >= 1 are to be honored. There is no dependency condition to be met between ranks `r` and `q` + + 2) The first `b` dimensions of the shape of `indices` tensor and `data` tensor must be equal. + + 3) b < min(q, r) is to be honored. + + 4) The `indices_shape[-1]` should have a value between 1 (inclusive) and rank `r-b` (inclusive) + + 5) All values in `indices` are expected to be within bounds [-s, s-1] along axis of size `s` (i.e.) `-data_shape[i] <= indices[...,i] <= data_shape[i] - 1`. + It is an error if any of the index values are out of bounds. + + The output is computed as follows: + + The output tensor is obtained by mapping each index-tuple in the `indices` tensor to the corresponding slice of the input `data`. + + 1) If `indices_shape[-1] > r-b` => error condition + + 2) If `indices_shape[-1] == r-b`, since the rank of `indices` is `q`, `indices` can be thought of as `N` `(q-b-1)`-dimensional tensors + containing 1-D tensors of dimension `r-b`, where `N` is an integer equals to the product of 1 and all the elements in the batch dimensions + of the indices_shape. Let us think of each such `r-b` ranked tensor as `indices_slice`. Each *scalar value* corresponding to `data[0:b-1,indices_slice]` + is filled into the corresponding location of the `(q-b-1)`-dimensional tensor to form the `output` tensor (Example 1 below) + + 3) If `indices_shape[-1] < r-b`, since the rank of `indices` is `q`, `indices` can be thought of as `N` `(q-b-1)`-dimensional tensor + containing 1-D tensors of dimension `< r-b`. Let us think of each such tensors as `indices_slice`. Each *tensor slice* corresponding + to `data[0:b-1, indices_slice , :]` is filled into the corresponding location of the `(q-b-1)`-dimensional tensor + to form the `output` tensor (Examples 2, 3, 4 and 5 below) + + This operator is the inverse of `ScatterND`. + + **Example 1** + + ``` + batch_dims = 0 + data = [[0,1],[2,3]] # data_shape = [2, 2] + indices = [[0,0],[1,1]] # indices_shape = [2, 2] + output = [0,3] # output_shape = [2] + ``` + + **Example 2** + + ``` + batch_dims = 0 + data = [[0,1],[2,3]] # data_shape = [2, 2] + indices = [[1],[0]] # indices_shape = [2, 1] + output = [[2,3],[0,1]] # output_shape = [2, 2] + ``` + + **Example 3** + + ``` + batch_dims = 0 + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + indices = [[0,1],[1,0]] # indices_shape = [2, 2] + output = [[2,3],[4,5]] # output_shape = [2, 2] + ``` + + **Example 4** + + ``` + batch_dims = 0 + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + indices = [[[0,1]],[[1,0]]] # indices_shape = [2, 1, 2] + output = [[[2,3]],[[4,5]]] # output_shape = [2, 1, 2] + ``` + + **Example 5** + + ``` + batch_dims = 1 + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + indices = [[1],[0]] # indices_shape = [2, 1] + output = [[2,3],[4,5]] # output_shape = [2, 2] + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
batch_dims : int (default is 0)
+
The number of batch dimensions. The gather of indexing starts from dimension of data[batch_dims:]
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : tensor(int64)
+
Tensor of rank q >= 1. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank q + r - indices_shape[-1] - 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
+ +### **Gemm-13** + + General Matrix multiplication: + https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3 + + * A' = transpose(A) if transA else A + * B' = transpose(B) if transB else B + + Compute Y = alpha * A' * B' + beta * C, where input tensor A has shape (M, K) or (K, M), + input tensor B has shape (K, N) or (N, K), input tensor C is broadcastable to shape (M, N), + and output tensor Y has shape (M, N). A will be transposed before doing the + computation if attribute transA is non-zero, same for B and transB. + This operator supports **unidirectional broadcasting** (tensor C should be unidirectional broadcastable to tensor A * B); for more details please check [the doc](Broadcasting.md). + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Scalar multiplier for the product of input tensors A * B.
+
beta : float (default is 1.0)
+
Scalar multiplier for input tensor C.
+
transA : int (default is 0)
+
Whether A should be transposed
+
transB : int (default is 0)
+
Whether B should be transposed
+
+ +#### Inputs (2 - 3) + +
+
A (differentiable) : T
+
Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
+
B (differentiable) : T
+
Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
+
C (optional, differentiable) : T
+
Optional input tensor C. If not specified, the computation is done as if C is a scalar 0. The shape of C should be unidirectional broadcastable to (M, N).
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor of shape (M, N).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(bfloat16)
+
Constrain input and output types to float/int tensors.
+
+ +### **Greater-13** + + Returns the tensor resulted from performing the `greater` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Hardmax-13** + + The operator computes the hardmax values for the given input: + + Hardmax(element in input, axis) = 1 if the element is the first maximum value along the specified axis, 0 otherwise + + The "axis" attribute indicates the dimension along which Hardmax + will be performed. The output tensor has the same shape + and contains the Hardmax values of the corresponding input. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
+Describes the dimension Hardmax will be performed on. +Negative value means counting dimensions +from the back. Accepted range is [-r, r-1] where r = rank(input). +
+
+ +#### Inputs + +
+
input (differentiable) : T
+
The input tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output values with the same shape as the input tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **Identity-13** + + Identity operator + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor to copy input into.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **If-13** + + If conditional + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
else_branch : graph (required)
+
Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
+
then_branch : graph (required)
+
Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
+
+ +#### Inputs + +
+
cond : B
+
Condition for the if. The tensor must contain a single element.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : V
+
Values that are live-out to the enclosing scope. The return values in the `then_branch` and `else_branch` must be of the same data type. The `then_branch` and `else_branch` may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if-node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of `then_branch` is typed float tensor with shape [2] and the first output of `else_branch` is another float tensor with shape [3], If's first output should have (a) no shape set, or (b) a shape of rank 1 with neither `dim_value` nor `dim_param` set, or (c) a shape of rank 1 with a unique `dim_param`. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
All Tensor and Sequence types
+
B : tensor(bool)
+
Only bool
+
+ +### **IsNaN-13** + + Returns which elements of the input are NaN. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T1
+
input
+
+ +#### Outputs + +
+
Y (non-differentiable) : T2
+
output
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types to float tensors.
+
T2 : tensor(bool)
+
Constrain output types to boolean tensors.
+
+ +### **LRN-13** + + Local Response Normalization proposed in the [AlexNet paper](https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf). + It normalizes over local input regions. + The local region is defined across the channels. For an element `X[n, c, d1, ..., dk]` in a tensor + of shape `(N x C x D1 x D2, ..., Dk)`, its region is + `{X[n, i, d1, ..., dk] | max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2))}`. + + `square_sum[n, c, d1, ..., dk] = sum(X[n, i, d1, ..., dk] ^ 2)`, + where `max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2))`. + + `Y[n, c, d1, ..., dk] = X[n, c, d1, ..., dk] / (bias + alpha / size * square_sum[n, c, d1, ..., dk] ) ^ beta` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 0.0001)
+
Scaling parameter.
+
beta : float (default is 0.75)
+
The exponent.
+
bias : float (default is 1.0)
+
+
size : int (required)
+
The number of channels to sum over
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor, which has the shape and type as input tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **Less-13** + + Returns the tensor resulted from performing the `less` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Log-13** + + Calculates the natural log of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The natural log of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **LogSoftmax-13** + + The operator computes the log of softmax values for the given input: + + LogSoftmax(input, axis) = Log(Softmax(input, axis=axis)) + + The "axis" attribute indicates the dimension along which LogSoftmax + will be performed. The output tensor has the same shape + and contains the LogSoftmax values of the corresponding input. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
+Describes the dimension LogSoftmax will be performed on. +Negative value means counting dimensions +from the back. Accepted range is [-r, r-1] where r = rank(input). +
+
+ +#### Inputs + +
+
input (differentiable) : T
+
The input tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output values with the same shape as the input tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **Loop-13** + + Generic Looping construct. This loop has multiple termination conditions: + + 1) Trip count. Iteration count specified at runtime. Set by + specifying the input M. Optional. Set to empty string to omit. + Note that a static trip count (specified at graph construction time) can be + specified by passing in a constant node for input M. + 2) Loop termination condition. This is an input to the op that determines + whether to run the first iteration and also a loop-carried dependency for + the body graph. The body graph must yield a value for the condition variable, + whether this input is provided or not. + + This table summarizes the operating modes of this operator with equivalent + C-style code: + + Operator inputs defined as (max_trip_count, condition_var). + + input ("", ""): + for (int i=0; ; ++i) { + cond = ... // Note this value is ignored, but is required in the body + } + + input ("", cond) // Note this is analogous to a while loop + bool cond = ...; + for (int i=0; cond; ++i) { + cond = ...; + } + + input ("", 1) // Note this is analogous to a do-while loop + bool cond = true + for (int i=0; cond; ++i) { + cond = ...; + } + + input (trip_count, "") // Note this is analogous to a for loop + int trip_count = ... + for (int i=0; i < trip_count; ++i) { + cond = ...; // ignored + } + + input (trip_count, cond) + int trip_count = ...; + bool cond = ...; + for (int i=0; i < trip_count && cond; ++i) { + cond = ...; + } + + + *Sample usage - cond as well as trip count* + + graph predict-net { + %a = Constant[value = ]() + %b = Constant[value = ]() + %keepgoing = Constant[value = ]() + %max_trip_count = Constant[value = ]() + %keepgoing_out, %b_out, %user_defined_vals = Loop[body = ](%max_trip_count, %keepgoing, %b) + return + } + + graph body-net ( + %i[INT32, scalar] // iteration number + %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used + %b_in[INT32, scalar] // incoming value of loop-carried-dependency b + ) { + %my_local = Add(%a, %b_in) + %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b + %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition + %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated + return %keepgoing_out, %b_out, %user_defined_val + } + + *Sample equivalent C code* + + { + /* User-defined code (enclosing scope) */ + int a = 3, b = 6; + bool keepgoing = true; // Analogous to input cond + /* End user-defined code */ + + /* Implicitly-defined code */ + const int max_trip_count = 10; // Analogous to input M + int user_defined_vals[]; // Imagine this is resizable + /* End implicitly-defined code */ + /* initialize loop-carried variables and scan-output variables */ + bool keepgoing_out = keepgoing + int b_out = b + + for (int i=0; i < max_trip_count && keepgoing_out; ++i) { + /* Implicitly-defined code: bind actual parameter values + to formal parameter variables of loop-body */ + bool keepgoing_in = keepgoing_out; + bool b_in = b_out; + + /* User-defined code (loop body) */ + int my_local = a + b_in; // Reading value "a" from the enclosing scope is fine + b_out = a - b_in; + keepgoing_out = my_local > b_out; + user_defined_val = b_in + b_in; // b_in and b_out are different variables + /* End user-defined code */ + + /* Implicitly defined-code */ + user_defined_vals[i] = user_defined_val // accumulate scan-output values + } + // int t = my_local; // Can't do this. my_local is not accessible here. + + // The values below are bound to the output variables of the loop and therefore accessible + // b_out; user_defined_vals; keepgoing_out; + } + + There are several things of note in this code snippet: + + 1) Values from the enclosing scope (i.e. variable "a" here) are in scope and can + be referenced in the inputs of the loop. + 2) Any values computed in the loop body that needs to be used in a subsequent + iteration or after the loop are modeled using a pair of variables in the loop-body, + consisting of an input variable (eg., b_in) and an output variable (eg., b_out). + These are referred to as loop-carried dependences. The loop operation node + supplies the input value of the input variable for the first iteration, and + returns the output value of the output variable produced by the final + iteration. + 3) Scan_output variables are used to implicitly concatenate values computed across + all the iterations. In the above example, the value of user_defined_val computed + over all iterations are concatenated and returned as the value of user_defined_vals + after the loop. + 4) Values created in the body cannot be accessed in the enclosing scope, + except using the mechanism described above. + + Note that the semantics of this op support "diagonal" or "wavefront" execution. + (See Step 3 here for an example: + https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). + Frontends should emit multi-layer RNNs as a series of While operators (with + time being the inner looping dimension), with each successive layer consuming + the scan_outputs from the previous layer, possibly going through several + point-wise operators (e.g. dropout, residual connections, linear layer). + + The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies...). It has 1+N+K outputs: (condition, loop carried dependencies..., scan_outputs...). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
+
+ +#### Inputs (2 - ∞) + +
+
M (optional) : I
+
A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
+
cond (optional) : B
+
A boolean termination condition. Optional. Pass empty string to skip.
+
v_initial (variadic, heterogeneous) : V
+
The initial values of any loop-carried dependencies (values that change across loop iterations)
+
+ +#### Outputs (1 - ∞) + +
+
v_final_and_scan_outputs (variadic, heterogeneous) : V
+
Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
All Tensor and Sequence types
+
I : tensor(int64)
+
tensor of int64, which should be a scalar.
+
B : tensor(bool)
+
tensor of bool, which should be a scalar.
+
+ +### **MatMul-13** + + Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
A (differentiable) : T
+
N-dimensional matrix A
+
B (differentiable) : T
+
N-dimensional matrix B
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Matrix multiply results from A * B
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(bfloat16)
+
Constrain input and output types to float/int tensors.
+
+ +### **Max-13** + + Element-wise max of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic, differentiable) : T
+
List of tensors for max.
+
+ +#### Outputs + +
+
max (differentiable) : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **Mean-13** + + Element-wise mean of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic, differentiable) : T
+
List of tensors for mean.
+
+ +#### Outputs + +
+
mean (differentiable) : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **MeanVarianceNormalization-13** + + A MeanVarianceNormalization Function: Perform mean variance normalization + on the input tensor X using formula: `(X-EX)/sqrt(E(X-EX)^2)` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints (default is ['0', '2', '3'])
+
A list of integers, along which to reduce. The default is to calculate along axes [0,2,3] for calculating mean and variance along each channel. Two variables with the same C-coordinate are associated with the same mean and variance.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ +### **Min-13** + + Element-wise min of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic, differentiable) : T
+
List of tensors for min.
+
+ +#### Outputs + +
+
min (differentiable) : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **Mod-13** + + Performs an element-wise binary modulo operation. + The semantics and supported data types depend on the value of the `fmod` attribute which must be `0` (default), or `1`. + + If the `fmod` attribute is set to `0`, `T` is constrained to integer data types and the semantics follow that of the Python `%`-operator. + The sign of the result is that of the divisor. + + If `fmod` is set to `1`, the behavior of this operator follows that of the `fmod` function in C and `T` is constrained to floating point data types. + The result of this operator is the remainder of the division operation `x / y` where `x` and `y` are respective elements of `A` and `B`. The result is exactly the value `x - n * y`, where `n` is `x / y` with its fractional part truncated. + The returned value has the same sign as `x` (except if `x` is `-0`) and is less or equal to `|y|` in magnitude. + The following special cases apply when `fmod` is set to `1`: + - If `x` is `-0` and `y` is greater than zero, either `+0` or `-0` may be returned. + - If `x` is `±∞` and `y` is not `NaN`, `NaN` is returned. + - If `y` is `±0` and `x` is not `NaN`, `NaN` should be returned. + - If `y` is `±∞` and `x` is finite, `x` is returned. + - If either argument is `NaN`, `NaN` is returned. + + This operator supports **multidirectional (i.e., NumPy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
fmod : int (default is 0)
+
Whether the operator should behave like fmod (default=0 meaning it will do integer mods); Set this to 1 to force fmod treatment
+
+ +#### Inputs + +
+
A (differentiable) : T
+
Dividend tensor
+
B (non-differentiable) : T
+
Divisor tensor
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Remainder tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Mul-13** + + Performs element-wise binary multiplication (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Neg-13** + + Neg takes one input data (Tensor) and produces one output data + (Tensor) where each element flipped sign, y = -x, is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(int32), tensor(int8), tensor(int16), tensor(int64), tensor(float16), tensor(double), tensor(bfloat16)
+
Constrain input and output types to signed numeric tensors.
+
+ +### **NegativeLogLikelihoodLoss-13** + + A NegativeLogLikelihoodLoss operator computes (weighted) negative log likelihood loss. + Its "input" tensor has the shape of (N, C, d1, d2, ..., dk) where k >= 0. + The "input" tensor contains log-probabilities for input[n, :, d_1, d_2,..., d_k] being in a class of [0, C). + The operator's "target" input tensor has the shape of (N, d1, d2, ..., dk). It encodes class labels (one of C classes) + or it may contain a special value (indicated by an attribute ignore_index) for N x d1 x d2 x ... x dk samples. + The loss value for input[n, :, d_1, d_2,...d_k] being classified as class c = target[n][d_1][d_2]...[d_k] is computed as: + + ``` + loss[n][d_1][d_2]...[d_k] = -input[n][c][d_1][d_2]...[d_k]. + ``` + + When an optional "weight" is provided, the sample loss is calculated as: + + ``` + loss[n][d_1][d_2]...[d_k] = -input[n][c][d_1][d_2]...[d_k] * weight[c]. + ``` + + loss is zero for the case when target-value equals ignore_index. + + ``` + loss[n][d_1][d_2]...[d_k] = 0, when target[n][d_1][d_2]...[d_k] = ignore_index + ``` + + If "reduction" attribute is set to "none", the operator's output will be the above loss with shape (N, d1, d2, ..., dk). + If "reduction" attribute is set to "mean" (the default attribute value), the output loss is (weight) averaged: + + ``` + mean(loss), if "weight" is not provided, + ``` + + or if weight is provided, + + ``` + sum(loss) / sum(weight[target[n][d_1][d_2]...[d_k]]]), for all samples. + ``` + + If "reduction" attribute is set to "sum", the output is a scalar: `sum(loss)`. + + See also https://pytorch.org/docs/stable/nn.html#torch.nn.NLLLoss. + + Example 1: + + ``` + // negative log likelihood loss, "none" reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + + loss = np.zeros((N, d1)) + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] + + // print(loss) + // [[-3. -2.] + // [-0. -2.]] + ``` + + Example 2: + + ``` + // weighted negative log likelihood loss, sum reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + weight = [0.2, 0.3, 0.1] + loss = np.zeros((N, d1)) + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] * weight[c] + + loss = np.sum(loss) + // print(loss) + // -1.1 + ``` + + Example 3: + + ``` + // weighted negative log likelihood loss, mean reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + weight = [0.2, 0.3, 0.1] + loss = np.zeros((N, d1)) + weight_total = 0 + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] * weight[c] + weight_total = weight_total + weight[c] + + loss = np.sum(loss) / weight_total + // print(loss) + // -1.57 + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
ignore_index : int
+
Specifies a target value that is ignored and does not contribute to the input gradient. It's an optional value.
+
reduction : string (default is mean)
+
Type of reduction to apply to loss: none, sum, mean (default). 'none': the output is the loss for each sample. 'sum': the output will be summed. 'mean': the sum of the output will be divided by the sum of applied weights.
+
+ +#### Inputs (2 - 3) + +
+
input (differentiable) : T
+
Input tensor of shape (N, C) or (N, C, d1, d2, ..., dk).
+
target (non-differentiable) : Tind
+
Target tensor of shape (N) or (N, d1, d2, ..., dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
+
weight (optional, non-differentiable) : T
+
Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
+
+ +#### Outputs + +
+
loss (differentiable) : T
+
The negative log likelihood loss
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input, weight, and output types to floating-point tensors.
+
Tind : tensor(int32), tensor(int64)
+
Constrain target to integer types
+
+ +### **NonZero-13** + + Returns the indices of the elements that are non-zero + (in row-major order - by dimension). + NonZero behaves similar to numpy.nonzero: + https://docs.scipy.org/doc/numpy/reference/generated/numpy.nonzero.html, + but for scalar input, NonZero produces output shape (0, N) instead of (1, N), which is different from Numpy's behavior. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
input
+
+ +#### Outputs + +
+
Y (non-differentiable) : tensor(int64)
+
output
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to all tensor types.
+
+ +### **Pad-13** + + Given a tensor containing the data to be padded (`data`), a tensor containing the number of start and end pad values for axis (`pads`), (optionally) a `mode`, and (optionally) `constant_value`, + a padded tensor (`output`) is generated. + + The three supported `modes` are (similar to corresponding modes supported by `numpy.pad`): + + 1) `constant`(default) - pads with a given constant value as specified by `constant_value` (which defaults to 0, empty string, or False) + + 2) `reflect` - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis + + 3) `edge` - pads with the edge values of array + + + Example 1 (`constant` mode): + Insert 0 pads to the beginning of the second dimension. + + data = + [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'constant' + + constant_value = 0.0 + + output = + [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ] + + + Example 2 (`reflect` mode): + data = + [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'reflect' + + output = + [ + [1.0, 1.2, 1.0, 1.2], + [2.3, 3.4, 2.3, 3.4], + [4.5, 5.7, 4.5, 5.7], + ] + + + Example 3 (`edge` mode): + data = + [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'edge' + + output = + [ + [1.0, 1.0, 1.0, 1.2], + [2.3, 2.3, 2.3, 3.4], + [4.5, 4.5, 4.5, 5.7], + ] + + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is constant)
+
Supported modes: `constant`(default), `reflect`, `edge`
+
+ +#### Inputs (2 - 3) + +
+
data (differentiable) : T
+
Input tensor.
+
pads (non-differentiable) : tensor(int64)
+
Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. `pads` should be a 1D tensor of shape [2 * input_rank]. `pads` format should be: [x1_begin, x2_begin,...,x1_end, x2_end,...], where xi_begin is the number of pad values added at the beginning of axis `i` and xi_end, the number of pad values added at the end of axis `i`.
+
constant_value (optional, non-differentiable) : T
+
(Optional) A scalar value to be used if the mode chosen is `constant` (by default it is 0, empty string or False).
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Pow-13** + + Pow takes input data (Tensor) and exponent Tensor, and + produces one output data (Tensor) where the function `f(x) = x^exponent`, + is applied to the data tensor elementwise. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
First operand, base of the exponent.
+
Y (differentiable) : T1
+
Second operand, power of the exponent.
+
+ +#### Outputs + +
+
Z (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input X and output types to float/int tensors.
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input Y types to float/int tensors.
+
+ +### **QuantizeLinear-13** + + The linear quantization operator. It consumes a high precision tensor, a scale, and a zero point to compute the low precision / quantized tensor. + The scale factor and zero point must have same shape, and can be either a scalar for per-tensor / per layer quantization, or a 1-D tensor for per-axis quantization. + The quantization formula is y = saturate ((x / y_scale) + y_zero_point). + For saturation, it saturates to [0, 255] if it's uint8, or [-128, 127] if it's int8. + For (x / y_scale), it's rounding to the nearest even. Refer to https://en.wikipedia.org/wiki/Rounding for details. 'y_zero_point' and 'y' must have same type. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the quantization dimension of the input tensor. Ignored for per-tensor quantization. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D full precision Input tensor to be quantized.
+
y_scale : tensor(float)
+
Scale for doing quantization to get 'y'. It can be a scalar, which means per-tensor/layer quantization, or a 1-D Tensor for per-axis quantization.
+
y_zero_point (optional) : T2
+
Zero point for doing quantization to get 'y'. Shape must match y_scale. Default is uint8 with zero point of 0 if it's not specified.
+
+ +#### Outputs + +
+
y : T2
+
N-D quantized output tensor. It has same shape as input 'x'.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(int32)
+
Constrain 'x' to float or int32 tensor.
+
T2 : tensor(int8), tensor(uint8)
+
Constrain 'y_zero_point' and 'y' to 8-bit integer tensor.
+
+ +### **Reciprocal-13** + + Reciprocal takes one input data (Tensor) and produces one output data + (Tensor) where the reciprocal is, y = 1/x, is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **ReduceL1-13** + + Computes the L1 norm of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceL2-13** + + Computes the L2 norm of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceLogSum-13** + + Computes the log sum of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or undefined otherwise. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceLogSumExp-13** + + Computes the log sum exponent of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or undefined otherwise. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceMax-13** + + Computes the max of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or the minimum value of the data type otherwise. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(uint8), tensor(int8)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceMean-13** + + Computes the mean of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields undefined. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceMin-13** + + Computes the min of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields plus infinity (if supported by the datatype) or the maximum value of the data type otherwise. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(uint8), tensor(int8)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceProd-13** + + Computes the product of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 1. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceSum-13** + + Computes the sum of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceSumSquare-13** + + Computes the sum square of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **Relu-13** + + Relu takes one input data (Tensor) and produces one output data + (Tensor) where the rectified linear function, y = max(0, x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **Reshape-13** + + Reshape the input tensor similar to numpy.reshape. + First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. + At most one dimension of the new shape can be -1. In this case, the value is + inferred from the size of the tensor and the remaining dimensions. A dimension + could also be 0, in which case the actual dimension value is unchanged (i.e. taken + from the input tensor). Shape (second input) could be an empty shape, which means converting to a scalar. + The input tensor's shape and the output tensor's shape are required to have the same number of elements. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
shape (non-differentiable) : tensor(int64)
+
Specified shape for output.
+
+ +#### Outputs + +
+
reshaped (differentiable) : T
+
Reshaped data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Resize-13** + + Resize the input tensor. In general, it calculates every value in the output tensor as a weighted average of neighborhood (a.k.a. sampling locations) in the input tensor. + Each dimension value of the output tensor is: + output_dimension = floor(input_dimension * (roi_end - roi_start) * scale) if input \"sizes\" is not specified. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
coordinate_transformation_mode : string (default is half_pixel)
+
+This attribute describes how to transform the coordinate in the resized tensor to the coordinate in the original tensor.
+ +The coordinate of each dimension is transformed individually. Let's describe a case using axis x as an example. +Denote x_resized as the coordinate of axis x in the resized tensor, x_original as the coordinate of axis x in the original tensor, length_original as the length of the original tensor in axis x, length_resized as the length of the resized tensor in axis x, roi_x = (start_x, end_x) of the axis x in input "roi", scale = length_resized / length_original,
+ +if coordinate_transformation_mode is "half_pixel",
+x_original = (x_resized + 0.5) / scale - 0.5,
+ +if coordinate_transformation_mode is "pytorch_half_pixel",
+x_original = length_resized > 1 ? (x_resized + 0.5) / scale - 0.5 : 0,
+ +if coordinate_transformation_mode is "align_corners",
+x_original = x_resized * (length_original - 1) / (length_resized - 1),
+ +if coordinate_transformation_mode is "asymmetric",
+x_original = x_resized / scale,
+ +if coordinate_transformation_mode is "tf_crop_and_resize",
+x_original = length_resized > 1 ? start_x * (length_original - 1) + x_resized * (end_x - start_x) * (length_original - 1) / (length_resized - 1) : 0.5 * (start_x + end_x) * (length_original - 1).
+
cubic_coeff_a : float (default is -0.75)
+
The coefficient 'a' used in cubic interpolation. Two common choice are -0.5 (in some cases of TensorFlow) and -0.75 (in PyTorch). Check out Equation (4) in https://ieeexplore.ieee.org/document/1163711 for the details. This attribute is valid only if "mode" is "cubic".
+
exclude_outside : int (default is 0)
+
If set to 1, the weight of sampling locations outside the tensor will be set to 0 and the weight will be renormalized so that their sum is 1.0. The default value is 0.
+
extrapolation_value : float (default is 0.0)
+
When coordinate_transformation_mode is "tf_crop_and_resize" and x_original is outside the range [0, length_original - 1], this value is used as the corresponding output value. Default is 0.0f.
+
mode : string (default is nearest)
+
Three interpolation modes: nearest (default), linear and cubic. The "linear" mode includes linear interpolation for 1D tensor and N-linear interpolation for N-D tensor (for example, bilinear interpolation for 2D tensor). The "cubic" mode includes cubic interpolation for 1D tensor and N-cubic interpolation for N-D tensor (for example, bicubic interpolation for 2D tensor).
+
nearest_mode : string (default is round_prefer_floor)
+
Four modes: round_prefer_floor (default, as known as round half down), round_prefer_ceil (as known as round half up), floor, ceil. Only used by nearest interpolation. It indicates how to get "nearest" pixel in input tensor from x_original, so this attribute is valid only if "mode" is "nearest".
+
+ +#### Inputs (1 - 4) + +
+
X (differentiable) : T1
+
N-D tensor
+
roi (optional, non-differentiable) : T2
+
1-D tensor given as [start1, ..., startN, end1, ..., endN], where N is the rank of X. The RoIs' coordinates are normalized in the coordinate system of the input image. It only takes effect when coordinate_transformation_mode is "tf_crop_and_resize"
+
scales (optional, non-differentiable) : tensor(float)
+
The scale array along each dimension. It takes value greater than 0. If it's less than 1, it's sampling down, otherwise, it's upsampling. The number of elements of 'scales' should be the same as the rank of input 'X'. One of 'scales' and 'sizes' MUST be specified and it is an error if both are specified. If 'sizes' is needed, the user can use an empty string as the name of 'scales' in this operator's input list.
+
sizes (optional, non-differentiable) : tensor(int64)
+
The size of the output tensor. The number of elements of 'sizes' should be the same as the rank of input 'X'. Only one of 'scales' and 'sizes' can be specified.
+
+ +#### Outputs + +
+
Y (differentiable) : T1
+
N-D tensor after resizing
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input 'X' and output 'Y' to all tensor types.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Constrain roi type to float or double.
+
+ +### **ScatterElements-13** + + ScatterElements takes three inputs `data`, `updates`, and `indices` of the same + rank r >= 1 and an optional attribute axis that identifies an axis of `data` + (by default, the outer-most axis, that is axis 0). The output of the operation + is produced by creating a copy of the input `data`, and then updating its value + to values specified by `updates` at specific index positions specified by + `indices`. Its output shape is the same as the shape of `data`. + + For each entry in `updates`, the target index in `data` is obtained by combining + the corresponding entry in `indices` with the index of the entry itself: the + index-value for dimension = axis is obtained from the value of the corresponding + entry in `indices` and the index-value for dimension != axis is obtained from the + index of the entry itself. + + For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry + is performed as below: + ``` + output[indices[i][j]][j] = updates[i][j] if axis = 0, + output[i][indices[i][j]] = updates[i][j] if axis = 1, + ``` + + This operator is the inverse of GatherElements. It is similar to Torch's Scatter operation. + + Example 1: + ``` + data = [ + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + ] + indices = [ + [1, 0, 2], + [0, 2, 1], + ] + updates = [ + [1.0, 1.1, 1.2], + [2.0, 2.1, 2.2], + ] + output = [ + [2.0, 1.1, 0.0] + [1.0, 0.0, 2.2] + [0.0, 2.1, 1.2] + ] + ``` + Example 2: + ``` + data = [[1.0, 2.0, 3.0, 4.0, 5.0]] + indices = [[1, 3]] + updates = [[1.1, 2.1]] + axis = 1 + output = [[1.0, 1.1, 3.0, 2.1, 5.0]] + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : Tind
+
Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
updates (differentiable) : T
+
Tensor of rank r >=1 (same rank and shape as indices)
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank r >= 1 (same rank as input).
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input and output types can be of any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **ScatterND-13** + + ScatterND takes three inputs `data` tensor of rank r >= 1, `indices` tensor of rank q >= 1, + and `updates` tensor of rank q + r - indices.shape[-1] - 1. The output of the operation + is produced by creating a copy of the input `data`, and then updating its value to values + specified by `updates` at specific index positions specified by `indices`. Its output shape + is the same as the shape of `data`. Note that `indices` should not have duplicate entries. + That is, two or more `updates` for the same index-location is not supported. + + `indices` is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of `indices`. + `indices` is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into `data`. + Hence, k can be a value at most the rank of `data`. When k equals rank(data), each update entry specifies an + update to a single element of the tensor. When k is less than rank(data) each update entry specifies an + update to a slice of the tensor. Index values are allowed to be negative, as per the usual + convention for counting backwards from the end, but are expected in the valid range. + + `updates` is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the + first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. + The remaining dimensions of `updates` correspond to the dimensions of the + replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, + corresponding to the trailing (r-k) dimensions of `data`. Thus, the shape of `updates` + must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation + of shapes. + + The `output` is calculated via the following equation: + + output = np.copy(data) + update_indices = indices.shape[:-1] + for idx in np.ndindex(update_indices): + output[tuple(indices[idx])] = updates[idx] + + The order of iteration in the above loop is not specified. + In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. + This ensures that the output value does not depend on the iteration order. + + This operator is the inverse of GatherND. + + Example 1: + ``` + data = [1, 2, 3, 4, 5, 6, 7, 8] + indices = [[4], [3], [1], [7]] + updates = [9, 10, 11, 12] + output = [1, 11, 3, 10, 9, 6, 7, 12] + ``` + + Example 2: + ``` + data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] + indices = [[0], [2]] + updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]] + output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : tensor(int64)
+
Tensor of rank q >= 1.
+
updates (differentiable) : T
+
Tensor of rank q + r - indices_shape[-1] - 1.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank r >= 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
+ +### **Shape-13** + + Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
shape (non-differentiable) : T1
+
Shape of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor.
+
+ +### **Sigmoid-13** + + Sigmoid takes one input data (Tensor) and produces one output data + (Tensor) where the sigmoid function, y = 1 / (1 + exp(-x)), is applied to the + tensor elementwise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **Sign-13** + + Calculate the sign of the given input tensor element-wise. + If input > 0, output 1. if input < 0, output -1. if input == 0, output 0. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
input (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (non-differentiable) : T
+
The sign of the input tensor computed element-wise. It has the same shape and type of the input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ +### **Size-13** + + Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
size (non-differentiable) : T1
+
Total number of elements of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor, which should be a scalar though.
+
+ +### **Slice-13** + + Produces a slice of the input tensor along multiple axes. Similar to numpy: + https://numpy.org/doc/stable/user/basics.indexing.html?highlight=slice#slicing-and-striding + + Slice uses the `starts`, `ends`, `axes` and `steps` inputs to select a sub-tensor + of its input `data` tensor. + + An effective `starts[i]`, `ends[i]`, and `steps[i]` must be computed for each `i` + in `[0, ... r-1]` where `r = rank(input)` as follows: + + If `axes` are omitted, they are set to `[0, ..., r-1]`. + If `steps` are omitted, they are set to `[1, ..., 1]` of length `len(starts)` + + The effective values are initialized as `start[i] = 0`, `ends[i] = dims[i]` where + `dims` are the dimensions of `input` and `steps[i] = 1`. + + All negative elements of `axes` are made non-negative by adding `r` to them, where + `r =rank(input)`. + + All negative values in `starts[i]` and `ends[i]` have `dims[axes[i]]` added to them, + where `dims` are the dimensions of `input`. Then `start[axes[i]]` is the adjusted + `starts[i]` is clamped into the range `[0, dims[axes[i]]]` for positive stepping + and `[0, dims[axes[i]]-1]` for negative stepping. + + The clamping for the adjusted `ends[i]` depends on the sign of `steps[i]` and must + accommodate copying 0 through `dims[axes[i]]` elements, so for positive stepping + `ends[axes[i]]` is clamped to `[0, dims[axes[i]]]`, while for negative stepping it + is clamped to `[-1, dims[axes[i]]-1]`. + + Finally, `steps[axes[i]] = steps[i]`. + + For slicing to the end of a dimension with unknown size, it is recommended to pass + in `INT_MAX` when slicing forward and 'INT_MIN' when slicing backward. + + Example 1: + + ``` + data = [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + axes = [0, 1] + starts = [1, 0] + ends = [2, 3] + steps = [1, 2] + result = [ + [5, 7], + ] + ``` + + Example 2: + + ``` + data = [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + starts = [0, 1] + ends = [-1, 1000] + result = [ + [2, 3, 4], + ] + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs (3 - 5) + +
+
data (differentiable) : T
+
Tensor of data to extract slices from.
+
starts (non-differentiable) : Tind
+
1-D tensor of starting indices of corresponding axis in `axes`
+
ends (non-differentiable) : Tind
+
1-D tensor of ending indices (exclusive) of corresponding axis in `axes`
+
axes (optional, non-differentiable) : Tind
+
1-D tensor of axes that `starts` and `ends` apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated.
+
steps (optional, non-differentiable) : Tind
+
1-D tensor of slice step of corresponding axis in `axes`. Negative value means slicing backward. 'steps' cannot be 0. Defaults to 1s.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Sliced data tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **Softmax-13** + + The operator computes the normalized exponential values for the given input: + + Softmax(input, axis) = Exp(input) / ReduceSum(Exp(input), axis=axis, keepdims=1) + + The "axis" attribute indicates the dimension along which Softmax + will be performed. The output tensor has the same shape + and contains the Softmax values of the corresponding input. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
+Describes the dimension Softmax will be performed on. +Negative value means counting dimensions +from the back. Accepted range is [-r, r-1] where r = rank(input). +
+
+ +#### Inputs + +
+
input (differentiable) : T
+
The input tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output values with the same shape as the input tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **SoftmaxCrossEntropyLoss-13** + + Loss function that measures the softmax cross entropy + between 'scores' and 'labels'. + This operator first computes a loss tensor whose shape is identical to the labels input. + If the input is 2-D with shape (N, C), the loss tensor may be a N-element vector L = (l_1, l_2, ..., l_N). + If the input is N-D tensor with shape (N, C, D1, D2, ..., Dk), + the loss tensor L may have (N, D1, D2, ..., Dk) as its shape and L[i,][j_1][j_2]...[j_k] denotes a scalar element in L. + After L is available, this operator can optionally do a reduction operator. + + * shape(scores): (N, C) where C is the number of classes, or (N, C, D1, D2,..., Dk), + with K >= 1 in case of K-dimensional loss. + * shape(labels): (N) where each value is 0 <= labels[i] <= C-1, or (N, D1, D2,..., Dk), + with K >= 1 in case of K-dimensional loss. + + The loss for one sample, l_i, can calculated as follows: + ``` + l[i][d1][d2]...[dk] = -y[i][c][d1][d2]..[dk], where i is the index of classes. + ``` + or + ``` + l[i][d1][d2]...[dk] = -y[i][c][d1][d2]..[dk] * weights[c], if 'weights' is provided. + ``` + + loss is zero for the case when label-value equals ignore_index. + ``` + l[i][d1][d2]...[dk] = 0, when labels[n][d1][d2]...[dk] = ignore_index + ``` + + where: + ``` + p = Softmax(scores) + y = Log(p) + c = labels[i][d1][d2]...[dk] + ``` + + Finally, L is optionally reduced: + + * If reduction = 'none', the output is L with shape (N, D1, D2, ..., Dk). + * If reduction = 'sum', the output is scalar: Sum(L). + * If reduction = 'mean', the output is scalar: ReduceMean(L), or if weight is provided: `ReduceSum(L) / ReduceSum(W)`, + where tensor W is of shape `(N, D1, D2, ..., Dk)` and `W[n][d1][d2]...[dk] = weights[labels[i][d1][d2]...[dk]]`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
ignore_index : int
+
Specifies a target value that is ignored and does not contribute to the input gradient. It's an optional value.
+
reduction : string (default is mean)
+
Type of reduction to apply to loss: none, sum, mean(default). 'none': no reduction will be applied, 'sum': the output will be summed. 'mean': the sum of the output will be divided by the number of elements in the output.
+
+ +#### Inputs (2 - 3) + +
+
scores (differentiable) : T
+
The predicted outputs with shape [batch_size, class_size], or [batch_size, class_size, D1, D2 , ..., Dk], where K is the number of dimensions.
+
labels (non-differentiable) : Tind
+
The ground truth output tensor, with shape [batch_size], or [batch_size, D1, D2, ..., Dk], where K is the number of dimensions. Labels element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the label values should either be in the range [0, C) or have the value ignore_index.
+
weights (optional, non-differentiable) : T
+
A manual rescaling weight given to each class. If given, it has to be a 1D Tensor assigning weight to each of the classes. Otherwise, it is treated as if having all ones.
+
+ +#### Outputs (1 - 2) + +
+
output (differentiable) : T
+
Weighted loss float Tensor. If reduction is 'none', this has the shape of [batch_size], or [batch_size, D1, D2, ..., Dk] in case of K-dimensional loss. Otherwise, it is a scalar.
+
log_prob (optional, differentiable) : T
+
Log probability tensor. If the output of softmax is prob, its value is log(prob).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
Tind : tensor(int32), tensor(int64)
+
Constrain target to integer types
+
+ +### **SpaceToDepth-13** + + SpaceToDepth rearranges blocks of spatial data into depth. More specifically, + this op outputs a copy of the input tensor where values from the height and width dimensions + are moved to the depth dimension. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
blocksize : int (required)
+
Blocks of [blocksize, blocksize] are moved.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor of [N, C * blocksize * blocksize, H/blocksize, W/blocksize].
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Split-13** + + Split a tensor into a list of tensors, along the specified + 'axis'. Lengths of the parts can be specified using input 'split'. + Otherwise, the tensor is split to equal sized parts. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1] where r = rank(input).
+
+ +#### Inputs (1 - 2) + +
+
input (differentiable) : T
+
The tensor to split
+
split (optional, non-differentiable) : tensor(int64)
+
Optional length of each output. Values should be >= 0.Sum of the values must be equal to the dim value at 'axis' specified.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, differentiable) : T
+
One or more outputs forming list of tensors after splitting
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Sqrt-13** + + Square root takes one input data (Tensor) and produces one output data + (Tensor) where the square root is, y = x^0.5, is applied to + the tensor elementwise. If x is negative, then it will return NaN. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **Squeeze-13** + + Remove single-dimensional entries from the shape of a tensor. + Takes an input `axes` with a list of axes to squeeze. + If `axes` is not provided, all the single dimensions will be removed from + the shape. If an axis is selected with shape entry not equal to one, an error is raised. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
Tensors with at least max(dims) dimensions.
+
axes (optional, non-differentiable) : tensor(int64)
+
List of integers indicating the dimensions to squeeze. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
squeezed (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Sub-13** + + Performs element-wise binary subtraction (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to high-precision numeric tensors.
+
+ +### **Sum-13** + + Element-wise sum of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
data_0 (variadic, differentiable) : T
+
List of tensors for sum.
+
+ +#### Outputs + +
+
sum (differentiable) : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **Tanh-13** + + Calculates the hyperbolic tangent of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic tangent values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Tile-13** + + Constructs a tensor by tiling a given tensor. + This is the same as function `tile` in Numpy, but no broadcast. + For example A = [[1, 2], [3, 4]], B = [1, 2], tile(A, B) = [[1, 2, 1, 2], [3, 4, 3, 4]] + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor of any shape.
+
repeats (non-differentiable) : T1
+
1D int64 tensor of the same length as input's dimension number, includes numbers of repeated copies along input's dimensions.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor of the same dimensions and type as tensor input. output_dim[i] = input_dim[i] * repeats[i]
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
T1 : tensor(int64)
+
Constrain repeat's type to int64 tensors.
+
+ +### **Transpose-13** + + Returns a transpose of the input tensor. (Similar to `numpy.transpose`). + The optional attribute `perm` must be a permutation of the dimensions of + the input tensor. Axis `i` of the output tensor corresponds to the axis + `perm[i]` of the input tensor. + For example, when perm=(1, 0, 2), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 1, 3). + When perm=(1, 2, 0), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 3, 1). + If the attribute `perm` is omitted, its default value is `(n-1, ..., 0)`, + where `n` is the rank of the input tensor. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Attributes + +
+
perm : list of ints
+
A list of integers. By default, reverse the dimensions, otherwise permute the axes according to the values given.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
transposed (differentiable) : T
+
Transposed output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Unsqueeze-13** + + Insert single-dimensional entries to the shape of an input tensor (`data`). + Takes one required input `axes` - which contains a list of dimension indices and this operator will insert a dimension of value `1` into the corresponding index of the output tensor (`expanded`). + + For example, given an input tensor (`data`) of shape [3, 4, 5], then + Unsqueeze(data, axes=[0, 4]) outputs a tensor (`expanded`) containing same data as `data` but with shape [1, 3, 4, 5, 1]. + + The input `axes` should not contain any duplicate entries. It is an error if it contains duplicates. + The rank of the output tensor (`output_rank`) is the rank of the input tensor (`data`) plus the number of values in `axes`. + Each value in `axes` should be within the (inclusive) range [-output_rank , output_rank - 1]. + The order of values in `axes` does not matter and can come in any order. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +#### Inputs + +
+
data (differentiable) : T
+
Original tensor
+
axes (non-differentiable) : tensor(int64)
+
List of integers indicating the dimensions to be inserted. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(expanded).
+
+ +#### Outputs + +
+
expanded (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +## Version 14 of the default ONNX operator set +### **Add-14** + + Performs element-wise binary addition (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + + (Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ +### **BatchNormalization-14** + + Carries out batch normalization as described in the paper + https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, + There are five required inputs 'X', 'scale', 'B', 'input_mean' and + 'input_var'. + Note that 'input_mean' and 'input_var' are expected to be the estimated + statistics in inference mode (training_mode=False, default), + and the running statistics in training mode (training_mode=True). + There are multiple cases for the number of outputs, which we list below: + + Output case #1: Y, running_mean, running_var (training_mode=True) + Output case #2: Y (training_mode=False) + + When training_mode=False, extra outputs are invalid. + The outputs are updated as follows when training_mode=True: + ``` + running_mean = input_mean * momentum + current_mean * (1 - momentum) + running_var = input_var * momentum + current_var * (1 - momentum) + + Y = (X - current_mean) / sqrt(current_var + epsilon) * scale + B + + where: + + current_mean = ReduceMean(X, axis=all_except_channel_index) + current_var = ReduceVar(X, axis=all_except_channel_index) + + Notice that ReduceVar refers to the population variance, and it equals to + sum(sqrd(x_i - x_avg)) / N + where N is the population size (this formula does not use sample size N - 1). + + ``` + + When training_mode=False: + ``` + Y = (X - input_mean) / sqrt(input_var + epsilon) * scale + B + ``` + + For previous (depreciated) non-spatial cases, implementors are suggested + to flatten the input shape to (N x C * D1 * D2 * ... * Dn) before a BatchNormalization Op. + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
momentum : float (default is 0.9)
+
Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum).
+
training_mode : int (default is 0)
+
If set to true, it indicates BatchNormalization is being used for training, and outputs 1, 2, 3, and 4 would be populated.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size, C is the number of channels. Statistics are computed for every channel of C over N and D1 to Dn dimensions. For image data, input dimensions become (N x C x H x W). The op also accepts single dimension input of size N in which case C is assumed to be 1
+
scale (differentiable) : T
+
Scale tensor of shape (C).
+
B (differentiable) : T
+
Bias tensor of shape (C).
+
input_mean (differentiable) : U
+
running (training) or estimated (testing) mean tensor of shape (C).
+
input_var (differentiable) : U
+
running (training) or estimated (testing) variance tensor of shape (C).
+
+ +#### Outputs (1 - 3) + +
+
Y (differentiable) : T
+
The output tensor of the same shape as X
+
running_mean (optional, non-differentiable) : U
+
The running mean after the BatchNormalization operator.
+
running_var (optional, non-differentiable) : U
+
The running variance after the BatchNormalization operator. This op uses the population size (N) for calculating variance, and not the sample size N-1.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
U : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain mean and variance types to float tensors. It allows all float type for U.
+
+ +### **CumSum-14** + + Performs cumulative sum of the input elements along the given axis. + By default, it will do the sum inclusively meaning the first element is copied as is. + Through an `exclusive` attribute, this behavior can change to exclude the first element. + It can also perform summation in the opposite direction of the axis. For that, set `reverse` attribute to 1. + + Example: + ``` + input_x = [1, 2, 3] + axis=0 + output = [1, 3, 6] + exclusive=1 + output = [0, 1, 3] + exclusive=0 + reverse=1 + output = [6, 5, 3] + exclusive=1 + reverse=1 + output = [5, 3, 0] + ``` + + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Attributes + +
+
exclusive : int (default is 0)
+
If set to 1 will return exclusive sum in which the top element is not included. In other terms, if set to 1, the j-th output element would be the sum of the first (j-1) elements. Otherwise, it would be the sum of the first j elements.
+
reverse : int (default is 0)
+
If set to 1 will perform the sums in reverse direction.
+
+ +#### Inputs + +
+
x (differentiable) : T
+
An input tensor that is to be processed.
+
axis (non-differentiable) : T2
+
A 0-D tensor. Must be in the range [-rank(x), rank(x)-1]. Negative value means counting dimensions from the back.
+
+ +#### Outputs + +
+
y (differentiable) : T
+
Output tensor of the same type as 'x' with cumulative sums of the x's elements
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
T2 : tensor(int32), tensor(int64)
+
axis tensor can be int32 or int64 only
+
+ +### **Div-14** + + Performs element-wise binary division (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + + For integer inputs, the result is computed using truncating division (rounding toward zero). + (Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ +### **GRU-14** + + Computes an one-layer GRU. This operator is usually supported via some custom + implementation such as CuDNN. + + Notations: + + * `X` - input tensor + * `z` - update gate + * `r` - reset gate + * `h` - hidden gate + * `t` - time step (t-1 means previous time step) + * `W[zrh]` - W parameter weight matrix for update, reset, and hidden gates + * `R[zrh]` - R recurrence weight matrix for update, reset, and hidden gates + * `Wb[zrh]` - W bias vectors for update, reset, and hidden gates + * `Rb[zrh]` - R bias vectors for update, reset, and hidden gates + * `WB[zrh]` - W parameter weight matrix for backward update, reset, and hidden gates + * `RB[zrh]` - R recurrence weight matrix for backward update, reset, and hidden gates + * `WBb[zrh]` - W bias vectors for backward update, reset, and hidden gates + * `RBb[zrh]` - R bias vectors for backward update, reset, and hidden gates + * `H` - Hidden state + * `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + * Relu(x) - max(0, x) + * Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + * Sigmoid(x) - 1/(1 + e^{-x}) + + NOTE: + Below are optional + + * Affine(x) - alpha * x + beta + * LeakyRelu(x) - x if x >= 0 else alpha * x + * ThresholdedRelu(x) - x if x >= alpha else 0 + * ScaledTanh(x) - alpha * Tanh(beta * x) + * HardSigmoid(x) - min(max(alpha * x + beta, 0), 1) + * Elu(x) - x if x >= 0 else alpha * (e^x - 1) + * Softsign(x) - x/(1 + |x|) + * Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh): + + * zt = f(Xt*(Wz^T) + Ht-1*(Rz^T) + Wbz + Rbz) + * rt = f(Xt*(Wr^T) + Ht-1*(Rr^T) + Wbr + Rbr) + * ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*(Rh^T) + Rbh + Wbh) # default, when linear_before_reset = 0 + * ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*(Rh^T) + Rbh)) + Wbh) # when linear_before_reset != 0 + * Ht = (1 - zt) (.) ht + zt (.) Ht-1 + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings
+
A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
layout : int (default is 0)
+
The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size].
+
linear_before_reset : int (default is 0)
+
When computing the output of the hidden gate, apply the linear transformation before multiplying by the output of the reset gate.
+
+ +#### Inputs (3 - 6) + +
+
X (differentiable) : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W (differentiable) : T
+
The weight tensor for the gates. Concatenation of `W[zrh]` and `WB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, input_size]`.
+
R (differentiable) : T
+
The recurrence weight tensor. Concatenation of `R[zrh]` and `RB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, hidden_size]`.
+
B (optional, differentiable) : T
+
The bias tensor for the gates. Concatenation of `[Wb[zrh], Rb[zrh]]` and `[WBb[zrh], RBb[zrh]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 6*hidden_size]`. Optional: If not specified - assumed to be 0
+
sequence_lens (optional, non-differentiable) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional, non-differentiable) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs (0 - 2) + +
+
Y (optional, differentiable) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional, differentiable) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **HardSwish-14** + + HardSwish takes one input data (Tensor) and produces one output data (Tensor) where + the HardSwish function, y = x * max(0, min(1, alpha * x + beta)) = x * HardSigmoid(x), + where alpha = 1/6 and beta = 0.5, is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Identity-14** + + Identity operator + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : V
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : V
+
Tensor to copy input into.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input and output types to all tensor and sequence types.
+
+ +### **LSTM-14** + + Computes an one-layer LSTM. This operator is usually supported via some + custom implementation such as CuDNN. + + Notations: + + * `X` - input tensor + * `i` - input gate + * `o` - output gate + * `f` - forget gate + * `c` - cell gate + * `t` - time step (t-1 means previous time step) + * `W[iofc]` - W parameter weight matrix for input, output, forget, and cell gates + * `R[iofc]` - R recurrence weight matrix for input, output, forget, and cell gates + * `Wb[iofc]` - W bias vectors for input, output, forget, and cell gates + * `Rb[iofc]` - R bias vectors for input, output, forget, and cell gates + * `P[iof]` - P peephole weight vector for input, output, and forget gates + * `WB[iofc]` - W parameter weight matrix for backward input, output, forget, and cell gates + * `RB[iofc]` - R recurrence weight matrix for backward input, output, forget, and cell gates + * `WBb[iofc]` - W bias vectors for backward input, output, forget, and cell gates + * `RBb[iofc]` - R bias vectors for backward input, output, forget, and cell gates + * `PB[iof]` - P peephole weight vector for backward input, output, and forget gates + * `H` - Hidden state + * `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + * Relu(x) - max(0, x) + * Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + * Sigmoid(x) - 1/(1 + e^{-x}) + + NOTE: Below are optional + + * Affine(x) - alpha*x + beta + * LeakyRelu(x) - x if x >= 0 else alpha * x + * ThresholdedRelu(x) - x if x >= alpha else 0 + * ScaledTanh(x) - alpha*Tanh(beta*x) + * HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + * Elu(x) - x if x >= 0 else alpha*(e^x - 1) + * Softsign(x) - x/(1 + |x|) + * Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh, h=Tanh): + + * it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi) + * ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf) + * ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc) + * Ct = ft (.) Ct-1 + it (.) ct + * ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Po (.) Ct + Wbo + Rbo) + * Ht = ot (.) h(Ct) + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings
+
A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
input_forget : int (default is 0)
+
Couple the input and forget gates if 1.
+
layout : int (default is 0)
+
The shape format of inputs X, initial_h, initial_c and outputs Y, Y_h, Y_c. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = initial_c.shape = Y_c.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = initial_c.shape = Y_c.shape = [batch_size, num_directions, hidden_size].
+
+ +#### Inputs (3 - 8) + +
+
X (differentiable) : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W (differentiable) : T
+
The weight tensor for the gates. Concatenation of `W[iofc]` and `WB[iofc]` (if bidirectional) along dimension 0. The tensor has shape `[num_directions, 4*hidden_size, input_size]`.
+
R (differentiable) : T
+
The recurrence weight tensor. Concatenation of `R[iofc]` and `RB[iofc]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 4*hidden_size, hidden_size]`.
+
B (optional, differentiable) : T
+
The bias tensor for input gate. Concatenation of `[Wb[iofc], Rb[iofc]]`, and `[WBb[iofc], RBb[iofc]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 8*hidden_size]`. Optional: If not specified - assumed to be 0.
+
sequence_lens (optional, non-differentiable) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional, non-differentiable) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
initial_c (optional, non-differentiable) : T
+
Optional initial value of the cell. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
P (optional, differentiable) : T
+
The weight tensor for peepholes. Concatenation of `P[iof]` and `PB[iof]` (if bidirectional) along dimension 0. It has shape `[num_directions, 3*hidde_size]`. Optional: If not specified - assumed to be 0.
+
+ +#### Outputs (0 - 3) + +
+
Y (optional, differentiable) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional, differentiable) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
Y_c (optional, differentiable) : T
+
The last output value of the cell. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **Mul-14** + + Performs element-wise binary multiplication (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + + (Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ +### **RNN-14** + + Computes an one-layer simple RNN. This operator is usually supported + via some custom implementation such as CuDNN. + + Notations: + + * `X` - input tensor + * `i` - input gate + * `t` - time step (t-1 means previous time step) + * `Wi` - W parameter weight matrix for input gate + * `Ri` - R recurrence weight matrix for input gate + * `Wbi` - W parameter bias vector for input gate + * `Rbi` - R parameter bias vector for input gate + * `WBi` - W parameter weight matrix for backward input gate + * `RBi` - R recurrence weight matrix for backward input gate + * `WBbi` - WR bias vectors for backward input gate + * `RBbi` - RR bias vectors for backward input gate + * `H` - Hidden state + * `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + * Relu(x) - max(0, x) + * Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + * Sigmoid(x) - 1/(1 + e^{-x}) + + NOTE: Below are optional + + * Affine(x) - alpha*x + beta + * LeakyRelu(x) - x if x >= 0 else alpha * x + * ThresholdedRelu(x) - x if x >= alpha else 0 + * ScaledTanh(x) - alpha*Tanh(beta*x) + * HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + * Elu(x) - x if x >= 0 else alpha*(e^x - 1) + * Softsign(x) - x/(1 + |x|) + * Softplus(x) - log(1 + e^x) + + Equations (Default: f=Tanh): + + * Ht = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Wbi + Rbi) + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings (default is ['Tanh', 'Tanh'])
+
One (or two if bidirectional) activation function for input gate. The activation function must be one of the activation functions specified above. Optional: Default `Tanh` if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
layout : int (default is 0)
+
The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size].
+
+ +#### Inputs (3 - 6) + +
+
X (differentiable) : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W (differentiable) : T
+
The weight tensor for input gate. Concatenation of `Wi` and `WBi` (if bidirectional). The tensor has shape `[num_directions, hidden_size, input_size]`.
+
R (differentiable) : T
+
The recurrence weight tensor. Concatenation of `Ri` and `RBi` (if bidirectional). The tensor has shape `[num_directions, hidden_size, hidden_size]`.
+
B (optional, differentiable) : T
+
The bias tensor for input gate. Concatenation of `[Wbi, Rbi]` and `[WBbi, RBbi]` (if bidirectional). The tensor has shape `[num_directions, 2*hidden_size]`. Optional: If not specified - assumed to be 0.
+
sequence_lens (optional, non-differentiable) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional, non-differentiable) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs (0 - 2) + +
+
Y (optional, differentiable) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional, differentiable) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **Relu-14** + + Relu takes one input data (Tensor) and produces one output data + (Tensor) where the rectified linear function, y = max(0, x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(int32), tensor(int8), tensor(int16), tensor(int64), tensor(float16), tensor(double), tensor(bfloat16)
+
Constrain input and output types to signed numeric tensors.
+
+ +### **Reshape-14** + + Reshape the input tensor similar to numpy.reshape. + First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. + At most one dimension of the new shape can be -1. In this case, the value is + inferred from the size of the tensor and the remaining dimensions. A dimension + could also be 0, in which case the actual dimension value is unchanged (i.e. taken + from the input tensor). If 'allowzero' is set, and the new shape includes 0, the + dimension will be set explicitly to zero (i.e. not taken from input tensor). + Shape (second input) could be an empty shape, which means converting to a scalar. + The input tensor's shape and the output tensor's shape are required to have the same number of elements. + + If the attribute 'allowzero' is set, it is invalid for the specified shape to + contain both a zero value and -1, as the value of the dimension corresponding + to -1 cannot be determined uniquely. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Attributes + +
+
allowzero : int (default is 0)
+
(Optional) By default, when any value in the 'shape' input is equal to zero the corresponding dimension value is copied from the input tensor dynamically. allowzero=1 indicates that if any value in the 'shape' input is set to zero, the zero value is honored, similar to NumPy.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
shape (non-differentiable) : tensor(int64)
+
Specified shape for output.
+
+ +#### Outputs + +
+
reshaped (differentiable) : T
+
Reshaped data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +### **Sub-14** + + Performs element-wise binary subtraction (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + + (Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ +### **Trilu-14** + + Given a 2-D matrix or batches of 2-D matrices, returns the upper or lower triangular part of the tensor(s). + The attribute "upper" determines whether the upper or lower part is retained. If set to true, + the upper triangular matrix is retained. Lower triangular matrix is retained otherwise. + Default value for the "upper" attribute is true. + Trilu takes one input tensor of shape [*, N, M], where * is zero or more batch dimensions. The upper triangular part consists + of the elements on and above the given diagonal (k). The lower triangular part consists of elements on and below the diagonal. + All other elements in the matrix are set to zero. + If k = 0, the triangular part on and above/below the main diagonal is retained. + If upper is set to true, a positive k retains the upper triangular matrix excluding the main diagonal and (k-1) diagonals above it. + A negative k value retains the main diagonal and |k| diagonals below it. + If upper is set to false, a positive k retains the lower triangular matrix including the main diagonal and k diagonals above it. + A negative k value excludes the main diagonal and (|k|-1) diagonals below it. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Attributes + +
+
upper : int (default is 1)
+
Boolean. Indicates whether upper or lower part of matrix is retained. Default is true.
+
+ +#### Inputs (1 - 2) + +
+
input (differentiable) : T
+
Input tensor of rank 2 or higher.
+
k (optional, non-differentiable) : tensor(int64)
+
A 0-D tensor containing a single value corresponding to the number diagonals above or below the main diagonal to exclude or include. Default value is 0 if it's not specified.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor of the same type and shape as the input tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +## Version 15 of the default ONNX operator set +### **BatchNormalization-15** + + Carries out batch normalization as described in the paper + https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, + There are five required inputs 'X', 'scale', 'B', 'input_mean' and + 'input_var'. + Note that 'input_mean' and 'input_var' are expected to be the estimated + statistics in inference mode (training_mode=False, default), + and the running statistics in training mode (training_mode=True). + There are multiple cases for the number of outputs, which we list below: + + * Output case #1: Y, running_mean, running_var (training_mode=True) + * Output case #2: Y (training_mode=False) + + When training_mode=False, extra outputs are invalid. + The outputs are updated as follows when training_mode=True: + ``` + running_mean = input_mean * momentum + current_mean * (1 - momentum) + running_var = input_var * momentum + current_var * (1 - momentum) + + Y = (X - current_mean) / sqrt(current_var + epsilon) * scale + B + ``` + where: + ``` + current_mean = ReduceMean(X, axis=all_except_channel_index) + current_var = ReduceVar(X, axis=all_except_channel_index) + ``` + Notice that `ReduceVar` refers to the population variance, and it equals to + `sum(sqrd(x_i - x_avg)) / N` + where `N` is the population size (this formula does not use sample size `N - 1`). + + The computation of ReduceMean and ReduceVar uses float to avoid overflow for float16 inputs. + + When training_mode=False: + ``` + Y = (X - input_mean) / sqrt(input_var + epsilon) * scale + B + ``` + + For previous (depreciated) non-spatial cases, implementors are suggested + to flatten the input shape to (N x C * D1 * D2 * ... * Dn) before a BatchNormalization Op. + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
momentum : float (default is 0.9)
+
Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum).
+
training_mode : int (default is 0)
+
If set to true, it indicates BatchNormalization is being used for training, and outputs 1 and 2 are to be computed.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size, C is the number of channels. Statistics are computed for every channel of C over N and D1 to Dn dimensions. For image data, input dimensions become (N x C x H x W). The op also accepts single dimension input of size N in which case C is assumed to be 1
+
scale (differentiable) : T1
+
Scale tensor of shape (C).
+
B (differentiable) : T1
+
Bias tensor of shape (C).
+
input_mean (differentiable) : T2
+
running (training) or estimated (testing) mean tensor of shape (C).
+
input_var (differentiable) : T2
+
running (training) or estimated (testing) variance tensor of shape (C).
+
+ +#### Outputs (1 - 3) + +
+
Y (differentiable) : T
+
The output tensor of the same shape as X
+
running_mean (optional, non-differentiable) : T2
+
The running mean after the BatchNormalization operator.
+
running_var (optional, non-differentiable) : T2
+
The running variance after the BatchNormalization operator. This op uses the population size (N) for calculating variance, and not the sample size N-1.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain scale and bias types to float tensors.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain mean and variance types to float tensors.
+
+ +### **Bernoulli-15** + + Draws binary random numbers (0 or 1) from a Bernoulli distribution. The input tensor should be a tensor + containing probabilities p (a value in the range [0,1]) to be used for drawing the binary random number, + where an output of 1 is produced with probability p and an output of 0 is produced with probability (1-p). + + This operator is non-deterministic and may not produce the same values in different + implementations (even if a seed is specified). + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int
+
The data type for the elements of the output tensor. if not specified, we will use the data type of the input tensor.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
All values in input have to be in the range:[0, 1].
+
+ +#### Outputs + +
+
output : T2
+
The returned output tensor only has values 0 or 1, same shape as input tensor.
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bool)
+
Constrain output types to all numeric tensors and bool tensors.
+
+ +### **CastLike-15** + + The operator casts the elements of a given input tensor (the first input) to + the same data type as the elements of the second input tensor. + See documentation of the Cast operator for further details. + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
target_type (non-differentiable) : T2
+
The (first) input tensor will be cast to produce a tensor of the same type as this (second input) tensor.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor produced by casting the first input tensor to have the same type as the second input tensor.
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string), tensor(bfloat16)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string), tensor(bfloat16)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **Optional-15** + + Constructs an optional-type value containing either an empty optional of a certain type specified by the attribute, + or a non-empty value containing the input element. + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +#### Attributes + +
+
type : type_proto
+
Type of the element in the optional output
+
+ +#### Inputs (0 - 1) + +
+
input (optional) : V
+
The input element.
+
+ +#### Outputs + +
+
output : O
+
The optional output enclosing the input element.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input type to all tensor and sequence types.
+
O : optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain output type to all optional tensor or optional sequence types.
+
+ +### **OptionalGetElement-15** + + Outputs the element in the optional-type input. It is an error if the input value does not have an element + and the behavior is undefined in this case. + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +#### Inputs + +
+
input : O
+
The optional input.
+
+ +#### Outputs + +
+
output : V
+
Output element in the optional input.
+
+ +#### Type Constraints + +
+
O : optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain input type to optional tensor and optional sequence types.
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain output type to all tensor or sequence types.
+
+ +### **OptionalHasElement-15** + + Returns true if the optional-type input contains an element. If it is an empty optional-type, this op returns false. + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +#### Inputs + +
+
input : O
+
The optional input.
+
+ +#### Outputs + +
+
output : B
+
A scalar boolean tensor. If true, it indicates that optional-type input contains an element. Otherwise, it is empty.
+
+ +#### Type Constraints + +
+
O : optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain input type to optional tensor and optional sequence types.
+
B : tensor(bool)
+
Constrain output to a boolean tensor.
+
+ +### **Pow-15** + + Pow takes input data (Tensor) and exponent Tensor, and + produces one output data (Tensor) where the function `f(x) = x^exponent`, + is applied to the data tensor elementwise. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
First operand, base of the exponent.
+
Y (differentiable) : T1
+
Second operand, power of the exponent.
+
+ +#### Outputs + +
+
Z (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input X and output types to float/int tensors.
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input Y types to float/int tensors.
+
+ +### **Shape-15** + + Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. + Optional attributes start and end can be used to compute a slice of the input tensor's shape. + If start axis is omitted, the slice starts from axis 0. + The end axis, if specified, is exclusive (and the returned value will not include the size of that axis). + If the end axis is omitted, the axes upto the last one will be included. + Negative axes indicate counting back from the last axis. + Note that axes will be clamped to the range [0, r], where r is the + rank of the input tensor if they are out-of-range (after adding r in the case of + negative axis). Thus, specifying any end value > r is equivalent to specifying an end + value of r, and specifying any start value < -r is equivalent to specifying a start + value of 0. If start > end, the result will be an empty shape. + + Examples: + + ``` + Input tensor with shape: [2, 3, 4] + No attributes specified. + Output: [2, 3, 4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: -1 + Output: [4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + end: -1 + Output: [2, 3] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: 1 + end: 2 + Output: [3] + ``` + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +#### Attributes + +
+
end : int
+
(Optional) Ending axis for slicing the shape. Negative value means counting dimensions from the back. If omitted, sizes of all axes upto (including) the last one will be included.
+
start : int (default is 0)
+
(Optional) Starting axis for slicing the shape. Default value is 0.Negative value means counting dimensions from the back.
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
shape (non-differentiable) : T1
+
Shape of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor.
+
+ +## Version 16 of the default ONNX operator set +### **GreaterOrEqual-16** + + Returns the tensor resulted from performing the `greater_equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **GridSample-16** + + Given an input `X` and a flow-field `grid`, computes the output `Y` using `X` values and pixel locations from `grid`. + Currently, only spatial (4-D) inputs are supported. For input `X` with shape (N, C, H, W) and `grid` with shape (N, H_out, W_out, 2), + the output `Y` will have shape (N, C, H_out, W_out). + + The tensor `X` contains values at centers of square pixels in a H by W 2-dimensional image. + The tensor `grid` describes normalized positions where the output `Y` is to be computed + using a specified interpolation method (the mode) and a padding mode (for grid positions falling outside the 2-dimensional image). + + Elements in `grid[N, H_out, W_out]` are size-2 vectors specifying positions in the 2-dimensional space of `X`. + They are used to interpolate output values of `Y[N, C, H_out, W_out]`. + + The GridSample operator is often used in doing grid generator and sampler in the [Spatial Transformer Networks](https://arxiv.org/abs/1506.02025). + See also in [torch.nn.functional.grid_sample](https://pytorch.org/docs/master/generated/torch.nn.functional.grid_sample.html#torch-nn-functional-grid-sample). + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Attributes + +
+
align_corners : int (default is 0)
+
If align_corners=1, the extrema (-1 and 1) are considered as referring to the center points of the input's corner pixels. If align_corners=0, they are instead considered as referring to the corner points of the input's corner pixels, making the sampling more resolution agnostic.
+
mode : string (default is bilinear)
+
Three interpolation modes: bilinear (default), nearest and bicubic.
+
padding_mode : string (default is zeros)
+
Support padding modes for outside grid values: `zeros`(default), `border`, `reflection`. zeros: use 0 for out-of-bound grid locations, border: use border values for out-of-bound grid locations, reflection: use values at locations reflected by the border for out-of-bound grid locations. If index 0 represents the margin pixel, the reflected value at index -1 will be the same as the value at index 1. For location far away from the border, it will keep being reflected until becoming in bound. If pixel location x = -3.5 reflects by border -1 and becomes x' = 1.5, then reflects by border 1 and becomes x'' = 0.5.
+
+ +#### Inputs + +
+
X (differentiable) : T1
+
4-D tensor of shape (N, C, H, W), where N is the batch size, C is the numbers of channels, H and W are the height and width of the input data.
+
grid (non-differentiable) : T2
+
Input offset, 4-D tensor of shape (N, H_out, W_out, 2), where H_out and W_out are the height and width of grid and output, Grid specifies the sampling pixel locations normalized by the input spatial dimensions. Therefore, it should have most values in the range of [-1, 1]. If grid has values outside the range of [-1, 1], the corresponding outputs will be handled as defined by padding_mode.
+
+ +#### Outputs + +
+
Y (differentiable) : T1
+
4-D tensor of shape (N, C, H_out, W_out) of sampled values. For integer input types, intermediate values are computed as floating point and cast to integer at the end.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input `X` and output `Y` types to all tensor types.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Constrain grid types to float tensors.
+
+ +### **Identity-16** + + Identity operator + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : V
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : V
+
Tensor to copy input into.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain input and output types to all tensor, sequence, and optional types.
+
+ +### **If-16** + + If conditional + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Attributes + +
+
else_branch : graph (required)
+
Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
+
then_branch : graph (required)
+
Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
+
+ +#### Inputs + +
+
cond : B
+
Condition for the if. The tensor must contain a single element.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : V
+
Values that are live-out to the enclosing scope. The return values in the `then_branch` and `else_branch` must be of the same data type. The `then_branch` and `else_branch` may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if-node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of `then_branch` is typed float tensor with shape [2] and the first output of `else_branch` is another float tensor with shape [3], If's first output should have (a) no shape set, or (b) a shape of rank 1 with neither `dim_value` nor `dim_param` set, or (c) a shape of rank 1 with a unique `dim_param`. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv4.
+
B : tensor(bool)
+
Only bool
+
+ +### **LeakyRelu-16** + + LeakyRelu takes input data (Tensor) and an argument alpha, and produces one + output data (Tensor) where the function `f(x) = alpha * x for x < 0`, + `f(x) = x for x >= 0`, is applied to the data tensor elementwise. + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 0.01)
+
Coefficient of leakage.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **LessOrEqual-16** + + Returns the tensor resulted from performing the `less_equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Loop-16** + + Generic Looping construct. This loop has multiple termination conditions: + + 1) Trip count. Iteration count specified at runtime. Set by + specifying the input M. Optional. Set to empty string to omit. + Note that a static trip count (specified at graph construction time) can be + specified by passing in a constant node for input M. + 2) Loop termination condition. This is an input to the op that determines + whether to run the first iteration and also a loop-carried dependency for + the body graph. The body graph must yield a value for the condition variable, + whether this input is provided or not. + + This table summarizes the operating modes of this operator with equivalent + C-style code: + + Operator inputs defined as (max_trip_count, condition_var). + + * input ("", ""): + for (int i=0; ; ++i) { + cond = ... // Note this value is ignored, but is required in the body + } + + * input ("", cond) // Note this is analogous to a while loop + bool cond = ...; + for (int i=0; cond; ++i) { + cond = ...; + } + + * input ("", 1) // Note this is analogous to a do-while loop + bool cond = true + for (int i=0; cond; ++i) { + cond = ...; + } + + * input (trip_count, "") // Note this is analogous to a for loop + int trip_count = ... + for (int i=0; i < trip_count; ++i) { + cond = ...; // ignored + } + + * input (trip_count, cond) + int trip_count = ...; + bool cond = ...; + for (int i=0; i < trip_count && cond; ++i) { + cond = ...; + } + + + *Sample usage - cond as well as trip count* + + graph predict-net { + %a = Constant[value = ]() + %b = Constant[value = ]() + %keepgoing = Constant[value = ]() + %max_trip_count = Constant[value = ]() + %keepgoing_out, %b_out, %user_defined_vals = Loop[body = ](%max_trip_count, %keepgoing, %b) + return + } + + graph body-net ( + %i[INT32, scalar] // iteration number + %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used + %b_in[INT32, scalar] // incoming value of loop-carried-dependency b + ) { + %my_local = Add(%a, %b_in) + %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b + %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition + %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated + return %keepgoing_out, %b_out, %user_defined_val + } + + *Sample equivalent C code* + + { + /* User-defined code (enclosing scope) */ + int a = 3, b = 6; + bool keepgoing = true; // Analogous to input cond + /* End user-defined code */ + + /* Implicitly-defined code */ + const int max_trip_count = 10; // Analogous to input M + int user_defined_vals[]; // Imagine this is resizable + /* End implicitly-defined code */ + /* initialize loop-carried variables and scan-output variables */ + bool keepgoing_out = keepgoing + int b_out = b + + for (int i=0; i < max_trip_count && keepgoing_out; ++i) { + /* Implicitly-defined code: bind actual parameter values + to formal parameter variables of loop-body */ + bool keepgoing_in = keepgoing_out; + bool b_in = b_out; + + /* User-defined code (loop body) */ + int my_local = a + b_in; // Reading value "a" from the enclosing scope is fine + b_out = a - b_in; + keepgoing_out = my_local > b_out; + user_defined_val = b_in + b_in; // b_in and b_out are different variables + /* End user-defined code */ + + /* Implicitly defined-code */ + user_defined_vals[i] = user_defined_val // accumulate scan-output values + } + // int t = my_local; // Can't do this. my_local is not accessible here. + + // The values below are bound to the output variables of the loop and therefore accessible + // b_out; user_defined_vals; keepgoing_out; + } + + There are several things of note in this code snippet: + + 1) Values from the enclosing scope (i.e. variable "a" here) are in scope and can + be referenced in the inputs of the loop. + 2) Any values computed in the loop body that needs to be used in a subsequent + iteration or after the loop are modeled using a pair of variables in the loop-body, + consisting of an input variable (eg., b_in) and an output variable (eg., b_out). + These are referred to as loop-carried dependences. The loop operation node + supplies the input value of the input variable for the first iteration, and + returns the output value of the output variable produced by the final + iteration. + 3) Scan_output variables are used to implicitly concatenate values computed across + all the iterations. In the above example, the value of user_defined_val computed + over all iterations are concatenated and returned as the value of user_defined_vals + after the loop. + 4) Values created in the body cannot be accessed in the enclosing scope, + except using the mechanism described above. + + Note that the semantics of this op support "diagonal" or "wavefront" execution. + (See Step 3 here for an example: + https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). + Frontends should emit multi-layer RNNs as a series of While operators (with + time being the inner looping dimension), with each successive layer consuming + the scan_outputs from the previous layer, possibly going through several + point-wise operators (e.g. dropout, residual connections, linear layer). + + The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order. + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies...). It has 1+N+K outputs: (condition, loop carried dependencies..., scan_outputs...). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
+
+ +#### Inputs (2 - ∞) + +
+
M (optional) : I
+
A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
+
cond (optional) : B
+
A boolean termination condition. Optional. Pass empty string to skip.
+
v_initial (variadic, heterogeneous) : V
+
The initial values of any loop-carried dependencies (values that change across loop iterations)
+
+ +#### Outputs (1 - ∞) + +
+
v_final_and_scan_outputs (variadic, heterogeneous) : V
+
Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv4.
+
I : tensor(int64)
+
tensor of int64, which should be a scalar.
+
B : tensor(bool)
+
tensor of bool, which should be a scalar.
+
+ +### **PRelu-16** + + PRelu takes input data (Tensor) and slope tensor as input, and produces one + output data (Tensor) where the function `f(x) = slope * x for x < 0`, + `f(x) = x for x >= 0`., is applied to the data tensor elementwise. + This operator supports **unidirectional broadcasting** (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
slope (differentiable) : T
+
Slope tensor. The shape of slope can be smaller than first input X; if so, its shape must be unidirectional broadcastable to X
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor (same size as X)
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64)
+
Constrain input and output types to float/int tensors.
+
+ +### **RoiAlign-16** + + Region of Interest (RoI) align operation described in the + [Mask R-CNN paper](https://arxiv.org/abs/1703.06870). + RoiAlign consumes an input tensor X and region of interests (rois) + to apply pooling across each RoI; it produces a 4-D tensor of shape + (num_rois, C, output_height, output_width). + + RoiAlign is proposed to avoid the misalignment by removing + quantizations while converting from original image into feature + map and from feature map into RoI feature; in each ROI bin, + the value of the sampled locations are computed directly + through bilinear interpolation. + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Attributes + +
+
coordinate_transformation_mode : string (default is half_pixel)
+
Allowed values are 'half_pixel' and 'output_half_pixel'. Use the value 'half_pixel' to pixel shift the input coordinates by -0.5 (the recommended behavior). Use the value 'output_half_pixel' to omit the pixel shift for the input (use this for a backward-compatible behavior).
+
mode : string (default is avg)
+
The pooling method. Two modes are supported: 'avg' and 'max'. Default is 'avg'.
+
output_height : int (default is 1)
+
default 1; Pooled output Y's height.
+
output_width : int (default is 1)
+
default 1; Pooled output Y's width.
+
sampling_ratio : int (default is 0)
+
Number of sampling points in the interpolation grid used to compute the output value of each pooled output bin. If > 0, then exactly sampling_ratio x sampling_ratio grid points are used. If == 0, then an adaptive number of grid points are used (computed as ceil(roi_width / output_width), and likewise for height). Default is 0.
+
spatial_scale : float (default is 1.0)
+
Multiplicative spatial scale factor to translate ROI coordinates from their input spatial scale to the scale used when pooling, i.e., spatial scale of the input feature map X relative to the input image. E.g.; default is 1.0f.
+
+ +#### Inputs + +
+
X : T1
+
Input data tensor from the previous operator; 4-D feature map of shape (N, C, H, W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
+
rois : T1
+
RoIs (Regions of Interest) to pool over; rois is 2-D input of shape (num_rois, 4) given as [[x1, y1, x2, y2], ...]. The RoIs' coordinates are in the coordinate system of the input image. Each coordinate set has a 1:1 correspondence with the 'batch_indices' input.
+
batch_indices : T2
+
1-D tensor of shape (num_rois,) with each element denoting the index of the corresponding image in the batch.
+
+ +#### Outputs + +
+
Y : T1
+
RoI pooled output, 4-D tensor of shape (num_rois, C, output_height, output_width). The r-th batch element Y[r-1] is a pooled feature map corresponding to the r-th RoI X[r-1].
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double)
+
Constrain types to float tensors.
+
T2 : tensor(int64)
+
Constrain types to int tensors.
+
+ +### **Scan-16** + + Scan can be used to iterate over one or more scan_input tensors, + constructing zero or more scan_output tensors. It combines ideas from general recurrences, + functional programming constructs such as scan, fold, map, and zip, and is intended to enable + generalizations of RNN-like constructs for sequence-to-sequence processing. + Other tensors (referred to as state_variables here) can be used to carry a state + when iterating from one element to another (similar to hidden-state in RNNs, also referred + to as loop-carried dependences in the context of loops). + Many common usages involve a single scan_input tensor (where functionality + similar to scan, fold and map can be obtained). When more than one scan_input is used, + a behavior similar to zip is obtained. + + The attribute body must be a graph, specifying the computation to be performed in + every iteration. It takes as input the current values of the state_variables and + the current iterated element of the scan_inputs. It must return the (updated) values + of the state_variables and zero or more scan_output_element tensors. The values of the + scan_output_element tensors are concatenated over all the iterations to produce the + scan_output values of the scan construct (similar to the concatenated intermediate + hidden-state values of RNN-like constructs). All the output tensors (state_variables as + well as scan_output_element tensors) are required to have the same shape in each iteration + of the loop (a restriction imposed to enable efficient memory allocation). + + Note that the iterated element passed to the body subgraph does not have a sequence + axis. It will have a rank one less than the rank of the corresponding scan_input. + + The scan operation returns the final values of the state_variables as well as the + scan_outputs. + + The optional attribute scan_input_directions specifies the direction (forward or backward) + for each scan input. If this attribute is omitted, all sequences are scanned in the forward + direction. A bidirectional scan may be performed by specifying the same tensor input twice + in the scan_inputs, once with a forward direction, and once with a backward direction. + + The scan_output of the operation is produced by concatenating the scan_output_element + values produced by the body in each iteration. The optional attribute scan_output_directions + specifies the direction in which scan_output is constructed (by appending or prepending the + scan_output_element to scan_output in each iteration) for each scan_output. If this attribute + is omitted, the scan_output_element is appended to the scan_output in each iteration. + + The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. + If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the + batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. + Note that scanning a non-zero axis may be less efficient than scanning axis zero. + + The optional attribute scan_output_axes specifies the axis along which the scan_outputs + are accumulated for each scan_output. For example, if axis 1 is the time axis (to be + scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis + value of 1. + + Note that because of the ONNX restriction that only the last parameter of an operator can + be variadic, the initial-states and scan-inputs are listed together as one input parameter. + Similarly, the final-states and scan-outputs are listed together as one output parameter. + The attribute num_scan_inputs indicates the number M of scan-inputs. + + The behavior of + + Scan < + num_scan_inputs = m, + body = loop-body, + scan_input_axes = [axis_1, ..., axis_m] + > (init_1, ..., init_n, scan_1, ..., scan_m) + + is equivalent to the following pseudo-code: + + // scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i + // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. + sequence_length = scan_1.shape[axis_1]; + + // initialize state-variables + st_1 = init_1; ... st_n = init_n; + // initialize scan-output variables: [] denotes an empty tensor + scan_out_1 = []; ...; scan_out_k = []; + // identify number of iterations: + + // execute loop + for (int t = 0; t < sequence_length; ++t) { + // generate the scan-input elements: the notation T[t] indicates the sub-tensor + // of rank one less than T obtained by indexing T at position t along axis k. + si_1 = scan_1[t]; + ... ; + si_m = scan_m[t]; + // execute loop-body + st_1, ..., st_n, so_1, ..., so_k = loop-body(st_1, ..., st_n, si_1, ..., si_m) + // accumulate the scan-output elements + scan_out_1 = Concat(scan_out_1, so_1); ... ; scan_out_k = Concat(scan_out_k, so_k); + } + + return st_1, ..., st_n, scan_out_1, ..., scan_out_k; + + *Sample usage: Encoding RNN using a Scan* + + The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, + recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can + be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes + %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these + values are computed in the outer graph, they need to be passed in as extra state_variables. + + graph rnn-encoding { + %H_0 = ... + %X = ... + %Y_h, %Y = Scan[body = , num_scan_inputs=1](%H_0, %X) + return %Y, %Y_h + } + + graph rnn-cell-1 ( + %H_tminus1[FLOAT, tensor] + %X_t[FLOAT, tensor] + ) { + %Wi = ... + %Ri = ... + %Wbi = ... + %Rbi = ... + %t1 = X_t * (Wi^T) + %t2 = H_tminus1*(Ri^T) + %t3 = Add(%t1, %t2) + %t4 = Add(%t3, %Wbi) + %t5 = Add(%t4, %Rbi) + %Ht = Tanh(%t5) + %Accumulate = Identity(%Ht) + return %Ht, %Accumulate + } + + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has N+M inputs: (loop state variables..., scan_input_elts...). It has N+K outputs: (loop state variables..., scan_output_elts...). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
+
num_scan_inputs : int (required)
+
An attribute specifying the number of scan_inputs M.
+
scan_input_axes : list of ints
+
An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
scan_input_directions : list of ints
+
An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
+
scan_output_axes : list of ints
+
An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
+
scan_output_directions : list of ints
+
An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
+
+ +#### Inputs (1 - ∞) + +
+
initial_state_and_scan_inputs (variadic, heterogeneous) : V
+
Initial values of the loop's N state variables followed by M scan_inputs
+
+ +#### Outputs (1 - ∞) + +
+
final_state_and_scan_outputs (variadic, heterogeneous) : V
+
Final values of the loop's N state variables followed by K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
All Tensor types up to IRv4.
+
+ +### **ScatterElements-16** + + ScatterElements takes three inputs `data`, `updates`, and `indices` of the same + rank r >= 1 and an optional attribute axis that identifies an axis of `data` + (by default, the outer-most axis, that is axis 0). The output of the operation + is produced by creating a copy of the input `data`, and then updating its value + to values specified by `updates` at specific index positions specified by + `indices`. Its output shape is the same as the shape of `data`. + For each entry in `updates`, the target index in `data` is obtained by combining + the corresponding entry in `indices` with the index of the entry itself: the + index-value for dimension = axis is obtained from the value of the corresponding + entry in `indices` and the index-value for dimension != axis is obtained from the + index of the entry itself. + `reduction` allows specification of an optional reduction operation, which is applied to all values in `updates` + tensor into `output` at the specified `indices`. + In cases where `reduction` is set to "none", indices should not have duplicate entries: that is, if idx1 != idx2, + then indices[idx1] != indices[idx2]. For instance, in a 2-D tensor case, the update + corresponding to the [i][j] entry is performed as below: + ``` + output[indices[i][j]][j] = updates[i][j] if axis = 0, + output[i][indices[i][j]] = updates[i][j] if axis = 1, + ``` + When `reduction` is set to "add", the update corresponding to the [i][j] entry is performed as below: + ``` + output[indices[i][j]][j] += updates[i][j] if axis = 0, + output[i][indices[i][j]] += updates[i][j] if axis = 1, + ``` + When `reduction` is set to "mul", the update corresponding to the [i][j] entry is performed as below: + ``` + output[indices[i][j]][j] *= updates[i][j] if axis = 0, + output[i][indices[i][j]] *= updates[i][j] if axis = 1, + ``` + This operator is the inverse of GatherElements. It is similar to Torch's Scatter operation. + Example 1: + ``` + data = [ + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + ] + indices = [ + [1, 0, 2], + [0, 2, 1], + ] + updates = [ + [1.0, 1.1, 1.2], + [2.0, 2.1, 2.2], + ] + output = [ + [2.0, 1.1, 0.0] + [1.0, 0.0, 2.2] + [0.0, 2.1, 1.2] + ] + ``` + Example 2: + ``` + data = [[1.0, 2.0, 3.0, 4.0, 5.0]] + indices = [[1, 3]] + updates = [[1.1, 2.1]] + axis = 1 + output = [[1.0, 1.1, 3.0, 2.1, 5.0]] + ``` + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
reduction : string (default is none)
+
Type of reduction to apply: none (default), add, mul. 'none': no reduction applied. 'add': reduction using the addition operation. 'mul': reduction using the multiplication operation.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : Tind
+
Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
updates (differentiable) : T
+
Tensor of rank r >=1 (same rank and shape as indices)
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank r >= 1 (same rank as input).
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input and output types can be of any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **ScatterND-16** + + ScatterND takes three inputs `data` tensor of rank r >= 1, `indices` tensor of rank q >= 1, + and `updates` tensor of rank q + r - indices.shape[-1] - 1. The output of the operation + is produced by creating a copy of the input `data`, and then updating its value to values + specified by `updates` at specific index positions specified by `indices`. Its output shape + is the same as the shape of `data`. + + `indices` is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of `indices`. + `indices` is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into `data`. + Hence, k can be a value at most the rank of `data`. When k equals rank(data), each update entry specifies an + update to a single element of the tensor. When k is less than rank(data) each update entry specifies an + update to a slice of the tensor. Index values are allowed to be negative, as per the usual + convention for counting backwards from the end, but are expected in the valid range. + + `updates` is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the + first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. + The remaining dimensions of `updates` correspond to the dimensions of the + replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, + corresponding to the trailing (r-k) dimensions of `data`. Thus, the shape of `updates` + must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation + of shapes. + + The `output` is calculated via the following equation: + output = np.copy(data) + update_indices = indices.shape[:-1] + for idx in np.ndindex(update_indices): + output[tuple(indices[idx])] = updates[idx] + The order of iteration in the above loop is not specified. + In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. + This ensures that the output value does not depend on the iteration order. + + `reduction` allows specification of an optional reduction operation, which is applied to all values in `updates` + tensor into `output` at the specified `indices`. + In cases where `reduction` is set to "none", indices should not have duplicate entries: that is, if idx1 != idx2, + then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order. + When `reduction` is set to "add", `output` is calculated as follows: + output = np.copy(data) + update_indices = indices.shape[:-1] + for idx in np.ndindex(update_indices): + output[tuple(indices[idx])] += updates[idx] + When `reduction` is set to "mul", `output` is calculated as follows: + output = np.copy(data) + update_indices = indices.shape[:-1] + for idx in np.ndindex(update_indices): + output[tuple(indices[idx])] *= updates[idx] + This operator is the inverse of GatherND. + Example 1: + ``` + data = [1, 2, 3, 4, 5, 6, 7, 8] + indices = [[4], [3], [1], [7]] + updates = [9, 10, 11, 12] + output = [1, 11, 3, 10, 9, 6, 7, 12] + ``` + Example 2: + ``` + data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] + indices = [[0], [2]] + updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]] + output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] + ``` + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Attributes + +
+
reduction : string (default is none)
+
Type of reduction to apply: none (default), add, mul. 'none': no reduction applied. 'add': reduction using the addition operation. 'mul': reduction using the multiplication operation.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : tensor(int64)
+
Tensor of rank q >= 1.
+
updates (differentiable) : T
+
Tensor of rank q + r - indices_shape[-1] - 1.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank r >= 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
+ +### **Where-16** + + Return elements, either from X or Y, depending on condition. + Where behaves like + [numpy.where](https://docs.scipy.org/doc/numpy/reference/generated/numpy.where.html) + with three parameters. + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +#### Inputs + +
+
condition (non-differentiable) : B
+
When True (nonzero), yield X, otherwise yield Y
+
X (differentiable) : T
+
values selected at indices where condition is True
+
Y (differentiable) : T
+
values selected at indices where condition is False
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of shape equal to the broadcasted shape of condition, X, and Y.
+
+ +#### Type Constraints + +
+
B : tensor(bool)
+
Constrain to boolean tensors.
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types (including bfloat).
+
+ +## Version 17 of the default ONNX operator set +### **BlackmanWindow-17** + + Generates a Blackman window as described in the paper https://ieeexplore.ieee.org/document/1455106. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
output_datatype : int (default is 1)
+
The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT.
+
periodic : int (default is 1)
+
If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When 'periodic' is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1.
+
+ +#### Inputs + +
+
size (non-differentiable) : T1
+
A scalar value indicating the length of the window.
+
+ +#### Outputs + +
+
output (non-differentiable) : T2
+
A Blackman window with length: size. The output has the shape: [size].
+
+ +#### Type Constraints + +
+
T1 : tensor(int32), tensor(int64)
+
Constrain the input size to int32_t or int64_t.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain output types to numeric tensors.
+
+ +### **DFT-17** + + Computes the discrete Fourier transform of input. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
The axis on which to perform the DFT. By default this value is set to 1, which corresponds to the first dimension after the batch index. Negative value means counting dimensions from the back. Accepted range is $[-r, -2] \cup [0, r-2]$ where `r = rank(input)`. The last dimension is for representing complex numbers and thus is an invalid axis.
+
inverse : int (default is 0)
+
Whether to perform the inverse discrete fourier transform. By default this value is set to 0, which corresponds to false.
+
onesided : int (default is 0)
+
If onesided is 1, only values for w in [0, 1, 2, ..., floor(n_fft/2) + 1] are used or returned because the real-to-complex Fourier transform satisfies the conjugate symmetry, i.e., X[m, w] = X[m, n_fft-w]*. When onesided=1 and inverse=0 (forward DFT), only real input is supported and a one-sided complex spectrum is returned (RFFT). When onesided=1 and inverse=1 (inverse DFT), only complex input is supported and a full real signal is returned (IRFFT). When invoked with real or complex valued input, the default value is 0. Values can be 0 or 1.
+
+ +#### Inputs (1 - 2) + +
+
input (non-differentiable) : T1
+
For real input, the following shape is expected: [batch_idx][signal_dim1][signal_dim2]...[signal_dimN][1]. For complex input, the following shape is expected: [batch_idx][signal_dim1][signal_dim2]...[signal_dimN][2]. The first dimension is the batch dimension. The following N dimensions correspond to the signal's dimensions. The final dimension represents the real and imaginary parts of the value in that order.
+
dft_length (optional, non-differentiable) : T2
+
The length of the signal as a scalar. If greater than the axis dimension, the signal will be zero-padded up to dft_length. If less than the axis dimension, only the first dft_length values will be used as the signal. If not provided, the default dft_length = signal_dim_axis, except for the IRFFT case (onesided=1, inverse=1), in which case the default dft_length is 2 * (signal_dim_axis - 1). It's an optional value.
+
+ +#### Outputs + +
+
output : T1
+
The Fourier Transform of the input vector. For standard DFT (onesided=0), the output shape is: [batch_idx][signal_dim1][signal_dim2]...[signal_dimN][2] (complex), with signal_dim_axis = dft_length. For RFFT (onesided=1, inverse=0), the output shape is: [batch_idx][signal_dim1][signal_dim2]...[signal_dimN][2] (one-sided complex), with signal_dim_axis = floor(dft_length/2) + 1. For IRFFT (onesided=1, inverse=1), the output shape is: [batch_idx][signal_dim1][signal_dim2]...[signal_dimN][1] (real), where signal_dim_axis = dft_length.
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
T2 : tensor(int32), tensor(int64)
+
Constrain scalar length types to int64_t.
+
+ +### **HammingWindow-17** + + Generates a Hamming window as described in the paper https://ieeexplore.ieee.org/document/1455106. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
output_datatype : int (default is 1)
+
The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT.
+
periodic : int (default is 1)
+
If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When 'periodic' is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1.
+
+ +#### Inputs + +
+
size (non-differentiable) : T1
+
A scalar value indicating the length of the window.
+
+ +#### Outputs + +
+
output (non-differentiable) : T2
+
A Hamming window with length: size. The output has the shape: [size].
+
+ +#### Type Constraints + +
+
T1 : tensor(int32), tensor(int64)
+
Constrain the input size to int32_t or int64_t.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain output types to numeric tensors.
+
+ +### **HannWindow-17** + + Generates a Hann window as described in the paper https://ieeexplore.ieee.org/document/1455106. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
output_datatype : int (default is 1)
+
The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT.
+
periodic : int (default is 1)
+
If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When 'periodic' is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1.
+
+ +#### Inputs + +
+
size (non-differentiable) : T1
+
A scalar value indicating the length of the window.
+
+ +#### Outputs + +
+
output (non-differentiable) : T2
+
A Hann window with length: size. The output has the shape: [size].
+
+ +#### Type Constraints + +
+
T1 : tensor(int32), tensor(int64)
+
Constrain the input size to int32_t or int64_t.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain output types to numeric tensors.
+
+ +### **LayerNormalization-17** + + This is layer normalization defined in ONNX as function. + The overall computation can be split into two stages. + The first stage is standardization, which makes the + normalized elements have zero mean and unit variances. + The computation required by standardization can be + described by the following equations. + ``` + Mean = ReduceMean(X) + D = Sub(X, Mean) + DD = Mul(D, D) + Var = ReduceMean(DD) + VarEps = Add(Var, epsilon) + StdDev = Sqrt(VarEps) + InvStdDev = Reciprocal(StdDev) + Normalized = Mul(D, InvStdDev) + ``` + where `normalized_axes` is `[axis, ..., rank of X - 1]`. + The variables `Var` and `StdDev` stand for variance and + standard deviation, respectively. The second output is + `Mean` and the last one is `InvStdDev`. + Depending on `stash_type` attribute, the actual computation + must happen in different floating-point precision. + For example, if `stash_type` is 1, this operator casts + all input variables to 32-bit float, perform the computation, and + finally cast `Normalized` back to the original type of `X`. + The second stage then scales and shifts the outcome of the + first stage using + ``` + NormalizedScaled = Mul(Normalized, Scale) + Y = Add(NormalizedScaled, B) + ``` + The second stage doesn't depends on `stash_type`. + All equations are in [this syntax](https://github.com/onnx/onnx/blob/main/docs/Syntax.md). + The same variable (i.e., input, output, and attribute) uses + the same name in the equations above and this operator's definition. + Let `d[i]` indicate the i-th dimension of `X`. + If `X`'s shape is `[d[0], ..., d[axis-1], d[axis], ..., d[rank-1]]`, + the shape of `Mean` and `InvStdDev` is `[d[0], ..., d[axis-1], 1, ..., 1]`. + `Y` and `X` have the same shape. This operator supports unidirectional broadcasting + (tensors `Scale` and `B` should be unidirectional broadcastable to tensor `X`); + for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
The first normalization dimension. If rank(X) is r, axis' allowed range is [-r, r). Negative value means counting dimensions from the back.
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
stash_type : int (default is 1)
+
Type of Mean and InvStdDev. This also specifies stage one's computation precision.
+
+ +#### Inputs (2 - 3) + +
+
X : T
+
Tensor to be normalized.
+
Scale : T
+
Scale tensor.
+
B (optional) : T
+
Bias tensor.
+
+ +#### Outputs (1 - 3) + +
+
Y : T
+
Normalized tensor.
+
Mean (optional) : U
+
Saved mean used during training to speed up gradient computation
+
InvStdDev (optional) : U
+
Saved inverse standard deviation used during training to speed up gradient computation.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types and output Y type to float tensors.
+
U : tensor(float), tensor(bfloat16)
+
Type of Mean and InvStdDev tensors.
+
+ +### **MelWeightMatrix-17** + + Generate a MelWeightMatrix that can be used to re-weight a Tensor containing a linearly sampled frequency spectra (from DFT or STFT) into num_mel_bins frequency information based on the [lower_edge_hertz, upper_edge_hertz] range on the mel scale. + This function defines the mel scale in terms of a frequency in hertz according to the following formula: + + mel(f) = 2595 * log10(1 + f/700) + + In the returned matrix, all the triangles (filterbanks) have a peak value of 1.0. + + The returned MelWeightMatrix can be used to right-multiply a spectrogram S of shape [frames, num_spectrogram_bins] of linear scale spectrum values (e.g. STFT magnitudes) to generate a "mel spectrogram" M of shape [frames, num_mel_bins]. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
output_datatype : int (default is 1)
+
The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T3. The default value is 1 = FLOAT.
+
+ +#### Inputs + +
+
num_mel_bins (non-differentiable) : T1
+
The number of bands in the mel spectrum.
+
dft_length (non-differentiable) : T1
+
The size of the original DFT. The size of the original DFT is used to infer the size of the onesided DFT, which is understood to be floor(dft_length/2) + 1, i.e. the spectrogram only contains the nonredundant DFT bins.
+
sample_rate (non-differentiable) : T1
+
Samples per second of the input signal used to create the spectrogram. Used to figure out the frequencies corresponding to each spectrogram bin, which dictates how they are mapped into the mel scale.
+
lower_edge_hertz (non-differentiable) : T2
+
Lower bound on the frequencies to be included in the mel spectrum. This corresponds to the lower edge of the lowest triangular band.
+
upper_edge_hertz (non-differentiable) : T2
+
The desired top edge of the highest frequency band.
+
+ +#### Outputs + +
+
output (non-differentiable) : T3
+
The Mel Weight Matrix. The output has the shape: [floor(dft_length/2) + 1][num_mel_bins].
+
+ +#### Type Constraints + +
+
T1 : tensor(int32), tensor(int64)
+
Constrain to integer tensors.
+
T2 : tensor(float), tensor(float16), tensor(double), tensor(bfloat16)
+
Constrain to float tensors
+
T3 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain to any numerical types.
+
+ +### **STFT-17** + + Computes the Short-time Fourier Transform of the signal. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
onesided : int (default is 1)
+
If onesided is 1, only values for w in [0, 1, 2, ..., floor(n_fft/2) + 1] are returned because the real-to-complex Fourier transform satisfies the conjugate symmetry, i.e., X[m, w] = X[m,w]=X[m,n_fft-w]*. Note if the input or window tensors are complex, then onesided output is not possible. Enabling onesided with real inputs performs a Real-valued fast Fourier transform (RFFT).When invoked with real or complex valued input, the default value is 1. Values can be 0 or 1.
+
+ +#### Inputs (2 - 4) + +
+
signal (non-differentiable) : T1
+
Input tensor representing a real or complex valued signal. For real input, the following shape is expected: [batch_size][signal_length][1]. For complex input, the following shape is expected: [batch_size][signal_length][2], where [batch_size][signal_length][0] represents the real component and [batch_size][signal_length][1] represents the imaginary component of the signal.
+
frame_step (non-differentiable) : T2
+
The number of samples to step between successive DFTs.
+
window (optional, non-differentiable) : T1
+
A tensor representing the window that will be slid over the signal.The window must have rank 1 with shape: [window_shape]. It's an optional value.
+
frame_length (optional, non-differentiable) : T2
+
A scalar representing the size of the DFT. It's an optional value.
+
+ +#### Outputs + +
+
output (non-differentiable) : T1
+
The Short-time Fourier Transform of the signals.If onesided is 1, the output has the shape: [batch_size][frames][dft_unique_bins][2], where dft_unique_bins is frame_length // 2 + 1 (the unique components of the DFT) If onesided is 0, the output has the shape: [batch_size][frames][frame_length][2], where frame_length is the length of the DFT.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(float16), tensor(double), tensor(bfloat16)
+
Constrain signal and output to float tensors.
+
T2 : tensor(int32), tensor(int64)
+
Constrain scalar length types to int64_t.
+
+ +### **SequenceMap-17** + + Applies a sub-graph to each sample in the input sequence(s). + + Inputs can be either tensors or sequences, with the exception of the first input which must + be a sequence. The length of the first input sequence will determine the number of samples in the + outputs. Any other sequence inputs should have the same number of samples. The number of inputs + and outputs, should match the one of the subgraph. + + For each i-th element in the output, a sample will be extracted from the input sequence(s) at + the i-th position and the sub-graph will be applied to it. + The outputs will contain the outputs of the sub-graph for each sample, in the same order as in + the input. + + This operator assumes that processing each sample is independent and could executed in parallel + or in any order. Users cannot expect any specific ordering in which each subgraph is computed. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph to be run for each sample in the sequence(s). It should have as many inputs and outputs as inputs and outputs to the SequenceMap function.
+
+ +#### Inputs (1 - ∞) + +
+
input_sequence : S
+
Input sequence.
+
additional_inputs (variadic, heterogeneous) : V
+
Additional inputs to the graph
+
+ +#### Outputs (1 - ∞) + +
+
out_sequence (variadic, heterogeneous) : S
+
Output sequence(s)
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input types to any sequence type.
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain to any tensor or sequence type.
+
+ +## Version 18 of the default ONNX operator set +### **BitwiseAnd-18** + + Returns the tensor resulting from performing the bitwise `and` operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the bitwise operator.
+
B (non-differentiable) : T
+
Second input operand for the bitwise operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64)
+
Constrain input to integer tensors.
+
+ +### **BitwiseNot-18** + + Returns the bitwise not of the input tensor element-wise. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64)
+
Constrain input/output to integer tensors.
+
+ +### **BitwiseOr-18** + + Returns the tensor resulting from performing the bitwise `or` operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the bitwise operator.
+
B (non-differentiable) : T
+
Second input operand for the bitwise operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64)
+
Constrain input to integer tensors.
+
+ +### **BitwiseXor-18** + + Returns the tensor resulting from performing the bitwise `xor` operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the bitwise operator.
+
B (non-differentiable) : T
+
Second input operand for the bitwise operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64)
+
Constrain input to integer tensors.
+
+ +### **CenterCropPad-18** + + Center crop or pad an input to given dimensions. + + The crop/pad dimensions can be specified for a subset of the `axes`; unspecified dimensions will remain unchanged. + + If the input dimensions are larger than the target crop dimensions, a centered cropping window will be extracted + from the input. The starting value for the cropping window is rounded down, which means that if the difference + between the input shape and the crop shape is odd, the cropping window will be shifted half a pixel to the left + of the input center. + + If the input dimensions are smaller than the target crop dimensions, the input will be padded equally on both sides + to center it in the output. In cases where the total number of padding pixels is odd, an additional pixel will be + added to the right side. + + The padding value used is zero. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
If provided, it specifies a subset of axes that 'shape' refer to. If not provided, all axes are assumed [0, 1, ..., r-1], where r = rank(data). Negative value means counting dimensions from the back. Accepted range is [-r, r-1], where r = rank(data). Behavior is undefined if an axis is repeated.
+
+ +#### Inputs + +
+
input_data (differentiable) : T
+
Input to extract the centered crop from.
+
shape (non-differentiable) : Tind
+
1-D tensor representing the cropping window dimensions.
+
+ +#### Outputs + +
+
output_data (differentiable) : T
+
Output data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **Col2Im-18** + + The operator rearranges column blocks back into a multidimensional image + + Col2Im behaves similarly to PyTorch's fold https://pytorch.org/docs/stable/generated/torch.nn.Fold.html, + but it only supports *batched* multi-dimensional image tensors. + Another implementation in Python with N-dimension support can be found at https://github.com/f-dangel/unfoldNd/. + + NOTE: + Although specifying image_shape looks redundant because it could be calculated from + convolution formulas, it is required as input for more advanced scenarios as explained + at PyTorch's implementation (https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/Col2Im.cpp#L10) + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
dilations : list of ints
+
1-dimensional tensor with dilation value along each spatial axis of the image. If not present, the dilation defaults to 1 along each spatial axis of the image.
+
pads : list of ints
+
1-dimensional tensor with padding value for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin is the number of pixels added at the beginning of axis `i` and xi_end is the number of pixels added at the end of axis `i`. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
1-dimensional tensor with stride value along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input data tensor to be rearranged from column blocks back into an image. This is a 3-dimensional tensor containing [N, C * n-ary-product(block_shape), L], where N is batch dimension, C is image channel dimension and L is number of blocks.The blocks are enumerated in increasing lexicographic-order of their indices.For example, with an image-size 10*20 and block-size 9*18, there would be 2*3 blocks, enumerated in the order block(0, 0), block(0, 1), block(0, 2), block(1, 0), block(1, 1), block(1, 2).
+
image_shape (non-differentiable) : tensor(int64)
+
The shape of the spatial dimensions of the image after rearranging the column blocks.This is a 1-dimensional tensor with size of at least 2, containing the value [H_img, W_img] for a 2-D image or [dim_i1, dim_i2, ..., dim_iN] for a N-D image.
+
block_shape (non-differentiable) : tensor(int64)
+
The shape of the block to apply on the input.This is a 1-dimensional tensor of size of at least 2, containing the value [H_block, W_block] for a 2-D image or [dim_b1, dim_b2, ..., dim_bN] for a N-D block.This is the block-shape before dilation is applied to it.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor produced by rearranging blocks into an image.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all numeric tensor types.
+
+ +### **GroupNormalization-18** (deprecated) + + A GroupNormalization function. Carries out group normalization as described in + the paper https://arxiv.org/abs/1803.08494 + + This operator transforms input according to + ``` + y = scale * (x - mean) / sqrt(variance + epsilon) + bias, + ``` + where the mean and variance are computed per instance per group of channels, and + `scale` and `bias` should be specified for each group of channels. The number of + groups `num_groups` should be divisible by the number of channels so that there are + an equal number of channels per group. + + When the number of groups is the same as the number of channels, this operator is + equivalent to InstanceNormalization. When there is only one group, this operator + is equivalent to LayerNormalization. + +#### Version + +This version of the operator has been deprecated since version 18 of the default ONNX operator set. + +### **LpPool-18** + + LpPool consumes an input tensor X and applies Lp pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + Lp pooling consisting of computing the Lp norm on all values of a subset + of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - {kernelSpatialShape}) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - {kernelSpatialShape}) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled `pad_shape[i]` is the sum of pads along axis `i`. + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - {kernelSpatialShape} + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + {kernelSpatialShape} - input_spatial_shape[i] + ``` + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
p : int (default is 2)
+
p value of the Lp norm used to pool over the input data.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Mish-18** + + Mish: A Self Regularized Non-Monotonic Neural Activation Function. + + Perform the linear unit element-wise on the input tensor X using formula: + + ``` + mish(x) = x * tanh(softplus(x)) = x * tanh(ln(1 + e^{x})) + ``` + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input X and output types to float tensors.
+
+ +### **OptionalGetElement-18** + + If the input is a tensor or sequence type, it returns the input. + If the input is an optional type, it outputs the element in the input. + It is an error if the input is an empty optional-type (i.e. does not have an element) and the behavior is undefined in this case. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs + +
+
input : O
+
The optional input.
+
+ +#### Outputs + +
+
output : V
+
Output element in the optional input.
+
+ +#### Type Constraints + +
+
O : optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input type to optional tensor and optional sequence types.
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain output type to all tensor or sequence types.
+
+ +### **OptionalHasElement-18** + + Returns true if (1) the input is an optional-type and contains an element, + or, (2) the input is a tensor or sequence type. + If the input is not provided or is an empty optional-type, this op returns false. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs (0 - 1) + +
+
input (optional) : O
+
The optional input.
+
+ +#### Outputs + +
+
output : B
+
A scalar boolean tensor. If true, it indicates that optional-type input contains an element. Otherwise, it is empty.
+
+ +#### Type Constraints + +
+
O : optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input type to optional tensor and optional sequence types.
+
B : tensor(bool)
+
Constrain output to a boolean tensor.
+
+ +### **Pad-18** + + Given a tensor containing the data to be padded (`data`), a tensor containing the number of start and end pad values for axis (`pads`), (optionally) a `mode`, and (optionally) `constant_value`, + a padded tensor (`output`) is generated. + + The three supported `modes` are (similar to corresponding modes supported by `numpy.pad`): + + 1) `constant`(default) - pads with a given constant value as specified by `constant_value` (which defaults to 0, empty string, or False) + + 2) `reflect` - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis + + 3) `edge` - pads with the edge values of array + + + Example 1 (`constant` mode): + + Insert 0 pads to the beginning of the second dimension. + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'constant' + + constant_value = 0.0 + + output = [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ] + ``` + + Example 2 (`reflect` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'reflect' + + output = [ + [1.0, 1.2, 1.0, 1.2], + [2.3, 3.4, 2.3, 3.4], + [4.5, 5.7, 4.5, 5.7], + ] + ``` + + Example 3 (`edge` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'edge' + + output = [ + [1.0, 1.0, 1.0, 1.2], + [2.3, 2.3, 2.3, 3.4], + [4.5, 4.5, 4.5, 5.7], + ] + ``` + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is constant)
+
Supported modes: `constant`(default), `reflect`, `edge`
+
+ +#### Inputs (2 - 4) + +
+
data (differentiable) : T
+
Input tensor.
+
pads (non-differentiable) : tensor(int64)
+
Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. `pads` should be a 1D tensor of shape [2 * num_axes] where `num_axes` refers to the number of elements in the `axes` input or the input rank if `axes` are not provided explicitly. `pads` format should be: [x1_begin, x2_begin, ..., x1_end, x2_end,...], where xi_begin is the number of pad values added at the beginning of axis `axes[i]` and xi_end, the number of pad values added at the end of axis `axes[i]`.
+
constant_value (optional, non-differentiable) : T
+
(Optional) A scalar value to be used if the mode chosen is `constant` (by default it is 0, empty string or False).
+
axes (optional, non-differentiable) : Tind
+
1-D tensor of axes that `pads` apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated. If not provided, all axes are assumed (`[0, 1, ..., input_rank-1]`).
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **ReduceL1-18** + + Computes the L1 norm of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceL2-18** + + Computes the L2 norm of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceLogSum-18** + + Computes the log sum of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or undefined otherwise. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceLogSumExp-18** + + Computes the log sum exponent of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or undefined otherwise. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceMax-18** + + Computes the max of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or the minimum value of the data type otherwise. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(uint8), tensor(int8)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceMean-18** + + Computes the mean of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields undefined. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceMin-18** + + Computes the min of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields plus infinity (if supported by the datatype) or the maximum value of the data type otherwise. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(uint8), tensor(int8)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceProd-18** + + Computes the product of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 1. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **ReduceSumSquare-18** + + Computes the sum square of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ +### **Resize-18** + + Resize the input tensor. In general, it calculates every value in the output tensor as a weighted average of neighborhood (a.k.a. sampling locations) in the input tensor. + Each dimension value of the output tensor is:
+ `output_dimension = floor(input_dimension * (roi_end - roi_start) * scale)`
+ if input \"sizes\" is not specified. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
antialias : int (default is 0)
+
If set to 1, "linear" and "cubic" interpolation modes will use an antialiasing filter when downscaling. Antialiasing is achieved by stretching the resampling filter by a factor max(1, 1 / scale), which means that when downsampling, more input pixels contribute to an output pixel.
+
axes : list of ints
+
If provided, it specifies a subset of axes that 'roi', 'scales' and 'sizes' refer to. If not provided, all axes are assumed [0, 1, ..., r-1], where r = rank(data). Non-specified dimensions are interpreted as non-resizable. Negative value means counting dimensions from the back. Accepted range is [-r, r-1], where r = rank(data). Behavior is undefined if an axis is repeated.
+
coordinate_transformation_mode : string (default is half_pixel)
+
+This attribute describes how to transform the coordinate in the resized tensor to the coordinate in the original tensor.
+ +The coordinate of each dimension is transformed individually. Let's describe a case using axis x as an example. +Denote x_resized as the coordinate of axis x in the resized tensor, x_original as the coordinate of axis x in the original tensor, `length_original` as the length of the original tensor in axis x, length_resized as the length of the resized tensor in axis x, roi_x = (start_x, end_x) of the axis x in input "roi", `scale = length_resized / length_original`,
+ +if coordinate_transformation_mode is `"half_pixel"`,
+`x_original = (x_resized + 0.5) / scale - 0.5`
+ +if coordinate_transformation_mode is `"pytorch_half_pixel"`,
+`x_original = length_resized > 1 ? (x_resized + 0.5) / scale - 0.5 : 0`
+ +if coordinate_transformation_mode is `"align_corners"`,
+`x_original = x_resized * (length_original - 1) / (length_resized - 1)`
+ +if coordinate_transformation_mode is `"asymmetric"`,
+`x_original = x_resized / scale`
+ +if coordinate_transformation_mode is `"tf_crop_and_resize"`,
+`x_original = length_resized > 1 ? start_x * (length_original - 1) + x_resized * (end_x - start_x) * (length_original - 1) / (length_resized - 1) : 0.5 * (start_x + end_x) * (length_original - 1)` +.
+
cubic_coeff_a : float (default is -0.75)
+
The coefficient 'a' used in cubic interpolation. Two common choice are -0.5 (in some cases of TensorFlow) and -0.75 (in PyTorch). Check out Equation (4) in https://ieeexplore.ieee.org/document/1163711 for the details. This attribute is valid only if mode is "cubic".
+
exclude_outside : int (default is 0)
+
If set to 1, the weight of sampling locations outside the tensor will be set to 0 and the weight will be renormalized so that their sum is 1.0. The default value is 0.
+
extrapolation_value : float (default is 0.0)
+
When coordinate_transformation_mode is "tf_crop_and_resize" and x_original is outside the range [0, length_original - 1], this value is used as the corresponding output value. Default is 0.0f.
+
keep_aspect_ratio_policy : string (default is stretch)
+
+This attribute describes how to interpret the `sizes` input with regard to keeping the original aspect ratio of the input, and it is not applicable when +the `scales` input is used.
+ +Given a set of `sizes`, associated with a subset of `axes` (explicitly provided or default), and assuming `d = axes[i]`, with `i` being the index of the provided `sizes`.
+ +If `keep_aspect_ratio_policy` is `"stretch"`, the original aspect ratio is disregarded, and the input is resized to the specified size:
+`out_size[d] = sizes[i]`
+ +If `keep_aspect_ratio_policy` is `"not_larger"`, the sizes are adjusted so that no extent of the output is larger than the specified size, while keeping the original aspect ratio:
+`scale = Min(sizes[i] / in_size[d])`
+`out_size[d] = round_int(scale * in_size[d])`
+ +If `keep_aspect_ratio_policy` is `"not_smaller"`, the sizes are adjusted so that no extent of the output is smaller than the specified size, while keeping the original aspect ratio:
+`scale = Max(sizes[i] / in_size[d])`
+`out_size[d] = round_int(scale * in_size[d])`
+ +For non-resizable axes (those not specified in `axes`), the output size will be equal to the input size. + +Note: `round_int` stands for computing the nearest integer value, rounding halfway cases up.
+
mode : string (default is nearest)
+
Three interpolation modes: "nearest" (default), "linear" and "cubic". The "linear" mode includes linear interpolation for 1D tensor and N-linear interpolation for N-D tensor (for example, bilinear interpolation for 2D tensor). The "cubic" mode includes cubic interpolation for 1D tensor and N-cubic interpolation for N-D tensor (for example, bicubic interpolation for 2D tensor).
+
nearest_mode : string (default is round_prefer_floor)
+
Four modes: "round_prefer_floor" (default, as known as round half down), "round_prefer_ceil" (as known as round half up), "floor", "ceil". Only used by nearest interpolation. It indicates how to get "nearest" pixel in input tensor from x_original, so this attribute is valid only if "mode" is "nearest".
+
+ +#### Inputs (1 - 4) + +
+
X (differentiable) : T1
+
N-D tensor
+
roi (optional, non-differentiable) : T2
+
1-D tensor given as [start1, ..., startN, end1, ..., endN], where N is the rank of X or the length of axes, if provided. The RoIs' coordinates are normalized in the coordinate system of the input image. It only takes effect when coordinate_transformation_mode is "tf_crop_and_resize"
+
scales (optional, non-differentiable) : tensor(float)
+
The scale array along each dimension. It takes value greater than 0. If it's less than 1, it's sampling down, otherwise, it's upsampling. The number of elements of 'scales' should be the same as the rank of input 'X' or the length of 'axes', if provided. One of 'scales' and 'sizes' MUST be specified and it is an error if both are specified. If 'sizes' is needed, the user can use an empty string as the name of 'scales' in this operator's input list.
+
sizes (optional, non-differentiable) : tensor(int64)
+
Target size of the output tensor. Its interpretation depends on the 'keep_aspect_ratio_policy' value.The number of elements of 'sizes' should be the same as the rank of input 'X', or the length of 'axes', if provided. Only one of 'scales' and 'sizes' can be specified.
+
+ +#### Outputs + +
+
Y (differentiable) : T1
+
N-D tensor after resizing
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input 'X' and output 'Y' to all tensor types.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Constrain roi type to float or double.
+
+ +### **ScatterElements-18** + + ScatterElements takes three inputs `data`, `updates`, and `indices` of the same + rank r >= 1 and an optional attribute axis that identifies an axis of `data` + (by default, the outer-most axis, that is axis 0). The output of the operation + is produced by creating a copy of the input `data`, and then updating its value + to values specified by `updates` at specific index positions specified by + `indices`. Its output shape is the same as the shape of `data`. + + For each entry in `updates`, the target index in `data` is obtained by combining + the corresponding entry in `indices` with the index of the entry itself: the + index-value for dimension = axis is obtained from the value of the corresponding + entry in `indices` and the index-value for dimension != axis is obtained from the + index of the entry itself. + + `reduction` allows specification of an optional reduction operation, which is applied to all values in `updates` + tensor into `output` at the specified `indices`. + In cases where `reduction` is set to "none", indices should not have duplicate entries: that is, if idx1 != idx2, + then indices[idx1] != indices[idx2]. For instance, in a 2-D tensor case, the update + corresponding to the [i][j] entry is performed as below: + ``` + output[indices[i][j]][j] = updates[i][j] if axis = 0, + output[i][indices[i][j]] = updates[i][j] if axis = 1, + ``` + When `reduction` is set to some reduction function `f`, the update corresponding to the [i][j] entry is performed as below: + ``` + output[indices[i][j]][j] = f(output[indices[i][j]][j], updates[i][j]) if axis = 0, + output[i][indices[i][j]] = f(output[i][indices[i][j]], updates[i][j]) if axis = 1, + ``` + where the `f` is `+`, `*`, `max` or `min` as specified. + + This operator is the inverse of GatherElements. It is similar to Torch's Scatter operation. + + (Opset 18 change): Adds max/min to the set of allowed reduction ops. + + Example 1: + ``` + data = [ + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + ] + indices = [ + [1, 0, 2], + [0, 2, 1], + ] + updates = [ + [1.0, 1.1, 1.2], + [2.0, 2.1, 2.2], + ] + output = [ + [2.0, 1.1, 0.0] + [1.0, 0.0, 2.2] + [0.0, 2.1, 1.2] + ] + ``` + Example 2: + ``` + data = [[1.0, 2.0, 3.0, 4.0, 5.0]] + indices = [[1, 3]] + updates = [[1.1, 2.1]] + axis = 1 + output = [[1.0, 1.1, 3.0, 2.1, 5.0]] + ``` + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
reduction : string (default is none)
+
Type of reduction to apply: none (default), add, mul, max, min. 'none': no reduction applied. 'add': reduction using the addition operation. 'mul': reduction using the multiplication operation.'max': reduction using the maximum operation.'min': reduction using the minimum operation.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : Tind
+
Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
updates (differentiable) : T
+
Tensor of rank r >=1 (same rank and shape as indices)
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank r >= 1 (same rank as input).
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input and output types can be of any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **ScatterND-18** + + ScatterND takes three inputs `data` tensor of rank r >= 1, `indices` tensor of rank q >= 1, + and `updates` tensor of rank q + r - indices.shape[-1] - 1. The output of the operation + is produced by creating a copy of the input `data`, and then updating its value to values + specified by `updates` at specific index positions specified by `indices`. Its output shape + is the same as the shape of `data`. + + `indices` is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of `indices`. + `indices` is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into `data`. + Hence, k can be a value at most the rank of `data`. When k equals rank(data), each update entry specifies an + update to a single element of the tensor. When k is less than rank(data) each update entry specifies an + update to a slice of the tensor. Index values are allowed to be negative, as per the usual + convention for counting backwards from the end, but are expected in the valid range. + + `updates` is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the + first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. + The remaining dimensions of `updates` correspond to the dimensions of the + replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, + corresponding to the trailing (r-k) dimensions of `data`. Thus, the shape of `updates` + must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation + of shapes. + + The `output` is calculated via the following equation: + + ``` + output = np.copy(data) + update_indices = indices.shape[:-1] + for idx in np.ndindex(update_indices): + output[tuple(indices[idx])] = updates[idx] + ``` + + The order of iteration in the above loop is not specified. + In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. + This ensures that the output value does not depend on the iteration order. + + `reduction` allows specification of an optional reduction operation, which is applied to all values in `updates` + tensor into `output` at the specified `indices`. + In cases where `reduction` is set to "none", indices should not have duplicate entries: that is, if idx1 != idx2, + then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order. + When `reduction` is set to some reduction function `f`, `output` is calculated as follows: + + ``` + output = np.copy(data) + update_indices = indices.shape[:-1] + for idx in np.ndindex(update_indices): + output[tuple(indices[idx])] = f(output[tuple(indices[idx])], updates[idx]) + ``` + + where the `f` is `+`, `*`, `max` or `min` as specified. + + This operator is the inverse of GatherND. + + (Opset 18 change): Adds max/min to the set of allowed reduction ops. + + Example 1: + ``` + data = [1, 2, 3, 4, 5, 6, 7, 8] + indices = [[4], [3], [1], [7]] + updates = [9, 10, 11, 12] + output = [1, 11, 3, 10, 9, 6, 7, 12] + ``` + + Example 2: + ``` + data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] + indices = [[0], [2]] + updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]] + output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] + ``` + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
reduction : string (default is none)
+
Type of reduction to apply: none (default), add, mul, max, min. 'none': no reduction applied. 'add': reduction using the addition operation. 'mul': reduction using the addition operation. 'max': reduction using the maximum operation.'min': reduction using the minimum operation.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : tensor(int64)
+
Tensor of rank q >= 1.
+
updates (differentiable) : T
+
Tensor of rank q + r - indices_shape[-1] - 1.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank r >= 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
+ +### **Split-18** + + Split a tensor into a list of tensors, along the specified 'axis'. + Either input 'split' or the attribute 'num_outputs' should be specified, but not both. + If the attribute 'num_outputs' is specified, then the tensor is split into equal sized parts. + If the tensor is not evenly splittable into `num_outputs`, the last chunk will be smaller. + If the input 'split' is specified, it indicates the sizes of each output in the split. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1] where r = rank(input).
+
num_outputs : int
+
Number of outputs to split parts of the tensor into. If the tensor is not evenly splittable the last chunk will be smaller.
+
+ +#### Inputs (1 - 2) + +
+
input (differentiable) : T
+
The tensor to split
+
split (optional, non-differentiable) : tensor(int64)
+
Optional length of each output. Values should be >= 0.Sum of the values must be equal to the dim value at 'axis' specified.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, differentiable) : T
+
One or more outputs forming list of tensors after splitting
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ +## Version 19 of the default ONNX operator set +### **AveragePool-19** + + AveragePool consumes an input tensor X and applies average pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + average pooling consisting of computing the average on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape is calculated differently + depending on whether explicit padding is used, where pads is employed, or auto padding is used, where auto_pad is utilized. + With explicit padding (https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html?highlight=maxpool#torch.nn.MaxPool2d): + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled. `pad_shape[i]` is the sum of pads along axis `i`. + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following when ceil_mode is enabled: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + or when ceil_mode is disabled (https://www.tensorflow.org/api_docs/python/tf/keras/layers/AveragePooling2D): + ``` + VALID: output_spatial_shape[i] = floor((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i]) + 1 + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = floor((input_spatial_shape[i] - 1) / strides_spatial_shape[i]) + 1 + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i] + ``` + The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero). + + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
count_include_pad : int (default is 0)
+
Whether include pad pixels when calculating values for the edges. Default is 0, doesn't count include pad.
+
dilations : list of ints
+
Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Cast-19** + + The operator casts the elements of a given input tensor to a data type + specified by the 'to' argument and returns an output tensor of the same size in + the converted type. The 'to' argument must be one of the data types specified + in the 'DataType' enum field in the TensorProto message. + + Casting from string tensor in plain (e.g., "3.14" and "1000") and scientific numeric representations + (e.g., "1e-5" and "1E8") to float types is supported. For example, converting string "100.5" to an integer may + yield result 100. There are some string literals reserved for special floating-point values; + "+INF" (and "INF"), "-INF", and "NaN" are positive infinity, negative infinity, and not-a-number, respectively. + Any string which can exactly match "+INF" in a case-insensitive way would be mapped to positive infinite. Similarly, + this case-insensitive rule is applied to "INF" and "NaN". When casting from numeric tensors + to string tensors, plain floating-point representation (such as "314.15926") would be used. + Converting non-numerical-literal string such as "Hello World!" is an undefined behavior. Cases + of converting string representing floating-point arithmetic value, such as "2.718", to INT is an undefined behavior. + + Conversion from a numerical type to any numerical type is always allowed. + User must be aware of precision loss and value change caused by range difference between two types. + For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting + an integer 36 to Boolean may produce 1 because we truncate bits which can't be stored in the targeted type. + + In more detail, the conversion among numerical types should follow these rules + if the destination type is not a float 8 type. + + * Casting from floating point to: + * floating point: +/- infinity if OOR (out of range). + * fixed point: undefined if OOR. + * bool: +/- 0.0 to False; all else to True. + * Casting from fixed point to: + * floating point: +/- infinity if OOR. (+ infinity in the case of uint) + * fixed point: when OOR, discard higher bits and reinterpret (with respect to two's complement representation for + signed types). For example, 200 (int16) -> -56 (int8). + * bool: zero to False; nonzero to True. + * Casting from bool to: + * floating point: `{1.0, 0.0}`. + * fixed point: `{1, 0}`. + * bool: no change. + + Float 8 type were introduced to speed up the training of + deep models. By default the conversion of a float *x* obeys + to the following rules. `[x]` means the value rounded to + the target mantissa width. + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | -------- | -------- | -------- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | Inf | FLT_MAX | NaN | FLT_MAX | NaN | + | -Inf | -FLT_MAX | NaN | -FLT_MAX | NaN | + | \[x\] > FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | + | \[x\] \< -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | + | else | RNE | RNE | RNE | RNE | + + The behavior changes if the parameter 'saturate' is set to False. + The rules then become: + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | ------ | -------- | ---- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | -NaN | -NaN | NaN | -NaN | NaN | + | Inf | NaN | NaN | Inf | NaN | + | -Inf | -NaN | NaN | -Inf | NaN | + | \[x\] > FLT_MAX | NaN | NaN | Inf | NaN | + | \[x\] \< -FLT_MAX | NaN | NaN | -Inf | NaN | + | else | RNE | RNE | RNE | RNE | + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. All cases are fully described in two tables inserted in the operator description.
+
to : int (required)
+
The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor with the same shape as input with type specified by the 'to' argument
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **CastLike-19** + + The operator casts the elements of a given input tensor (the first input) to + the same data type as the elements of the second input tensor. + See documentation of the Cast operator for further details. + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. Please refer to operator Cast description for further details.
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
target_type (non-differentiable) : T2
+
The (first) input tensor will be cast to produce a tensor of the same type as this (second input) tensor.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor produced by casting the first input tensor to have the same type as the second input tensor.
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **Constant-19** + + This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, + or value_* must be specified. + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
sparse_value : sparse_tensor
+
The value for the elements of the output tensor in sparse format.
+
value : tensor
+
The value for the elements of the output tensor.
+
value_float : float
+
The value for the sole element for the scalar, float32, output tensor.
+
value_floats : list of floats
+
The values for the elements for the 1D, float32, output tensor.
+
value_int : int
+
The value for the sole element for the scalar, int64, output tensor.
+
value_ints : list of ints
+
The values for the elements for the 1D, int64, output tensor.
+
value_string : string
+
The value for the sole element for the scalar, UTF-8 string, output tensor.
+
value_strings : list of strings
+
The values for the elements for the 1D, UTF-8 string, output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input and output types to all tensor types.
+
+ +### **DeformConv-19** + + Performs deformable convolution as described in https://arxiv.org/abs/1703.06211 and https://arxiv.org/abs/1811.11168. + This operator specification supports the general N-D case. Note that most common use cases have 2D or 3D data. + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
dilations : list of ints
+
Dilation value along each spatial axis of the kernel. Default is 1 along each axis.
+
group : int (default is 1)
+
Number of groups the input and output channels, C and oC, are divided into. C and oC must both be divisible by group. Default is 1.
+
kernel_shape : list of ints
+
Shape of the convolution kernel. If not present, it is inferred from the shape of input W.
+
offset_group : int (default is 1)
+
Number of groups of offset. C must be divisible by offset_group. Default is 1.
+
pads : list of ints
+
Padding for the beginning and end along each spatial axis. The values represent the number of pixels added to the beginning and end of the corresponding axis and can take any nonnegative value. The format should be as follows: [x1_begin, x2_begin, ..., x1_end, x2_end, ...], where xi_begin is the number of pixels added at the beginning of axis `i` and xi_end is the number of pixels added at the end of axis `i`. Default is 0 along each axis.
+
strides : list of ints
+
Stride along each spatial axis. Default is 1 along each axis.
+
+ +#### Inputs (3 - 5) + +
+
X : T
+
Input data tensor. For 2D image data, it has shape (N, C, H, W) where N is the batch size, C is the number of input channels, and H and W are the height and width. In general, the shape is (N, C, D1, D2, ... , Dn) for n-dimensional data, where D1 to Dn are the spatial dimension sizes. Most common use cases have n = 2 or 3.
+
W : T
+
Weight tensor that will be used in the convolutions. It has shape (oC, C/group, kH, kW), where oC is the number of output channels and kH and kW are the kernel height and width. For more than 2 dimensions, it has shape (oC, C/group, k1, k2, ... , kn).
+
offset : T
+
Offset tensor denoting the offset for the sampling locations in the convolution kernel. It has shape (N, offset_group * kH * kW * 2, oH, oW) for 2D data or (N, offset_group * k1 * k2 * ... * kn * n, o1, o2, ... , on) for nD data. Use linear interpolationfor fractional offset values. Sampling locations outside of the padded input tensor gives zero.
+
B (optional) : T
+
Optional 1D bias of length oC to be added to the convolution. Default is a tensor of zeros.
+
mask (optional) : T
+
The mask tensor to be applied to each position in the convolution kernel. It has shape (N, offset_group * kH * kW, oH, oW) for 2D data or (N, offset_group * k1 * k2 * ... * kn * n, o1, o2, ... , on) for nD data. Default is a tensor of ones.
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor that contains the result of convolution. It has shape (N, oC, oH, oW) for 2D data or (N, oC, o1, o2, ..., on) for nD data
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **DequantizeLinear-19** + + The linear dequantization operator. It consumes a quantized tensor, a scale, and a zero point to compute the full precision tensor. + The dequantization formula is `y = (x - x_zero_point) * x_scale`. `x_scale` and `x_zero_point` must have same shape, and can be either a scalar + for per-tensor / per layer quantization, or a 1-D tensor for per-axis quantization. + `x_zero_point` and `x` must have same type. `x` and `y` must have same shape. In the case of dequantizing int32, + there's no zero point (zero point is supposed to be 0). + `zero-point` is usually not used in the case of float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz quantization, + but the dequantization formula remains the same for consistency and 'x_scale' still determines the output type. + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used only for per-axis quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`. When the rank of the input is 1, per-tensor quantization is applied, rendering the axis unnecessary in this scenario.
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D quantized input tensor to be de-quantized.
+
x_scale : T2
+
Scale for input 'x'. It can be a scalar, which means a per-tensor/layer dequantization, or a 1-D tensor for per-axis dequantization.
+
x_zero_point (optional) : T1
+
Zero point for input 'x'. Shape must match x_scale. It's optional. Zero point is 0 when it's not specified.
+
+ +#### Outputs + +
+
y : T2
+
N-D full precision output tensor. It has same shape as input 'x'.
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8), tensor(int32), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain 'x_zero_point' and 'x' to 8-bit integer or float, or /32-bit integer tensor.
+
T2 : tensor(float), tensor(float16), tensor(bfloat16)
+
'x_scale' determines the output type.
+
+ +### **Equal-19** + + Returns the tensor resulted from performing the `equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(string)
+
Constrain input types to all (non-complex) tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ +### **Identity-19** + + Identity operator + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : V
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : V
+
Tensor to copy input into.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain input and output types to all tensor, sequence, and optional types.
+
+ +### **If-19** + + If conditional + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
else_branch : graph (required)
+
Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
+
then_branch : graph (required)
+
Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
+
+ +#### Inputs + +
+
cond : B
+
Condition for the if. The tensor must contain a single element.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : V
+
Values that are live-out to the enclosing scope. The return values in the `then_branch` and `else_branch` must be of the same data type. The `then_branch` and `else_branch` may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if-node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of `then_branch` is typed float tensor with shape [2] and the first output of `else_branch` is another float tensor with shape [3], If's first output should have (a) no shape set, or (b) a shape of rank 1 with neither `dim_value` nor `dim_param` set, or (c) a shape of rank 1 with a unique `dim_param`. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv9.
+
B : tensor(bool)
+
Only bool
+
+ +### **Loop-19** + + Generic Looping construct. This loop has multiple termination conditions: + + 1) Trip count. Iteration count specified at runtime. Set by + specifying the input M. Optional. Set to empty string to omit. + Note that a static trip count (specified at graph construction time) can be + specified by passing in a constant node for input M. + 2) Loop termination condition. This is an input to the op that determines + whether to run the first iteration and also a loop-carried dependency for + the body graph. The body graph must yield a value for the condition variable, + whether this input is provided or not. + + This table summarizes the operating modes of this operator with equivalent + C-style code: + + Operator inputs defined as (max_trip_count, condition_var). + + * input ("", ""): + for (int i=0; ; ++i) { + cond = ... // Note this value is ignored, but is required in the body + } + + * input ("", cond) // Note this is analogous to a while loop + bool cond = ...; + for (int i=0; cond; ++i) { + cond = ...; + } + + * input ("", 1) // Note this is analogous to a do-while loop + bool cond = true + for (int i=0; cond; ++i) { + cond = ...; + } + + * input (trip_count, "") // Note this is analogous to a for loop + int trip_count = ... + for (int i=0; i < trip_count; ++i) { + cond = ...; // ignored + } + + * input (trip_count, cond) + int trip_count = ...; + bool cond = ...; + for (int i=0; i < trip_count && cond; ++i) { + cond = ...; + } + + + *Sample usage - cond as well as trip count* + + graph predict-net { + %a = Constant[value = ]() + %b = Constant[value = ]() + %keepgoing = Constant[value = ]() + %max_trip_count = Constant[value = ]() + %keepgoing_out, %b_out, %user_defined_vals = Loop[body = ](%max_trip_count, %keepgoing, %b) + return + } + + graph body-net ( + %i[INT32, scalar] // iteration number + %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used + %b_in[INT32, scalar] // incoming value of loop-carried-dependency b + ) { + %my_local = Add(%a, %b_in) + %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b + %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition + %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated + return %keepgoing_out, %b_out, %user_defined_val + } + + *Sample equivalent C code* + + { + /* User-defined code (enclosing scope) */ + int a = 3, b = 6; + bool keepgoing = true; // Analogous to input cond + /* End user-defined code */ + + /* Implicitly-defined code */ + const int max_trip_count = 10; // Analogous to input M + int user_defined_vals[]; // Imagine this is resizable + /* End implicitly-defined code */ + /* initialize loop-carried variables and scan-output variables */ + bool keepgoing_out = keepgoing + int b_out = b + + for (int i=0; i < max_trip_count && keepgoing_out; ++i) { + /* Implicitly-defined code: bind actual parameter values + to formal parameter variables of loop-body */ + bool keepgoing_in = keepgoing_out; + bool b_in = b_out; + + /* User-defined code (loop body) */ + int my_local = a + b_in; // Reading value "a" from the enclosing scope is fine + b_out = a - b_in; + keepgoing_out = my_local > b_out; + user_defined_val = b_in + b_in; // b_in and b_out are different variables + /* End user-defined code */ + + /* Implicitly defined-code */ + user_defined_vals[i] = user_defined_val // accumulate scan-output values + } + // int t = my_local; // Can't do this. my_local is not accessible here. + + // The values below are bound to the output variables of the loop and therefore accessible + // b_out; user_defined_vals; keepgoing_out; + } + + There are several things of note in this code snippet: + + 1) Values from the enclosing scope (i.e. variable "a" here) are in scope and can + be referenced in the inputs of the loop. + 2) Any values computed in the loop body that needs to be used in a subsequent + iteration or after the loop are modeled using a pair of variables in the loop-body, + consisting of an input variable (eg., b_in) and an output variable (eg., b_out). + These are referred to as loop-carried dependences. The loop operation node + supplies the input value of the input variable for the first iteration, and + returns the output value of the output variable produced by the final + iteration. + 3) Scan_output variables are used to implicitly concatenate values computed across + all the iterations. In the above example, the value of user_defined_val computed + over all iterations are concatenated and returned as the value of user_defined_vals + after the loop. + 4) Values created in the body cannot be accessed in the enclosing scope, + except using the mechanism described above. + + Note that the semantics of this op support "diagonal" or "wavefront" execution. + (See Step 3 here for an example: + https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). + Frontends should emit multi-layer RNNs as a series of While operators (with + time being the inner looping dimension), with each successive layer consuming + the scan_outputs from the previous layer, possibly going through several + point-wise operators (e.g. dropout, residual connections, linear layer). + + The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order. + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies...). It has 1+N+K outputs: (condition, loop carried dependencies..., scan_outputs...). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
+
+ +#### Inputs (2 - ∞) + +
+
M (optional) : I
+
A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
+
cond (optional) : B
+
A boolean termination condition. Optional. Pass empty string to skip.
+
v_initial (variadic, heterogeneous) : V
+
The initial values of any loop-carried dependencies (values that change across loop iterations)
+
+ +#### Outputs (1 - ∞) + +
+
v_final_and_scan_outputs (variadic, heterogeneous) : V
+
Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv9.
+
I : tensor(int64)
+
tensor of int64, which should be a scalar.
+
B : tensor(bool)
+
tensor of bool, which should be a scalar.
+
+ +### **Pad-19** + + Given a tensor containing the data to be padded (`data`), a tensor containing the number of start and end pad values for axis (`pads`), (optionally) a `mode`, and (optionally) `constant_value`, + a padded tensor (`output`) is generated. + + The four supported `modes` are (similar to corresponding modes supported by `numpy.pad`): + + 1) `constant`(default) - pads with a given constant value as specified by `constant_value` (which defaults to 0, empty string, or False) + + 2) `reflect` - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis + + 3) `edge` - pads with the edge values of array + + 4) `wrap` - wrap-around padding as if the data tensor forms a torus + + + Example 1 (`constant` mode): + + Insert 0 pads to the beginning of the second dimension. + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'constant' + + constant_value = 0.0 + + output = [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ] + ``` + + Example 2 (`reflect` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'reflect' + + output = [ + [1.0, 1.2, 1.0, 1.2], + [2.3, 3.4, 2.3, 3.4], + [4.5, 5.7, 4.5, 5.7], + ] + ``` + + Example 3 (`edge` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'edge' + + output = [ + [1.0, 1.0, 1.0, 1.2], + [2.3, 2.3, 2.3, 3.4], + [4.5, 4.5, 4.5, 5.7], + ] + ``` + + Example 4 (`wrap` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [2, 1, 1, 1] + + mode = 'wrap' + + output = [ + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + ] + ``` + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is constant)
+
Supported modes: `constant`(default), `reflect`, `edge`, `wrap`
+
+ +#### Inputs (2 - 4) + +
+
data (differentiable) : T
+
Input tensor.
+
pads (non-differentiable) : tensor(int64)
+
Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. `pads` should be a 1D tensor of shape [2 * num_axes] where `num_axes` refers to the number of elements in the `axes` input or the input rank if `axes` are not provided explicitly. `pads` format should be: [x1_begin, x2_begin, ..., x1_end, x2_end,...], where xi_begin is the number of pad values added at the beginning of axis `axes[i]` and xi_end, the number of pad values added at the end of axis `axes[i]`.
+
constant_value (optional, non-differentiable) : T
+
(Optional) A scalar value to be used if the mode chosen is `constant` (by default it is 0, empty string or False).
+
axes (optional, non-differentiable) : Tind
+
1-D tensor of axes that `pads` apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated. If not provided, all axes are assumed (`[0, 1, ..., input_rank-1]`).
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **QuantizeLinear-19** + + The linear quantization operator. It consumes a high precision tensor, a scale, and a zero point to compute the low precision / quantized tensor. + The scale factor and zero point must have same shape, and can be either a scalar for per-tensor / per layer quantization, or a 1-D tensor for per-axis quantization. + The quantization formula is `y = saturate ((x / y_scale) + y_zero_point)`. + For saturation, it saturates to [0, 255] if it's uint8, or [-128, 127] if it's int8. + For (x / y_scale), it's rounding to the nearest even. Refer to https://en.wikipedia.org/wiki/Rounding for details. + 'y_zero_point' and 'y' must have same type. + 'y_zero_point' is usually not used for quantization to float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz, + but the quantization formula remains the same for consistency and + the type of the attribute 'y_zero_point' still determines the quantization type. + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the quantization dimension of the input tensor. Ignored for per-tensor quantization. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 quantization (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. All cases are fully described in two tables inserted in the operator description.
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D full precision Input tensor to be quantized.
+
y_scale : T1
+
Scale for doing quantization to get 'y'. It can be a scalar, which means per-tensor/layer quantization, or a 1-D Tensor for per-axis quantization.
+
y_zero_point (optional) : T2
+
Zero point for doing quantization to get 'y'. Shape must match y_scale. Default is uint8 with zero point of 0 if it's not specified.
+
+ +#### Outputs + +
+
y : T2
+
N-D quantized output tensor. It has same shape as input 'x'.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(float16), tensor(bfloat16), tensor(int32)
+
Constrain 'x' to float, float16, bfloat16 or int32 tensor.
+
T2 : tensor(int8), tensor(uint8), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain 'y_zero_point' and 'y' to 8-bit integer/float tensor.
+
+ +### **Reshape-19** + + Reshape the input tensor similar to numpy.reshape. + First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. + At most one dimension of the new shape can be -1. In this case, the value is + inferred from the size of the tensor and the remaining dimensions. A dimension + could also be 0, in which case the actual dimension value is unchanged (i.e. taken + from the input tensor). If 'allowzero' is set, and the new shape includes 0, the + dimension will be set explicitly to zero (i.e. not taken from input tensor). + Shape (second input) could be an empty shape, which means converting to a scalar. + The input tensor's shape and the output tensor's shape are required to have the same number of elements. + + If the attribute 'allowzero' is set, it is invalid for the specified shape to + contain both a zero value and -1, as the value of the dimension corresponding + to -1 cannot be determined uniquely. + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
allowzero : int (default is 0)
+
(Optional) By default, when any value in the 'shape' input is equal to zero the corresponding dimension value is copied from the input tensor dynamically. allowzero=1 indicates that if any value in the 'shape' input is set to zero, the zero value is honored, similar to NumPy.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
shape (non-differentiable) : tensor(int64)
+
Specified shape for output.
+
+ +#### Outputs + +
+
reshaped (differentiable) : T
+
Reshaped data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input and output types to all tensor types.
+
+ +### **Resize-19** + + Resize the input tensor. In general, it calculates every value in the output tensor as a weighted average of neighborhood (a.k.a. sampling locations) in the input tensor. + Each dimension value of the output tensor is: + ``` + output_dimension = floor(input_dimension * (roi_end - roi_start) * scale) + ``` + if input \"sizes\" is not specified. + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
antialias : int (default is 0)
+
If set to 1, "linear" and "cubic" interpolation modes will use an antialiasing filter when downscaling. Antialiasing is achieved by stretching the resampling filter by a factor max(1, 1 / scale), which means that when downsampling, more input pixels contribute to an output pixel.
+
axes : list of ints
+
If provided, it specifies a subset of axes that 'roi', 'scales' and 'sizes' refer to. If not provided, all axes are assumed [0, 1, ..., r-1], where r = rank(data). Non-specified dimensions are interpreted as non-resizable. Negative value means counting dimensions from the back. Accepted range is [-r, r-1], where r = rank(data). Behavior is undefined if an axis is repeated.
+
coordinate_transformation_mode : string (default is half_pixel)
+
+This attribute describes how to transform the coordinate in the resized tensor to the coordinate in the original tensor. + +The coordinate of each dimension is transformed individually. Let's describe a case using axis x as an example. +Denote `x_resized` as the coordinate of axis x in the resized tensor, + `x_original` as the coordinate of axis x in the original tensor, + `length_original` as the length of the original tensor in axis x, + `length_resized` as the length of the resized tensor in axis x, + `scale = length_resized / length_original`, + `output_width` the target length on the axis x which can be a fractional number when it is calculated out of a scale factor, + and `output_width_int` the effective output width as an integer. + +if coordinate_transformation_mode is `"half_pixel"`, +``` +x_original = (x_resized + 0.5) / scale - 0.5 +``` + +if coordinate_transformation_mode is `"half_pixel_symmetric"`, +``` +adjustment = output_width_int / output_width +center = input_width / 2 +offset = center * (1 - adjustment) +x_ori = offset + (x + 0.5) / scale - 0.5 +``` + +if coordinate_transformation_mode is `"pytorch_half_pixel"`, +``` +x_original = length_resized > 1 ? (x_resized + 0.5) / scale - 0.5 : 0 +``` + +if coordinate_transformation_mode is `"align_corners"`, +``` +x_original = x_resized * (length_original - 1) / (length_resized - 1) +``` + +if coordinate_transformation_mode is `"asymmetric"`, +``` +x_original = x_resized / scale +``` + +if coordinate_transformation_mode is `"tf_crop_and_resize"`, +``` +x_original = length_resized > 1 ? start_x * (length_original - 1) + x_resized * (end_x - start_x) * (length_original - 1) / (length_resized - 1) : 0.5 * (start_x + end_x) * (length_original - 1) +``` +.
+
cubic_coeff_a : float (default is -0.75)
+
The coefficient 'a' used in cubic interpolation. Two common choice are -0.5 (in some cases of TensorFlow) and -0.75 (in PyTorch). Check out Equation (4) in https://ieeexplore.ieee.org/document/1163711 for the details. This attribute is valid only if mode is "cubic".
+
exclude_outside : int (default is 0)
+
If set to 1, the weight of sampling locations outside the tensor will be set to 0 and the weight will be renormalized so that their sum is 1.0. The default value is 0.
+
extrapolation_value : float (default is 0.0)
+
When coordinate_transformation_mode is "tf_crop_and_resize" and x_original is outside the range [0, length_original - 1], this value is used as the corresponding output value. Default is 0.0f.
+
keep_aspect_ratio_policy : string (default is stretch)
+
+This attribute describes how to interpret the `sizes` input with regard to keeping the original aspect ratio of the input, and it is not applicable when +the `scales` input is used. + +Given a set of `sizes`, associated with a subset of `axes` (explicitly provided or default), and assuming `d = axes[i]`, with `i` being the index of the provided `sizes`. + +If `keep_aspect_ratio_policy` is `"stretch"`, the original aspect ratio is disregarded, and the input is resized to the specified size: +`out_size[d] = sizes[i]` + +If `keep_aspect_ratio_policy` is `"not_larger"`, the sizes are adjusted so that no extent of the output is larger than the specified size, while keeping the original aspect ratio: +``` +scale = Min(sizes[i] / in_size[d]) +out_size[d] = round_int(scale * in_size[d]) +``` + +If `keep_aspect_ratio_policy` is `"not_smaller"`, the sizes are adjusted so that no extent of the output is smaller than the specified size, while keeping the original aspect ratio: +``` +scale = Max(sizes[i] / in_size[d]) +out_size[d] = round_int(scale * in_size[d]) +``` + +For non-resizable axes (those not specified in `axes`), the output size will be equal to the input size. + +Note: `round_int` stands for computing the nearest integer value, rounding halfway cases up.
+
mode : string (default is nearest)
+
Three interpolation modes: "nearest" (default), "linear" and "cubic". The "linear" mode includes linear interpolation for 1D tensor and N-linear interpolation for N-D tensor (for example, bilinear interpolation for 2D tensor). The "cubic" mode includes cubic interpolation for 1D tensor and N-cubic interpolation for N-D tensor (for example, bicubic interpolation for 2D tensor).
+
nearest_mode : string (default is round_prefer_floor)
+
Four modes: "round_prefer_floor" (default, as known as round half down), "round_prefer_ceil" (as known as round half up), "floor", "ceil". Only used by nearest interpolation. It indicates how to get "nearest" pixel in input tensor from x_original, so this attribute is valid only if "mode" is "nearest".
+
+ +#### Inputs (1 - 4) + +
+
X (differentiable) : T1
+
N-D tensor
+
roi (optional, non-differentiable) : T2
+
1-D tensor given as [start1, ..., startN, end1, ..., endN], where N is the rank of X or the length of axes, if provided. The RoIs' coordinates are normalized in the coordinate system of the input image. It only takes effect when coordinate_transformation_mode is "tf_crop_and_resize"
+
scales (optional, non-differentiable) : tensor(float)
+
The scale array along each dimension. It takes value greater than 0. If it's less than 1, it's sampling down, otherwise, it's upsampling. The number of elements of 'scales' should be the same as the rank of input 'X' or the length of 'axes', if provided. One of 'scales' and 'sizes' MUST be specified and it is an error if both are specified. If 'sizes' is needed, the user can use an empty string as the name of 'scales' in this operator's input list.
+
sizes (optional, non-differentiable) : tensor(int64)
+
Target size of the output tensor. Its interpretation depends on the 'keep_aspect_ratio_policy' value.The number of elements of 'sizes' should be the same as the rank of input 'X', or the length of 'axes', if provided. Only one of 'scales' and 'sizes' can be specified.
+
+ +#### Outputs + +
+
Y (differentiable) : T1
+
N-D tensor after resizing
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input 'X' and output 'Y' to all tensor types.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Constrain roi type to float or double.
+
+ +### **Scan-19** + + Scan can be used to iterate over one or more scan_input tensors, + constructing zero or more scan_output tensors. It combines ideas from general recurrences, + functional programming constructs such as scan, fold, map, and zip, and is intended to enable + generalizations of RNN-like constructs for sequence-to-sequence processing. + Other tensors (referred to as state_variables here) can be used to carry a state + when iterating from one element to another (similar to hidden-state in RNNs, also referred + to as loop-carried dependences in the context of loops). + Many common usages involve a single scan_input tensor (where functionality + similar to scan, fold and map can be obtained). When more than one scan_input is used, + a behavior similar to zip is obtained. + + The attribute body must be a graph, specifying the computation to be performed in + every iteration. It takes as input the current values of the state_variables and + the current iterated element of the scan_inputs. It must return the (updated) values + of the state_variables and zero or more scan_output_element tensors. The values of the + scan_output_element tensors are concatenated over all the iterations to produce the + scan_output values of the scan construct (similar to the concatenated intermediate + hidden-state values of RNN-like constructs). All the output tensors (state_variables as + well as scan_output_element tensors) are required to have the same shape in each iteration + of the loop (a restriction imposed to enable efficient memory allocation). + + Note that the iterated element passed to the body subgraph does not have a sequence + axis. It will have a rank one less than the rank of the corresponding scan_input. + + The scan operation returns the final values of the state_variables as well as the + scan_outputs. + + The optional attribute scan_input_directions specifies the direction (forward or backward) + for each scan input. If this attribute is omitted, all sequences are scanned in the forward + direction. A bidirectional scan may be performed by specifying the same tensor input twice + in the scan_inputs, once with a forward direction, and once with a backward direction. + + The scan_output of the operation is produced by concatenating the scan_output_element + values produced by the body in each iteration. The optional attribute scan_output_directions + specifies the direction in which scan_output is constructed (by appending or prepending the + scan_output_element to scan_output in each iteration) for each scan_output. If this attribute + is omitted, the scan_output_element is appended to the scan_output in each iteration. + + The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. + If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the + batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. + Note that scanning a non-zero axis may be less efficient than scanning axis zero. + + The optional attribute scan_output_axes specifies the axis along which the scan_outputs + are accumulated for each scan_output. For example, if axis 1 is the time axis (to be + scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis + value of 1. + + Note that because of the ONNX restriction that only the last parameter of an operator can + be variadic, the initial-states and scan-inputs are listed together as one input parameter. + Similarly, the final-states and scan-outputs are listed together as one output parameter. + The attribute num_scan_inputs indicates the number M of scan-inputs. + + The behavior of + + Scan < + num_scan_inputs = m, + body = loop-body, + scan_input_axes = [axis_1, ..., axis_m] + > (init_1, ..., init_n, scan_1, ..., scan_m) + + is equivalent to the following pseudo-code: + + // scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i + // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. + sequence_length = scan_1.shape[axis_1]; + + // initialize state-variables + st_1 = init_1; ... st_n = init_n; + // initialize scan-output variables: [] denotes an empty tensor + scan_out_1 = []; ...; scan_out_k = []; + // identify number of iterations: + + // execute loop + for (int t = 0; t < sequence_length; ++t) { + // generate the scan-input elements: the notation T[t] indicates the sub-tensor + // of rank one less than T obtained by indexing T at position t along axis k. + si_1 = scan_1[t]; + ... ; + si_m = scan_m[t]; + // execute loop-body + st_1, ..., st_n, so_1, ..., so_k = loop-body(st_1, ..., st_n, si_1, ..., si_m) + // accumulate the scan-output elements + scan_out_1 = Concat(scan_out_1, so_1); ... ; scan_out_k = Concat(scan_out_k, so_k); + } + + return st_1, ..., st_n, scan_out_1, ..., scan_out_k; + + *Sample usage: Encoding RNN using a Scan* + + The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, + recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can + be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes + %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these + values are computed in the outer graph, they need to be passed in as extra state_variables. + + graph rnn-encoding { + %H_0 = ... + %X = ... + %Y_h, %Y = Scan[body = , num_scan_inputs=1](%H_0, %X) + return %Y, %Y_h + } + + graph rnn-cell-1 ( + %H_tminus1[FLOAT, tensor] + %X_t[FLOAT, tensor] + ) { + %Wi = ... + %Ri = ... + %Wbi = ... + %Rbi = ... + %t1 = X_t * (Wi^T) + %t2 = H_tminus1*(Ri^T) + %t3 = Add(%t1, %t2) + %t4 = Add(%t3, %Wbi) + %t5 = Add(%t4, %Rbi) + %Ht = Tanh(%t5) + %Accumulate = Identity(%Ht) + return %Ht, %Accumulate + } + + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has N+M inputs: (loop state variables..., scan_input_elts...). It has N+K outputs: (loop state variables..., scan_output_elts...). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
+
num_scan_inputs : int (required)
+
An attribute specifying the number of scan_inputs M.
+
scan_input_axes : list of ints
+
An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
scan_input_directions : list of ints
+
An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
+
scan_output_axes : list of ints
+
An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
+
scan_output_directions : list of ints
+
An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
+
+ +#### Inputs (1 - ∞) + +
+
initial_state_and_scan_inputs (variadic, heterogeneous) : V
+
Initial values of the loop's N state variables followed by M scan_inputs
+
+ +#### Outputs (1 - ∞) + +
+
final_state_and_scan_outputs (variadic, heterogeneous) : V
+
Final values of the loop's N state variables followed by K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
All Tensor types up to IRv9.
+
+ +### **Shape-19** + + Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. + Optional attributes start and end can be used to compute a slice of the input tensor's shape. + If start axis is omitted, the slice starts from axis 0. + The end axis, if specified, is exclusive (and the returned value will not include the size of that axis). + If the end axis is omitted, the axes upto the last one will be included. + Negative axes indicate counting back from the last axis. + Note that axes will be clamped to the range [0, r], where r is the + rank of the input tensor if they are out-of-range (after adding r in the case of + negative axis). Thus, specifying any end value > r is equivalent to specifying an end + value of r, and specifying any start value < -r is equivalent to specifying a start + value of 0. If start > end, the result will be an empty shape. + + Examples: + + ``` + Input tensor with shape: [2, 3, 4] + No attributes specified. + Output: [2, 3, 4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: -1 + Output: [4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + end: -1 + Output: [2, 3] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: 1 + end: 2 + Output: [3] + ``` + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Attributes + +
+
end : int
+
(Optional) Ending axis for slicing the shape. Negative value means counting dimensions from the back. If omitted, sizes of all axes upto (including) the last one will be included.
+
start : int (default is 0)
+
(Optional) Starting axis for slicing the shape. Default value is 0.Negative value means counting dimensions from the back.
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
shape (non-differentiable) : T1
+
Shape of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor.
+
+ +### **Size-19** + + Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor. + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
size (non-differentiable) : T1
+
Total number of elements of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor, which should be a scalar though.
+
+ +## Version 20 of the default ONNX operator set +### **AffineGrid-20** + + Generates a 2D or 3D flow field (sampling grid), given a batch of affine matrices theta + (https://pytorch.org/docs/stable/generated/torch.nn.functional.affine_grid.html). + An affine matrix `theta` is applied to a position tensor represented in its homogeneous expression. Here is an example in 3D: + ``` + [r00, r01, r02, t0] [x] [x'] + [r10, r11, r12, t1] * [y] = [y'] + [r20, r21, r22, t2] [z] [z'] + [0, 0, 0, 1 ] [1] [1 ] + ``` + where `(x, y, z)` is the position in the original space, `(x', y', z')` is the position in the output space. + The last row is always `[0, 0, 0, 1]` and is not stored in the affine matrix. Therefore we have `theta` of shape `(N, 2, 3)` for 2D or `(N, 3, 4)` for 3D. + + Input `size` is used to define grid of positions evenly spaced in the original 2D or 3D space, with dimensions ranging from `-1` to `1`. + The output `grid` contains positions in the output space. + + When `align_corners=1`, consider `-1` and `1` to refer to the centers of the corner pixels (mark `v` in illustration). + ``` + v v v v + |-------------------|------------------| + -1 0 1 + ``` + When `align_corners=0`, consider `-1` and `1` to refer to the outer edge of the corner pixels. + ``` + v v v v + |------------------|-------------------| + -1 0 1 + ``` + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
align_corners : int (default is 0)
+
if align_corners=1, consider -1 and 1 to refer to the centers of the corner pixels. if align_corners=0, consider -1 and 1 to refer to the outer edge the corner pixels.
+
+ +#### Inputs + +
+
theta (non-differentiable) : T1
+
input batch of affine matrices with shape (N, 2, 3) for 2D or (N, 3, 4) for 3D
+
size (non-differentiable) : T2
+
the target output image size (N, C, H, W) for 2D or (N, C, D, H, W) for 3D
+
+ +#### Outputs + +
+
grid (differentiable) : T1
+
output tensor of shape (N, H, W, 2) of 2D sample coordinates or (N, D, H, W, 3) of 3D sample coordinates.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain grid types to float tensors.
+
T2 : tensor(int64)
+
Constrain size's type to int64 tensors.
+
+ +### **ConstantOfShape-20** + + Generate a tensor with given value and shape. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
value : tensor
+
(Optional) The value of the output elements.Should be a one-element tensor. If not specified, it defaults to a tensor of value 0 and datatype float32
+
+ +#### Inputs + +
+
input : T1
+
1D tensor. The shape of the expected output tensor. If empty tensor is given, the output would be a scalar. All values must be >= 0.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of shape specified by 'input'.If attribute 'value' is specified, the value and datatype of the output tensor is taken from 'value'.If attribute 'value' is not specified, the value in the output defaults to 0, and the datatype defaults to float32.
+
+ +#### Type Constraints + +
+
T1 : tensor(int64)
+
Constrain input types.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain output types to be numerics.
+
+ +### **DFT-20** + + Computes the discrete Fourier Transform (DFT) of the input. + + Assuming the input has shape `[M, N]`, where `N` is the dimension over which the + DFT is computed and `M` denotes the conceptual "all other dimensions," + the DFT `y[m, k]` of shape `[M, N]` is defined as + + $$y[m, k] = \sum_{n=0}^{N-1} e^{-2 \pi j \frac{k n}{N} } x[m, n] ,$$ + + and the inverse transform is defined as + + $$x[m, n] = \frac{1}{N} \sum_{k=0}^{N-1} e^{2 \pi j \frac{k n}{N} } y[m, k] ,$$ + + where $j$ is the imaginary unit. + + The actual shape of the output is specified in the "output" section. + + Reference: https://docs.scipy.org/doc/scipy/tutorial/fft.html + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
inverse : int (default is 0)
+
Whether to perform the inverse discrete Fourier Transform. Default is 0, which corresponds to `false`.
+
onesided : int (default is 0)
+
If `onesided` is `1`, only values for `k` in `[0, 1, 2, ..., floor(n_fft/2) + 1]` are used or returned because the real-to-complex Fourier transform satisfies the conjugate symmetry, i.e., `X[m, k] = X[m, n_fft-k]*`, where `m` denotes "all other dimensions" DFT was not applied on. When `onesided=1` and `inverse=0` (forward DFT), only real input is supported and a one-sided complex spectrum is returned (RFFT). When `onesided=1` and `inverse=1` (inverse DFT), only complex input is supported and a full real signal is returned (IRFFT). Value can be `0` or `1`. Default is `0`.
+
+ +#### Inputs (1 - 3) + +
+
input (non-differentiable) : T1
+
For real input, the following shape is expected: `[signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][1]`. For complex input, the following shape is expected: `[signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][2]`. The final dimension represents the real and imaginary parts of the value in that order.
+
dft_length (optional, non-differentiable) : T2
+
The length of the signal as a scalar. If greater than the axis dimension, the signal will be zero-padded up to `dft_length`. If less than the axis dimension, only the first `dft_length` values will be used as the signal. If not provided, the default `dft_length = signal_dim_axis`, except for the IRFFT case (`onesided=1`, `inverse=1`), in which case the default dft_length is `2 * (signal_dim_axis - 1)`.
+
axis (optional, non-differentiable) : tensor(int64)
+
The axis as a scalar on which to perform the DFT. Default is `-2` (last signal axis). Negative value means counting dimensions from the back. Accepted range is $[-r, -2] \cup [0, r-2]$ where `r = rank(input)`. The last dimension is for representing complex numbers and thus is an invalid axis.
+
+ +#### Outputs + +
+
output : T1
+
The Fourier Transform of the input vector. For standard DFT (`onesided=0`), the output shape is: `[signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][2]` (complex), with `signal_dim_axis = dft_length`. For RFFT (`onesided=1`, `inverse=0`), the output shape is: `[signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][2]` (one-sided complex), with `signal_dim_axis = floor(dft_length/2) + 1`. For IRFFT (`onesided=1`, `inverse=1`), the output shape is: `[signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][1]` (real), where `signal_dim_axis = dft_length`.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T2 : tensor(int32), tensor(int64)
+
Constrain scalar length types to integers.
+
+ +### **Gelu-20** + + Gelu takes one input data (Tensor) and produces one + output data (Tensor) where the gaussian error linear units function, + $y = 0.5 * x * (1 + erf(x/sqrt(2)))$ is applied to the tensor elementwise. + If the attribute "approximate" is set to "tanh", the function estimation, + $y = 0.5 * x * (1 + Tanh(sqrt(2/\pi) * (x + 0.044715 * x^3)))$ is used and applied + to the tensor elementwise. + + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
approximate : string (default is none)
+
Gelu approximation algorithm: `"tanh"`, `"none"`(default).`"none"`: do not use approximation.`"tanh"`: use tanh approximation.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **GridSample-20** + + Given an input `X` and a flow-field `grid`, computes the output `Y` using `X` values and pixel locations from the `grid`. + For spatial input `X` with shape (N, C, H, W), the `grid` will have shape (N, H_out, W_out, 2), + the output `Y` will have shape (N, C, H_out, W_out). For volumetric input `X` with shape (N, C, D, H, W), + the `grid` will have shape (N, D_out, H_out, W_out, 3), the output `Y` will have shape (N, C, D_out, H_out, W_out). + More generally, for an input `X` of rank r+2 with shape (N, C, d1, d2, ..., dr), + the `grid` will have shape (N, D1_out, D2_out, ..., Dr_out, r), the output `Y` will have shape (N, C, D1_out, D2_out, ..., Dr_out). + + The tensor `X` contains values at centers of square pixels (voxels, etc) locations such as (n, c, d1_in, d2_in, ..., dr_in). + The (n, d1_out, d2_out, ..., dr_out, :) values from the tensor `grid` are the normalized positions for interpolating the values + at the (n, c, d1_out, d2_out, ..., dr_out) locations from the output tensor `Y` using a specified interpolation method (the mode) + and a padding mode (for `grid` positions falling outside the 2-dimensional image). + + For example, the values in `grid[n, h_out, w_out, :]` are size-2 vectors specifying normalized positions in the 2-dimensional space of `X`. + They are used to interpolate output values of `Y[n, c, h_out, w_out]`. + + The GridSample operator is often used in doing grid generator and sampler in the + [Spatial Transformer Networks](https://arxiv.org/abs/1506.02025). + See also in [torch.nn.functional.grid_sample](https://pytorch.org/docs/stable/generated/torch.nn.functional.grid_sample.html). + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
align_corners : int (default is 0)
+
If align_corners=1, the extrema (-1 and 1) are considered as referring to the center points of the input's corner pixels (voxels, etc.). If align_corners=0, they are instead considered as referring to the corner points of the input's corner pixels (voxels, etc.), making the sampling more resolution agnostic.
+
mode : string (default is linear)
+
Three interpolation modes: linear (default), nearest and cubic. The "linear" mode includes linear and N-linear interpolation modes depending on the number of spatial dimensions of the input tensor (i.e. linear for 1 spatial dimension, bilinear for 2 spatial dimensions, etc.). The "cubic" mode also includes N-cubic interpolation modes following the same rules. The "nearest" mode rounds to the nearest even index when the sampling point falls halfway between two indices.
+
padding_mode : string (default is zeros)
+
Support padding modes for outside grid values: `zeros`(default), `border`, `reflection`. zeros: use 0 for out-of-bound grid locations, border: use border values for out-of-bound grid locations, reflection: use values at locations reflected by the border for out-of-bound grid locations. If index 0 represents the margin pixel, the reflected value at index -1 will be the same as the value at index 1. For location far away from the border, it will keep being reflected until becoming in bound. If pixel location x = -3.5 reflects by border -1 and becomes x' = 1.5, then reflects by border 1 and becomes x'' = 0.5.
+
+ +#### Inputs + +
+
X (differentiable) : T1
+
Input tensor of rank r+2 that has shape (N, C, D1, D2, ..., Dr), where N is the batch size, C is the number of channels, D1, D2, ..., Dr are the spatial dimensions.
+
grid (non-differentiable) : T2
+
Input offset of shape (N, D1_out, D2_out, ..., Dr_out, r), where D1_out, D2_out, ..., Dr_out are the spatial dimensions of the grid and output, and r is the number of spatial dimensions. Grid specifies the sampling locations normalized by the input spatial dimensions. Therefore, it should have most values in the range of [-1, 1]. If the grid has values outside the range of [-1, 1], the corresponding outputs will be handled as defined by padding_mode. Following computer vision convention, the coordinates in the length-r location vector are listed from the innermost tensor dimension to the outermost, the opposite of regular tensor indexing.
+
+ +#### Outputs + +
+
Y (differentiable) : T1
+
Output tensor of rank r+2 that has shape (N, C, D1_out, D2_out, ..., Dr_out) of the sampled values. For integer input types, intermediate values are computed as floating point and cast to integer at the end.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input `X` and output `Y` types to all tensor types.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Constrain grid types to float tensors.
+
+ +### **ImageDecoder-20** + + Loads and decodes and image from a file. If it can't decode for any reason (e.g. corrupted encoded + stream, invalid format, it will return an empty matrix). + The following image formats are supported: + * BMP + * JPEG (note: Lossless JPEG support is optional) + * JPEG2000 + * TIFF + * PNG + * WebP + * Portable image format (PBM, PGM, PPM, PXM, PNM) + Decoded images follow a channel-last layout: (Height, Width, Channels). + **JPEG chroma upsampling method:** + When upsampling the chroma components by a factor of 2, the pixels are linearly interpolated so that the + centers of the output pixels are 1/4 and 3/4 of the way between input pixel centers. + When rounding, 0.5 is rounded down and up at alternative pixels locations to prevent bias towards + larger values (ordered dither pattern). + Considering adjacent input pixels A, B, and C, B is upsampled to pixels B0 and B1 so that + ``` + B0 = round_half_down((1/4) * A + (3/4) * B) + B1 = round_half_up((3/4) * B + (1/4) * C) + ``` + This method, is the default chroma upsampling method in the well-established libjpeg-turbo library, + also referred as "smooth" or "fancy" upsampling. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
pixel_format : string (default is RGB)
+
Pixel format. Can be one of "RGB", "BGR", or "Grayscale".
+
+ +#### Inputs + +
+
encoded_stream (non-differentiable) : T1
+
Encoded stream
+
+ +#### Outputs + +
+
image (non-differentiable) : T2
+
Decoded image
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8)
+
Constrain input types to 8-bit unsigned integer tensor.
+
T2 : tensor(uint8)
+
Constrain output types to 8-bit unsigned integer tensor.
+
+ +### **IsInf-20** + + Map infinity to true and other values to false. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
detect_negative : int (default is 1)
+
(Optional) Whether map negative infinity to true. Default to 1 so that negative infinity induces true. Set this attribute to 0 if negative infinity should be mapped to false.
+
detect_positive : int (default is 1)
+
(Optional) Whether map positive infinity to true. Default to 1 so that positive infinity induces true. Set this attribute to 0 if positive infinity should be mapped to false.
+
+ +#### Inputs + +
+
X (non-differentiable) : T1
+
input
+
+ +#### Outputs + +
+
Y (non-differentiable) : T2
+
output
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input types to float tensors.
+
T2 : tensor(bool)
+
Constrain output types to boolean tensors.
+
+ +### **IsNaN-20** + + Returns which elements of the input are NaN. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T1
+
input
+
+ +#### Outputs + +
+
Y (non-differentiable) : T2
+
output
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input types to float tensors.
+
T2 : tensor(bool)
+
Constrain output types to boolean tensors.
+
+ +### **ReduceMax-20** + + Computes the max of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or the minimum value of the data type otherwise. + + + If the input data type is Boolean, the comparison should consider `False < True`. + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(uint8), tensor(int8), tensor(bool)
+
Constrain input and output types to numeric and Boolean tensors.
+
+ +### **ReduceMin-20** + + Computes the min of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields plus infinity (if supported by the datatype) or the maximum value of the data type otherwise. + + + If the input data type is Boolean, the comparison should consider `False < True`. + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(uint8), tensor(int8), tensor(bool)
+
Constrain input and output types to numeric and Boolean tensors.
+
+ +### **RegexFullMatch-20** + + RegexFullMatch performs a full regex match on each element of the input tensor. If an element fully matches the regex pattern specified as an attribute, the corresponding element in the output is True and it is False otherwise. [RE2](https://github.com/google/re2/wiki/Syntax) regex syntax is used. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
pattern : string
+
Regex pattern to match on. This must be valid RE2 syntax.
+
+ +#### Inputs + +
+
X (non-differentiable) : T1
+
Tensor with strings to match on.
+
+ +#### Outputs + +
+
Y (non-differentiable) : T2
+
Tensor of bools indicating if each input string fully matches the regex pattern specified.
+
+ +#### Type Constraints + +
+
T1 : tensor(string)
+
Inputs must be UTF-8 strings
+
T2 : tensor(bool)
+
Outputs are bools and are True where there is a full regex match and False otherwise.
+
+ +### **StringConcat-20** + + StringConcat concatenates string tensors elementwise (with NumPy-style broadcasting support) + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
Tensor to prepend in concatenation
+
Y (non-differentiable) : T
+
Tensor to append in concatenation
+
+ +#### Outputs + +
+
Z (non-differentiable) : T
+
Concatenated string tensor
+
+ +#### Type Constraints + +
+
T : tensor(string)
+
Inputs and outputs must be UTF-8 strings
+
+ +### **StringSplit-20** + + StringSplit splits a string tensor's elements into substrings based on a delimiter attribute and a maxsplit attribute. + + The first output of this operator is a tensor of strings representing the substrings from splitting each input string on the `delimiter` substring. This tensor has one additional rank compared to the input tensor in order to store the substrings for each input element (where the input tensor is not empty). Note that, in order to ensure the same number of elements are present in the final dimension, this tensor will pad empty strings as illustrated in the examples below. Consecutive delimiters are not grouped together and are deemed to delimit empty strings, except if the `delimiter` is unspecified or is the empty string (""). In the case where the `delimiter` is unspecified or the empty string, consecutive whitespace characters are regarded as a single separator and leading or trailing whitespace is removed in the output. + + The second output tensor represents the number of substrings generated. `maxsplit` can be used to limit the number of splits performed - after the `maxsplit`th split if the string is not fully split, the trailing suffix of input string after the final split point is also added. For elements where fewer splits are possible than specified in `maxsplit`, it has no effect. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
delimiter : string
+
Delimiter to split on. If left unset or set to the empty string (""), the input is split on consecutive whitespace.
+
maxsplit : int
+
Maximum number of splits (from left to right). If left unset (or if the number of possible splits are less than maxsplit), it will make as many splits as possible. Note that the maximum possible number of substrings returned with `maxsplit` specified is `maxsplit+1` since the remaining suffix after the `maxsplit`th split is included in the output.
+
+ +#### Inputs + +
+
X (non-differentiable) : T1
+
Tensor of strings to split.
+
+ +#### Outputs + +
+
Y (non-differentiable) : T2
+
Tensor of substrings representing the outcome of splitting the strings in the input on the delimiter. Note that to ensure the same number of elements are present in the final rank, this tensor will pad any necessary empty strings.
+
Z (non-differentiable) : T3
+
The number of substrings generated for each input element.
+
+ +#### Type Constraints + +
+
T1 : tensor(string)
+
The input must be a UTF-8 string tensor
+
T2 : tensor(string)
+
Tensor of substrings.
+
T3 : tensor(int64)
+
The number of substrings generated.
+
+ +## Version 21 of the default ONNX operator set +### **Cast-21** + + The operator casts the elements of a given input tensor to a data type + specified by the 'to' argument and returns an output tensor of the same size in + the converted type. The 'to' argument must be one of the data types specified + in the 'DataType' enum field in the TensorProto message. + + Casting from string tensor in plain (e.g., "3.14" and "1000") and scientific numeric representations + (e.g., "1e-5" and "1E8") to float types is supported. For example, converting string "100.5" to an integer may + yield result 100. There are some string literals reserved for special floating-point values; + "+INF" (and "INF"), "-INF", and "NaN" are positive infinity, negative infinity, and not-a-number, respectively. + Any string which can exactly match "+INF" in a case-insensitive way would be mapped to positive infinite. Similarly, + this case-insensitive rule is applied to "INF" and "NaN". When casting from numeric tensors + to string tensors, plain floating-point representation (such as "314.15926") would be used. + Converting non-numerical-literal string such as "Hello World!" is an undefined behavior. Cases + of converting string representing floating-point arithmetic value, such as "2.718", to INT is an undefined behavior. + + Conversion from a numerical type to any numerical type is always allowed. + User must be aware of precision loss and value change caused by range difference between two types. + For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting + an integer 36 to Boolean may produce 1 because we truncate bits which can't be stored in the targeted type. + + In more detail, the conversion among numerical types should follow these rules + if the destination type is not a float 8 type. + + * Casting from floating point to: + * floating point: +/- infinity if OOR (out of range). + * fixed point: undefined if OOR. + * bool: +/- 0.0 to False; all else to True. + * Casting from fixed point to: + * floating point: +/- infinity if OOR. (+ infinity in the case of uint) + * fixed point: when OOR, discard higher bits and reinterpret (with respect to two's complement representation for + signed types). For example, 200 (int16) -> -56 (int8). + * bool: zero to False; nonzero to True. + * Casting from bool to: + * floating point: `{1.0, 0.0}`. + * fixed point: `{1, 0}`. + * bool: no change. + + Float 8 type were introduced to speed up the training of + deep models. By default the conversion of a float *x* obeys + to the following rules. `[x]` means the value rounded to + the target mantissa width. + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | -------- | -------- | -------- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | Inf | FLT_MAX | NaN | FLT_MAX | NaN | + | -Inf | -FLT_MAX | NaN | -FLT_MAX | NaN | + | \[x\] > FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | + | \[x\] \< -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | + | else | RNE | RNE | RNE | RNE | + + The behavior changes if the parameter 'saturate' is set to False. + The rules then become: + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | ------ | -------- | ---- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | -NaN | -NaN | NaN | -NaN | NaN | + | Inf | NaN | NaN | Inf | NaN | + | -Inf | -NaN | NaN | -Inf | NaN | + | \[x\] > FLT_MAX | NaN | NaN | Inf | NaN | + | \[x\] \< -FLT_MAX | NaN | NaN | -Inf | NaN | + | else | RNE | RNE | RNE | RNE | + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. All cases are fully described in two tables inserted in the operator description.
+
to : int (required)
+
The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor with the same shape as input with type specified by the 'to' argument
+
+ +#### Type Constraints + +
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(bool), tensor(string), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **CastLike-21** + + The operator casts the elements of a given input tensor (the first input) to + the same data type as the elements of the second input tensor. + See documentation of the Cast operator for further details. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. Please refer to operator Cast description for further details.
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
target_type (non-differentiable) : T2
+
The (first) input tensor will be cast to produce a tensor of the same type as this (second input) tensor.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor produced by casting the first input tensor to have the same type as the second input tensor.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **Constant-21** + + This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, + or value_* must be specified. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
sparse_value : sparse_tensor
+
The value for the elements of the output tensor in sparse format.
+
value : tensor
+
The value for the elements of the output tensor.
+
value_float : float
+
The value for the sole element for the scalar, float32, output tensor.
+
value_floats : list of floats
+
The values for the elements for the 1D, float32, output tensor.
+
value_int : int
+
The value for the sole element for the scalar, int64, output tensor.
+
value_ints : list of ints
+
The values for the elements for the 1D, int64, output tensor.
+
value_string : string
+
The value for the sole element for the scalar, UTF-8 string, output tensor.
+
value_strings : list of strings
+
The values for the elements for the 1D, UTF-8 string, output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain input and output types to all tensor types.
+
+ +### **ConstantOfShape-21** + + Generate a tensor with given value and shape. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
value : tensor
+
(Optional) The value of the output elements.Should be a one-element tensor. If not specified, it defaults to a tensor of value 0 and datatype float32
+
+ +#### Inputs + +
+
input : T1
+
1D tensor. The shape of the expected output tensor. If empty tensor is given, the output would be a scalar. All values must be >= 0.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of shape specified by 'input'.If attribute 'value' is specified, the value and datatype of the output tensor is taken from 'value'.If attribute 'value' is not specified, the value in the output defaults to 0, and the datatype defaults to float32.
+
+ +#### Type Constraints + +
+
T1 : tensor(int64)
+
Constrain input types.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint4), tensor(int4), tensor(bool), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain output types to be numerics or boolean.
+
+ +### **DequantizeLinear-21** + + The linear dequantization operator. It consumes a quantized tensor, a scale, and a zero point to compute the + full-precision tensor. The dequantization formula is `y = (x - x_zero_point) * x_scale`. `x_scale` and `x_zero_point` + must have the same shape, determining the quantization's granularity: a scalar for per-tensor/per-layer quantization, + a 1-D tensor for per-axis quantization, or have a rank identical to the input for blocked quantization. + See QuantizeLinear for details on quantization granularity. + `x_zero_point` and `x` must have the same type. `x` and `y` must have the same shape. In the case of dequantizing + `int32`, there's no zero point (zero point is supposed to be 0). + `zero-point` is usually not used in the case of float8 types quantization, but the dequantization formula remains the same + for consistency, and `x_scale` still determines the output type. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`.
+
block_size : int (default is 0)
+
(Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given `x` shape `(D0, ..., Di, ..., Dn)`, `y_scale` shape `(S0, ... Si, ...Sn)` and `axis=i`, the accepted range is `[ceil(Di/Si), ceil(Di/(Si-1))-1]`
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D quantized input tensor to be de-quantized.
+
x_scale : T2
+
Scale for input `x`. For per-tensor/layer dequantization the scale is a scalar, for per per-axis dequantization it is a 1-D Tensor and for blocked dequantization it has the same shape as the input, except for one dimension in which blocking is performed.
+
x_zero_point (optional) : T1
+
Zero point for input `x`. Shape must match x_scale. It's optional. Zero point is 0 when it's not specified.
+
+ +#### Outputs + +
+
y : T2
+
N-D full precision output tensor. It has same shape as input `x`.
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(int32), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
The type of the inputs 'x_zero_point' and 'x'.
+
T2 : tensor(float), tensor(float16), tensor(bfloat16)
+
'x_scale' determines the output type.
+
+ +### **Flatten-21** + + Flattens the input tensor into a 2D matrix. If input tensor has shape + (d_0, d_1, ... d_n) then the output will have shape + (d_0 X d_1 ... d_(axis-1), d_axis X d_(axis+1) ... X dn). + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 ... d_n), where the shape of the input tensor is (d_0, d_1, ... d_n).
+
+ +#### Inputs + +
+
input (differentiable) : T
+
A tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain input and output to all tensor types up to IRv10.
+
+ +### **GroupNormalization-21** + + A GroupNormalization function. Carries out group normalization as described in + the paper https://arxiv.org/abs/1803.08494 + + This operator transforms input according to + ``` + y = scale * (x - mean) / sqrt(variance + epsilon) + bias, + ``` + where the mean and variance are computed per instance per group of channels, and + `scale` and `bias` should be specified for each channel. The number of + groups `num_groups` should be divisible by the number of channels so that there are + an equal number of channels per group. + + The overall computation has two stages: the first stage normalizes the elements to + have zero mean and unit variance for each instance in each group, and the second + stage scales and shifts the results of the first stage. The floating-point precision + used in the first stage is determined by the `stash_type` attribute. For example, + if `stash_type` is 1, the operator casts all input variables to 32-bit float, + performs the computation, and finally casts the normalized results back to the + original type of `X`. The second stage does not depend on `stash_type`. + + When the number of groups is the same as the number of channels, this operator is + equivalent to InstanceNormalization. When there is only one group, this operator + is equivalent to LayerNormalization. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
num_groups : int (required)
+
The number of groups of channels. It should be a divisor of the number of channels `C`.
+
stash_type : int (default is 1)
+
The floating-point precision used in stage one of the computation.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor. Dimensions for image cases are `(N x C x H x W)`, where `N` is the batch size, `C` is the number of channels, and `H` and `W` are the height and width of the data. Statistics are computed for every group of channels over `C`, `H`, and `W`. For non-image cases, the dimensions are in the form of `(N x C x D1 x D2 ... Dn)`.
+
scale (differentiable) : T
+
Scale tensor of shape `(C)`.
+
bias (differentiable) : T
+
Bias tensor of shape `(C)`.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
The output tensor of the same shape as `X`.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Identity-21** + + Identity operator + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : V
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : V
+
Tensor to copy input into.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain input and output types to all tensor, sequence, and optional types.
+
+ +### **If-21** + + If conditional + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
else_branch : graph (required)
+
Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
+
then_branch : graph (required)
+
Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
+
+ +#### Inputs + +
+
cond : B
+
Condition for the if. The tensor must contain a single element.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : V
+
Values that are live-out to the enclosing scope. The return values in the `then_branch` and `else_branch` must be of the same data type. The `then_branch` and `else_branch` may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if-node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of `then_branch` is typed float tensor with shape [2] and the first output of `else_branch` is another float tensor with shape [3], If's first output should have (a) no shape set, or (b) a shape of rank 1 with neither `dim_value` nor `dim_param` set, or (c) a shape of rank 1 with a unique `dim_param`. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv10.
+
B : tensor(bool)
+
Only bool
+
+ +### **Loop-21** + + Generic Looping construct. This loop has multiple termination conditions: + + 1) Trip count. Iteration count specified at runtime. Set by + specifying the input M. Optional. Set to empty string to omit. + Note that a static trip count (specified at graph construction time) can be + specified by passing in a constant node for input M. + 2) Loop termination condition. This is an input to the op that determines + whether to run the first iteration and also a loop-carried dependency for + the body graph. The body graph must yield a value for the condition variable, + whether this input is provided or not. + + This table summarizes the operating modes of this operator with equivalent + C-style code: + + Operator inputs defined as (max_trip_count, condition_var). + + * input ("", ""): + for (int i=0; ; ++i) { + cond = ... // Note this value is ignored, but is required in the body + } + + * input ("", cond) // Note this is analogous to a while loop + bool cond = ...; + for (int i=0; cond; ++i) { + cond = ...; + } + + * input ("", 1) // Note this is analogous to a do-while loop + bool cond = true + for (int i=0; cond; ++i) { + cond = ...; + } + + * input (trip_count, "") // Note this is analogous to a for loop + int trip_count = ... + for (int i=0; i < trip_count; ++i) { + cond = ...; // ignored + } + + * input (trip_count, cond) + int trip_count = ...; + bool cond = ...; + for (int i=0; i < trip_count && cond; ++i) { + cond = ...; + } + + + *Sample usage - cond as well as trip count* + + graph predict-net { + %a = Constant[value = ]() + %b = Constant[value = ]() + %keepgoing = Constant[value = ]() + %max_trip_count = Constant[value = ]() + %keepgoing_out, %b_out, %user_defined_vals = Loop[body = ](%max_trip_count, %keepgoing, %b) + return + } + + graph body-net ( + %i[INT32, scalar] // iteration number + %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used + %b_in[INT32, scalar] // incoming value of loop-carried-dependency b + ) { + %my_local = Add(%a, %b_in) + %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b + %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition + %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated + return %keepgoing_out, %b_out, %user_defined_val + } + + *Sample equivalent C code* + + { + /* User-defined code (enclosing scope) */ + int a = 3, b = 6; + bool keepgoing = true; // Analogous to input cond + /* End user-defined code */ + + /* Implicitly-defined code */ + const int max_trip_count = 10; // Analogous to input M + int user_defined_vals[]; // Imagine this is resizable + /* End implicitly-defined code */ + /* initialize loop-carried variables and scan-output variables */ + bool keepgoing_out = keepgoing + int b_out = b + + for (int i=0; i < max_trip_count && keepgoing_out; ++i) { + /* Implicitly-defined code: bind actual parameter values + to formal parameter variables of loop-body */ + bool keepgoing_in = keepgoing_out; + bool b_in = b_out; + + /* User-defined code (loop body) */ + int my_local = a + b_in; // Reading value "a" from the enclosing scope is fine + b_out = a - b_in; + keepgoing_out = my_local > b_out; + user_defined_val = b_in + b_in; // b_in and b_out are different variables + /* End user-defined code */ + + /* Implicitly defined-code */ + user_defined_vals[i] = user_defined_val // accumulate scan-output values + } + // int t = my_local; // Can't do this. my_local is not accessible here. + + // The values below are bound to the output variables of the loop and therefore accessible + // b_out; user_defined_vals; keepgoing_out; + } + + There are several things of note in this code snippet: + + 1) Values from the enclosing scope (i.e. variable "a" here) are in scope and can + be referenced in the inputs of the loop. + 2) Any values computed in the loop body that needs to be used in a subsequent + iteration or after the loop are modeled using a pair of variables in the loop-body, + consisting of an input variable (eg., b_in) and an output variable (eg., b_out). + These are referred to as loop-carried dependences. The loop operation node + supplies the input value of the input variable for the first iteration, and + returns the output value of the output variable produced by the final + iteration. + 3) Scan_output variables are used to implicitly concatenate values computed across + all the iterations. In the above example, the value of user_defined_val computed + over all iterations are concatenated and returned as the value of user_defined_vals + after the loop. + 4) Values created in the body cannot be accessed in the enclosing scope, + except using the mechanism described above. + + Note that the semantics of this op support "diagonal" or "wavefront" execution. + (See Step 3 here for an example: + https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). + Frontends should emit multi-layer RNNs as a series of While operators (with + time being the inner looping dimension), with each successive layer consuming + the scan_outputs from the previous layer, possibly going through several + point-wise operators (e.g. dropout, residual connections, linear layer). + + The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies...). It has 1+N+K outputs: (condition, loop carried dependencies..., scan_outputs...). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
+
+ +#### Inputs (2 - ∞) + +
+
M (optional) : I
+
A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
+
cond (optional) : B
+
A boolean termination condition. Optional. Pass empty string to skip.
+
v_initial (variadic, heterogeneous) : V
+
The initial values of any loop-carried dependencies (values that change across loop iterations)
+
+ +#### Outputs (1 - ∞) + +
+
v_final_and_scan_outputs (variadic, heterogeneous) : V
+
Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv10.
+
I : tensor(int64)
+
tensor of int64, which should be a scalar.
+
B : tensor(bool)
+
tensor of bool, which should be a scalar.
+
+ +### **Pad-21** + + Given a tensor containing the data to be padded (`data`), a tensor containing the number of start and end pad values for axis (`pads`), (optionally) a `mode`, and (optionally) `constant_value`, + a padded tensor (`output`) is generated. + + The four supported `modes` are (similar to corresponding modes supported by `numpy.pad`): + + 1) `constant`(default) - pads with a given constant value as specified by `constant_value` (which defaults to 0, empty string, or False) + + 2) `reflect` - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis + + 3) `edge` - pads with the edge values of array + + 4) `wrap` - wrap-around padding as if the data tensor forms a torus + + + Example 1 (`constant` mode): + + Insert 0 pads to the beginning of the second dimension. + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'constant' + + constant_value = 0.0 + + output = [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ] + ``` + + Example 2 (`reflect` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'reflect' + + output = [ + [1.0, 1.2, 1.0, 1.2], + [2.3, 3.4, 2.3, 3.4], + [4.5, 5.7, 4.5, 5.7], + ] + ``` + + Example 3 (`edge` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'edge' + + output = [ + [1.0, 1.0, 1.0, 1.2], + [2.3, 2.3, 2.3, 3.4], + [4.5, 4.5, 4.5, 5.7], + ] + ``` + + Example 4 (`wrap` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [2, 1, 1, 1] + + mode = 'wrap' + + output = [ + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + ] + ``` + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is constant)
+
Supported modes: `constant`(default), `reflect`, `edge`, `wrap`
+
+ +#### Inputs (2 - 4) + +
+
data (differentiable) : T
+
Input tensor.
+
pads (non-differentiable) : tensor(int64)
+
Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. `pads` should be a 1D tensor of shape [2 * num_axes] where `num_axes` refers to the number of elements in the `axes` input or the input rank if `axes` are not provided explicitly. `pads` format should be: [x1_begin, x2_begin, ..., x1_end, x2_end,...], where xi_begin is the number of pad values added at the beginning of axis `axes[i]` and xi_end, the number of pad values added at the end of axis `axes[i]`.
+
constant_value (optional, non-differentiable) : T
+
(Optional) A scalar value to be used if the mode chosen is `constant` (by default it is 0, empty string or False).
+
axes (optional, non-differentiable) : Tind
+
1-D tensor of axes that `pads` apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated. If not provided, all axes are assumed (`[0, 1, ..., input_rank-1]`).
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain input and output types to all tensor types up to IRv10.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **QLinearMatMul-21** + + Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html). + It consumes two quantized input tensors, their scales and zero points, scale and zero point of output, + and computes the quantized output. The quantization formula is y = saturate((x / y_scale) + y_zero_point). + For (x / y_scale), it is rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details. + Scale and zero point must have same shape. They must be either scalar (per tensor) or N-D tensor + (per row for 'a' and per column for 'b'). Scalar refers to per tensor quantization whereas N-D refers to per row + or per column quantization. If the input is 2D of shape [M, K] then zero point and scale tensor may be + an M element vector [v_1, v_2, ..., v_M] for per row quantization and K element vector of shape [v_1, v_2, ..., v_K] + for per column quantization. If the input is N-D tensor with shape [D1, D2, M, K] then zero point and scale tensor may + have shape [D1, D2, M, 1] for per row quantization and shape [D1, D2, 1, K] for per column quantization. + Production must never overflow, and accumulation may overflow if and only if in 32 bits. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Inputs + +
+
a (non-differentiable) : T1
+
N-dimensional quantized matrix a
+
a_scale (non-differentiable) : TS
+
scale of quantized input a
+
a_zero_point (non-differentiable) : T1
+
zero point of quantized input a
+
b (non-differentiable) : T2
+
N-dimensional quantized matrix b
+
b_scale (non-differentiable) : TS
+
scale of quantized input b
+
b_zero_point (non-differentiable) : T2
+
zero point of quantized input b
+
y_scale (non-differentiable) : TS
+
scale of quantized output y
+
y_zero_point (non-differentiable) : T3
+
zero point of quantized output y
+
+ +#### Outputs + +
+
y (non-differentiable) : T3
+
Quantized matrix multiply results from a * b
+
+ +#### Type Constraints + +
+
TS : tensor(float), tensor(float16), tensor(bfloat16)
+
Constrain scales.
+
T1 : tensor(int8), tensor(uint8), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
The type of input a and its zeropoint.
+
T2 : tensor(int8), tensor(uint8), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
The type of input b and its zeropoint.
+
T3 : tensor(int8), tensor(uint8), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
The type of the output and its zeropoint.
+
+ +### **QuantizeLinear-21** + + The linear quantization operator consumes a high-precision tensor, a scale, and a zero point to compute the + low-precision/quantized tensor. The scale factor and zero point must have the same shape, determining the quantization + granularity. The quantization formula is `y = saturate((x / y_scale) + y_zero_point)`. + Saturation is done according to: + - uint16: [0, 65535] + - int16: [-32768, 32767] + - uint8: [0, 255] + - int8: [-128, 127] + - uint4: [0, 15] + - int4: [-8, 7] + For `(x / y_scale)`, it rounds to the nearest even. Refer to https://en.wikipedia.org/wiki/Rounding for details. + `y_zero_point` and `y` must have the same type. `y_zero_point` is usually not used for quantization to float8 types, but the quantization + formula remains the same for consistency, and the type of the attribute `y_zero_point` still determines the quantization type. + There are three supported quantization granularities, determined by the shape of `y_scale`. + In all cases, `y_zero_point` must have the same shape as `y_scale`. + - Per-tensor (per-layer) quantization: `y_scale` is a scalar. + - Per-axis quantization: The scale must be a 1-D tensor, with the length of the quantization axis. For an input shape + `(D0, ..., Di, ..., Dn)` and `axis=i`, `y_scale` is a 1-D tensor of length `Di`. + - Blocked quantization: The scale's shape is identical to the input's shape, except for one dimension, in which + blocking is performed. Given `x` shape `(D0, ..., Di, ..., Dn)`, `axis=i`, and block size `B`: `y_scale` shape is + `(D0, ..., ceil(Di/B), ..., Dn)`. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used only for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`. When the rank of the input is 1, per-tensor quantization is applied, rendering the axis unnecessary in this scenario.
+
block_size : int (default is 0)
+
(Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given `x` shape `(D0, ..., Di, ..., Dn)`, `y_scale` shape `(S0, ... Si, ...Sn)` and `axis=i`, the accepted range is `[ceil(Di/Si), ceil(Di/(Si-1))-1]`
+
output_dtype : int (default is 0)
+
(Optional) The output data type. If not supplied, the output data type is inferred from `y_zero_point` data type (`T2`). If neither `output_dtype` nor `y_zero_point` are supplied, output data type is uint8. If both `output_dtype` and `y_zero_point` are specified, `output_dtype` must be `T2`.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 quantization (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. All cases are fully described in two tables inserted in the operator description.
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D full precision Input tensor to be quantized.
+
y_scale : T1
+
Scale for doing quantization to get `y`. For per-tensor/layer quantization the scale is a scalar, for per-axis quantization it is a 1-D Tensor and for blocked quantization it has the same shape as the input, except for one dimension in which blocking is performed.
+
y_zero_point (optional) : T2
+
Zero point for doing quantization to get `y`. Shape must match `y_scale`.Default is uint8 with zero point of 0 if it's not specified.
+
+ +#### Outputs + +
+
y : T2
+
N-D quantized output tensor. It has same shape as input `x`.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(float16), tensor(bfloat16), tensor(int32)
+
The type of the input 'x'.
+
T2 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
The type of the input `y_zero_point` and the output `y`.
+
+ +### **Reshape-21** + + Reshape the input tensor similar to numpy.reshape. + First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. + At most one dimension of the new shape can be -1. In this case, the value is + inferred from the size of the tensor and the remaining dimensions. A dimension + could also be 0, in which case the actual dimension value is unchanged (i.e. taken + from the input tensor). If 'allowzero' is set, and the new shape includes 0, the + dimension will be set explicitly to zero (i.e. not taken from input tensor). + Shape (second input) could be an empty shape, which means converting to a scalar. + The input tensor's shape and the output tensor's shape are required to have the same number of elements. + + If the attribute 'allowzero' is set, it is invalid for the specified shape to + contain both a zero value and -1, as the value of the dimension corresponding + to -1 cannot be determined uniquely. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
allowzero : int (default is 0)
+
(Optional) By default, when any value in the 'shape' input is equal to zero the corresponding dimension value is copied from the input tensor dynamically. allowzero=1 indicates that if any value in the 'shape' input is set to zero, the zero value is honored, similar to NumPy.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
shape (non-differentiable) : tensor(int64)
+
Specified shape for output.
+
+ +#### Outputs + +
+
reshaped (differentiable) : T
+
Reshaped data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain input and output types to all tensor types.
+
+ +### **Scan-21** + + Scan can be used to iterate over one or more scan_input tensors, + constructing zero or more scan_output tensors. It combines ideas from general recurrences, + functional programming constructs such as scan, fold, map, and zip, and is intended to enable + generalizations of RNN-like constructs for sequence-to-sequence processing. + Other tensors (referred to as state_variables here) can be used to carry a state + when iterating from one element to another (similar to hidden-state in RNNs, also referred + to as loop-carried dependences in the context of loops). + Many common usages involve a single scan_input tensor (where functionality + similar to scan, fold and map can be obtained). When more than one scan_input is used, + a behavior similar to zip is obtained. + + The attribute body must be a graph, specifying the computation to be performed in + every iteration. It takes as input the current values of the state_variables and + the current iterated element of the scan_inputs. It must return the (updated) values + of the state_variables and zero or more scan_output_element tensors. The values of the + scan_output_element tensors are concatenated over all the iterations to produce the + scan_output values of the scan construct (similar to the concatenated intermediate + hidden-state values of RNN-like constructs). All the output tensors (state_variables as + well as scan_output_element tensors) are required to have the same shape in each iteration + of the loop (a restriction imposed to enable efficient memory allocation). + + Note that the iterated element passed to the body subgraph does not have a sequence + axis. It will have a rank one less than the rank of the corresponding scan_input. + + The scan operation returns the final values of the state_variables as well as the + scan_outputs. + + The optional attribute scan_input_directions specifies the direction (forward or backward) + for each scan input. If this attribute is omitted, all sequences are scanned in the forward + direction. A bidirectional scan may be performed by specifying the same tensor input twice + in the scan_inputs, once with a forward direction, and once with a backward direction. + + The scan_output of the operation is produced by concatenating the scan_output_element + values produced by the body in each iteration. The optional attribute scan_output_directions + specifies the direction in which scan_output is constructed (by appending or prepending the + scan_output_element to scan_output in each iteration) for each scan_output. If this attribute + is omitted, the scan_output_element is appended to the scan_output in each iteration. + + The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. + If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the + batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. + Note that scanning a non-zero axis may be less efficient than scanning axis zero. + + The optional attribute scan_output_axes specifies the axis along which the scan_outputs + are accumulated for each scan_output. For example, if axis 1 is the time axis (to be + scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis + value of 1. + + Note that because of the ONNX restriction that only the last parameter of an operator can + be variadic, the initial-states and scan-inputs are listed together as one input parameter. + Similarly, the final-states and scan-outputs are listed together as one output parameter. + The attribute num_scan_inputs indicates the number M of scan-inputs. + + The behavior of + + Scan < + num_scan_inputs = m, + body = loop-body, + scan_input_axes = [axis_1, ..., axis_m] + > (init_1, ..., init_n, scan_1, ..., scan_m) + + is equivalent to the following pseudo-code: + + // scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i + // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. + sequence_length = scan_1.shape[axis_1]; + + // initialize state-variables + st_1 = init_1; ... st_n = init_n; + // initialize scan-output variables: [] denotes an empty tensor + scan_out_1 = []; ...; scan_out_k = []; + // identify number of iterations: + + // execute loop + for (int t = 0; t < sequence_length; ++t) { + // generate the scan-input elements: the notation T[t] indicates the sub-tensor + // of rank one less than T obtained by indexing T at position t along axis k. + si_1 = scan_1[t]; + ... ; + si_m = scan_m[t]; + // execute loop-body + st_1, ..., st_n, so_1, ..., so_k = loop-body(st_1, ..., st_n, si_1, ..., si_m) + // accumulate the scan-output elements + scan_out_1 = Concat(scan_out_1, so_1); ... ; scan_out_k = Concat(scan_out_k, so_k); + } + + return st_1, ..., st_n, scan_out_1, ..., scan_out_k; + + *Sample usage: Encoding RNN using a Scan* + + The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, + recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can + be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes + %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these + values are computed in the outer graph, they need to be passed in as extra state_variables. + + graph rnn-encoding { + %H_0 = ... + %X = ... + %Y_h, %Y = Scan[body = , num_scan_inputs=1](%H_0, %X) + return %Y, %Y_h + } + + graph rnn-cell-1 ( + %H_tminus1[FLOAT, tensor] + %X_t[FLOAT, tensor] + ) { + %Wi = ... + %Ri = ... + %Wbi = ... + %Rbi = ... + %t1 = X_t * (Wi^T) + %t2 = H_tminus1*(Ri^T) + %t3 = Add(%t1, %t2) + %t4 = Add(%t3, %Wbi) + %t5 = Add(%t4, %Rbi) + %Ht = Tanh(%t5) + %Accumulate = Identity(%Ht) + return %Ht, %Accumulate + } + + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has N+M inputs: (loop state variables..., scan_input_elts...). It has N+K outputs: (loop state variables..., scan_output_elts...). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
+
num_scan_inputs : int (required)
+
An attribute specifying the number of scan_inputs M.
+
scan_input_axes : list of ints
+
An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
scan_input_directions : list of ints
+
An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
+
scan_output_axes : list of ints
+
An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
+
scan_output_directions : list of ints
+
An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
+
+ +#### Inputs (1 - ∞) + +
+
initial_state_and_scan_inputs (variadic, heterogeneous) : V
+
Initial values of the loop's N state variables followed by M scan_inputs
+
+ +#### Outputs (1 - ∞) + +
+
final_state_and_scan_outputs (variadic, heterogeneous) : V
+
Final values of the loop's N state variables followed by K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
All Tensor types up to IRv10.
+
+ +### **Shape-21** + + Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. + Optional attributes start and end can be used to compute a slice of the input tensor's shape. + If start axis is omitted, the slice starts from axis 0. + The end axis, if specified, is exclusive (and the returned value will not include the size of that axis). + If the end axis is omitted, the axes upto the last one will be included. + Negative axes indicate counting back from the last axis. + Note that axes will be clamped to the range [0, r], where r is the + rank of the input tensor if they are out-of-range (after adding r in the case of + negative axis). Thus, specifying any end value > r is equivalent to specifying an end + value of r, and specifying any start value < -r is equivalent to specifying a start + value of 0. If start > end, the result will be an empty shape. + + Examples: + + ``` + Input tensor with shape: [2, 3, 4] + No attributes specified. + Output: [2, 3, 4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: -1 + Output: [4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + end: -1 + Output: [2, 3] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: 1 + end: 2 + Output: [3] + ``` + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
end : int
+
(Optional) Ending axis for slicing the shape. Negative value means counting dimensions from the back. If omitted, sizes of all axes upto (including) the last one will be included.
+
start : int (default is 0)
+
(Optional) Starting axis for slicing the shape. Default value is 0.Negative value means counting dimensions from the back.
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
shape (non-differentiable) : T1
+
Shape of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor.
+
+ +### **Size-21** + + Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
size (non-differentiable) : T1
+
Total number of elements of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor, which should be a scalar though.
+
+ +### **Squeeze-21** + + Remove single-dimensional entries from the shape of a tensor. + Takes an input `axes` with a list of axes to squeeze. + If `axes` is not provided, all the single dimensions will be removed from + the shape. If an axis is selected with shape entry not equal to one, an error is raised. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
Tensors with at least max(dims) dimensions.
+
axes (optional, non-differentiable) : tensor(int64)
+
List of integers indicating the dimensions to squeeze. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
squeezed (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain input and output types to all tensor types up to IRv10.
+
+ +### **Transpose-21** + + Returns a transpose of the input tensor. (Similar to `numpy.transpose`). + The optional attribute `perm` must be a permutation of the dimensions of + the input tensor. Axis `i` of the output tensor corresponds to the axis + `perm[i]` of the input tensor. + For example, when perm=(1, 0, 2), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 1, 3). + When perm=(1, 2, 0), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 3, 1). + If the attribute `perm` is omitted, its default value is `(n-1, ..., 0)`, + where `n` is the rank of the input tensor. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Attributes + +
+
perm : list of ints
+
A list of integers. By default, reverse the dimensions, otherwise permute the axes according to the values given. Its length must be equal to the rank of the input.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
transposed (differentiable) : T
+
Transposed output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain input and output types to all tensor types.
+
+ +### **Unsqueeze-21** + + Insert single-dimensional entries to the shape of an input tensor (`data`). + Takes one required input `axes` - which contains a list of dimension indices and this operator will insert a dimension of value `1` into the corresponding index of the output tensor (`expanded`). + + For example, given an input tensor (`data`) of shape [3, 4, 5], then + Unsqueeze(data, axes=[0, 4]) outputs a tensor (`expanded`) containing same data as `data` but with shape [1, 3, 4, 5, 1]. + + The input `axes` should not contain any duplicate entries. It is an error if it contains duplicates. + The rank of the output tensor (`output_rank`) is the rank of the input tensor (`data`) plus the number of values in `axes`. + Each value in `axes` should be within the (inclusive) range [-output_rank , output_rank - 1]. + The order of values in `axes` does not matter and can come in any order. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +#### Inputs + +
+
data (differentiable) : T
+
Original tensor
+
axes (non-differentiable) : tensor(int64)
+
List of integers indicating the dimensions to be inserted. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(expanded).
+
+ +#### Outputs + +
+
expanded (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4)
+
Constrain input and output types to all tensor types up to IRv10.
+
+ +## Version 22 of the default ONNX operator set +### **Acos-22** + + Calculates the arccosine (inverse of cosine) of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The arccosine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Acosh-22** + + Calculates the hyperbolic arccosine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic arccosine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Asin-22** + + Calculates the arcsine (inverse of sine) of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The arcsine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Asinh-22** + + Calculates the hyperbolic arcsine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic arcsine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Atan-22** + + Calculates the arctangent (inverse of tangent) of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The arctangent of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Atanh-22** + + Calculates the hyperbolic arctangent of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic arctangent values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **AveragePool-22** + + AveragePool consumes an input tensor X and applies average pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + average pooling consisting of computing the average on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape is calculated differently + depending on whether explicit padding is used, where pads is employed, or auto padding is used, where auto_pad is utilized. + With explicit padding (https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html?highlight=maxpool#torch.nn.MaxPool2d): + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled. `pad_shape[i]` is the sum of pads along axis `i`. Sliding windows that would start in the right padded region are ignored. + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following when ceil_mode is enabled: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + or when ceil_mode is disabled (https://www.tensorflow.org/api_docs/python/tf/keras/layers/AveragePooling2D): + ``` + VALID: output_spatial_shape[i] = floor((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i]) + 1 + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = floor((input_spatial_shape[i] - 1) / strides_spatial_shape[i]) + 1 + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i] + ``` + The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero). + + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
count_include_pad : int (default is 0)
+
Whether include pad pixels when calculating values for the edges. Default is 0, doesn't count include pad.
+
dilations : list of ints
+
Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Bernoulli-22** + + Draws binary random numbers (0 or 1) from a Bernoulli distribution. The input tensor should be a tensor + containing probabilities p (a value in the range [0,1]) to be used for drawing the binary random number, + where an output of 1 is produced with probability p and an output of 0 is produced with probability (1-p). + + This operator is non-deterministic and may not produce the same values in different + implementations (even if a seed is specified). + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int
+
The data type for the elements of the output tensor. if not specified, we will use the data type of the input tensor.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
All values in input have to be in the range:[0, 1].
+
+ +#### Outputs + +
+
output : T2
+
The returned output tensor only has values 0 or 1, same shape as input tensor.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool)
+
Constrain output types to all numeric tensors and bool tensors.
+
+ +### **Conv-22** + + The convolution operator consumes an input tensor and a filter, and + computes the output. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input W.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults is 1 along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X (differentiable) : T
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
W (differentiable) : T
+
The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.
+
B (optional, differentiable) : T
+
Optional 1D bias to be added to the convolution, has size of M.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **ConvTranspose-22** + + The convolution transpose operator consumes an input tensor and a filter, + and computes the output. + + If the pads parameter is provided the shape of the output is calculated via the following equation: + + output_shape[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - pads[start_i] - pads[end_i] + + output_shape can also be explicitly specified in which case pads values are auto generated using these equations: + + total_padding[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - output_shape[i] + If (auto_pads == SAME_UPPER): pads[start_i] = total_padding[i]/2; pads[end_i] = total_padding[i] - (total_padding[i]/2) + Else: pads[start_i] = total_padding[i] - (total_padding[i]/2); pads[end_i] = (total_padding[i]/2). + + + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = input_shape[i] * strides[i]` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each spatial axis.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input W.
+
output_padding : list of ints
+
Additional elements added to the side with higher coordinate indices in the output. Each padding value in "output_padding" must be less than the corresponding stride/dilation dimension. By default, this attribute is a zero vector. Note that this attribute doesn't directly affect the computed output values. It only controls the selection of the computed values, so changing this attribute only adds or removes output elements. If "output_shape" is explicitly provided, "output_padding" does not contribute additional size to "output_shape" but participates in the computation of the needed padding amount. This is also called adjs or adjustment in some frameworks.
+
output_shape : list of ints
+
The shape of the output can be explicitly set which will cause pads values to be auto generated. If output_shape is specified pads values are ignored. See doc for details for equations to generate pads. Note that the output_shape attribute value should not include dimensions for batch size and channels, which are automatically inferred.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X (differentiable) : T
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn)
+
W (differentiable) : T
+
The weight tensor that will be used in the convolutions; has size (C x M/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the weight shape will be (C x M/group x k1 x k2 x ... x kn), where (k1 x k2 x ... x kn) is the dimension of the kernel. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
+
B (optional, differentiable) : T
+
Optional 1D bias to be added to the convolution, has size of M.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, pad lengths and group count. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Cos-22** + + Calculates the cosine of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The cosine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Cosh-22** + + Calculates the hyperbolic cosine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic cosine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **DeformConv-22** + + Performs deformable convolution as described in https://arxiv.org/abs/1703.06211 and https://arxiv.org/abs/1811.11168. + This operator specification supports the general N-D case. Note that most common use cases have 2D or 3D data. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
dilations : list of ints
+
Dilation value along each spatial axis of the kernel. Default is 1 along each axis.
+
group : int (default is 1)
+
Number of groups the input and output channels, C and oC, are divided into. C and oC must both be divisible by group. Default is 1.
+
kernel_shape : list of ints
+
Shape of the convolution kernel. If not present, it is inferred from the shape of input W.
+
offset_group : int (default is 1)
+
Number of groups of offset. C must be divisible by offset_group. Default is 1.
+
pads : list of ints
+
Padding for the beginning and end along each spatial axis. The values represent the number of pixels added to the beginning and end of the corresponding axis and can take any nonnegative value. The format should be as follows: [x1_begin, x2_begin, ..., x1_end, x2_end, ...], where xi_begin is the number of pixels added at the beginning of axis `i` and xi_end is the number of pixels added at the end of axis `i`. Default is 0 along each axis.
+
strides : list of ints
+
Stride along each spatial axis. Default is 1 along each axis.
+
+ +#### Inputs (3 - 5) + +
+
X : T
+
Input data tensor. For 2D image data, it has shape (N, C, H, W) where N is the batch size, C is the number of input channels, and H and W are the height and width. In general, the shape is (N, C, D1, D2, ... , Dn) for n-dimensional data, where D1 to Dn are the spatial dimension sizes. Most common use cases have n = 2 or 3.
+
W : T
+
Weight tensor that will be used in the convolutions. It has shape (oC, C/group, kH, kW), where oC is the number of output channels and kH and kW are the kernel height and width. For more than 2 dimensions, it has shape (oC, C/group, k1, k2, ... , kn).
+
offset : T
+
Offset tensor denoting the offset for the sampling locations in the convolution kernel. It has shape (N, offset_group * kH * kW * 2, oH, oW) for 2D data or (N, offset_group * k1 * k2 * ... * kn * n, o1, o2, ... , on) for nD data. Use linear interpolationfor fractional offset values. Sampling locations outside of the padded input tensor gives zero.
+
B (optional) : T
+
Optional 1D bias of length oC to be added to the convolution. Default is a tensor of zeros.
+
mask (optional) : T
+
The mask tensor to be applied to each position in the convolution kernel. It has shape (N, offset_group * kH * kW, oH, oW) for 2D data or (N, offset_group * k1 * k2 * ... * kn * n, o1, o2, ... , on) for nD data. Default is a tensor of ones.
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor that contains the result of convolution. It has shape (N, oC, oH, oW) for 2D data or (N, oC, o1, o2, ..., on) for nD data
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Det-22** + + Det calculates determinant of a square matrix or batches of square matrices. + Det takes one input tensor of shape `[*, M, M]`, where `*` is zero or more batch dimensions, + and the inner-most 2 dimensions form square matrices. + The output is a tensor of shape `[*]`, containing the determinants of all input submatrices. + e.g., When the input is 2-D, the output is a scalar(shape is empty: `[]`). + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to floating-point tensors.
+
+ +### **Dropout-22** + + Dropout takes an input floating-point tensor, an optional input ratio (floating-point scalar) and an optional input training_mode (boolean scalar). It produces two tensor outputs, + output (floating-point tensor) and mask (optional `Tensor`). If `training_mode` is true then the output Y will be a random dropout; + Note that this Dropout scales the masked input data by the following equation, so to convert the trained model into inference mode, + the user can simply not pass `training_mode` input or set it to false. + ``` + output = scale * data * mask, + ``` + where + ``` + scale = 1. / (1. - ratio). + ``` + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
seed : int
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs (1 - 3) + +
+
data (differentiable) : T
+
The input data as Tensor.
+
ratio (optional, non-differentiable) : T1
+
The ratio of random dropout, with value in [0, 1). If set to 0, the output would be a simple copy of the input. If it's non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
+
training_mode (optional, non-differentiable) : T2
+
If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
+
+ +#### Outputs (1 - 2) + +
+
output (differentiable) : T
+
The output.
+
mask (optional, non-differentiable) : T2
+
The output mask.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input and output types to float tensors.
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input 'ratio' types to float tensors.
+
T2 : tensor(bool)
+
Constrain output 'mask' types to boolean tensors.
+
+ +### **Elu-22** + + Elu takes one input data (Tensor) and produces one output data + (Tensor) where the function `f(x) = alpha * (exp(x) - 1.) for x < + 0`, `f(x) = x for x >= 0`., is applied to the tensor elementwise. + + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Coefficient of ELU.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **EyeLike-22** + + Generate a 2D tensor (matrix) with ones on the diagonal and zeros everywhere else. Only 2D + tensors are supported, i.e. input T1 must be of rank 2. The shape of the output tensor is the + same as the input tensor. The data type can be specified by the 'dtype' argument. If + 'dtype' is not specified, then the type of input tensor is used. By default, the main diagonal + is populated with ones, but attribute 'k' can be used to populate upper or lower diagonals. + The 'dtype' argument must be one of the data types specified in the 'DataType' enum field in the + TensorProto message and be valid as an output type. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int
+
(Optional) The data type for the elements of the output tensor. If not specified, the data type of the input tensor T1 is used.
+
k : int (default is 0)
+
(Optional) Index of the diagonal to be populated with ones. Default is 0. If T2 is the output, this op sets T2[i, i+k] = 1. k = 0 populates the main diagonal, k > 0 populates an upper diagonal, and k < 0 populates a lower diagonal.
+
+ +#### Inputs + +
+
input : T1
+
2D input tensor to copy shape, and optionally, type information from.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor, same shape as input tensor T1.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool)
+
Constrain input types. Strings and complex are not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool)
+
Constrain output types. Strings and complex are not supported.
+
+ +### **GRU-22** + + Computes an one-layer GRU. This operator is usually supported via some custom + implementation such as CuDNN. + + Notations: + + * `X` - input tensor + * `z` - update gate + * `r` - reset gate + * `h` - hidden gate + * `t` - time step (t-1 means previous time step) + * `W[zrh]` - W parameter weight matrix for update, reset, and hidden gates + * `R[zrh]` - R recurrence weight matrix for update, reset, and hidden gates + * `Wb[zrh]` - W bias vectors for update, reset, and hidden gates + * `Rb[zrh]` - R bias vectors for update, reset, and hidden gates + * `WB[zrh]` - W parameter weight matrix for backward update, reset, and hidden gates + * `RB[zrh]` - R recurrence weight matrix for backward update, reset, and hidden gates + * `WBb[zrh]` - W bias vectors for backward update, reset, and hidden gates + * `RBb[zrh]` - R bias vectors for backward update, reset, and hidden gates + * `H` - Hidden state + * `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + * Relu(x) - max(0, x) + * Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + * Sigmoid(x) - 1/(1 + e^{-x}) + + NOTE: + Below are optional + + * Affine(x) - alpha * x + beta + * LeakyRelu(x) - x if x >= 0 else alpha * x + * ThresholdedRelu(x) - x if x >= alpha else 0 + * ScaledTanh(x) - alpha * Tanh(beta * x) + * HardSigmoid(x) - min(max(alpha * x + beta, 0), 1) + * Elu(x) - x if x >= 0 else alpha * (e^x - 1) + * Softsign(x) - x/(1 + |x|) + * Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh): + + * zt = f(Xt*(Wz^T) + Ht-1*(Rz^T) + Wbz + Rbz) + * rt = f(Xt*(Wr^T) + Ht-1*(Rr^T) + Wbr + Rbr) + * ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*(Rh^T) + Rbh + Wbh) # default, when linear_before_reset = 0 + * ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*(Rh^T) + Rbh)) + Wbh) # when linear_before_reset != 0 + * Ht = (1 - zt) (.) ht + zt (.) Ht-1 + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings
+
A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
layout : int (default is 0)
+
The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size].
+
linear_before_reset : int (default is 0)
+
When computing the output of the hidden gate, apply the linear transformation before multiplying by the output of the reset gate.
+
+ +#### Inputs (3 - 6) + +
+
X (differentiable) : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W (differentiable) : T
+
The weight tensor for the gates. Concatenation of `W[zrh]` and `WB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, input_size]`.
+
R (differentiable) : T
+
The recurrence weight tensor. Concatenation of `R[zrh]` and `RB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, hidden_size]`.
+
B (optional, differentiable) : T
+
The bias tensor for the gates. Concatenation of `[Wb[zrh], Rb[zrh]]` and `[WBb[zrh], RBb[zrh]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 6*hidden_size]`. Optional: If not specified - assumed to be 0
+
sequence_lens (optional, non-differentiable) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional, non-differentiable) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs (0 - 2) + +
+
Y (optional, differentiable) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional, differentiable) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **GlobalAveragePool-22** + + GlobalAveragePool consumes an input tensor X and applies average pooling across + the values in the same channel. This is equivalent to AveragePool with kernel size + equal to the spatial dimension of input tensor. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **GlobalLpPool-22** + + GlobalLpPool consumes an input tensor X and applies lp pool pooling across + the values in the same channel. This is equivalent to LpPool with kernel size + equal to the spatial dimension of input tensor. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
p : int (default is 2)
+
p value of the Lp norm used to pool over the input data.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **GlobalMaxPool-22** + + GlobalMaxPool consumes an input tensor X and applies max pooling across + the values in the same channel. This is equivalent to MaxPool with kernel size + equal to the spatial dimension of input tensor. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **GridSample-22** + + Given an input `X` and a flow-field `grid`, computes the output `Y` using `X` values and pixel locations from the `grid`. + For spatial input `X` with shape (N, C, H, W), the `grid` will have shape (N, H_out, W_out, 2), + the output `Y` will have shape (N, C, H_out, W_out). For volumetric input `X` with shape (N, C, D, H, W), + the `grid` will have shape (N, D_out, H_out, W_out, 3), the output `Y` will have shape (N, C, D_out, H_out, W_out). + More generally, for an input `X` of rank r+2 with shape (N, C, d1, d2, ..., dr), + the `grid` will have shape (N, D1_out, D2_out, ..., Dr_out, r), the output `Y` will have shape (N, C, D1_out, D2_out, ..., Dr_out). + + The tensor `X` contains values at centers of square pixels (voxels, etc) locations such as (n, c, d1_in, d2_in, ..., dr_in). + The (n, d1_out, d2_out, ..., dr_out, :) values from the tensor `grid` are the normalized positions for interpolating the values + at the (n, c, d1_out, d2_out, ..., dr_out) locations from the output tensor `Y` using a specified interpolation method (the mode) + and a padding mode (for `grid` positions falling outside the 2-dimensional image). + + For example, the values in `grid[n, h_out, w_out, :]` are size-2 vectors specifying normalized positions in the 2-dimensional space of `X`. + They are used to interpolate output values of `Y[n, c, h_out, w_out]`. + + The GridSample operator is often used in doing grid generator and sampler in the + [Spatial Transformer Networks](https://arxiv.org/abs/1506.02025). + See also in [torch.nn.functional.grid_sample](https://pytorch.org/docs/stable/generated/torch.nn.functional.grid_sample.html). + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
align_corners : int (default is 0)
+
If align_corners=1, the extrema (-1 and 1) are considered as referring to the center points of the input's corner pixels (voxels, etc.). If align_corners=0, they are instead considered as referring to the corner points of the input's corner pixels (voxels, etc.), making the sampling more resolution agnostic.
+
mode : string (default is linear)
+
Three interpolation modes: linear (default), nearest and cubic. The "linear" mode includes linear and N-linear interpolation modes depending on the number of spatial dimensions of the input tensor (i.e. linear for 1 spatial dimension, bilinear for 2 spatial dimensions, etc.). The "cubic" mode also includes N-cubic interpolation modes following the same rules. The "nearest" mode rounds to the nearest even index when the sampling point falls halfway between two indices.
+
padding_mode : string (default is zeros)
+
Support padding modes for outside grid values: `zeros`(default), `border`, `reflection`. zeros: use 0 for out-of-bound grid locations, border: use border values for out-of-bound grid locations, reflection: use values at locations reflected by the border for out-of-bound grid locations. If index 0 represents the margin pixel, the reflected value at index -1 will be the same as the value at index 1. For location far away from the border, it will keep being reflected until becoming in bound. If pixel location x = -3.5 reflects by border -1 and becomes x' = 1.5, then reflects by border 1 and becomes x'' = 0.5.
+
+ +#### Inputs + +
+
X (differentiable) : T1
+
Input tensor of rank r+2 that has shape (N, C, D1, D2, ..., Dr), where N is the batch size, C is the number of channels, D1, D2, ..., Dr are the spatial dimensions.
+
grid (non-differentiable) : T2
+
Input offset of shape (N, D1_out, D2_out, ..., Dr_out, r), where D1_out, D2_out, ..., Dr_out are the spatial dimensions of the grid and output, and r is the number of spatial dimensions. Grid specifies the sampling locations normalized by the input spatial dimensions. Therefore, it should have most values in the range of [-1, 1]. If the grid has values outside the range of [-1, 1], the corresponding outputs will be handled as defined by padding_mode. Following computer vision convention, the coordinates in the length-r location vector are listed from the innermost tensor dimension to the outermost, the opposite of regular tensor indexing.
+
+ +#### Outputs + +
+
Y (differentiable) : T1
+
Output tensor of rank r+2 that has shape (N, C, D1_out, D2_out, ..., Dr_out) of the sampled values. For integer input types, intermediate values are computed as floating point and cast to integer at the end.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input `X` and output `Y` types to all tensor types.
+
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain grid types to float tensors.
+
+ +### **HardSigmoid-22** + + HardSigmoid takes one input data (Tensor) and produces one output data + (Tensor) where the HardSigmoid function, y = max(0, min(1, alpha * x + beta)), + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 0.2)
+
Value of alpha.
+
beta : float (default is 0.5)
+
Value of beta.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **HardSwish-22** + + HardSwish takes one input data (Tensor) and produces one output data (Tensor) where + the HardSwish function, y = x * max(0, min(1, alpha * x + beta)) = x * HardSigmoid(x), + where alpha = 1/6 and beta = 0.5, is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **InstanceNormalization-22** + + Carries out instance normalization as described in the paper + https://arxiv.org/abs/1607.08022. + + y = scale * (x - mean) / sqrt(variance + epsilon) + B, + where mean and variance are computed per instance per channel. + + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
scale (differentiable) : T
+
The input 1-dimensional scale tensor of size C.
+
B (differentiable) : T
+
The input 1-dimensional bias tensor of size C.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output tensor of the same shape as input.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **LSTM-22** + + Computes an one-layer LSTM. This operator is usually supported via some + custom implementation such as CuDNN. + + Notations: + + * `X` - input tensor + * `i` - input gate + * `o` - output gate + * `f` - forget gate + * `c` - cell gate + * `t` - time step (t-1 means previous time step) + * `W[iofc]` - W parameter weight matrix for input, output, forget, and cell gates + * `R[iofc]` - R recurrence weight matrix for input, output, forget, and cell gates + * `Wb[iofc]` - W bias vectors for input, output, forget, and cell gates + * `Rb[iofc]` - R bias vectors for input, output, forget, and cell gates + * `P[iof]` - P peephole weight vector for input, output, and forget gates + * `WB[iofc]` - W parameter weight matrix for backward input, output, forget, and cell gates + * `RB[iofc]` - R recurrence weight matrix for backward input, output, forget, and cell gates + * `WBb[iofc]` - W bias vectors for backward input, output, forget, and cell gates + * `RBb[iofc]` - R bias vectors for backward input, output, forget, and cell gates + * `PB[iof]` - P peephole weight vector for backward input, output, and forget gates + * `H` - Hidden state + * `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + * Relu(x) - max(0, x) + * Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + * Sigmoid(x) - 1/(1 + e^{-x}) + + NOTE: Below are optional + + * Affine(x) - alpha*x + beta + * LeakyRelu(x) - x if x >= 0 else alpha * x + * ThresholdedRelu(x) - x if x >= alpha else 0 + * ScaledTanh(x) - alpha*Tanh(beta*x) + * HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + * Elu(x) - x if x >= 0 else alpha*(e^x - 1) + * Softsign(x) - x/(1 + |x|) + * Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh, h=Tanh): + + * it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi) + * ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf) + * ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc) + * Ct = ft (.) Ct-1 + it (.) ct + * ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Po (.) Ct + Wbo + Rbo) + * Ht = ot (.) h(Ct) + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings
+
A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
input_forget : int (default is 0)
+
Couple the input and forget gates if 1.
+
layout : int (default is 0)
+
The shape format of inputs X, initial_h, initial_c and outputs Y, Y_h, Y_c. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = initial_c.shape = Y_c.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = initial_c.shape = Y_c.shape = [batch_size, num_directions, hidden_size].
+
+ +#### Inputs (3 - 8) + +
+
X (differentiable) : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W (differentiable) : T
+
The weight tensor for the gates. Concatenation of `W[iofc]` and `WB[iofc]` (if bidirectional) along dimension 0. The tensor has shape `[num_directions, 4*hidden_size, input_size]`.
+
R (differentiable) : T
+
The recurrence weight tensor. Concatenation of `R[iofc]` and `RB[iofc]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 4*hidden_size, hidden_size]`.
+
B (optional, differentiable) : T
+
The bias tensor for input gate. Concatenation of `[Wb[iofc], Rb[iofc]]`, and `[WBb[iofc], RBb[iofc]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 8*hidden_size]`. Optional: If not specified - assumed to be 0.
+
sequence_lens (optional, non-differentiable) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional, non-differentiable) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
initial_c (optional, non-differentiable) : T
+
Optional initial value of the cell. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
P (optional, differentiable) : T
+
The weight tensor for peepholes. Concatenation of `P[iof]` and `PB[iof]` (if bidirectional) along dimension 0. It has shape `[num_directions, 3*hidde_size]`. Optional: If not specified - assumed to be 0.
+
+ +#### Outputs (0 - 3) + +
+
Y (optional, differentiable) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional, differentiable) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
Y_c (optional, differentiable) : T
+
The last output value of the cell. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **LpNormalization-22** + + Given a matrix, apply Lp-normalization along the provided axis. + The output is computed as: `output = input / Lp_norm(input, axis)`. + When the Lp norm is zero (i.e., all elements along the axis are zero), + the output is defined to be zero to avoid division by zero. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
The axis on which to apply normalization, -1 mean last axis.
+
p : int (default is 2)
+
The order of the normalization, only 1 or 2 are supported.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input matrix
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Matrix after normalization
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **LpPool-22** + + LpPool consumes an input tensor X and applies Lp pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + Lp pooling consisting of computing the Lp norm on all values of a subset + of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - {kernelSpatialShape}) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - {kernelSpatialShape}) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled `pad_shape[i]` is the sum of pads along axis `i`. + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - {kernelSpatialShape} + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + {kernelSpatialShape} - input_spatial_shape[i] + ``` + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
p : int (default is 2)
+
p value of the Lp norm used to pool over the input data.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **MaxPool-22** + + MaxPool consumes an input tensor X and applies max pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + max pooling consisting of computing the max on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape is calculated differently + depending on whether explicit padding is used, where pads is employed, or auto padding is used, where auto_pad is utilized. + With explicit padding (https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html?highlight=maxpool#torch.nn.MaxPool2d): + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled. `pad_shape[i]` is the sum of pads along axis `i`. Sliding windows that would start in the right padded region are ignored. + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following when ceil_mode is enabled: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + or when ceil_mode is disabled (https://www.tensorflow.org/api_docs/python/tf/keras/layers/AveragePooling2D): + ``` + VALID: output_spatial_shape[i] = floor((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i]) + 1 + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = floor((input_spatial_shape[i] - 1) / strides_spatial_shape[i]) + 1 + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i] + ``` + The output of each pooling window is maximum number of elements exclude pad. + + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
dilations : list of ints
+
Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
storage_order : int (default is 0)
+
The storage order of the tensor. 0 is row major, and 1 is column major. This attribute is used only to convert an n-tuple index value into a single integer value for producing the second output.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs (1 - 2) + +
+
Y (differentiable) : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
Indices (optional, non-differentiable) : I
+
Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x ... x Dn).
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(uint8)
+
Constrain input and output types to float and 8 bit tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **MaxRoiPool-22** + + ROI max pool consumes an input tensor X and region of interests (RoIs) to + apply max pooling across each RoI, to produce output 4-D tensor of shape + (num_rois, channels, pooled_shape[0], pooled_shape[1]). + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
pooled_shape : list of ints (required)
+
ROI pool output shape (height, width).
+
spatial_scale : float (default is 1.0)
+
Multiplicative spatial scale factor to translate ROI coordinates from their input scale to the scale used when pooling.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
+
rois (non-differentiable) : T
+
RoIs (Regions of Interest) to pool over. Should be a 2-D tensor of shape (num_rois, 5) given as [[batch_id, x1, y1, x2, y2], ...].
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
RoI pooled output 4-D tensor of shape (num_rois, channels, pooled_shape[0], pooled_shape[1]).
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **MaxUnpool-22** + + MaxUnpool essentially computes the partial inverse of the MaxPool op. + The input information to this op is typically the output information from a MaxPool op. The first + input tensor X is the tensor that needs to be unpooled, which is typically the pooled tensor (first output) + from MaxPool. The second input tensor, I, contains the indices to the (locally maximal) elements corresponding + to the elements in the first input tensor X. Input tensor I is typically the second output of the MaxPool op. + The third (optional) input is a tensor that specifies the output size of the unpooling operation. + + MaxUnpool is intended to do 'partial' inverse of the MaxPool op. 'Partial' because all the non-maximal + values from the original input to MaxPool are set to zero in the output of the MaxUnpool op. Pooling + the result of an unpooling operation should give back the original input to the unpooling op. + + MaxUnpool can produce the same output size for several input sizes, which makes unpooling op ambiguous. + The third input argument, output_size, is meant to disambiguate the op and produce output tensor of + known/predictable size. + + In addition to the inputs, MaxUnpool takes three attributes, namely kernel_shape, strides, and pads, + which define the exact unpooling op. The attributes typically have the same values as the corresponding + pooling op that the unpooling op is trying to invert. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X (differentiable) : T1
+
Input data tensor that has to be unpooled. This tensor is typically the first output of the MaxPool op.Dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non-image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
I (non-differentiable) : T2
+
Input data tensor containing the indices corresponding to elements in the first input tensor X.This tensor is typically the second output of the MaxPool op.Dimensions must be the same as input tensor X. The indices are linear, i.e. computed considering the tensor as flattened 1-D tensor, assuming row-major storage. Also, the linear indices should not consider padding. So the values in indices are in the range [0, N x C x D1 x ... x Dn).
+
output_shape (optional, non-differentiable) : T2
+
The shape of the output can be explicitly set which will cause pads values to be auto generated. If 'output_shape' is specified, 'pads' values are ignored.
+
+ +#### Outputs + +
+
output (differentiable) : T1
+
Output data tensor that contains the result of the unpooling.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T2 : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **Mish-22** + + Mish: A Self Regularized Non-Monotonic Neural Activation Function. + + Perform the linear unit element-wise on the input tensor X using formula: + + ``` + mish(x) = x * tanh(softplus(x)) = x * tanh(ln(1 + e^{x})) + ``` + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input X and output types to float tensors.
+
+ +### **Multinomial-22** + + Generate a tensor of samples from a multinomial distribution according to the probabilities + of each of the possible outcomes. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int (default is 6)
+
(Optional) The data type for the elements of the output tensor, if not specified, we will use int32.
+
sample_size : int (default is 1)
+
Number of times to sample.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
Input tensor with shape [batch_size, class_size], where class_size is the number of all possible outcomes. Each value along the axis zero represents the unnormalized log-probability of each corresponding outcome in a batch.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor with shape [batch_size, sample_size], where sample_size is the number of times to sample. Each value along the axis zero represents the outcome of the corresponding sample in a batch.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
T2 : tensor(int32), tensor(int64)
+
Constrain output types to integral tensors.
+
+ +### **NegativeLogLikelihoodLoss-22** + + A NegativeLogLikelihoodLoss operator computes (weighted) negative log likelihood loss. + Its "input" tensor has the shape of (N, C, d1, d2, ..., dk) where k >= 0. + The "input" tensor contains log-probabilities for input[n, :, d_1, d_2,..., d_k] being in a class of [0, C). + The operator's "target" input tensor has the shape of (N, d1, d2, ..., dk). It encodes class labels (one of C classes) + or it may contain a special value (indicated by an attribute ignore_index) for N x d1 x d2 x ... x dk samples. + The loss value for input[n, :, d_1, d_2,...d_k] being classified as class c = target[n][d_1][d_2]...[d_k] is computed as: + + ``` + loss[n][d_1][d_2]...[d_k] = -input[n][c][d_1][d_2]...[d_k]. + ``` + + When an optional "weight" is provided, the sample loss is calculated as: + + ``` + loss[n][d_1][d_2]...[d_k] = -input[n][c][d_1][d_2]...[d_k] * weight[c]. + ``` + + loss is zero for the case when target-value equals ignore_index. + + ``` + loss[n][d_1][d_2]...[d_k] = 0, when target[n][d_1][d_2]...[d_k] = ignore_index + ``` + + If "reduction" attribute is set to "none", the operator's output will be the above loss with shape (N, d1, d2, ..., dk). + If "reduction" attribute is set to "mean" (the default attribute value), the output loss is (weight) averaged: + + ``` + mean(loss), if "weight" is not provided, + ``` + + or if weight is provided, + + ``` + sum(loss) / sum(weight[target[n][d_1][d_2]...[d_k]]]), for all samples. + ``` + + If "reduction" attribute is set to "sum", the output is a scalar: `sum(loss)`. + + See also https://pytorch.org/docs/stable/nn.html#torch.nn.NLLLoss. + + Example 1: + + ``` + // negative log likelihood loss, "none" reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + + loss = np.zeros((N, d1)) + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] + + // print(loss) + // [[-3. -2.] + // [-0. -2.]] + ``` + + Example 2: + + ``` + // weighted negative log likelihood loss, sum reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + weight = [0.2, 0.3, 0.1] + loss = np.zeros((N, d1)) + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] * weight[c] + + loss = np.sum(loss) + // print(loss) + // -1.1 + ``` + + Example 3: + + ``` + // weighted negative log likelihood loss, mean reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + weight = [0.2, 0.3, 0.1] + loss = np.zeros((N, d1)) + weight_total = 0 + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] * weight[c] + weight_total = weight_total + weight[c] + + loss = np.sum(loss) / weight_total + // print(loss) + // -1.57 + ``` + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
ignore_index : int
+
Specifies a target value that is ignored and does not contribute to the input gradient. It's an optional value.
+
reduction : string (default is mean)
+
Type of reduction to apply to loss: none, sum, mean (default). 'none': the output is the loss for each sample. 'sum': the output will be summed. 'mean': the sum of the output will be divided by the sum of applied weights.
+
+ +#### Inputs (2 - 3) + +
+
input (differentiable) : T
+
Input tensor of shape (N, C) or (N, C, d1, d2, ..., dk).
+
target (non-differentiable) : Tind
+
Target tensor of shape (N) or (N, d1, d2, ..., dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
+
weight (optional, non-differentiable) : T
+
Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
+
+ +#### Outputs + +
+
loss (differentiable) : T
+
The negative log likelihood loss
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input, weight, and output types to floating-point tensors.
+
Tind : tensor(int32), tensor(int64)
+
Constrain target to integer types
+
+ +### **RNN-22** + + Computes an one-layer simple RNN. This operator is usually supported + via some custom implementation such as CuDNN. + + Notations: + + * `X` - input tensor + * `i` - input gate + * `t` - time step (t-1 means previous time step) + * `Wi` - W parameter weight matrix for input gate + * `Ri` - R recurrence weight matrix for input gate + * `Wbi` - W parameter bias vector for input gate + * `Rbi` - R parameter bias vector for input gate + * `WBi` - W parameter weight matrix for backward input gate + * `RBi` - R recurrence weight matrix for backward input gate + * `WBbi` - WR bias vectors for backward input gate + * `RBbi` - RR bias vectors for backward input gate + * `H` - Hidden state + * `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + * Relu(x) - max(0, x) + * Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + * Sigmoid(x) - 1/(1 + e^{-x}) + + NOTE: Below are optional + + * Affine(x) - alpha*x + beta + * LeakyRelu(x) - x if x >= 0 else alpha * x + * ThresholdedRelu(x) - x if x >= alpha else 0 + * ScaledTanh(x) - alpha*Tanh(beta*x) + * HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + * Elu(x) - x if x >= 0 else alpha*(e^x - 1) + * Softsign(x) - x/(1 + |x|) + * Softplus(x) - log(1 + e^x) + + Equations (Default: f=Tanh): + + * Ht = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Wbi + Rbi) + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings (default is ['Tanh', 'Tanh'])
+
One (or two if bidirectional) activation function for input gate. The activation function must be one of the activation functions specified above. Optional: Default `Tanh` if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
layout : int (default is 0)
+
The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size].
+
+ +#### Inputs (3 - 6) + +
+
X (differentiable) : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W (differentiable) : T
+
The weight tensor for input gate. Concatenation of `Wi` and `WBi` (if bidirectional). The tensor has shape `[num_directions, hidden_size, input_size]`.
+
R (differentiable) : T
+
The recurrence weight tensor. Concatenation of `Ri` and `RBi` (if bidirectional). The tensor has shape `[num_directions, hidden_size, hidden_size]`.
+
B (optional, differentiable) : T
+
The bias tensor for input gate. Concatenation of `[Wbi, Rbi]` and `[WBbi, RBbi]` (if bidirectional). The tensor has shape `[num_directions, 2*hidden_size]`. Optional: If not specified - assumed to be 0.
+
sequence_lens (optional, non-differentiable) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional, non-differentiable) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs (0 - 2) + +
+
Y (optional, differentiable) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional, differentiable) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ +### **RandomNormal-22** + + Generate a tensor with random values drawn from a normal distribution. The shape + of the tensor is specified by the `shape` argument and the parameter of the normal distribution + specified by `mean` and `scale`. + + The data type is specified by the 'dtype' argument. The 'dtype' argument must + be one of the data types specified in the 'DataType' enum field in the + TensorProto message. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int (default is 1)
+
The data type for the elements of the output tensor. Default is TensorProto::FLOAT.
+
mean : float (default is 0.0)
+
The mean of the normal distribution.
+
scale : float (default is 1.0)
+
The standard deviation of the normal distribution.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
shape : list of ints (required)
+
The shape of the output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor of random values drawn from normal distribution
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ +### **RandomNormalLike-22** + + Generate a tensor with random values drawn from a normal distribution. + The shape of the output tensor is copied from the shape of the input tensor, + and the parameters of the normal distribution are specified by `mean` and `scale`. + + The data type is specified by the 'dtype' argument, or copied from the input tensor if not provided. + The 'dtype' argument must be one of the data types specified in the 'DataType' enum field in the + TensorProto message, and be valid as an output type. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int
+
(Optional) The data type for the elements of the output tensor, if not specified, we will use the data type of the input tensor.
+
mean : float (default is 0.0)
+
The mean of the normal distribution.
+
scale : float (default is 1.0)
+
The standard deviation of the normal distribution.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
Input tensor to copy shape and optionally type information from.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of random values drawn from normal distribution
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type. If the dtype attribute is not provided this must be a valid output type.
+
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ +### **RandomUniform-22** + + Generate a tensor with random values drawn from a uniform distribution. The shape + of the tensor is specified by the `shape` argument and the range by `low` and `high`. + + The data type is specified by the 'dtype' argument. The 'dtype' argument must + be one of the data types specified in the 'DataType' enum field in the + TensorProto message. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int (default is 1)
+
The data type for the elements of the output tensor. If not specified, default is TensorProto::FLOAT.
+
high : float (default is 1.0)
+
Upper boundary of the output values.
+
low : float (default is 0.0)
+
Lower boundary of the output values.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
shape : list of ints (required)
+
The shape of the output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor of random values drawn from uniform distribution
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ +### **RandomUniformLike-22** + + Generate a tensor with random values drawn from a uniform distribution. + The shape of the output tensor is copied from the shape of the input tensor, + and the parameters of the uniform distribution are specified by `low` and `high`. + + The data type is specified by the 'dtype' argument, or copied from the input tensor if not provided. + The 'dtype' argument must be one of the data types specified in the 'DataType' enum field in the + TensorProto message and be valid as an output type. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int
+
(Optional) The data type for the elements of the output tensor, if not specified, we will use the data type of the input tensor.
+
high : float (default is 1.0)
+
Upper boundary of the output values.
+
low : float (default is 0.0)
+
Lower boundary of the output values.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
Input tensor to copy shape and optionally type information from.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of random values drawn from uniform distribution
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type. If the dtype attribute is not provided this must be a valid output type.
+
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ +### **RoiAlign-22** + + Region of Interest (RoI) align operation described in the + [Mask R-CNN paper](https://arxiv.org/abs/1703.06870). + RoiAlign consumes an input tensor X and region of interests (rois) + to apply pooling across each RoI; it produces a 4-D tensor of shape + (num_rois, C, output_height, output_width). + + RoiAlign is proposed to avoid the misalignment by removing + quantizations while converting from original image into feature + map and from feature map into RoI feature; in each ROI bin, + the value of the sampled locations are computed directly + through bilinear interpolation. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
coordinate_transformation_mode : string (default is half_pixel)
+
Allowed values are 'half_pixel' and 'output_half_pixel'. Use the value 'half_pixel' to pixel shift the input coordinates by -0.5 (the recommended behavior). Use the value 'output_half_pixel' to omit the pixel shift for the input (use this for a backward-compatible behavior).
+
mode : string (default is avg)
+
The pooling method. Two modes are supported: 'avg' and 'max'. Default is 'avg'.
+
output_height : int (default is 1)
+
default 1; Pooled output Y's height.
+
output_width : int (default is 1)
+
default 1; Pooled output Y's width.
+
sampling_ratio : int (default is 0)
+
Number of sampling points in the interpolation grid used to compute the output value of each pooled output bin. If > 0, then exactly sampling_ratio x sampling_ratio grid points are used. If == 0, then an adaptive number of grid points are used (computed as ceil(roi_width / output_width), and likewise for height). Default is 0.
+
spatial_scale : float (default is 1.0)
+
Multiplicative spatial scale factor to translate ROI coordinates from their input spatial scale to the scale used when pooling, i.e., spatial scale of the input feature map X relative to the input image. E.g.; default is 1.0f.
+
+ +#### Inputs + +
+
X : T1
+
Input data tensor from the previous operator; 4-D feature map of shape (N, C, H, W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
+
rois : T1
+
RoIs (Regions of Interest) to pool over; rois is 2-D input of shape (num_rois, 4) given as [[x1, y1, x2, y2], ...]. The RoIs' coordinates are in the coordinate system of the input image. Each coordinate set has a 1:1 correspondence with the 'batch_indices' input.
+
batch_indices : T2
+
1-D tensor of shape (num_rois,) with each element denoting the index of the corresponding image in the batch.
+
+ +#### Outputs + +
+
Y : T1
+
RoI pooled output, 4-D tensor of shape (num_rois, C, output_height, output_width). The r-th batch element Y[r-1] is a pooled feature map corresponding to the r-th RoI X[r-1].
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain types to float tensors.
+
T2 : tensor(int64)
+
Constrain types to int tensors.
+
+ +### **Round-22** + + Round takes one input Tensor and rounds the values, element-wise, meaning + it finds the nearest integer for each value. + In case of halves, the rule is to round them to the nearest even integer. + If input x is integral, +0, -0, NaN, or infinite, x itself is returned. + The output tensor has the same shape and type as the input. + + Examples: + ``` + round([0.9]) = [1.0] + round([2.5]) = [2.0] + round([2.3]) = [2.0] + round([1.5]) = [2.0] + round([-4.5]) = [-4.0] + ``` + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Selu-22** + + Selu takes one input data (Tensor) and produces one output data + (Tensor) where the scaled exponential linear unit function, + `y = gamma * (alpha * e^x - alpha) for x <= 0`, `y = gamma * x for x > 0`, + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.67326)
+
Coefficient of SELU default to 1.67326319217681884765625 (i.e., float32 approximation of 1.6732632423543772848170429916717).
+
gamma : float (default is 1.0507)
+
Coefficient of SELU default to 1.05070102214813232421875 (i.e., float32 approximation of 1.0507009873554804934193349852946).
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Sin-22** + + Calculates the sine of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The sine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Sinh-22** + + Calculates the hyperbolic sine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic sine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Softplus-22** + + Softplus takes one input data (Tensor) and produces one output data + (Tensor) where the softplus function, y = ln(exp(x) + 1), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Softsign-22** + + Calculates the softsign (x/(1+|x|)) of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The softsign (x/(1+|x|)) values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **Tan-22** + + Calculates the tangent of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The tangent of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **ThresholdedRelu-22** + + ThresholdedRelu takes one input data (Tensor) and produces one output data + (Tensor) where the rectified linear function, y = x for x > alpha, y = 0 otherwise, + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Threshold value
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +## Version 23 of the default ONNX operator set +### **Attention-23** + + Computes scaled dot product attention on query, key and value tensors, using an optional attention mask if passed. + + This operator covers self and cross variants of the attention operation based on sequence lengths of K, Q and V. + + For self attention, `kv_sequence_length` equals to `q_sequence_length`. + + For cross attention, query and key might have different lengths. + + This operator also covers the 3 following variants based on the number of heads: + 1) Multi-headed Attention (MHA): Described in the paper https://arxiv.org/pdf/1706.03762, `q_num_heads = kv_num_heads`. + 2) Group-query Attention (GQA): Described in the paper https://arxiv.org/pdf/2305.13245, `q_num_heads > kv_num_heads`, `q_num_heads % kv_num_heads == 0`. + 3) Multi-query Attention (MQA): Described in the paper https://arxiv.org/pdf/1911.02150, `q_num_heads > kv_num_heads`, `kv_num_heads=1`. + + Attention bias to be added is calculated based on `attn_mask` input and `is_causal` attribute: + 1) `attn_mask`: A boolean mask where a value of `True` indicates that the element should take part in attention or a float mask of the same type as query, key, value that is added to the attention score. + 2) If `is_causal` is set to `1`, causal masking is applied with bottom-right (offset-aware) alignment: query `i` attends key `j` iff `j <= i + past_sequence_length` (the count of cached keys in `past_key`); for a square Q/K this is the standard lower-triangular mask. The causal frontier is computed independently of the `attn_mask` input and is then composed with it additively, by summing their attention biases: a boolean `attn_mask` intersects the allowed set (its disallowed positions contribute `-inf` to the bias), while a float `attn_mask` is added to the attention scores rather than disabling positions. A fully-masked query row (every key's combined additive bias is `-inf`, e.g. an all-`False` boolean `attn_mask` row) produces a zero output row (matching prevailing runtime practice), not `NaN`. + + Errata (in-place behavioral correction, no opset bump): a fully-masked query row (e.g. an all-`False` boolean `attn_mask` row) now produces a zero output row instead of `NaN`, and the same zero-row guard applies to the mode-`3` `qk_matmul_output` debug output; this only replaces previously-`NaN` outputs. The mode-`3` `qk_matmul_output` is also now emitted at the operator's output precision (`T1`), matching the reference implementation, which affects only its dtype and only when `softmax_precision` differs from `T1`. No numerically useful, well-defined result of the released opset is otherwise changed. + + Both past and present state key/values are optional. They shall be used together, and not allowed to use only one of them. + The following pattern is applied to the Q, K and V inputs after appropriate reshaping of K and V inputs based on sequence lengths and num heads provided: + + ``` + The following pattern is applied by this operator: + Q K V + | | | + Q*sqrt(scale) K*sqrt(scale) | + | | | + | Transpose | + | | | + ---MatMul--- | + | | + softcap (if provided) | + | | + at_mask---Add | + | | + Softmax | + | | + -----MatMul------ + | + Y + ``` + + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
is_causal : int (default is 0)
+
If set to `1`, causal masking is applied with bottom-right (offset-aware) alignment: query `i` attends key `j` iff `j <= i + past_sequence_length` (the count of cached keys in `past_key`); for a square Q/K this is the standard lower-triangular mask.
+
kv_num_heads : int
+
Number of heads of key and value. Must be used with 3D inputs of Q, K and V.
+
q_num_heads : int
+
Number of heads of query. Must be used with 3D inputs of Q, K and V.
+
qk_matmul_output_mode : int (default is 0)
+
If set to `0`, qk_matmul_output is the output of qk matmul. If set to `1`, qk_matmul_output is the output after the softcap operation (before mask addition). If set to `2`, qk_matmul_output includes the attention mask and softcap (if provided) applied to the output of qk matmul. If set to `3`, qk_matmul_output is the output after the softmax operation. In mode `3`, a fully-masked query row (every key disallowed, e.g. an all-`False` boolean `attn_mask` row) is a zero row, consistent with the corresponding row of the primary output `Y`: the fully-masked-row guard is applied before this output is produced. The mode-`3` output is emitted at the operator's output precision (`T1`); when `softmax_precision` differs from `T1` this is a cast of the softmax result to `T1`. Default value is 0.
+
scale : float
+
Scaling factor applied to $Q*K^T$. Default value is `1/sqrt(head_size)`. To prevent [numerical overflow](https://tinyurl.com/sudb9s96), scale `Q`, `K` by `sqrt(scale)` before matmul.
+
softcap : float (default is 0.0)
+
Softcap value for attention weights. Default value is 0.
+
softmax_precision : int
+
The floating-point precision used in softmax computation. If softmax precision is not provided, the same precision as the input of softmax (Q and K) is used.
+
+ +#### Inputs (3 - 6) + +
+
Q : T1
+
Query tensor. 4D tensor with shape `(batch_size, q_num_heads, q_sequence_length, head_size)` or 3D tensor with shape `(batch_size, q_sequence_length, q_hidden_size)`. For cases with a 3D input tensor, `q_hidden_size = q_num_heads * head_size`
+
K : T1
+
Key tensor. 4D tensor with shape `(batch_size, kv_num_heads, kv_sequence_length, head_size)` or 3D tensor with shape `(batch_size, kv_sequence_length, k_hidden_size)`. For cases with a 3D input tensor, `k_hidden_size = kv_num_heads * head_size`
+
V : T2
+
Value tensor. 4D tensor with shape `(batch_size, kv_num_heads, kv_sequence_length, v_head_size)` or 3D tensor with shape `(batch_size, kv_sequence_length, v_hidden_size)`. For cases with a 3D input tensor, `v_hidden_size = kv_num_heads * v_head_size`
+
attn_mask (optional) : U
+
Attention mask. Shape must be broadcastable to 4D tensor with shape `(batch_size, q_num_heads, q_sequence_length, total_sequence_length)` where `total_sequence_length = past_sequence_length + kv_sequence_length.` Two types of masks are supported. A boolean mask where a value of `True` indicates that the element should take part in attention. Also supports a float mask of the same type as query, key, value that is added to the attention score.
+
past_key (optional) : T1
+
past state cache for key with shape `(batch_size, kv_num_heads, past_sequence_length, head_size)`
+
past_value (optional) : T2
+
past state cache for value with shape `(batch_size, kv_num_heads, past_sequence_length, v_head_size)`
+
+ +#### Outputs (1 - 4) + +
+
Y : T1
+
The output tensor . 4D tensor with shape `(batch_size, q_num_heads, q_sequence_length, v_head_size)` or 3D tensor with shape `(batch_size, q_sequence_length, hidden_size)`. For cases with a 3D input tensor, `hidden_size = q_num_heads * v_head_size`
+
present_key (optional) : T1
+
Updated key cache with shape `(batch_size, kv_num_heads, total_sequence_length, head_size)` where `total_sequence_length = past_sequence_length + kv_sequence_length`.
+
present_value (optional) : T2
+
Updated value cache with shape `(batch_size, kv_num_heads, total_sequence_length, v_head_size)` where `total_sequence_length = past_sequence_length + kv_sequence_length`.
+
qk_matmul_output (optional) : T1
+
The output of QK matmul. 4D tensor with shape `(batch_size, q_num_heads, q_sequence_length, total_sequence_length)` where `total_sequence_length = past_sequence_length + kv_sequence_length`.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain Q and K inputs types to float tensors.
+
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain V input types to float tensors.
+
U : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool)
+
Constrain output 'mask' types to boolean tensors and input types.
+
+ +### **Cast-23** + + The operator casts the elements of a given input tensor to a data type + specified by the 'to' argument and returns an output tensor of the same size in + the converted type. The 'to' argument must be one of the data types specified + in the 'DataType' enum field in the TensorProto message. + + Casting from string tensor in plain (e.g., "3.14" and "1000") and scientific numeric representations + (e.g., "1e-5" and "1E8") to float types is supported. For example, converting string "100.5" to an integer may + yield result 100. There are some string literals reserved for special floating-point values; + "+INF" (and "INF"), "-INF", and "NaN" are positive infinity, negative infinity, and not-a-number, respectively. + Any string which can exactly match "+INF" in a case-insensitive way would be mapped to positive infinite. Similarly, + this case-insensitive rule is applied to "INF" and "NaN". When casting from numeric tensors + to string tensors, plain floating-point representation (such as "314.15926") would be used. + Converting non-numerical-literal string such as "Hello World!" is an undefined behavior. Cases + of converting string representing floating-point arithmetic value, such as "2.718", to INT is an undefined behavior. + + Conversion from a numerical type to any numerical type is always allowed. + User must be aware of precision loss and value change caused by range difference between two types. + For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting + an integer 36 to Boolean may produce 1 because we truncate bits which can't be stored in the targeted type. + + In more detail, the conversion among numerical types should follow these rules + if the destination type is not a float 8 type. + + * Casting from floating point to: + * floating point: +/- infinity if OOR (out of range). + * fixed point: undefined if OOR. + * bool: +/- 0.0 to False; all else to True. + * Casting from fixed point to: + * floating point: +/- infinity if OOR. (+ infinity in the case of uint) + * fixed point: when OOR, discard higher bits and reinterpret (with respect to two's complement representation for + signed types). For example, 200 (int16) -> -56 (int8). + * bool: zero to False; nonzero to True. + * Casting from bool to: + * floating point: `{1.0, 0.0}`. + * fixed point: `{1, 0}`. + * bool: no change. + + Float 8 type were introduced to speed up the training of + deep models. By default the conversion of a float *x* obeys + to the following rules. `[x]` means the value rounded to + the target mantissa width. + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | -------- | -------- | -------- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | Inf | FLT_MAX | NaN | FLT_MAX | NaN | + | -Inf | -FLT_MAX | NaN | -FLT_MAX | NaN | + | \[x\] > FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | + | \[x\] \< -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | + | else | RNE | RNE | RNE | RNE | + + The behavior changes if the parameter 'saturate' is set to False. + The rules then become: + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | ------ | -------- | ---- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | -NaN | -NaN | NaN | -NaN | NaN | + | Inf | NaN | NaN | Inf | NaN | + | -Inf | -NaN | NaN | -Inf | NaN | + | \[x\] > FLT_MAX | NaN | NaN | Inf | NaN | + | \[x\] \< -FLT_MAX | NaN | NaN | -Inf | NaN | + | else | RNE | RNE | RNE | RNE | + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. All cases are fully described in two tables inserted in the operator description.
+
to : int (required)
+
The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor with the same shape as input with type specified by the 'to' argument
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **CastLike-23** + + The operator casts the elements of a given input tensor (the first input) to + the same data type as the elements of the second input tensor. + See documentation of the Cast operator for further details. + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. Please refer to operator Cast description for further details.
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
target_type (non-differentiable) : T2
+
The (first) input tensor will be cast to produce a tensor of the same type as this (second input) tensor.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor produced by casting the first input tensor to have the same type as the second input tensor.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **Constant-23** + + This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, + or value_* must be specified. + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
sparse_value : sparse_tensor
+
The value for the elements of the output tensor in sparse format.
+
value : tensor
+
The value for the elements of the output tensor.
+
value_float : float
+
The value for the sole element for the scalar, float32, output tensor.
+
value_floats : list of floats
+
The values for the elements for the 1D, float32, output tensor.
+
value_int : int
+
The value for the sole element for the scalar, int64, output tensor.
+
value_ints : list of ints
+
The values for the elements for the 1D, int64, output tensor.
+
value_string : string
+
The value for the sole element for the scalar, UTF-8 string, output tensor.
+
value_strings : list of strings
+
The values for the elements for the 1D, UTF-8 string, output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain input and output types to all tensor types.
+
+ +### **ConstantOfShape-23** + + Generate a tensor with given value and shape. + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
value : tensor
+
(Optional) The value of the output elements.Should be a one-element tensor. If not specified, it defaults to a tensor of value 0 and datatype float32
+
+ +#### Inputs + +
+
input : T1
+
1D tensor. The shape of the expected output tensor. If empty tensor is given, the output would be a scalar. All values must be >= 0.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of shape specified by 'input'.If attribute 'value' is specified, the value and datatype of the output tensor is taken from 'value'.If attribute 'value' is not specified, the value in the output defaults to 0, and the datatype defaults to float32.
+
+ +#### Type Constraints + +
+
T1 : tensor(int64)
+
Constrain input types.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint4), tensor(int4), tensor(bool), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(float4e2m1)
+
Constrain output types to be numerics or boolean.
+
+ +### **DequantizeLinear-23** + + The linear dequantization operator. It consumes a quantized tensor, a scale, and a zero point to compute the + full-precision tensor. The dequantization formula is `y = (x - x_zero_point) * x_scale`. `x_scale` and `x_zero_point` + must have the same shape, determining the quantization's granularity: a scalar for per-tensor/per-layer quantization, + a 1-D tensor for per-axis quantization, or have a rank identical to the input for blocked quantization. + See QuantizeLinear for details on quantization granularity. + + `x_zero_point` and `x` must have the same type. `x` and `y` must have the same shape. In the case of dequantizing + `int32`, there's no zero point (zero point is supposed to be 0). + `zero-point` is usually not used in the case of float8 and 4-bit types quantization, but the dequantization formula remains the same + for consistency. The output type is determined by the attribute `output_dtype`. If `output_dtype` is not supplied then the output type + is the same as `x_scale`. The output type also determines the precision of the multiplication operation. + + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`.
+
block_size : int (default is 0)
+
(Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given `x` shape `(D0, ..., Di, ..., Dn)`, `y_scale` shape `(S0, ... Si, ...Sn)` and `axis=i`, the accepted range is `[ceil(Di/Si), ceil(Di/(Si-1))-1]`
+
output_dtype : int (default is 0)
+
(Optional) The output data type. If not supplied, the output data type is inferred from `x_scale` data type (`T2`)
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D quantized input tensor to be de-quantized.
+
x_scale : T2
+
Scale for input `x`. For per-tensor/layer dequantization the scale is a scalar, for per per-axis dequantization it is a 1-D Tensor and for blocked dequantization it has the same shape as the input, except for one dimension in which blocking is performed.
+
x_zero_point (optional) : T1
+
Zero point for input `x`. Shape must match x_scale. It's optional. Zero point is 0 when it's not specified.
+
+ +#### Outputs + +
+
y : T3
+
N-D full precision output tensor. It has the same shape as input `x`. The data type is specified by the `output_dtype` attribute or, in its absence, the type of `x_scale`.
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(int32), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
The type of the inputs 'x_zero_point' and 'x'.
+
T2 : tensor(float), tensor(float16), tensor(bfloat16)
+
The type of the input 'x_scale'.
+
T3 : tensor(float), tensor(float16), tensor(bfloat16)
+
The type of the output 'y'.
+
+ +### **Flatten-23** + + Flattens the input tensor into a 2D matrix. If input tensor has shape + (d_0, d_1, ... d_n) then the output will have shape + (d_0 X d_1 ... d_(axis-1), d_axis X d_(axis+1) ... X dn). + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 ... d_n), where the shape of the input tensor is (d_0, d_1, ... d_n).
+
+ +#### Inputs + +
+
input (differentiable) : T
+
A tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain input and output to all tensor types up to IRv10.
+
+ +### **Identity-23** + + Identity operator + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : V
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : V
+
Tensor to copy input into.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain input and output types to all tensor, sequence, and optional types.
+
+ +### **If-23** + + If conditional + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
else_branch : graph (required)
+
Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
+
then_branch : graph (required)
+
Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
+
+ +#### Inputs + +
+
cond : B
+
Condition for the if. The tensor must contain a single element.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : V
+
Values that are live-out to the enclosing scope. The return values in the `then_branch` and `else_branch` must be of the same data type. The `then_branch` and `else_branch` may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if-node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of `then_branch` is typed float tensor with shape [2] and the first output of `else_branch` is another float tensor with shape [3], If's first output should have (a) no shape set, or (b) a shape of rank 1 with neither `dim_value` nor `dim_param` set, or (c) a shape of rank 1 with a unique `dim_param`. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), seq(tensor(float4e2m1)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4)), optional(tensor(float4e2m1))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv11.
+
B : tensor(bool)
+
Only bool
+
+ +### **Loop-23** + + Generic Looping construct. This loop has multiple termination conditions: + + 1) Trip count. Iteration count specified at runtime. Set by + specifying the input M. Optional. Set to empty string to omit. + Note that a static trip count (specified at graph construction time) can be + specified by passing in a constant node for input M. + 2) Loop termination condition. This is an input to the op that determines + whether to run the first iteration and also a loop-carried dependency for + the body graph. The body graph must yield a value for the condition variable, + whether this input is provided or not. + + This table summarizes the operating modes of this operator with equivalent + C-style code: + + Operator inputs defined as (max_trip_count, condition_var). + + * input ("", ""): + for (int i=0; ; ++i) { + cond = ... // Note this value is ignored, but is required in the body + } + + * input ("", cond) // Note this is analogous to a while loop + bool cond = ...; + for (int i=0; cond; ++i) { + cond = ...; + } + + * input ("", 1) // Note this is analogous to a do-while loop + bool cond = true + for (int i=0; cond; ++i) { + cond = ...; + } + + * input (trip_count, "") // Note this is analogous to a for loop + int trip_count = ... + for (int i=0; i < trip_count; ++i) { + cond = ...; // ignored + } + + * input (trip_count, cond) + int trip_count = ...; + bool cond = ...; + for (int i=0; i < trip_count && cond; ++i) { + cond = ...; + } + + + *Sample usage - cond as well as trip count* + + graph predict-net { + %a = Constant[value = ]() + %b = Constant[value = ]() + %keepgoing = Constant[value = ]() + %max_trip_count = Constant[value = ]() + %keepgoing_out, %b_out, %user_defined_vals = Loop[body = ](%max_trip_count, %keepgoing, %b) + return + } + + graph body-net ( + %i[INT32, scalar] // iteration number + %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used + %b_in[INT32, scalar] // incoming value of loop-carried-dependency b + ) { + %my_local = Add(%a, %b_in) + %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b + %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition + %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated + return %keepgoing_out, %b_out, %user_defined_val + } + + *Sample equivalent C code* + + { + /* User-defined code (enclosing scope) */ + int a = 3, b = 6; + bool keepgoing = true; // Analogous to input cond + /* End user-defined code */ + + /* Implicitly-defined code */ + const int max_trip_count = 10; // Analogous to input M + int user_defined_vals[]; // Imagine this is resizable + /* End implicitly-defined code */ + /* initialize loop-carried variables and scan-output variables */ + bool keepgoing_out = keepgoing + int b_out = b + + for (int i=0; i < max_trip_count && keepgoing_out; ++i) { + /* Implicitly-defined code: bind actual parameter values + to formal parameter variables of loop-body */ + bool keepgoing_in = keepgoing_out; + bool b_in = b_out; + + /* User-defined code (loop body) */ + int my_local = a + b_in; // Reading value "a" from the enclosing scope is fine + b_out = a - b_in; + keepgoing_out = my_local > b_out; + user_defined_val = b_in + b_in; // b_in and b_out are different variables + /* End user-defined code */ + + /* Implicitly defined-code */ + user_defined_vals[i] = user_defined_val // accumulate scan-output values + } + // int t = my_local; // Can't do this. my_local is not accessible here. + + // The values below are bound to the output variables of the loop and therefore accessible + // b_out; user_defined_vals; keepgoing_out; + } + + There are several things of note in this code snippet: + + 1) Values from the enclosing scope (i.e. variable "a" here) are in scope and can + be referenced in the inputs of the loop. + 2) Any values computed in the loop body that needs to be used in a subsequent + iteration or after the loop are modeled using a pair of variables in the loop-body, + consisting of an input variable (eg., b_in) and an output variable (eg., b_out). + These are referred to as loop-carried dependences. The loop operation node + supplies the input value of the input variable for the first iteration, and + returns the output value of the output variable produced by the final + iteration. + 3) Scan_output variables are used to implicitly concatenate values computed across + all the iterations. In the above example, the value of user_defined_val computed + over all iterations are concatenated and returned as the value of user_defined_vals + after the loop. + 4) Values created in the body cannot be accessed in the enclosing scope, + except using the mechanism described above. + + Note that the semantics of this op support "diagonal" or "wavefront" execution. + (See Step 3 here for an example: + https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). + Frontends should emit multi-layer RNNs as a series of While operators (with + time being the inner looping dimension), with each successive layer consuming + the scan_outputs from the previous layer, possibly going through several + point-wise operators (e.g. dropout, residual connections, linear layer). + + The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order. + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies...). It has 1+N+K outputs: (condition, loop carried dependencies..., scan_outputs...). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
+
+ +#### Inputs (2 - ∞) + +
+
M (optional) : I
+
A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
+
cond (optional) : B
+
A boolean termination condition. Optional. Pass empty string to skip.
+
v_initial (variadic, heterogeneous) : V
+
The initial values of any loop-carried dependencies (values that change across loop iterations)
+
+ +#### Outputs (1 - ∞) + +
+
v_final_and_scan_outputs (variadic, heterogeneous) : V
+
Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), seq(tensor(float4e2m1)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4)), optional(tensor(float4e2m1))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv11.
+
I : tensor(int64)
+
tensor of int64, which should be a scalar.
+
B : tensor(bool)
+
tensor of bool, which should be a scalar.
+
+ +### **Pad-23** + + Given a tensor containing the data to be padded (`data`), a tensor containing the number of start and end pad values for axis (`pads`), (optionally) a `mode`, and (optionally) `constant_value`, + a padded tensor (`output`) is generated. + + The four supported `modes` are (similar to corresponding modes supported by `numpy.pad`): + + 1) `constant`(default) - pads with a given constant value as specified by `constant_value` (which defaults to 0, empty string, or False) + + 2) `reflect` - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis + + 3) `edge` - pads with the edge values of array + + 4) `wrap` - wrap-around padding as if the data tensor forms a torus + + + Example 1 (`constant` mode): + + Insert 0 pads to the beginning of the second dimension. + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'constant' + + constant_value = 0.0 + + output = [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ] + ``` + + Example 2 (`reflect` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'reflect' + + output = [ + [1.0, 1.2, 1.0, 1.2], + [2.3, 3.4, 2.3, 3.4], + [4.5, 5.7, 4.5, 5.7], + ] + ``` + + Example 3 (`edge` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'edge' + + output = [ + [1.0, 1.0, 1.0, 1.2], + [2.3, 2.3, 2.3, 3.4], + [4.5, 4.5, 4.5, 5.7], + ] + ``` + + Example 4 (`wrap` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [2, 1, 1, 1] + + mode = 'wrap' + + output = [ + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + ] + ``` + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is constant)
+
Supported modes: `constant`(default), `reflect`, `edge`, `wrap`
+
+ +#### Inputs (2 - 4) + +
+
data (differentiable) : T
+
Input tensor.
+
pads (non-differentiable) : tensor(int64)
+
Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. `pads` should be a 1D tensor of shape [2 * num_axes] where `num_axes` refers to the number of elements in the `axes` input or the input rank if `axes` are not provided explicitly. `pads` format should be: [x1_begin, x2_begin, ..., x1_end, x2_end,...], where xi_begin is the number of pad values added at the beginning of axis `axes[i]` and xi_end, the number of pad values added at the end of axis `axes[i]`.
+
constant_value (optional, non-differentiable) : T
+
(Optional) A scalar value to be used if the mode chosen is `constant` (by default it is 0, empty string or False).
+
axes (optional, non-differentiable) : Tind
+
1-D tensor of axes that `pads` apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated. If not provided, all axes are assumed (`[0, 1, ..., input_rank-1]`).
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain input and output types to all tensor types up to IRv11.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **QuantizeLinear-23** + + The linear quantization operator consumes a high-precision tensor, a scale, and a zero point to compute the + low-precision/quantized tensor. The scale factor and zero point must have the same shape, determining the quantization + granularity. The quantization formula is `y = saturate((x / y_scale) + y_zero_point)`. + + Saturation is done according to: + - uint16: [0, 65535] + - int16: [-32768, 32767] + - uint8: [0, 255] + - int8: [-128, 127] + - uint4: [0, 15] + - int4: [-8, 7] + + For `(x / y_scale)`, it rounds to the nearest even. Refer to https://en.wikipedia.org/wiki/Rounding for details. + + `y_zero_point` and `y` must have the same type. `y_zero_point` is usually not used for quantization to float8 and 4bit types, but the quantization + formula remains the same for consistency, and the type of the attribute `y_zero_point` still determines the quantization type. + `x` and `y_scale` are allowed to have different types. The type of `y_scale` determines the precision of the division operation between `x` and + `y_scale`, unless the `precision` attribute is specified. + + There are three supported quantization granularities, determined by the shape of `y_scale`. + In all cases, `y_zero_point` must have the same shape as `y_scale`. + - Per-tensor (per-layer) quantization: `y_scale` is a scalar. + - Per-axis quantization: The scale must be a 1-D tensor, with the length of the quantization axis. For an input shape + `(D0, ..., Di, ..., Dn)` and `axis=i`, `y_scale` is a 1-D tensor of length `Di`. + - Blocked quantization: The scale's shape is identical to the input's shape, except for one dimension, in which + blocking is performed. Given `x` shape `(D0, ..., Di, ..., Dn)`, `axis=i`, and block size `B`: `y_scale` shape is + `(D0, ..., ceil(Di/B), ..., Dn)`. + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used only for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`. When the rank of the input is 1, per-tensor quantization is applied, rendering the axis unnecessary in this scenario.
+
block_size : int (default is 0)
+
(Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given `x` shape `(D0, ..., Di, ..., Dn)`, `y_scale` shape `(S0, ... Si, ...Sn)` and `axis=i`, the accepted range is `[ceil(Di/Si), ceil(Di/(Si-1))-1]`
+
output_dtype : int (default is 0)
+
(Optional) The output data type. If not supplied, the output data type is inferred from `y_zero_point` data type (`T3`). If neither `output_dtype` nor `y_zero_point` are supplied, output data type is uint8. If both `output_dtype` and `y_zero_point` are specified, `output_dtype` must be `T3`.
+
precision : int (default is 0)
+
(Optional) The precision of the division operation between `x` and `y_scale`. If not provided, it will be the same as the type of `y_scale`.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 quantization (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. All cases are fully described in two tables inserted in the operator description.
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D full precision Input tensor to be quantized.
+
y_scale : T2
+
Scale for doing quantization to get `y`. For per-tensor/layer quantization the scale is a scalar, for per-axis quantization it is a 1-D Tensor and for blocked quantization it has the same shape as the input, except for one dimension in which blocking is performed.
+
y_zero_point (optional) : T3
+
Zero point for doing quantization to get `y`. Shape must match `y_scale`.Default is uint8 with zero point of 0 if it's not specified.
+
+ +#### Outputs + +
+
y : T3
+
N-D quantized output tensor. It has same shape as input `x`.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(float16), tensor(bfloat16), tensor(int32)
+
The type of the input 'x'.
+
T2 : tensor(float), tensor(float16), tensor(bfloat16), tensor(int32)
+
The type of the input 'y_scale'.
+
T3 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
The type of the input `y_zero_point` and the output `y`.
+
+ +### **RMSNormalization-23** + + This is RMS normalization defined in ONNX as function as described in the paper https://arxiv.org/pdf/1910.07467. + The overall computation can be split into two stages. The root mean squared norm is taken over the last D dimensions, + where D is the dimension of normalized_shape. For example, if normalized_shape is (3, 5) (a 2-dimensional shape), + the rms norm is computed over the last 2 dimensions of the input. The computation required by standardization can be + described by the following equations. + ``` + XSquared = Mul(X, X) + XSquaredMean = ReduceMean(XSquared) + MeanSquareEpsilon = Add(XSquaredMean, epsilon) + RMS = Sqrt(MeanSquareEpsilon) + Normalized = Div(X, RMS) + ``` + where `normalized_axes` is `[axis, ..., rank of X - 1]`. The variables `RMS` stand for root mean square, + Depending on `stash_type` attribute, the actual computation + must happen in different floating-point precision. + For example, if `stash_type` is 1, this operator casts + all input variables to 32-bit float, perform the computation, and + finally cast `Normalized` back to the original type of `X`. + The second stage then scales the outcome of the first stage using: + ``` + Y= Mul(Normalized, Scale) + ``` + Let `d[i]` indicate the i-th dimension of `X`. + If `X`'s shape is `[d[0], ..., d[axis-1], d[axis], ..., d[rank-1]]`, + the shape of `RMS` is `[d[0], ..., d[axis-1], 1, ..., 1]`. + `Y` and `X` have the same shape. This operator supports unidirectional broadcasting + (`Scale` should be unidirectional broadcastable to tensor `X`); + for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
The first normalization dimension. If rank(X) is r, axis' allowed range is [-r, r). Negative value means counting dimensions from the back.
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
stash_type : int (default is 1)
+
The floating-point precision used in stage one of the computation.
+
+ +#### Inputs + +
+
X : T
+
The input tensor to be normalized. In general, the shape is (D1, D2, ... , Dn) for n-dimensional data, where the root mean squared norm is taken over the last D dimensions, D is determined by the axis attribute.
+
scale : V
+
Scale tensor. Scale tensor shape should be broadcastable to the normalized shape.
+
+ +#### Outputs + +
+
Y : V
+
Output data tensor. Same shape as X
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input X type to float tensors.
+
V : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain output Y and scale type to float tensors.
+
+ +### **Reshape-23** + + Reshape the input tensor similar to numpy.reshape. + First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. + At most one dimension of the new shape can be -1. In this case, the value is + inferred from the size of the tensor and the remaining dimensions. A dimension + could also be 0, in which case the actual dimension value is unchanged (i.e. taken + from the input tensor). If 'allowzero' is set, and the new shape includes 0, the + dimension will be set explicitly to zero (i.e. not taken from input tensor). + Shape (second input) could be an empty shape, which means converting to a scalar. + The input tensor's shape and the output tensor's shape are required to have the same number of elements. + + If the attribute 'allowzero' is set, it is invalid for the specified shape to + contain both a zero value and -1, as the value of the dimension corresponding + to -1 cannot be determined uniquely. + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
allowzero : int (default is 0)
+
(Optional) By default, when any value in the 'shape' input is equal to zero the corresponding dimension value is copied from the input tensor dynamically. allowzero=1 indicates that if any value in the 'shape' input is set to zero, the zero value is honored, similar to NumPy.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
shape (non-differentiable) : tensor(int64)
+
Specified shape for output.
+
+ +#### Outputs + +
+
reshaped (differentiable) : T
+
Reshaped data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain input and output types to all tensor types.
+
+ +### **RotaryEmbedding-23** + + RotaryEmbedding is the implementation of rotary positional embeddings (RoPE) based on the paper https://arxiv.org/pdf/2104.09864. + The key advantage of RoPE is that it allows the model to understand both the absolute position of a token and the relative distances + between tokens. This is achieved through a rotational mechanism where the extent of rotation is computed based on the token's absolute position (position_ids). + + The rotational mechanism is defined by sine and cosine functions that are used to represent the rotation angles. + For each token in the sequence, its positional embedding is computed by rotating its embedding vector. This is done by splitting the + embedding vector either into two halves or interleaving every alternate token and applying the rotation matrix to each half of the embedding vector. + The rotation matrix is parameterized by the token's position in the sequence. The rotated halves of the embedding vector are concatenated + to form the final positional embedding for each token. The rotated positional embeddings are used in the self-attention mechanism. + The rotation ensures that the model captures both absolute and relative positional information. + + Rotary embeddings are defined using the following algorithm: + + ```python + def rotary_embedding( + input: np.ndarray, + cos_cache: np.ndarray, + sin_cache: np.ndarray, + position_ids: np.ndarray | None = None, + interleaved=None, + rotary_embedding_dim=None, + num_heads=None, + ) -> np.ndarray: + original_input_shape = input.shape + # First ensure input to be processed has shape [batch_size, seq_len, num_heads, head_size] + if len(input.shape) == 4: + input = np.transpose(input, (0, 2, 1, 3)) + batch_size = input.shape[0] + sequence_length = input.shape[1] + if len(input.shape) == 3: + hidden_size = input.shape[2] + assert num_heads != 0 + head_size = int(hidden_size / num_heads) + new_shape = [batch_size, sequence_length, num_heads, head_size] + input = np.reshape(input, new_shape) + assert len(input.shape) == 4 + head_size = input.shape[3] + + # Fully or partially perform rotation on input based on rotary_embedding_dim attribute + if rotary_embedding_dim is None or rotary_embedding_dim == 0: + # If rotary_embedding_dim not provided, perform full rotation by using head_size + rotary_embedding_dim = head_size + x_rotate = input[:, :, :, :rotary_embedding_dim] + x_not_rotate = input[:, :, :, rotary_embedding_dim:] + rotary_embedding_dim_half = int(rotary_embedding_dim / 2) + + # Retrieve sin and cos caches using position ids + if position_ids is not None: + cos_cache = cos_cache[ + position_ids + ] # Shape: [batch_size, sequence_length, rotary_embedding_dim/2] + sin_cache = sin_cache[ + position_ids + ] # Shape: [batch_size, sequence_length, rotary_embedding_dim/2] + + # Shape: [batch_size, sequence_length, rotary_embedding_dim/2] + if cos_cache.shape[-1] != rotary_embedding_dim_half: + raise ValueError( + f"Last dimension of cos cache ({cos_cache.shape[-1]}) does not match rotary_embedding_dim/2 ({rotary_embedding_dim_half})." + ) + if sin_cache.shape[-1] != rotary_embedding_dim_half: + raise ValueError( + f"Last dimension of sin cache ({sin_cache.shape[-1]}) does not match rotary_embedding_dim/2 ({rotary_embedding_dim_half})." + ) + + cos_cache = np.expand_dims( + cos_cache, axis=2 + ) # Shape: [batch_size, sequence_length, 1, rotary_embedding_dim/2] + sin_cache = np.expand_dims( + sin_cache, axis=2 + ) # Shape: [batch_size, sequence_length, 1, rotary_embedding_dim/2] + + # Either divide the input in halves or interleave (based on interleaved attribute) + if interleaved: + x1 = x_rotate[:, :, :, 0::2] + x2 = x_rotate[:, :, :, 1::2] + else: + x1, x2 = np.split(x_rotate, 2, axis=-1) + + # Calculate real and imaginary values + real = (cos_cache * x1) - (sin_cache * x2) + imag = (sin_cache * x1) + (cos_cache * x2) + + # Inserted rotated embeddings back to the original input + if interleaved: + # x_rotate[:, :, :, 0::2] = real + # x_rotate[:, :, :, 1::2] = imag + real = np.expand_dims(real, axis=-1) + imag = np.expand_dims(imag, axis=-1) + x_rotate_concat = np.concatenate((real, imag), axis=-1) + x_rotate = np.reshape(x_rotate_concat, x_rotate.shape) + else: + x_rotate = np.concatenate((real, imag), axis=-1) + output = np.concatenate((x_rotate, x_not_rotate), axis=-1) + if len(original_input_shape) == 3: + output = np.reshape(output, original_input_shape) + else: + output = np.transpose(output, (0, 2, 1, 3)) + return output + ``` + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
interleaved : int (default is 0)
+
Rotate using interleaved pattern. Default value is 0 (False).
+
num_heads : int
+
Number of attention heads. Must be provided when input is a 3D tensor.
+
rotary_embedding_dim : int (default is 0)
+
Rotary embedding dimension used to apply partial rotary embeddings.
+
+ +#### Inputs (3 - 4) + +
+
X : T
+
The input tensor representing the token embeddings. 4D tensor with shape `(batch_size, num_heads, sequence_length, head_size)` or 3D tensor with shape `(batch_size, sequence_length, hidden_size)`. For cases with a 4D input tensor, `head_size` has to be even. For cases with a 3D input tensor, `num_heads` attribute must be provided and `hidden_size` must be an even multiple of `num_heads` where `hidden_size = num_heads * head_size`
+
cos_cache : T
+
The cosine values for the rotation. 2D tensor with shape `(max_position_id_plus_1, head_size / 2)` for full rotation or `(max_position_id_plus_1, rotary_embedding_dim / 2)` for partial rotation when `position_ids` are provided. 3D tensor with shape `(batch_size, sequence_length, head_size / 2)` for full rotation or `(batch_size, sequence_length, rotary_embedding_dim / 2)` for partial rotation when `position_ids` are not provided. `max_position_id_plus_1` is a parameter to the model.
+
sin_cache : T
+
The sine values for the rotation. 2D tensor with shape `(max_position_id_plus_1, head_size / 2)` for full rotation or `(max_position_id_plus_1, rotary_embedding_dim / 2)` for partial rotation when `position_ids` are provided. 3D tensor with shape `(batch_size, sequence_length, head_size / 2)` for full rotation or `(batch_size, sequence_length, rotary_embedding_dim / 2)` for partial rotation when `position_ids` are not provided. `max_position_id_plus_1` is a parameter to the model.
+
position_ids (optional) : M
+
The position indices for the tokens. 2D tensor with shape `(batch_size, sequence_length)`
+
+ +#### Outputs + +
+
Y : T
+
Tensor with same shape as input.
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(float16), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
M : tensor(int64)
+
Constrain input and output types to integer tensors.
+
+ +### **Scan-23** + + Scan can be used to iterate over one or more scan_input tensors, + constructing zero or more scan_output tensors. It combines ideas from general recurrences, + functional programming constructs such as scan, fold, map, and zip, and is intended to enable + generalizations of RNN-like constructs for sequence-to-sequence processing. + Other tensors (referred to as state_variables here) can be used to carry a state + when iterating from one element to another (similar to hidden-state in RNNs, also referred + to as loop-carried dependences in the context of loops). + Many common usages involve a single scan_input tensor (where functionality + similar to scan, fold and map can be obtained). When more than one scan_input is used, + a behavior similar to zip is obtained. + + The attribute body must be a graph, specifying the computation to be performed in + every iteration. It takes as input the current values of the state_variables and + the current iterated element of the scan_inputs. It must return the (updated) values + of the state_variables and zero or more scan_output_element tensors. The values of the + scan_output_element tensors are concatenated over all the iterations to produce the + scan_output values of the scan construct (similar to the concatenated intermediate + hidden-state values of RNN-like constructs). All the output tensors (state_variables as + well as scan_output_element tensors) are required to have the same shape in each iteration + of the loop (a restriction imposed to enable efficient memory allocation). + + Note that the iterated element passed to the body subgraph does not have a sequence + axis. It will have a rank one less than the rank of the corresponding scan_input. + + The scan operation returns the final values of the state_variables as well as the + scan_outputs. + + The optional attribute scan_input_directions specifies the direction (forward or backward) + for each scan input. If this attribute is omitted, all sequences are scanned in the forward + direction. A bidirectional scan may be performed by specifying the same tensor input twice + in the scan_inputs, once with a forward direction, and once with a backward direction. + + The scan_output of the operation is produced by concatenating the scan_output_element + values produced by the body in each iteration. The optional attribute scan_output_directions + specifies the direction in which scan_output is constructed (by appending or prepending the + scan_output_element to scan_output in each iteration) for each scan_output. If this attribute + is omitted, the scan_output_element is appended to the scan_output in each iteration. + + The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. + If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the + batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. + Note that scanning a non-zero axis may be less efficient than scanning axis zero. + + The optional attribute scan_output_axes specifies the axis along which the scan_outputs + are accumulated for each scan_output. For example, if axis 1 is the time axis (to be + scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis + value of 1. + + Note that because of the ONNX restriction that only the last parameter of an operator can + be variadic, the initial-states and scan-inputs are listed together as one input parameter. + Similarly, the final-states and scan-outputs are listed together as one output parameter. + The attribute num_scan_inputs indicates the number M of scan-inputs. + + The behavior of + + Scan < + num_scan_inputs = m, + body = loop-body, + scan_input_axes = [axis_1, ..., axis_m] + > (init_1, ..., init_n, scan_1, ..., scan_m) + + is equivalent to the following pseudo-code: + + // scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i + // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. + sequence_length = scan_1.shape[axis_1]; + + // initialize state-variables + st_1 = init_1; ... st_n = init_n; + // initialize scan-output variables: [] denotes an empty tensor + scan_out_1 = []; ...; scan_out_k = []; + // identify number of iterations: + + // execute loop + for (int t = 0; t < sequence_length; ++t) { + // generate the scan-input elements: the notation T[t] indicates the sub-tensor + // of rank one less than T obtained by indexing T at position t along axis k. + si_1 = scan_1[t]; + ... ; + si_m = scan_m[t]; + // execute loop-body + st_1, ..., st_n, so_1, ..., so_k = loop-body(st_1, ..., st_n, si_1, ..., si_m) + // accumulate the scan-output elements + scan_out_1 = Concat(scan_out_1, so_1); ... ; scan_out_k = Concat(scan_out_k, so_k); + } + + return st_1, ..., st_n, scan_out_1, ..., scan_out_k; + + *Sample usage: Encoding RNN using a Scan* + + The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, + recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can + be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes + %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these + values are computed in the outer graph, they need to be passed in as extra state_variables. + + graph rnn-encoding { + %H_0 = ... + %X = ... + %Y_h, %Y = Scan[body = , num_scan_inputs=1](%H_0, %X) + return %Y, %Y_h + } + + graph rnn-cell-1 ( + %H_tminus1[FLOAT, tensor] + %X_t[FLOAT, tensor] + ) { + %Wi = ... + %Ri = ... + %Wbi = ... + %Rbi = ... + %t1 = X_t * (Wi^T) + %t2 = H_tminus1*(Ri^T) + %t3 = Add(%t1, %t2) + %t4 = Add(%t3, %Wbi) + %t5 = Add(%t4, %Rbi) + %Ht = Tanh(%t5) + %Accumulate = Identity(%Ht) + return %Ht, %Accumulate + } + + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has N+M inputs: (loop state variables..., scan_input_elts...). It has N+K outputs: (loop state variables..., scan_output_elts...). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
+
num_scan_inputs : int (required)
+
An attribute specifying the number of scan_inputs M.
+
scan_input_axes : list of ints
+
An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
scan_input_directions : list of ints
+
An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
+
scan_output_axes : list of ints
+
An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
+
scan_output_directions : list of ints
+
An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
+
+ +#### Inputs (1 - ∞) + +
+
initial_state_and_scan_inputs (variadic, heterogeneous) : V
+
Initial values of the loop's N state variables followed by M scan_inputs
+
+ +#### Outputs (1 - ∞) + +
+
final_state_and_scan_outputs (variadic, heterogeneous) : V
+
Final values of the loop's N state variables followed by K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
All Tensor types up to IRv11.
+
+ +### **Shape-23** + + Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. + Optional attributes start and end can be used to compute a slice of the input tensor's shape. + If start axis is omitted, the slice starts from axis 0. + The end axis, if specified, is exclusive (and the returned value will not include the size of that axis). + If the end axis is omitted, the axes upto the last one will be included. + Negative axes indicate counting back from the last axis. + Note that axes will be clamped to the range [0, r], where r is the + rank of the input tensor if they are out-of-range (after adding r in the case of + negative axis). Thus, specifying any end value > r is equivalent to specifying an end + value of r, and specifying any start value < -r is equivalent to specifying a start + value of 0. If start > end, the result will be an empty shape. + + Examples: + + ``` + Input tensor with shape: [2, 3, 4] + No attributes specified. + Output: [2, 3, 4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: -1 + Output: [4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + end: -1 + Output: [2, 3] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: 1 + end: 2 + Output: [3] + ``` + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
end : int
+
(Optional) Ending axis for slicing the shape. Negative value means counting dimensions from the back. If omitted, sizes of all axes upto (including) the last one will be included.
+
start : int (default is 0)
+
(Optional) Starting axis for slicing the shape. Default value is 0.Negative value means counting dimensions from the back.
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
shape (non-differentiable) : T1
+
Shape of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor.
+
+ +### **Size-23** + + Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor. + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
size (non-differentiable) : T1
+
Total number of elements of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor, which should be a scalar though.
+
+ +### **Squeeze-23** + + Remove single-dimensional entries from the shape of a tensor. + Takes an input `axes` with a list of axes to squeeze. + If `axes` is not provided, all the single dimensions will be removed from + the shape. If an axis is selected with shape entry not equal to one, an error is raised. + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
Tensors with at least max(dims) dimensions.
+
axes (optional, non-differentiable) : tensor(int64)
+
1D tensor of integers indicating the dimensions to squeeze. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
squeezed (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain input and output types to all tensor types up to IRv11.
+
+ +### **Transpose-23** + + Returns a transpose of the input tensor. (Similar to `numpy.transpose`). + The optional attribute `perm` must be a permutation of the dimensions of + the input tensor. Axis `i` of the output tensor corresponds to the axis + `perm[i]` of the input tensor. + For example, when perm=(1, 0, 2), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 1, 3). + When perm=(1, 2, 0), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 3, 1). + If the attribute `perm` is omitted, its default value is `(n-1, ..., 0)`, + where `n` is the rank of the input tensor. + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
perm : list of ints
+
A list of integers. By default, reverse the dimensions, otherwise permute the axes according to the values given. Its length must be equal to the rank of the input.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
transposed (differentiable) : T
+
Transposed output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain input and output types to all tensor types.
+
+ +### **Unsqueeze-23** + + Insert single-dimensional entries to the shape of an input tensor (`data`). + Takes one required input `axes` - which contains a list of dimension indices and this operator will insert a dimension of value `1` into the corresponding index of the output tensor (`expanded`). + + For example, given an input tensor (`data`) of shape [3, 4, 5], then + Unsqueeze(data, axes=[0, 4]) outputs a tensor (`expanded`) containing same data as `data` but with shape [1, 3, 4, 5, 1]. + + The input `axes` should not contain any duplicate entries. It is an error if it contains duplicates. + The rank of the output tensor (`output_rank`) is the rank of the input tensor (`data`) plus the number of values in `axes`. + Each value in `axes` should be within the (inclusive) range [-output_rank , output_rank - 1]. + The order of values in `axes` does not matter and can come in any order. + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Inputs + +
+
data (differentiable) : T
+
Original tensor
+
axes (non-differentiable) : tensor(int64)
+
1D tensor of integers indicating the dimensions to be inserted. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(expanded).
+
+ +#### Outputs + +
+
expanded (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
Constrain input and output types to all tensor types up to IRv11.
+
+ +## Version 24 of the default ONNX operator set +### **Attention-24** + + Computes scaled dot product attention on query, key and value tensors, using an optional attention mask if passed. + + This operator covers self and cross variants of the attention operation based on sequence lengths of K, Q and V. + + For self attention, `kv_sequence_length` equals to `q_sequence_length`. + + For cross attention, query and key might have different lengths. + + This operator also covers the 3 following variants based on the number of heads: + 1) Multi-headed Attention (MHA): Described in the paper https://arxiv.org/pdf/1706.03762, `q_num_heads = kv_num_heads`. + 2) Group-query Attention (GQA): Described in the paper https://arxiv.org/pdf/2305.13245, `q_num_heads > kv_num_heads`, `q_num_heads % kv_num_heads == 0`. + 3) Multi-query Attention (MQA): Described in the paper https://arxiv.org/pdf/1911.02150, `q_num_heads > kv_num_heads`, `kv_num_heads=1`. + + Attention bias to be added is calculated based on `attn_mask` input and `is_causal` attribute: + 1) `attn_mask`: A boolean mask where a value of `True` indicates that the element should take part in attention or a float mask of the same type as query, key, value that is added to the attention score. + 2) If `is_causal` is set to `1`, causal masking is applied with bottom-right (offset-aware) alignment: query `i` attends key `j` iff `j <= i + offset`, as illustrated below. + + ``` + 2D causal mask for Attention (PR onnx/onnx#8068) + S_q=4 queries, S_k=8 keys + Rule: query i attends key j iff j <= i + offset + offset = nonpad_kv_seqlen - S_q + + nonpad_kv_seqlen=4, offset=4-4=0 + + k0 k1 k2 k3 k4 k5 k6 k7 + +----+----+----+----+----+----+----+----+ + q0 | ## | | | | | | | | + +----+----+----+----+----+----+----+----+ + q1 | ## | ## | | | | | | | + +----+----+----+----+----+----+----+----+ + q2 | ## | ## | ## | | | | | | + +----+----+----+----+----+----+----+----+ + q3 | ## | ## | ## | ## | | | | | + +----+----+----+----+----+----+----+----+ + + + nonpad_kv_seqlen=8, offset=8-4=4 + + k0 k1 k2 k3 k4 k5 k6 k7 + +----+----+----+----+----+----+----+----+ + q0 | ## | ## | ## | ## | ## | | | | + +----+----+----+----+----+----+----+----+ + q1 | ## | ## | ## | ## | ## | ## | | | + +----+----+----+----+----+----+----+----+ + q2 | ## | ## | ## | ## | ## | ## | ## | | + +----+----+----+----+----+----+----+----+ + q3 | ## | ## | ## | ## | ## | ## | ## | ## | + +----+----+----+----+----+----+----+----+ + ``` + + With `nonpad_kv_seqlen=4` (offset=0), the mask is the standard lower-triangular. With `nonpad_kv_seqlen=8` (offset=4), the diagonal shifts right by 4, so each query sees the 4 additional valid cached keys. + + `offset` is the count of valid keys preceding the current query block: `offset = past_sequence_length` when `past_key` is provided; `offset = nonpad_kv_seqlen - q_sequence_length` (per batch) when an external cache is indicated by `nonpad_kv_seqlen` without `past_key`; `offset = 0` when neither is provided (the no-cache case, which reduces to the standard lower-triangular mask). When `offset < 0` (`nonpad_kv_seqlen < q_sequence_length`, i.e. more query tokens than cached keys) the leading query rows have an empty key set (no key satisfies `j <= i + offset`) and are fully masked. The causal frontier is computed independently of `attn_mask` and is then composed with it additively: a boolean `attn_mask` intersects the allowed set (its disallowed positions contribute `-inf` to the bias), while a float `attn_mask` is added to the attention scores rather than disabling positions. A fully-masked query row (no key attended, including the negative-offset leading rows) produces a zero output row, not `NaN`, for both `Y` and the mode-`3` `qk_matmul_output` debug output; the mode-`3` `qk_matmul_output` is emitted at the operator's output precision (`T1`). + + Errata (in-place behavioral correction, no opset bump): the reference implementation and backend tests were incorrect when `nonpad_kv_seqlen != q_sequence_length` (nonzero bottom-right offset, top-left instead of bottom-right causal alignment) and produced `NaN` for fully-masked rows; corrected in version 1.23. This fixed three behaviors described above: external-cache bottom-right causal alignment (`offset = nonpad_kv_seqlen - q_sequence_length`), zero (non-`NaN`) output for fully-masked rows including the mode-`3` `qk_matmul_output`, and the mode-`3` `qk_matmul_output` precision (`T1`). + + With respect to KV cache update, this operator allows the following two use cases: + + 1) Cache update happens inside the Attention operator. In this case, the `K` and `V` inputs contain only the incoming + tokens for the current autoregressive step, and the four optional inputs/outputs past and present key and value are + all needed. The Attention op performs a Concat operation on the past and incoming key and value to form the present + key and value, respectively. Note that this only works correctly for the special case where the past key and value + do not contain padded tokens. + 2) Cache update happens outside the Attention operator (for example, through the `TensorScatter` operator). In this + case, the `K` and `V` inputs correspond to the entire cache tensor, so the four optional inputs/outputs past and + present key and value should not be used. An additional input `nonpad_kv_seqlen` of shape (batch_size,) may be + provided to indicate the number of non-padding tokens in each sample of the batch to save unnecessary computation. + Here, the kv_sequence dimension of `attn_mask` can be shorter than `K` and `V`, but still needs to be at least as long + as the maximum value of `nonpad_kv_seqlen`. + + Both past and present state key/values are optional. They shall be used together, and not allowed to use only one of them. + The following pattern is applied to the Q, K and V inputs after appropriate reshaping of K and V inputs based on sequence lengths and num heads provided: + + ``` + The following pattern is applied by this operator: + Q K V + | | | + Q*sqrt(scale) K*sqrt(scale) | + | | | + | Transpose | + | | | + ---MatMul--- | + | | + softcap (if provided) | + | | + at_mask---Add | + | | + Softmax | + | | + -----MatMul------ + | + Y + ``` + + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
is_causal : int (default is 0)
+
If set to `1`, causal masking is applied. For a square Q/K (no cache offset) this is a lower-triangular matrix. In general the mask is bottom-right (offset-aware): query in-block index `i` attends key `j` iff `j <= i + offset`, where `offset` is the count of valid keys preceding the query block (`past_sequence_length` for an internal `past_key` cache, or `nonpad_kv_seqlen - q_sequence_length` per batch for an external cache). When `offset = 0` this reduces to the lower-triangular (top-left) mask.
+
kv_num_heads : int
+
Number of heads of key and value. Must be used with 3D inputs of Q, K and V.
+
q_num_heads : int
+
Number of heads of query. Must be used with 3D inputs of Q, K and V.
+
qk_matmul_output_mode : int (default is 0)
+
If set to `0`, qk_matmul_output is the output of qk matmul. If set to `1`, qk_matmul_output is the output after the softcap operation (before mask addition). If set to `2`, qk_matmul_output includes the attention mask and softcap (if provided) applied to the output of qk matmul. If set to `3`, qk_matmul_output is the output after the softmax operation. In mode `3`, a fully-masked query row (every key disallowed) is a zero row, consistent with the corresponding row of the primary output `Y`: the fully-masked-row guard is applied before this output is produced. The mode-`3` output is emitted at the operator's output precision (`T1`); when `softmax_precision` differs from `T1` this is a cast of the softmax result to `T1`. Default value is 0.
+
scale : float
+
Scaling factor applied to $Q*K^T$. Default value is `1/sqrt(head_size)`. To prevent [numerical overflow](https://tinyurl.com/sudb9s96), scale `Q`, `K` by `sqrt(scale)` before matmul.
+
softcap : float (default is 0.0)
+
Softcap value for attention weights. Default value is 0.
+
softmax_precision : int
+
The floating-point precision used in softmax computation. If softmax precision is not provided, the same precision as the input of softmax (Q and K) is used.
+
+ +#### Inputs (3 - 7) + +
+
Q : T1
+
Query tensor. 4D tensor with shape `(batch_size, q_num_heads, q_sequence_length, head_size)` or 3D tensor with shape `(batch_size, q_sequence_length, q_hidden_size)`. For cases with a 3D input tensor, `q_hidden_size = q_num_heads * head_size`
+
K : T1
+
Key tensor. 4D tensor with shape `(batch_size, kv_num_heads, kv_sequence_length, head_size)` or 3D tensor with shape `(batch_size, kv_sequence_length, k_hidden_size)`. For cases with a 3D input tensor, `k_hidden_size = kv_num_heads * head_size`
+
V : T2
+
Value tensor. 4D tensor with shape `(batch_size, kv_num_heads, kv_sequence_length, v_head_size)` or 3D tensor with shape `(batch_size, kv_sequence_length, v_hidden_size)`. For cases with a 3D input tensor, `v_hidden_size = kv_num_heads * v_head_size`
+
attn_mask (optional) : U
+
Attention mask. Shape must be broadcastable to `(batch_size, q_num_heads, q_sequence_length, total_sequence_length)` where `total_sequence_length = past_sequence_length + kv_sequence_length.` The last dimension can also be shorter than `total_sequence_length` and will be padded to `total_sequence_length` with negative infinity. Two types of masks are supported: a boolean mask where a value of `True` indicates that the element should take part in attention, or a float mask of the same type as query, key, value that is added to the attention score.
+
past_key (optional) : T1
+
past state cache for key with shape `(batch_size, kv_num_heads, past_sequence_length, head_size)`
+
past_value (optional) : T2
+
past state cache for value with shape `(batch_size, kv_num_heads, past_sequence_length, v_head_size)`
+
nonpad_kv_seqlen (optional) : tensor(int64)
+
A vector of integers of shape `(batch_size,)` that indicates the number of valid (ie, non-padding) tokens in each sample. A padding mask can be derived from this. This should not be used together with `past_key` and `past_value` inputs or `present_key` and `present_value` outputs (See the KV cache use cases in the operator description).
+
+ +#### Outputs (1 - 4) + +
+
Y : T1
+
The output tensor . 4D tensor with shape `(batch_size, q_num_heads, q_sequence_length, v_head_size)` or 3D tensor with shape `(batch_size, q_sequence_length, hidden_size)`. For cases with a 3D input tensor, `hidden_size = q_num_heads * v_head_size`
+
present_key (optional) : T1
+
Updated key cache with shape `(batch_size, kv_num_heads, total_sequence_length, head_size)` where `total_sequence_length = past_sequence_length + kv_sequence_length`.
+
present_value (optional) : T2
+
Updated value cache with shape `(batch_size, kv_num_heads, total_sequence_length, v_head_size)` where `total_sequence_length = past_sequence_length + kv_sequence_length`.
+
qk_matmul_output (optional) : T1
+
The output of QK matmul. 4D tensor with shape `(batch_size, q_num_heads, q_sequence_length, total_sequence_length)` where `total_sequence_length = past_sequence_length + kv_sequence_length`.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain Q and K inputs types to float tensors.
+
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain V input types to float tensors.
+
U : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool)
+
Constrain output 'mask' types to boolean tensors and input types.
+
+ +### **Cast-24** + + The operator casts the elements of a given input tensor to a data type + specified by the 'to' argument and returns an output tensor of the same size in + the converted type. The 'to' argument must be one of the data types specified + in the 'DataType' enum field in the TensorProto message. + + Casting from string tensor in plain (e.g., "3.14" and "1000") and scientific numeric representations + (e.g., "1e-5" and "1E8") to float types is supported. For example, converting string "100.5" to an integer may + yield result 100. There are some string literals reserved for special floating-point values; + "+INF" (and "INF"), "-INF", and "NaN" are positive infinity, negative infinity, and not-a-number, respectively. + Any string which can exactly match "+INF" in a case-insensitive way would be mapped to positive infinite. Similarly, + this case-insensitive rule is applied to "INF" and "NaN". When casting from numeric tensors + to string tensors, plain floating-point representation (such as "314.15926") would be used. + Converting non-numerical-literal string such as "Hello World!" is an undefined behavior. Cases + of converting string representing floating-point arithmetic value, such as "2.718", to INT is an undefined behavior. + + Conversion from a numerical type to any numerical type is always allowed. + User must be aware of precision loss and value change caused by range difference between two types. + For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting + an integer 36 to Boolean may produce 1 because we truncate bits which can't be stored in the targeted type. + + In more detail, the conversion among numerical types should follow these rules + if the destination type is not a float 8 type. + + * Casting from floating point to: + * floating point: +/- infinity if OOR (out of range). + * fixed point: undefined if OOR. + * bool: +/- 0.0 to False; all else to True. + * Casting from fixed point to: + * floating point: +/- infinity if OOR. (+ infinity in the case of uint) + * fixed point: when OOR, discard higher bits and reinterpret (with respect to two's complement representation for + signed types). For example, 200 (int16) -> -56 (int8). + * bool: zero to False; nonzero to True. + * Casting from bool to: + * floating point: `{1.0, 0.0}`. + * fixed point: `{1, 0}`. + * bool: no change. + + Float 8 types (E4M3FN, E4M3FNUZ, E5M2, E5M2FNUZ) were introduced to speed up the training of + deep models. By default the conversion of a float *x* obeys + to the following rules. `[x]` means the value rounded to + the target mantissa width. + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | -------- | -------- | -------- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | Inf | FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | + | -Inf | -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | + | \[x\] > FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | + | \[x\] \< -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | + | else | RNE | RNE | RNE | RNE | + + The behavior changes if the parameter 'saturate' is set to False. + The rules then become: + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | ------ | -------- | ---- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | -NaN | -NaN | NaN | -NaN | NaN | + | Inf | NaN | NaN | Inf | NaN | + | -Inf | -NaN | NaN | -Inf | NaN | + | \[x\] > FLT_MAX | NaN | NaN | Inf | NaN | + | \[x\] \< -FLT_MAX | NaN | NaN | -Inf | NaN | + | else | RNE | RNE | RNE | RNE | + + FLOAT8E8M0 type was introduced to enable [Microscaling (MX) formats](https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf). + When casting to FLOAT8E8M0, the rounding behavior can be specified using the `round_mode` and `saturate` attributes. + The current CUDA behavior is to round up and saturate. Casting negative values to FLOAT8E8M0 gives undefined behavior. + The following table describes the casting behavior of special values to FLOAT8E8M0 in the two most common cases. + + | x | saturate + up | non-saturate + nearest | + | ----------------- | ------------- | --------------------- | + | 0 | 0 | NaN | + | -0 | Unspecified | Unspecified | + | NaN | NaN | NaN | + | Inf | E8M0_MAX | NaN | + | x > E8M0_MAX | E8M0_MAX | NaN | + | x < E8M0_MIN | E8M0_MIN | NaN | + | x < 0 | Unspecified | Unspecified | + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
round_mode : string (default is up)
+
Rounding mode for conversion to float8e8m0. It only applies to casting to float8e8m0 and is `up` by default. `up`: round to nearest value away from zero, `down`: round to nearest value towards zero, `nearest`: round to nearest value and ties round up.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz, float8e8m0). It is true by default. All cases are fully described in the tables inserted in the operator description.
+
to : int (required)
+
The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor with the same shape as input with type specified by the 'to' argument
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **CastLike-24** + + The operator casts the elements of a given input tensor (the first input) to + the same data type as the elements of the second input tensor. + See documentation of the Cast operator for further details. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
round_mode : string (default is up)
+
Rounding mode for conversion to float8e8m0. It only applies to casting to float8e8m0 and is `up` by default. `up`: round to nearest value away from zero, `down`: round to nearest value towards zero, `nearest`: round to nearest value and ties round up. Please refer to operator Cast description for further details.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz, float8e8m0). It is true by default. Please refer to operator Cast description for further details.
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
target_type (non-differentiable) : T2
+
The (first) input tensor will be cast to produce a tensor of the same type as this (second input) tensor.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor produced by casting the first input tensor to have the same type as the second input tensor.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **Constant-24** + + This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, + or value_* must be specified. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
sparse_value : sparse_tensor
+
The value for the elements of the output tensor in sparse format.
+
value : tensor
+
The value for the elements of the output tensor.
+
value_float : float
+
The value for the sole element for the scalar, float32, output tensor.
+
value_floats : list of floats
+
The values for the elements for the 1D, float32, output tensor.
+
value_int : int
+
The value for the sole element for the scalar, int64, output tensor.
+
value_ints : list of ints
+
The values for the elements for the 1D, int64, output tensor.
+
value_string : string
+
The value for the sole element for the scalar, UTF-8 string, output tensor.
+
value_strings : list of strings
+
The values for the elements for the 1D, UTF-8 string, output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input and output types to all tensor types.
+
+ +### **ConstantOfShape-24** + + Generate a tensor with given value and shape. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
value : tensor
+
(Optional) The value of the output elements.Should be a one-element tensor. If not specified, it defaults to a tensor of value 0 and datatype float32
+
+ +#### Inputs + +
+
input : T1
+
1D tensor. The shape of the expected output tensor. If empty tensor is given, the output would be a scalar. All values must be >= 0.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of shape specified by 'input'.If attribute 'value' is specified, the value and datatype of the output tensor is taken from 'value'.If attribute 'value' is not specified, the value in the output defaults to 0, and the datatype defaults to float32.
+
+ +#### Type Constraints + +
+
T1 : tensor(int64)
+
Constrain input types.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint4), tensor(int4), tensor(bool), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain output types to be numerics or boolean.
+
+ +### **DequantizeLinear-24** + + The linear dequantization operator. It consumes a quantized tensor, a scale, and a zero point to compute the + full-precision tensor. The dequantization formula is `y = (x - x_zero_point) * x_scale`. `x_scale` and `x_zero_point` + must have the same shape, determining the quantization's granularity: a scalar for per-tensor/per-layer quantization, + a 1-D tensor for per-axis quantization, or have a rank identical to the input for blocked quantization. + See QuantizeLinear for details on quantization granularity. + + `x_zero_point` and `x` must have the same type. `x` and `y` must have the same shape. In the case of dequantizing + `int32`, there's no zero point (zero point is supposed to be 0). + `zero-point` is usually not used in the case of float8 and 4-bit types quantization, but the dequantization formula remains the same + for consistency. The output type is determined by the attribute `output_dtype`. If `output_dtype` is not supplied then the output type + is the same as `x_scale`. The output type also determines the precision of the multiplication operation. + + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`.
+
block_size : int (default is 0)
+
(Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given `x` shape `(D0, ..., Di, ..., Dn)`, `y_scale` shape `(S0, ... Si, ...Sn)` and `axis=i`, the accepted range is `[ceil(Di/Si), ceil(Di/(Si-1))-1]`
+
output_dtype : int (default is 0)
+
(Optional) The output data type. If not supplied, the output data type is inferred from `x_scale` data type (`T2`)
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D quantized input tensor to be de-quantized.
+
x_scale : T2
+
Scale for input `x`. For per-tensor/layer dequantization the scale is a scalar, for per per-axis dequantization it is a 1-D Tensor and for blocked dequantization it has the same shape as the input, except for one dimension in which blocking is performed.
+
x_zero_point (optional) : T1
+
Zero point for input `x`. Shape must match x_scale. It's optional. Zero point is 0 when it's not specified.
+
+ +#### Outputs + +
+
y : T3
+
N-D full precision output tensor. It has the same shape as input `x`. The data type is specified by the `output_dtype` attribute or, in its absence, the type of `x_scale`.
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(int32), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
The type of the inputs 'x_zero_point' and 'x'.
+
T2 : tensor(float), tensor(float16), tensor(bfloat16), tensor(float8e8m0)
+
The type of the input 'x_scale'.
+
T3 : tensor(float), tensor(float16), tensor(bfloat16)
+
The type of the output 'y'.
+
+ +### **Flatten-24** + + Flattens the input tensor into a 2D matrix. If input tensor has shape + (d_0, d_1, ... d_n) then the output will have shape + (d_0 X d_1 ... d_(axis-1), d_axis X d_(axis+1) ... X dn). + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 ... d_n), where the shape of the input tensor is (d_0, d_1, ... d_n).
+
+ +#### Inputs + +
+
input (differentiable) : T
+
A tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input and output to all tensor types up to IRv12.
+
+ +### **Identity-24** + + Identity operator + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : V
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : V
+
Tensor to copy input into.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain input and output types to all tensor, sequence, and optional types.
+
+ +### **If-24** + + If conditional + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
else_branch : graph (required)
+
Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
+
then_branch : graph (required)
+
Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
+
+ +#### Inputs + +
+
cond : B
+
Condition for the if. The tensor must contain a single element.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : V
+
Values that are live-out to the enclosing scope. The return values in the `then_branch` and `else_branch` must be of the same data type. The `then_branch` and `else_branch` may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if-node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of `then_branch` is typed float tensor with shape [2] and the first output of `else_branch` is another float tensor with shape [3], If's first output should have (a) no shape set, or (b) a shape of rank 1 with neither `dim_value` nor `dim_param` set, or (c) a shape of rank 1 with a unique `dim_param`. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), seq(tensor(float4e2m1)), seq(tensor(float8e8m0)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4)), optional(tensor(float4e2m1)), optional(tensor(float8e8m0))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv11.
+
B : tensor(bool)
+
Only bool
+
+ +### **Loop-24** + + Generic Looping construct. This loop has multiple termination conditions: + + 1) Trip count. Iteration count specified at runtime. Set by + specifying the input M. Optional. Set to empty string to omit. + Note that a static trip count (specified at graph construction time) can be + specified by passing in a constant node for input M. + 2) Loop termination condition. This is an input to the op that determines + whether to run the first iteration and also a loop-carried dependency for + the body graph. The body graph must yield a value for the condition variable, + whether this input is provided or not. + + This table summarizes the operating modes of this operator with equivalent + C-style code: + + Operator inputs defined as (max_trip_count, condition_var). + + * input ("", ""): + for (int i=0; ; ++i) { + cond = ... // Note this value is ignored, but is required in the body + } + + * input ("", cond) // Note this is analogous to a while loop + bool cond = ...; + for (int i=0; cond; ++i) { + cond = ...; + } + + * input ("", 1) // Note this is analogous to a do-while loop + bool cond = true + for (int i=0; cond; ++i) { + cond = ...; + } + + * input (trip_count, "") // Note this is analogous to a for loop + int trip_count = ... + for (int i=0; i < trip_count; ++i) { + cond = ...; // ignored + } + + * input (trip_count, cond) + int trip_count = ...; + bool cond = ...; + for (int i=0; i < trip_count && cond; ++i) { + cond = ...; + } + + + *Sample usage - cond as well as trip count* + + graph predict-net { + %a = Constant[value = ]() + %b = Constant[value = ]() + %keepgoing = Constant[value = ]() + %max_trip_count = Constant[value = ]() + %keepgoing_out, %b_out, %user_defined_vals = Loop[body = ](%max_trip_count, %keepgoing, %b) + return + } + + graph body-net ( + %i[INT32, scalar] // iteration number + %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used + %b_in[INT32, scalar] // incoming value of loop-carried-dependency b + ) { + %my_local = Add(%a, %b_in) + %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b + %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition + %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated + return %keepgoing_out, %b_out, %user_defined_val + } + + *Sample equivalent C code* + + { + /* User-defined code (enclosing scope) */ + int a = 3, b = 6; + bool keepgoing = true; // Analogous to input cond + /* End user-defined code */ + + /* Implicitly-defined code */ + const int max_trip_count = 10; // Analogous to input M + int user_defined_vals[]; // Imagine this is resizable + /* End implicitly-defined code */ + /* initialize loop-carried variables and scan-output variables */ + bool keepgoing_out = keepgoing + int b_out = b + + for (int i=0; i < max_trip_count && keepgoing_out; ++i) { + /* Implicitly-defined code: bind actual parameter values + to formal parameter variables of loop-body */ + bool keepgoing_in = keepgoing_out; + bool b_in = b_out; + + /* User-defined code (loop body) */ + int my_local = a + b_in; // Reading value "a" from the enclosing scope is fine + b_out = a - b_in; + keepgoing_out = my_local > b_out; + user_defined_val = b_in + b_in; // b_in and b_out are different variables + /* End user-defined code */ + + /* Implicitly defined-code */ + user_defined_vals[i] = user_defined_val // accumulate scan-output values + } + // int t = my_local; // Can't do this. my_local is not accessible here. + + // The values below are bound to the output variables of the loop and therefore accessible + // b_out; user_defined_vals; keepgoing_out; + } + + There are several things of note in this code snippet: + + 1) Values from the enclosing scope (i.e. variable "a" here) are in scope and can + be referenced in the inputs of the loop. + 2) Any values computed in the loop body that needs to be used in a subsequent + iteration or after the loop are modeled using a pair of variables in the loop-body, + consisting of an input variable (eg., b_in) and an output variable (eg., b_out). + These are referred to as loop-carried dependences. The loop operation node + supplies the input value of the input variable for the first iteration, and + returns the output value of the output variable produced by the final + iteration. + 3) Scan_output variables are used to implicitly concatenate values computed across + all the iterations. In the above example, the value of user_defined_val computed + over all iterations are concatenated and returned as the value of user_defined_vals + after the loop. + 4) Values created in the body cannot be accessed in the enclosing scope, + except using the mechanism described above. + + Note that the semantics of this op support "diagonal" or "wavefront" execution. + (See Step 3 here for an example: + https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). + Frontends should emit multi-layer RNNs as a series of While operators (with + time being the inner looping dimension), with each successive layer consuming + the scan_outputs from the previous layer, possibly going through several + point-wise operators (e.g. dropout, residual connections, linear layer). + + The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies...). It has 1+N+K outputs: (condition, loop carried dependencies..., scan_outputs...). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
+
+ +#### Inputs (2 - ∞) + +
+
M (optional) : I
+
A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
+
cond (optional) : B
+
A boolean termination condition. Optional. Pass empty string to skip.
+
v_initial (variadic, heterogeneous) : V
+
The initial values of any loop-carried dependencies (values that change across loop iterations)
+
+ +#### Outputs (1 - ∞) + +
+
v_final_and_scan_outputs (variadic, heterogeneous) : V
+
Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), seq(tensor(float4e2m1)), seq(tensor(float8e8m0)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4)), optional(tensor(float4e2m1)), optional(tensor(float8e8m0))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv11.
+
I : tensor(int64)
+
tensor of int64, which should be a scalar.
+
B : tensor(bool)
+
tensor of bool, which should be a scalar.
+
+ +### **Pad-24** + + Given a tensor containing the data to be padded (`data`), a tensor containing the number of start and end pad values for axis (`pads`), (optionally) a `mode`, and (optionally) `constant_value`, + a padded tensor (`output`) is generated. + + The four supported `modes` are (similar to corresponding modes supported by `numpy.pad`): + + 1) `constant`(default) - pads with a given constant value as specified by `constant_value` (which defaults to 0, empty string, or False) + + 2) `reflect` - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis + + 3) `edge` - pads with the edge values of array + + 4) `wrap` - wrap-around padding as if the data tensor forms a torus + + + Example 1 (`constant` mode): + + Insert 0 pads to the beginning of the second dimension. + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'constant' + + constant_value = 0.0 + + output = [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ] + ``` + + Example 2 (`reflect` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'reflect' + + output = [ + [1.0, 1.2, 1.0, 1.2], + [2.3, 3.4, 2.3, 3.4], + [4.5, 5.7, 4.5, 5.7], + ] + ``` + + Example 3 (`edge` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'edge' + + output = [ + [1.0, 1.0, 1.0, 1.2], + [2.3, 2.3, 2.3, 3.4], + [4.5, 4.5, 4.5, 5.7], + ] + ``` + + Example 4 (`wrap` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [2, 1, 1, 1] + + mode = 'wrap' + + output = [ + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + ] + ``` + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is constant)
+
Supported modes: `constant`(default), `reflect`, `edge`, `wrap`
+
+ +#### Inputs (2 - 4) + +
+
data (differentiable) : T
+
Input tensor.
+
pads (non-differentiable) : tensor(int64)
+
Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. `pads` should be a 1D tensor of shape [2 * num_axes] where `num_axes` refers to the number of elements in the `axes` input or the input rank if `axes` are not provided explicitly. `pads` format should be: [x1_begin, x2_begin, ..., x1_end, x2_end,...], where xi_begin is the number of pad values added at the beginning of axis `axes[i]` and xi_end, the number of pad values added at the end of axis `axes[i]`.
+
constant_value (optional, non-differentiable) : T
+
(Optional) A scalar value to be used if the mode chosen is `constant` (by default it is 0, empty string or False).
+
axes (optional, non-differentiable) : Tind
+
1-D tensor of axes that `pads` apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated. If not provided, all axes are assumed (`[0, 1, ..., input_rank-1]`).
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input and output types to all tensor types up to IRv12.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **QuantizeLinear-24** + + The linear quantization operator consumes a high-precision tensor, a scale, and a zero point to compute the + low-precision/quantized tensor. The scale factor and zero point must have the same shape, determining the quantization + granularity. The quantization formula is `y = saturate((x / y_scale) + y_zero_point)`. + + Saturation is done according to: + - uint16: [0, 65535] + - int16: [-32768, 32767] + - uint8: [0, 255] + - int8: [-128, 127] + - uint4: [0, 15] + - int4: [-8, 7] + + For `(x / y_scale)`, it rounds to the nearest even. Refer to https://en.wikipedia.org/wiki/Rounding for details. + + `y_zero_point` and `y` must have the same type. `y_zero_point` is usually not used for quantization to float8 and 4bit types, but the quantization + formula remains the same for consistency, and the type of the attribute `y_zero_point` still determines the quantization type. + `x` and `y_scale` are allowed to have different types. The type of `y_scale` determines the precision of the division operation between `x` and + `y_scale`, unless the `precision` attribute is specified. + + There are three supported quantization granularities, determined by the shape of `y_scale`. + In all cases, `y_zero_point` must have the same shape as `y_scale`. + - Per-tensor (per-layer) quantization: `y_scale` is a scalar. + - Per-axis quantization: The scale must be a 1-D tensor, with the length of the quantization axis. For an input shape + `(D0, ..., Di, ..., Dn)` and `axis=i`, `y_scale` is a 1-D tensor of length `Di`. + - Blocked quantization: The scale's shape is identical to the input's shape, except for one dimension, in which + blocking is performed. Given `x` shape `(D0, ..., Di, ..., Dn)`, `axis=i`, and block size `B`: `y_scale` shape is + `(D0, ..., ceil(Di/B), ..., Dn)`. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used only for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`. When the rank of the input is 1, per-tensor quantization is applied, rendering the axis unnecessary in this scenario.
+
block_size : int (default is 0)
+
(Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given `x` shape `(D0, ..., Di, ..., Dn)`, `y_scale` shape `(S0, ... Si, ...Sn)` and `axis=i`, the accepted range is `[ceil(Di/Si), ceil(Di/(Si-1))-1]`
+
output_dtype : int (default is 0)
+
(Optional) The output data type. If not supplied, the output data type is inferred from `y_zero_point` data type (`T3`). If neither `output_dtype` nor `y_zero_point` are supplied, output data type is uint8. If both `output_dtype` and `y_zero_point` are specified, `output_dtype` must be `T3`.
+
precision : int (default is 0)
+
(Optional) The precision of the division operation between `x` and `y_scale`. If not provided, it will be the same as the type of `y_scale`.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 quantization (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. All cases are fully described in two tables inserted in the operator description.
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D full precision Input tensor to be quantized.
+
y_scale : T2
+
Scale for doing quantization to get `y`. For per-tensor/layer quantization the scale is a scalar, for per-axis quantization it is a 1-D Tensor and for blocked quantization it has the same shape as the input, except for one dimension in which blocking is performed.
+
y_zero_point (optional) : T3
+
Zero point for doing quantization to get `y`. Shape must match `y_scale`. Default is uint8 with zero point of 0 if it's not specified.
+
+ +#### Outputs + +
+
y : T3
+
N-D quantized output tensor. It has same shape as input `x`.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(float16), tensor(bfloat16), tensor(int32)
+
The type of the input 'x'.
+
T2 : tensor(float), tensor(float16), tensor(bfloat16), tensor(int32), tensor(float8e8m0)
+
The type of the input 'y_scale'.
+
T3 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1)
+
The type of the input `y_zero_point` and the output `y`.
+
+ +### **Reshape-24** + + Reshape the input tensor similar to numpy.reshape. + First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. + At most one dimension of the new shape can be -1. In this case, the value is + inferred from the size of the tensor and the remaining dimensions. A dimension + could also be 0, in which case the actual dimension value is unchanged (i.e. taken + from the input tensor). If 'allowzero' is set, and the new shape includes 0, the + dimension will be set explicitly to zero (i.e. not taken from input tensor). + Shape (second input) could be an empty shape, which means converting to a scalar. + The input tensor's shape and the output tensor's shape are required to have the same number of elements. + + If the attribute 'allowzero' is set, it is invalid for the specified shape to + contain both a zero value and -1, as the value of the dimension corresponding + to -1 cannot be determined uniquely. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
allowzero : int (default is 0)
+
(Optional) By default, when any value in the 'shape' input is equal to zero the corresponding dimension value is copied from the input tensor dynamically. allowzero=1 indicates that if any value in the 'shape' input is set to zero, the zero value is honored, similar to NumPy.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
shape (non-differentiable) : tensor(int64)
+
Specified shape for output.
+
+ +#### Outputs + +
+
reshaped (differentiable) : T
+
Reshaped data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input and output types to all tensor types.
+
+ +### **Scan-24** + + Scan can be used to iterate over one or more scan_input tensors, + constructing zero or more scan_output tensors. It combines ideas from general recurrences, + functional programming constructs such as scan, fold, map, and zip, and is intended to enable + generalizations of RNN-like constructs for sequence-to-sequence processing. + Other tensors (referred to as state_variables here) can be used to carry a state + when iterating from one element to another (similar to hidden-state in RNNs, also referred + to as loop-carried dependences in the context of loops). + Many common usages involve a single scan_input tensor (where functionality + similar to scan, fold and map can be obtained). When more than one scan_input is used, + a behavior similar to zip is obtained. + + The attribute body must be a graph, specifying the computation to be performed in + every iteration. It takes as input the current values of the state_variables and + the current iterated element of the scan_inputs. It must return the (updated) values + of the state_variables and zero or more scan_output_element tensors. The values of the + scan_output_element tensors are concatenated over all the iterations to produce the + scan_output values of the scan construct (similar to the concatenated intermediate + hidden-state values of RNN-like constructs). All the output tensors (state_variables as + well as scan_output_element tensors) are required to have the same shape in each iteration + of the loop (a restriction imposed to enable efficient memory allocation). + + Note that the iterated element passed to the body subgraph does not have a sequence + axis. It will have a rank one less than the rank of the corresponding scan_input. + + The scan operation returns the final values of the state_variables as well as the + scan_outputs. + + The optional attribute scan_input_directions specifies the direction (forward or backward) + for each scan input. If this attribute is omitted, all sequences are scanned in the forward + direction. A bidirectional scan may be performed by specifying the same tensor input twice + in the scan_inputs, once with a forward direction, and once with a backward direction. + + The scan_output of the operation is produced by concatenating the scan_output_element + values produced by the body in each iteration. The optional attribute scan_output_directions + specifies the direction in which scan_output is constructed (by appending or prepending the + scan_output_element to scan_output in each iteration) for each scan_output. If this attribute + is omitted, the scan_output_element is appended to the scan_output in each iteration. + + The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. + If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the + batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. + Note that scanning a non-zero axis may be less efficient than scanning axis zero. + + The optional attribute scan_output_axes specifies the axis along which the scan_outputs + are accumulated for each scan_output. For example, if axis 1 is the time axis (to be + scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis + value of 1. + + Note that because of the ONNX restriction that only the last parameter of an operator can + be variadic, the initial-states and scan-inputs are listed together as one input parameter. + Similarly, the final-states and scan-outputs are listed together as one output parameter. + The attribute num_scan_inputs indicates the number M of scan-inputs. + + The behavior of + + Scan < + num_scan_inputs = m, + body = loop-body, + scan_input_axes = [axis_1, ..., axis_m] + > (init_1, ..., init_n, scan_1, ..., scan_m) + + is equivalent to the following pseudo-code: + + // scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i + // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. + sequence_length = scan_1.shape[axis_1]; + + // initialize state-variables + st_1 = init_1; ... st_n = init_n; + // initialize scan-output variables: [] denotes an empty tensor + scan_out_1 = []; ...; scan_out_k = []; + // identify number of iterations: + + // execute loop + for (int t = 0; t < sequence_length; ++t) { + // generate the scan-input elements: the notation T[t] indicates the sub-tensor + // of rank one less than T obtained by indexing T at position t along axis k. + si_1 = scan_1[t]; + ... ; + si_m = scan_m[t]; + // execute loop-body + st_1, ..., st_n, so_1, ..., so_k = loop-body(st_1, ..., st_n, si_1, ..., si_m) + // accumulate the scan-output elements + scan_out_1 = Concat(scan_out_1, so_1); ... ; scan_out_k = Concat(scan_out_k, so_k); + } + + return st_1, ..., st_n, scan_out_1, ..., scan_out_k; + + *Sample usage: Encoding RNN using a Scan* + + The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, + recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can + be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes + %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these + values are computed in the outer graph, they need to be passed in as extra state_variables. + + graph rnn-encoding { + %H_0 = ... + %X = ... + %Y_h, %Y = Scan[body = , num_scan_inputs=1](%H_0, %X) + return %Y, %Y_h + } + + graph rnn-cell-1 ( + %H_tminus1[FLOAT, tensor] + %X_t[FLOAT, tensor] + ) { + %Wi = ... + %Ri = ... + %Wbi = ... + %Rbi = ... + %t1 = X_t * (Wi^T) + %t2 = H_tminus1*(Ri^T) + %t3 = Add(%t1, %t2) + %t4 = Add(%t3, %Wbi) + %t5 = Add(%t4, %Rbi) + %Ht = Tanh(%t5) + %Accumulate = Identity(%Ht) + return %Ht, %Accumulate + } + + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has N+M inputs: (loop state variables..., scan_input_elts...). It has N+K outputs: (loop state variables..., scan_output_elts...). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
+
num_scan_inputs : int (required)
+
An attribute specifying the number of scan_inputs M.
+
scan_input_axes : list of ints
+
An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
scan_input_directions : list of ints
+
An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
+
scan_output_axes : list of ints
+
An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
+
scan_output_directions : list of ints
+
An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
+
+ +#### Inputs (1 - ∞) + +
+
initial_state_and_scan_inputs (variadic, heterogeneous) : V
+
Initial values of the loop's N state variables followed by M scan_inputs
+
+ +#### Outputs (1 - ∞) + +
+
final_state_and_scan_outputs (variadic, heterogeneous) : V
+
Final values of the loop's N state variables followed by K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
All Tensor types up to IRv12.
+
+ +### **Shape-24** + + Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. + Optional attributes start and end can be used to compute a slice of the input tensor's shape. + If start axis is omitted, the slice starts from axis 0. + The end axis, if specified, is exclusive (and the returned value will not include the size of that axis). + If the end axis is omitted, the axes upto the last one will be included. + Negative axes indicate counting back from the last axis. + Note that axes will be clamped to the range [0, r], where r is the + rank of the input tensor if they are out-of-range (after adding r in the case of + negative axis). Thus, specifying any end value > r is equivalent to specifying an end + value of r, and specifying any start value < -r is equivalent to specifying a start + value of 0. If start > end, the result will be an empty shape. + + Examples: + + ``` + Input tensor with shape: [2, 3, 4] + No attributes specified. + Output: [2, 3, 4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: -1 + Output: [4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + end: -1 + Output: [2, 3] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: 1 + end: 2 + Output: [3] + ``` + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
end : int
+
(Optional) Ending axis for slicing the shape. Negative value means counting dimensions from the back. If omitted, sizes of all axes upto (including) the last one will be included.
+
start : int (default is 0)
+
(Optional) Starting axis for slicing the shape. Default value is 0.Negative value means counting dimensions from the back.
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
shape (non-differentiable) : T1
+
Shape of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor.
+
+ +### **Size-24** + + Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
size (non-differentiable) : T1
+
Total number of elements of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor, which should be a scalar though.
+
+ +### **SplitToSequence-24** + + Split a tensor into a sequence of tensors, along the specified 'axis'. + Lengths of the parts can be specified using the optional argument 'split'. + If the argument `split' is not specified, a default scalar value of 1 + is used as the value of `split'. + 'split' must contain only positive numbers. + 'split' is either a scalar (tensor of empty shape), or a 1-D tensor. + If 'split' is a scalar, then 'input' will be split into chunks all of size 'split' + if possible. The last chunk alone may be smaller than 'split' if the 'input' size + along the given axis 'axis' is not divisible by 'split'. + If 'split' is a 1-dimensional tensor, the input tensor is split into 'size(split)' chunks, + with lengths of the parts on 'axis' specified in 'split'. In this scenario, the sum of entries + in 'split' must be equal to the dimension size of input tensor on 'axis'. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1].
+
keepdims : int (default is 1)
+
Keep the split dimension or not. Default 1, which means we keep split dimension. If input 'split' is specified, this attribute is ignored.
+
+ +#### Inputs (1 - 2) + +
+
input : T
+
The tensor to split
+
split (optional) : I
+
Length of each output. It can be either a scalar(tensor of empty shape), or a 1-D tensor. All values must be >= 0.
+
+ +#### Outputs + +
+
output_sequence : S
+
One or more outputs forming a sequence of tensors after splitting
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input types to all tensor types.
+
I : tensor(int32), tensor(int64)
+
Constrain split size to integral tensor.
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain output types to all tensor types.
+
+ +### **Squeeze-24** + + Remove single-dimensional entries from the shape of a tensor. + Takes an input `axes` with a list of axes to squeeze. + If `axes` is not provided, all the single dimensions will be removed from + the shape. If an axis is selected with shape entry not equal to one, an error is raised. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
Tensors with at least max(dims) dimensions.
+
axes (optional, non-differentiable) : tensor(int64)
+
1D tensor of integers indicating the dimensions to squeeze. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
squeezed (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input and output types to all tensor types up to IRv12.
+
+ +### **Swish-24** + + Swish function takes one input data (Tensor) and produces one output data (Tensor) of the same shape, + where $Swish(x) = x * sigmoid(alpha * x)$. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Coefficient to multiply with input before sigmoid.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(bfloat16), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **TensorScatter-24** + + TensorScatter is a generic tensor update operation, motivated by the requirements for KV cache updates for Attention + ops commonly found in LLMs. It is a functional operation that models an in-place update to a KV cache buffer. + + The past and present cache tensors have the same shape (batch_size, D1, D2, ..., max_sequence_length, ..., Dn), with + the sequence dimension (indicated by the `axis` attribute) being max_sequence_length, so the sizes of these tensors do + not need to grow between iterations. The `update` tensor's shape only differs from the cache tensors in the sequence + dimension: (batch_size, D1, D2, ..., sequence_length, ..., Dn), where sequence_length <= max_sequence_length. + + The optional `write_indices` input indicates the write index for each sample in the batch, assumed to be zero + if not provided. When the `mode` attribute is set to "circular", the write index is modulo max_sequence_length. + The operation can be described using the following pseudocode: + + ``` + for prefix_idx in np.ndindex(past_cache.shape[:axis]): + batch_idx = prefix_idx[0] + for sequence_idx in range(sequence_length): + cache_idx = (*prefix_idx, write_indices[batch_idx] + sequence_idx) + if mode == "circular": + cache_idx = tuple(np.mod(np.asarray(cache_idx), max_sequence_length)) + update_idx = (*prefix_idx, sequence_idx) + present_cache[cache_idx] = update[update_idx] + ``` + + During the prefill phase of attention, only the first two inputs are needed. During the decode phase, `write_indices` + is also needed so that the incoming key or value update can be appended after the last valid token for each sample + in the batch. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -2)
+
Sequence dimension of the `past_cache` and `update` tensors. It cannot be 0 (the batch dimension). Default is -2.
+
mode : string (default is linear)
+
Write mode of cache update. Supported modes include `linear` and `circular`. `linear` mode requires write_indices+sequence_length<=max_sequence_length. For `circular` mode, the updates happen in wrap-around fashion, ie, the update index is modulo `max_sequence_length`
+
+ +#### Inputs (2 - 3) + +
+
past_cache (differentiable) : T
+
Past state cache for key or value with shape `(batch_size, D1, D2, ..., max_sequence_length, ..., Dn)`.
+
update (differentiable) : T
+
New update tensor with shape `(batch_size, D1, D2, ..., sequence_length, ..., Dn)`.
+
write_indices (optional, non-differentiable) : tensor(int64)
+
Write indices for the incoming update tensor in the cache. Shape is `(batch_size,)`. Assumed to be all zeros if not provided.
+
+ +#### Outputs + +
+
present_cache (differentiable) : T
+
Updated cache. Same shape as `past_cache`.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input and output types to any tensor type.
+
+ +### **TopK-24** + + Retrieve the top-K largest or smallest elements along a specified axis. Given an input tensor of + shape [a_0, a_1, ..., a_{n-1}] and integer argument k, return two outputs: + + * Value tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] + which contains the values of the top k elements along the specified axis + * Index tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] which + contains the indices of the top k elements (original indices from the input + tensor). + + * If "largest" is 1 (the default value) then the k largest elements are returned. + * If "sorted" is 1 (the default value) then the resulting k elements will be sorted. + * If "sorted" is 0, order of returned 'Values' and 'Indices' are undefined. + + Given two equivalent values, this operator uses the indices along the axis as + a tiebreaker. That is, the element with the lower index will appear first. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
Dimension on which to do the sort. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
largest : int (default is 1)
+
Whether to return the top-K largest or smallest elements.
+
sorted : int (default is 1)
+
Whether to return the elements in sorted order.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Tensor of shape [a_0, a_1, ..., a_{n-1}]
+
K (non-differentiable) : tensor(int64)
+
A 1-D tensor containing a single positive value corresponding to the number of top elements to retrieve
+
+ +#### Outputs + +
+
Values (differentiable) : T
+
Tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] containing top K values from the input tensor
+
Indices (non-differentiable) : I
+
Tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] containing the corresponding input tensor indices for the top K values.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ +### **Transpose-24** + + Returns a transpose of the input tensor. (Similar to `numpy.transpose`). + The optional attribute `perm` must be a permutation of the dimensions of + the input tensor. Axis `i` of the output tensor corresponds to the axis + `perm[i]` of the input tensor. + For example, when perm=(1, 0, 2), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 1, 3). + When perm=(1, 2, 0), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 3, 1). + If the attribute `perm` is omitted, its default value is `(n-1, ..., 0)`, + where `n` is the rank of the input tensor. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
perm : list of ints
+
A list of integers. By default, reverse the dimensions, otherwise permute the axes according to the values given. Its length must be equal to the rank of the input.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
transposed (differentiable) : T
+
Transposed output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input and output types to all tensor types.
+
+ +### **Unsqueeze-24** + + Insert single-dimensional entries to the shape of an input tensor (`data`). + Takes one required input `axes` - which contains a list of dimension indices and this operator will insert a dimension of value `1` into the corresponding index of the output tensor (`expanded`). + + For example, given an input tensor (`data`) of shape [3, 4, 5], then + Unsqueeze(data, axes=[0, 4]) outputs a tensor (`expanded`) containing same data as `data` but with shape [1, 3, 4, 5, 1]. + + The input `axes` should not contain any duplicate entries. It is an error if it contains duplicates. + The rank of the output tensor (`output_rank`) is the rank of the input tensor (`data`) plus the number of values in `axes`. + Each value in `axes` should be within the (inclusive) range [-output_rank , output_rank - 1]. + The order of values in `axes` does not matter and can come in any order. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Inputs + +
+
data (differentiable) : T
+
Original tensor
+
axes (non-differentiable) : tensor(int64)
+
1D tensor of integers indicating the dimensions to be inserted. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(expanded).
+
+ +#### Outputs + +
+
expanded (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input and output types to all tensor types up to IRv12.
+
+ +## Version 25 of the default ONNX operator set +### **Cast-25** + + The operator casts the elements of a given input tensor to a data type + specified by the 'to' argument and returns an output tensor of the same size in + the converted type. The 'to' argument must be one of the data types specified + in the 'DataType' enum field in the TensorProto message. + + Casting from string tensor in plain (e.g., "3.14" and "1000") and scientific numeric representations + (e.g., "1e-5" and "1E8") to float types is supported. For example, converting string "100.5" to an integer may + yield result 100. There are some string literals reserved for special floating-point values; + "+INF" (and "INF"), "-INF", and "NaN" are positive infinity, negative infinity, and not-a-number, respectively. + Any string which can exactly match "+INF" in a case-insensitive way would be mapped to positive infinite. Similarly, + this case-insensitive rule is applied to "INF" and "NaN". When casting from numeric tensors + to string tensors, plain floating-point representation (such as "314.15926") would be used. + Converting non-numerical-literal string such as "Hello World!" is an undefined behavior. Cases + of converting string representing floating-point arithmetic value, such as "2.718", to INT is an undefined behavior. + + Conversion from a numerical type to any numerical type is always allowed. + User must be aware of precision loss and value change caused by range difference between two types. + For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting + an integer 36 to Boolean may produce 1 because we truncate bits which can't be stored in the targeted type. + + In more detail, the conversion among numerical types should follow these rules + if the destination type is not a float 8 type. + + * Casting from floating point to: + * floating point: +/- infinity if OOR (out of range). + * fixed point: undefined if OOR. + * bool: +/- 0.0 to False; all else to True. + * Casting from fixed point to: + * floating point: +/- infinity if OOR. (+ infinity in the case of uint) + * fixed point: when OOR, discard higher bits and reinterpret (with respect to two's complement representation for + signed types). For example, 200 (int16) -> -56 (int8). + * bool: zero to False; nonzero to True. + * Casting from bool to: + * floating point: `{1.0, 0.0}`. + * fixed point: `{1, 0}`. + * bool: no change. + + Float 8 types (E4M3FN, E4M3FNUZ, E5M2, E5M2FNUZ) were introduced to speed up the training of + deep models. By default the conversion of a float *x* obeys + to the following rules. `[x]` means the value rounded to + the target mantissa width. + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | -------- | -------- | -------- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | Inf | FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | + | -Inf | -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | + | \[x\] > FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | + | \[x\] \< -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | + | else | RNE | RNE | RNE | RNE | + + The behavior changes if the parameter 'saturate' is set to False. + The rules then become: + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | ------ | -------- | ---- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | -NaN | -NaN | NaN | -NaN | NaN | + | Inf | NaN | NaN | Inf | NaN | + | -Inf | -NaN | NaN | -Inf | NaN | + | \[x\] > FLT_MAX | NaN | NaN | Inf | NaN | + | \[x\] \< -FLT_MAX | NaN | NaN | -Inf | NaN | + | else | RNE | RNE | RNE | RNE | + + FLOAT8E8M0 type was introduced to enable [Microscaling (MX) formats](https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf). + When casting to FLOAT8E8M0, the rounding behavior can be specified using the `round_mode` and `saturate` attributes. + The current CUDA behavior is to round up and saturate. Casting negative values to FLOAT8E8M0 gives undefined behavior. + The following table describes the casting behavior of special values to FLOAT8E8M0 in the two most common cases. + + | x | saturate + up | non-saturate + nearest | + | ----------------- | ------------- | --------------------- | + | 0 | 0 | NaN | + | -0 | Unspecified | Unspecified | + | NaN | NaN | NaN | + | Inf | E8M0_MAX | NaN | + | x > E8M0_MAX | E8M0_MAX | NaN | + | x < E8M0_MIN | E8M0_MIN | NaN | + | x < 0 | Unspecified | Unspecified | + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
round_mode : string (default is up)
+
Rounding mode for conversion to float8e8m0. It only applies to casting to float8e8m0 and is `up` by default. `up`: round to nearest value away from zero, `down`: round to nearest value towards zero, `nearest`: round to nearest value and ties round up.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz, float8e8m0). It is true by default. All cases are fully described in the tables inserted in the operator description.
+
to : int (required)
+
The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor with the same shape as input with type specified by the 'to' argument
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **CastLike-25** + + The operator casts the elements of a given input tensor (the first input) to + the same data type as the elements of the second input tensor. + See documentation of the Cast operator for further details. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
round_mode : string (default is up)
+
Rounding mode for conversion to float8e8m0. It only applies to casting to float8e8m0 and is `up` by default. `up`: round to nearest value away from zero, `down`: round to nearest value towards zero, `nearest`: round to nearest value and ties round up. Please refer to operator Cast description for further details.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz, float8e8m0). It is true by default. Please refer to operator Cast description for further details.
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
target_type (non-differentiable) : T2
+
The (first) input tensor will be cast to produce a tensor of the same type as this (second input) tensor.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor produced by casting the first input tensor to have the same type as the second input tensor.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain output types. Casting to complex is not supported.
+
+ +### **Constant-25** + + This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, + or value_* must be specified. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
sparse_value : sparse_tensor
+
The value for the elements of the output tensor in sparse format.
+
value : tensor
+
The value for the elements of the output tensor.
+
value_float : float
+
The value for the sole element for the scalar, float32, output tensor.
+
value_floats : list of floats
+
The values for the elements for the 1D, float32, output tensor.
+
value_int : int
+
The value for the sole element for the scalar, int64, output tensor.
+
value_ints : list of ints
+
The values for the elements for the 1D, int64, output tensor.
+
value_string : string
+
The value for the sole element for the scalar, UTF-8 string, output tensor.
+
value_strings : list of strings
+
The values for the elements for the 1D, UTF-8 string, output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types.
+
+ +### **ConstantOfShape-25** + + Generate a tensor with given value and shape. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
value : tensor
+
(Optional) The value of the output elements.Should be a one-element tensor. If not specified, it defaults to a tensor of value 0 and datatype float32
+
+ +#### Inputs + +
+
input : T1
+
1D tensor. The shape of the expected output tensor. If empty tensor is given, the output would be a scalar. All values must be >= 0.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of shape specified by 'input'.If attribute 'value' is specified, the value and datatype of the output tensor is taken from 'value'.If attribute 'value' is not specified, the value in the output defaults to 0, and the datatype defaults to float32.
+
+ +#### Type Constraints + +
+
T1 : tensor(int64)
+
Constrain input types.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint4), tensor(int4), tensor(bool), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain output types to be numerics or boolean.
+
+ +### **DequantizeLinear-25** + + The linear dequantization operator. It consumes a quantized tensor, a scale, and a zero point to compute the + full-precision tensor. The dequantization formula is `y = (x - x_zero_point) * x_scale`. `x_scale` and `x_zero_point` + must have the same shape, determining the quantization's granularity: a scalar for per-tensor/per-layer quantization, + a 1-D tensor for per-axis quantization, or have a rank identical to the input for blocked quantization. + See QuantizeLinear for details on quantization granularity. + + `x_zero_point` and `x` must have the same type. `x` and `y` must have the same shape. In the case of dequantizing + `int32`, there's no zero point (zero point is supposed to be 0). + `zero-point` is usually not used in the case of float8 and 4-bit types quantization, but the dequantization formula remains the same + for consistency. The output type is determined by the attribute `output_dtype`. If `output_dtype` is not supplied then the output type + is the same as `x_scale`. The output type also determines the precision of the multiplication operation. + + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`.
+
block_size : int (default is 0)
+
(Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given `x` shape `(D0, ..., Di, ..., Dn)`, `y_scale` shape `(S0, ... Si, ...Sn)` and `axis=i`, the accepted range is `[ceil(Di/Si), ceil(Di/(Si-1))-1]`
+
output_dtype : int (default is 0)
+
(Optional) The output data type. If not supplied, the output data type is inferred from `x_scale` data type (`T2`)
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D quantized input tensor to be de-quantized.
+
x_scale : T2
+
Scale for input `x`. For per-tensor/layer dequantization the scale is a scalar, for per per-axis dequantization it is a 1-D Tensor and for blocked dequantization it has the same shape as the input, except for one dimension in which blocking is performed.
+
x_zero_point (optional) : T1
+
Zero point for input `x`. Shape must match x_scale. It's optional. Zero point is 0 when it's not specified.
+
+ +#### Outputs + +
+
y : T3
+
N-D full precision output tensor. It has the same shape as input `x`. The data type is specified by the `output_dtype` attribute or, in its absence, the type of `x_scale`.
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(int32), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(uint2), tensor(int2)
+
The type of the inputs 'x_zero_point' and 'x'.
+
T2 : tensor(float), tensor(float16), tensor(bfloat16), tensor(float8e8m0)
+
The type of the input 'x_scale'.
+
T3 : tensor(float), tensor(float16), tensor(bfloat16)
+
The type of the output 'y'.
+
+ +### **Flatten-25** + + Flattens the input tensor into a 2D matrix. If input tensor has shape + (d_0, d_1, ... d_n) then the output will have shape + (d_0 X d_1 ... d_(axis-1), d_axis X d_(axis+1) ... X dn). + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 ... d_n), where the shape of the input tensor is (d_0, d_1, ... d_n).
+
+ +#### Inputs + +
+
input (differentiable) : T
+
A tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output to all tensor types up to IRv13.
+
+ +### **Identity-25** + + Identity operator + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Inputs + +
+
input (differentiable) : V
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : V
+
Tensor to copy input into.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain input and output types to all tensor, sequence, and optional types.
+
+ +### **If-25** + + If conditional + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
else_branch : graph (required)
+
Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
+
then_branch : graph (required)
+
Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
+
+ +#### Inputs + +
+
cond : B
+
Condition for the if. The tensor must contain a single element.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : V
+
Values that are live-out to the enclosing scope. The return values in the `then_branch` and `else_branch` must be of the same data type. The `then_branch` and `else_branch` may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if-node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of `then_branch` is typed float tensor with shape [2] and the first output of `else_branch` is another float tensor with shape [3], If's first output should have (a) no shape set, or (b) a shape of rank 1 with neither `dim_value` nor `dim_param` set, or (c) a shape of rank 1 with a unique `dim_param`. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), seq(tensor(float4e2m1)), seq(tensor(float8e8m0)), seq(tensor(uint2)), seq(tensor(int2)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4)), optional(tensor(float4e2m1)), optional(tensor(float8e8m0)), optional(tensor(uint2)), optional(tensor(int2))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv13.
+
B : tensor(bool)
+
Only bool
+
+ +### **Loop-25** + + Generic Looping construct. This loop has multiple termination conditions: + + 1) Trip count. Iteration count specified at runtime. Set by + specifying the input M. Optional. Set to empty string to omit. + Note that a static trip count (specified at graph construction time) can be + specified by passing in a constant node for input M. + 2) Loop termination condition. This is an input to the op that determines + whether to run the first iteration and also a loop-carried dependency for + the body graph. The body graph must yield a value for the condition variable, + whether this input is provided or not. + + This table summarizes the operating modes of this operator with equivalent + C-style code: + + Operator inputs defined as (max_trip_count, condition_var). + + * input ("", ""): + for (int i=0; ; ++i) { + cond = ... // Note this value is ignored, but is required in the body + } + + * input ("", cond) // Note this is analogous to a while loop + bool cond = ...; + for (int i=0; cond; ++i) { + cond = ...; + } + + * input ("", 1) // Note this is analogous to a do-while loop + bool cond = true + for (int i=0; cond; ++i) { + cond = ...; + } + + * input (trip_count, "") // Note this is analogous to a for loop + int trip_count = ... + for (int i=0; i < trip_count; ++i) { + cond = ...; // ignored + } + + * input (trip_count, cond) + int trip_count = ...; + bool cond = ...; + for (int i=0; i < trip_count && cond; ++i) { + cond = ...; + } + + + *Sample usage - cond as well as trip count* + + graph predict-net { + %a = Constant[value = ]() + %b = Constant[value = ]() + %keepgoing = Constant[value = ]() + %max_trip_count = Constant[value = ]() + %keepgoing_out, %b_out, %user_defined_vals = Loop[body = ](%max_trip_count, %keepgoing, %b) + return + } + + graph body-net ( + %i[INT32, scalar] // iteration number + %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used + %b_in[INT32, scalar] // incoming value of loop-carried-dependency b + ) { + %my_local = Add(%a, %b_in) + %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b + %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition + %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated + return %keepgoing_out, %b_out, %user_defined_val + } + + *Sample equivalent C code* + + { + /* User-defined code (enclosing scope) */ + int a = 3, b = 6; + bool keepgoing = true; // Analogous to input cond + /* End user-defined code */ + + /* Implicitly-defined code */ + const int max_trip_count = 10; // Analogous to input M + int user_defined_vals[]; // Imagine this is resizable + /* End implicitly-defined code */ + /* initialize loop-carried variables and scan-output variables */ + bool keepgoing_out = keepgoing + int b_out = b + + for (int i=0; i < max_trip_count && keepgoing_out; ++i) { + /* Implicitly-defined code: bind actual parameter values + to formal parameter variables of loop-body */ + bool keepgoing_in = keepgoing_out; + bool b_in = b_out; + + /* User-defined code (loop body) */ + int my_local = a + b_in; // Reading value "a" from the enclosing scope is fine + b_out = a - b_in; + keepgoing_out = my_local > b_out; + user_defined_val = b_in + b_in; // b_in and b_out are different variables + /* End user-defined code */ + + /* Implicitly defined-code */ + user_defined_vals[i] = user_defined_val // accumulate scan-output values + } + // int t = my_local; // Can't do this. my_local is not accessible here. + + // The values below are bound to the output variables of the loop and therefore accessible + // b_out; user_defined_vals; keepgoing_out; + } + + There are several things of note in this code snippet: + + 1) Values from the enclosing scope (i.e. variable "a" here) are in scope and can + be referenced in the inputs of the loop. + 2) Any values computed in the loop body that needs to be used in a subsequent + iteration or after the loop are modeled using a pair of variables in the loop-body, + consisting of an input variable (eg., b_in) and an output variable (eg., b_out). + These are referred to as loop-carried dependences. The loop operation node + supplies the input value of the input variable for the first iteration, and + returns the output value of the output variable produced by the final + iteration. + 3) Scan_output variables are used to implicitly concatenate values computed across + all the iterations. In the above example, the value of user_defined_val computed + over all iterations are concatenated and returned as the value of user_defined_vals + after the loop. + 4) Values created in the body cannot be accessed in the enclosing scope, + except using the mechanism described above. + + Note that the semantics of this op support "diagonal" or "wavefront" execution. + (See Step 3 here for an example: + https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). + Frontends should emit multi-layer RNNs as a series of While operators (with + time being the inner looping dimension), with each successive layer consuming + the scan_outputs from the previous layer, possibly going through several + point-wise operators (e.g. dropout, residual connections, linear layer). + + The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies...). It has 1+N+K outputs: (condition, loop carried dependencies..., scan_outputs...). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
+
+ +#### Inputs (2 - ∞) + +
+
M (optional) : I
+
A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
+
cond (optional) : B
+
A boolean termination condition. Optional. Pass empty string to skip.
+
v_initial (variadic, heterogeneous) : V
+
The initial values of any loop-carried dependencies (values that change across loop iterations)
+
+ +#### Outputs (1 - ∞) + +
+
v_final_and_scan_outputs (variadic, heterogeneous) : V
+
Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), seq(tensor(float4e2m1)), seq(tensor(float8e8m0)), seq(tensor(uint2)), seq(tensor(int2)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4)), optional(tensor(float4e2m1)), optional(tensor(float8e8m0)), optional(tensor(uint2)), optional(tensor(int2))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv13.
+
I : tensor(int64)
+
tensor of int64, which should be a scalar.
+
B : tensor(bool)
+
tensor of bool, which should be a scalar.
+
+ +### **Pad-25** + + Given a tensor containing the data to be padded (`data`), a tensor containing the number of start and end pad values for axis (`pads`), (optionally) a `mode`, and (optionally) `constant_value`, + a padded tensor (`output`) is generated. + + The four supported `modes` are (similar to corresponding modes supported by `numpy.pad`): + + 1) `constant`(default) - pads with a given constant value as specified by `constant_value` (which defaults to 0, empty string, or False) + + 2) `reflect` - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis + + 3) `edge` - pads with the edge values of array + + 4) `wrap` - wrap-around padding as if the data tensor forms a torus + + + Example 1 (`constant` mode): + + Insert 0 pads to the beginning of the second dimension. + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'constant' + + constant_value = 0.0 + + output = [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ] + ``` + + Example 2 (`reflect` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'reflect' + + output = [ + [1.0, 1.2, 1.0, 1.2], + [2.3, 3.4, 2.3, 3.4], + [4.5, 5.7, 4.5, 5.7], + ] + ``` + + Example 3 (`edge` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'edge' + + output = [ + [1.0, 1.0, 1.0, 1.2], + [2.3, 2.3, 2.3, 3.4], + [4.5, 4.5, 4.5, 5.7], + ] + ``` + + Example 4 (`wrap` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [2, 1, 1, 1] + + mode = 'wrap' + + output = [ + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + ] + ``` + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
mode : string (default is constant)
+
Supported modes: `constant`(default), `reflect`, `edge`, `wrap`
+
+ +#### Inputs (2 - 4) + +
+
data (differentiable) : T
+
Input tensor.
+
pads (non-differentiable) : tensor(int64)
+
Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. `pads` should be a 1D tensor of shape [2 * num_axes] where `num_axes` refers to the number of elements in the `axes` input or the input rank if `axes` are not provided explicitly. `pads` format should be: [x1_begin, x2_begin, ..., x1_end, x2_end,...], where xi_begin is the number of pad values added at the beginning of axis `axes[i]` and xi_end, the number of pad values added at the end of axis `axes[i]`.
+
constant_value (optional, non-differentiable) : T
+
(Optional) A scalar value to be used if the mode chosen is `constant` (by default it is 0, empty string or False).
+
axes (optional, non-differentiable) : Tind
+
1-D tensor of axes that `pads` apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated. If not provided, all axes are assumed (`[0, 1, ..., input_rank-1]`).
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types up to IRv13.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ +### **QuantizeLinear-25** + + The linear quantization operator consumes a high-precision tensor, a scale, and a zero point to compute the + low-precision/quantized tensor. The scale factor and zero point must have the same shape, determining the quantization + granularity. The quantization formula is `y = saturate((x / y_scale) + y_zero_point)`. + + Saturation is done according to: + - uint16: [0, 65535] + - int16: [-32768, 32767] + - uint8: [0, 255] + - int8: [-128, 127] + - uint4: [0, 15] + - int4: [-8, 7] + - uint2: [0, 3] + - int2: [-2, 1] + + For `(x / y_scale)`, it rounds to the nearest even. Refer to https://en.wikipedia.org/wiki/Rounding for details. + + `y_zero_point` and `y` must have the same type. `y_zero_point` is usually not used for quantization to float8 and 4bit types, but the quantization + formula remains the same for consistency, and the type of the attribute `y_zero_point` still determines the quantization type. + `x` and `y_scale` are allowed to have different types. The type of `y_scale` determines the precision of the division operation between `x` and + `y_scale`, unless the `precision` attribute is specified. + + There are three supported quantization granularities, determined by the shape of `y_scale`. + In all cases, `y_zero_point` must have the same shape as `y_scale`. + - Per-tensor (per-layer) quantization: `y_scale` is a scalar. + - Per-axis quantization: The scale must be a 1-D tensor, with the length of the quantization axis. For an input shape + `(D0, ..., Di, ..., Dn)` and `axis=i`, `y_scale` is a 1-D tensor of length `Di`. + - Blocked quantization: The scale's shape is identical to the input's shape, except for one dimension, in which + blocking is performed. Given `x` shape `(D0, ..., Di, ..., Dn)`, `axis=i`, and block size `B`: `y_scale` shape is + `(D0, ..., ceil(Di/B), ..., Dn)`. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used only for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`. When the rank of the input is 1, per-tensor quantization is applied, rendering the axis unnecessary in this scenario.
+
block_size : int (default is 0)
+
(Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given `x` shape `(D0, ..., Di, ..., Dn)`, `y_scale` shape `(S0, ... Si, ...Sn)` and `axis=i`, the accepted range is `[ceil(Di/Si), ceil(Di/(Si-1))-1]`
+
output_dtype : int (default is 0)
+
(Optional) The output data type. If not supplied, the output data type is inferred from `y_zero_point` data type (`T3`). If neither `output_dtype` nor `y_zero_point` are supplied, output data type is uint8. If both `output_dtype` and `y_zero_point` are specified, `output_dtype` must be `T3`.
+
precision : int (default is 0)
+
(Optional) The precision of the division operation between `x` and `y_scale`. If not provided, it will be the same as the type of `y_scale`.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 quantization (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. All cases are fully described in two tables inserted in the operator description.
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D full precision Input tensor to be quantized.
+
y_scale : T2
+
Scale for doing quantization to get `y`. For per-tensor/layer quantization the scale is a scalar, for per-axis quantization it is a 1-D Tensor and for blocked quantization it has the same shape as the input, except for one dimension in which blocking is performed.
+
y_zero_point (optional) : T3
+
Zero point for doing quantization to get `y`. Shape must match `y_scale`. Default is uint8 with zero point of 0 if it's not specified.
+
+ +#### Outputs + +
+
y : T3
+
N-D quantized output tensor. It has same shape as input `x`.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(float16), tensor(bfloat16), tensor(int32)
+
The type of the input 'x'.
+
T2 : tensor(float), tensor(float16), tensor(bfloat16), tensor(int32), tensor(float8e8m0)
+
The type of the input 'y_scale'.
+
T3 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(uint2), tensor(int2)
+
The type of the input `y_zero_point` and the output `y`.
+
+ +### **Reshape-25** + + Reshape the input tensor similar to numpy.reshape. + First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. + At most one dimension of the new shape can be -1. In this case, the value is + inferred from the size of the tensor and the remaining dimensions. A dimension + could also be 0, in which case the actual dimension value is unchanged (i.e. taken + from the input tensor). If 'allowzero' is set, and the new shape includes 0, the + dimension will be set explicitly to zero (i.e. not taken from input tensor). + Shape (second input) could be an empty shape, which means converting to a scalar. + The input tensor's shape and the output tensor's shape are required to have the same number of elements. + + If the attribute 'allowzero' is set, it is invalid for the specified shape to + contain both a zero value and -1, as the value of the dimension corresponding + to -1 cannot be determined uniquely. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
allowzero : int (default is 0)
+
(Optional) By default, when any value in the 'shape' input is equal to zero the corresponding dimension value is copied from the input tensor dynamically. allowzero=1 indicates that if any value in the 'shape' input is set to zero, the zero value is honored, similar to NumPy.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
shape (non-differentiable) : tensor(int64)
+
Specified shape for output.
+
+ +#### Outputs + +
+
reshaped (differentiable) : T
+
Reshaped data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types.
+
+ +### **Scan-25** + + Scan can be used to iterate over one or more scan_input tensors, + constructing zero or more scan_output tensors. It combines ideas from general recurrences, + functional programming constructs such as scan, fold, map, and zip, and is intended to enable + generalizations of RNN-like constructs for sequence-to-sequence processing. + Other tensors (referred to as state_variables here) can be used to carry a state + when iterating from one element to another (similar to hidden-state in RNNs, also referred + to as loop-carried dependences in the context of loops). + Many common usages involve a single scan_input tensor (where functionality + similar to scan, fold and map can be obtained). When more than one scan_input is used, + a behavior similar to zip is obtained. + + The attribute body must be a graph, specifying the computation to be performed in + every iteration. It takes as input the current values of the state_variables and + the current iterated element of the scan_inputs. It must return the (updated) values + of the state_variables and zero or more scan_output_element tensors. The values of the + scan_output_element tensors are concatenated over all the iterations to produce the + scan_output values of the scan construct (similar to the concatenated intermediate + hidden-state values of RNN-like constructs). All the output tensors (state_variables as + well as scan_output_element tensors) are required to have the same shape in each iteration + of the loop (a restriction imposed to enable efficient memory allocation). + + Note that the iterated element passed to the body subgraph does not have a sequence + axis. It will have a rank one less than the rank of the corresponding scan_input. + + The scan operation returns the final values of the state_variables as well as the + scan_outputs. + + The optional attribute scan_input_directions specifies the direction (forward or backward) + for each scan input. If this attribute is omitted, all sequences are scanned in the forward + direction. A bidirectional scan may be performed by specifying the same tensor input twice + in the scan_inputs, once with a forward direction, and once with a backward direction. + + The scan_output of the operation is produced by concatenating the scan_output_element + values produced by the body in each iteration. The optional attribute scan_output_directions + specifies the direction in which scan_output is constructed (by appending or prepending the + scan_output_element to scan_output in each iteration) for each scan_output. If this attribute + is omitted, the scan_output_element is appended to the scan_output in each iteration. + + The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. + If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the + batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. + Note that scanning a non-zero axis may be less efficient than scanning axis zero. + + The optional attribute scan_output_axes specifies the axis along which the scan_outputs + are accumulated for each scan_output. For example, if axis 1 is the time axis (to be + scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis + value of 1. + + Note that because of the ONNX restriction that only the last parameter of an operator can + be variadic, the initial-states and scan-inputs are listed together as one input parameter. + Similarly, the final-states and scan-outputs are listed together as one output parameter. + The attribute num_scan_inputs indicates the number M of scan-inputs. + + The behavior of + + Scan < + num_scan_inputs = m, + body = loop-body, + scan_input_axes = [axis_1, ..., axis_m] + > (init_1, ..., init_n, scan_1, ..., scan_m) + + is equivalent to the following pseudo-code: + + // scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i + // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. + sequence_length = scan_1.shape[axis_1]; + + // initialize state-variables + st_1 = init_1; ... st_n = init_n; + // initialize scan-output variables: [] denotes an empty tensor + scan_out_1 = []; ...; scan_out_k = []; + // identify number of iterations: + + // execute loop + for (int t = 0; t < sequence_length; ++t) { + // generate the scan-input elements: the notation T[t] indicates the sub-tensor + // of rank one less than T obtained by indexing T at position t along axis k. + si_1 = scan_1[t]; + ... ; + si_m = scan_m[t]; + // execute loop-body + st_1, ..., st_n, so_1, ..., so_k = loop-body(st_1, ..., st_n, si_1, ..., si_m) + // accumulate the scan-output elements + scan_out_1 = Concat(scan_out_1, so_1); ... ; scan_out_k = Concat(scan_out_k, so_k); + } + + return st_1, ..., st_n, scan_out_1, ..., scan_out_k; + + *Sample usage: Encoding RNN using a Scan* + + The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, + recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can + be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes + %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these + values are computed in the outer graph, they need to be passed in as extra state_variables. + + graph rnn-encoding { + %H_0 = ... + %X = ... + %Y_h, %Y = Scan[body = , num_scan_inputs=1](%H_0, %X) + return %Y, %Y_h + } + + graph rnn-cell-1 ( + %H_tminus1[FLOAT, tensor] + %X_t[FLOAT, tensor] + ) { + %Wi = ... + %Ri = ... + %Wbi = ... + %Rbi = ... + %t1 = X_t * (Wi^T) + %t2 = H_tminus1*(Ri^T) + %t3 = Add(%t1, %t2) + %t4 = Add(%t3, %Wbi) + %t5 = Add(%t4, %Rbi) + %Ht = Tanh(%t5) + %Accumulate = Identity(%Ht) + return %Ht, %Accumulate + } + + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has N+M inputs: (loop state variables..., scan_input_elts...). It has N+K outputs: (loop state variables..., scan_output_elts...). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
+
num_scan_inputs : int (required)
+
An attribute specifying the number of scan_inputs M.
+
scan_input_axes : list of ints
+
An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
scan_input_directions : list of ints
+
An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
+
scan_output_axes : list of ints
+
An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
+
scan_output_directions : list of ints
+
An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
+
+ +#### Inputs (1 - ∞) + +
+
initial_state_and_scan_inputs (variadic, heterogeneous) : V
+
Initial values of the loop's N state variables followed by M scan_inputs
+
+ +#### Outputs (1 - ∞) + +
+
final_state_and_scan_outputs (variadic, heterogeneous) : V
+
Final values of the loop's N state variables followed by K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
All Tensor types up to IRv13.
+
+ +### **Shape-25** + + Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. + Optional attributes start and end can be used to compute a slice of the input tensor's shape. + If start axis is omitted, the slice starts from axis 0. + The end axis, if specified, is exclusive (and the returned value will not include the size of that axis). + If the end axis is omitted, the axes upto the last one will be included. + Negative axes indicate counting back from the last axis. + Note that axes will be clamped to the range [0, r], where r is the + rank of the input tensor if they are out-of-range (after adding r in the case of + negative axis). Thus, specifying any end value > r is equivalent to specifying an end + value of r, and specifying any start value < -r is equivalent to specifying a start + value of 0. If start > end, the result will be an empty shape. + + Examples: + + ``` + Input tensor with shape: [2, 3, 4] + No attributes specified. + Output: [2, 3, 4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: -1 + Output: [4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + end: -1 + Output: [2, 3] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: 1 + end: 2 + Output: [3] + ``` + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
end : int
+
(Optional) Ending axis for slicing the shape. Negative value means counting dimensions from the back. If omitted, sizes of all axes upto (including) the last one will be included.
+
start : int (default is 0)
+
(Optional) Starting axis for slicing the shape. Default value is 0.Negative value means counting dimensions from the back.
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
shape (non-differentiable) : T1
+
Shape of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor.
+
+ +### **Size-25** + + Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
size (non-differentiable) : T1
+
Total number of elements of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor, which should be a scalar though.
+
+ +### **Squeeze-25** + + Remove single-dimensional entries from the shape of a tensor. + Takes an input `axes` with a list of axes to squeeze. + If `axes` is not provided, all the single dimensions will be removed from + the shape. If an axis is selected with shape entry not equal to one, an error is raised. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
Tensors with at least max(dims) dimensions.
+
axes (optional, non-differentiable) : tensor(int64)
+
1D tensor of integers indicating the dimensions to squeeze. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
squeezed (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types up to IRv13.
+
+ +### **Transpose-25** + + Returns a transpose of the input tensor. (Similar to `numpy.transpose`). + The optional attribute `perm` must be a permutation of the dimensions of + the input tensor. Axis `i` of the output tensor corresponds to the axis + `perm[i]` of the input tensor. + For example, when perm=(1, 0, 2), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 1, 3). + When perm=(1, 2, 0), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 3, 1). + If the attribute `perm` is omitted, its default value is `(n-1, ..., 0)`, + where `n` is the rank of the input tensor. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Attributes + +
+
perm : list of ints
+
A list of integers. By default, reverse the dimensions, otherwise permute the axes according to the values given. Its length must be equal to the rank of the input.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
transposed (differentiable) : T
+
Transposed output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types.
+
+ +### **Unsqueeze-25** + + Insert single-dimensional entries to the shape of an input tensor (`data`). + Takes one required input `axes` - which contains a list of dimension indices and this operator will insert a dimension of value `1` into the corresponding index of the output tensor (`expanded`). + + For example, given an input tensor (`data`) of shape [3, 4, 5], then + Unsqueeze(data, axes=[0, 4]) outputs a tensor (`expanded`) containing same data as `data` but with shape [1, 3, 4, 5, 1]. + + The input `axes` should not contain any duplicate entries. It is an error if it contains duplicates. + The rank of the output tensor (`output_rank`) is the rank of the input tensor (`data`) plus the number of values in `axes`. + Each value in `axes` should be within the (inclusive) range [-output_rank , output_rank - 1]. + The order of values in `axes` does not matter and can come in any order. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +#### Inputs + +
+
data (differentiable) : T
+
Original tensor
+
axes (non-differentiable) : tensor(int64)
+
1D tensor of integers indicating the dimensions to be inserted. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(expanded).
+
+ +#### Outputs + +
+
expanded (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types up to IRv13.
+
+ +## Version 26 of the default ONNX operator set +### **BitCast-26** + + Reinterprets the binary representation of a tensor as a different data type, + specified by the 'to' attribute. Unlike Cast, BitCast preserves the exact bit + pattern without any value conversion. + + The target data type must have the same bit-width as the input data type. + The output tensor has the same shape as the input tensor. + All types except string are supported. Implementations must treat the + underlying bytes as little endian. + +#### Version + +This version of the operator has been available since version 26 of the default ONNX operator set. + +#### Attributes + +
+
to : int (required)
+
The data type to which the input tensor is bitwise reinterpreted. Must be one of the non-string types from DataType enum in TensorProto. The target type must have the same bit-width as the input type.
+
+ +#### Inputs + +
+
input (non-differentiable) : T1
+
Input tensor to be bitcast.
+
+ +#### Outputs + +
+
output (non-differentiable) : T2
+
Output tensor with the same shape as the input.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input types. Bitcasting from string is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain output types. Bitcasting to string is not supported.
+
+ +### **CumProd-26** + + Performs cumulative product of the input elements along the given axis. + By default, it will do the product inclusively meaning the first element is copied as is. + Through an `exclusive` attribute, this behavior can change to exclude the first element. + It can also perform product in the opposite direction of the axis. For that, set `reverse` attribute to 1. + + Example: + ``` + input_x = [1, 2, 3] + axis=0 + output = [1, 2, 6] + exclusive=1 + output = [1, 1, 2] + exclusive=0 + reverse=1 + output = [6, 6, 3] + exclusive=1 + reverse=1 + output = [6, 3, 1] + ``` + + +#### Version + +This version of the operator has been available since version 26 of the default ONNX operator set. + +#### Attributes + +
+
exclusive : int (default is 0)
+
If set to 1 will return exclusive product in which the top element is not included. In other terms, if set to 1, the j-th output element would be the product of the first (j-1) elements. Otherwise, it would be the product of the first j elements.
+
reverse : int (default is 0)
+
If set to 1 will perform the products in reverse direction.
+
+ +#### Inputs + +
+
x (differentiable) : T
+
An input tensor that is to be processed.
+
axis (non-differentiable) : T2
+
A 0-D tensor. Must be in the range [-rank(x), rank(x)-1]. Negative value means counting dimensions from the back.
+
+ +#### Outputs + +
+
y (differentiable) : T
+
Output tensor of the same type as 'x' with cumulative products of the x's elements
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
T2 : tensor(int32), tensor(int64)
+
axis tensor can be int32 or int64 only
+
+ +## Version 27 of the default ONNX operator set +### **CausalConvWithState-27** + + Stateful causal 1D depthwise convolution. + + Used by Gated DeltaNet (Qwen3.5) and Mamba (Jamba, FalconMamba) as a preprocessing step. + Replaces the 3-op pattern (Concat + Conv + Slice) with a single fused operation. + + The convolution is causal (looks only at current and past positions) and depthwise + (each channel is convolved independently with its own kernel). + + The input, weight, past_state, output, and present_state tensors are rank-3 with + shape (batch_size, channels, length). The optional bias input is rank-1 with + shape (channels). For higher-dimensional data, use Reshape nodes before and + after this operator to pack extra dimensions into the batch or channel axis. + + Weight layout: (channels, 1, k) for depthwise convolution. + The carry state stores the last (k-1) positions for incremental decode. + + The optional activation attribute supports fused SiLU/Swish activation. + + +#### Version + +This version of the operator has been available since version 27 of the default ONNX operator set. + +#### Attributes + +
+
activation : string (default is none)
+
Fused activation function. One of: 'silu', 'swish', 'none'. Default is 'none'.
+
+ +#### Inputs (2 - 4) + +
+
input (differentiable) : T
+
Input tensor with shape (batch_size, channels, length). Channels-first layout.
+
weight (differentiable) : T
+
Depthwise convolution kernel with shape (channels, 1, k) where k is the kernel size. The middle dim of size 1 follows the ONNX `Conv` weight layout `(M, C/group, k1, ..., kn)`: since this op is always depthwise, `group = channels`, so `C/group = 1`. Keeping this layout makes the weight tensor a drop-in for a depthwise `Conv(group=channels)` weight, so `Conv` <-> `CausalConvWithState` rewrites require no reshape.
+
bias (optional, differentiable) : T
+
Optional per-channel bias with shape (channels).
+
past_state (optional, non-differentiable) : T
+
Carry state from previous step with shape (batch_size, channels, k - 1). If not provided, padding is zero.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Convolution output with same shape as input.
+
present_state (non-differentiable) : T
+
Updated carry state with shape (batch_size, channels, k - 1). Contains the last (k - 1) values of the effective padded/concatenated sequence along the causal axis, including any values from past_state or zero-padding when the current input is shorter than k - 1.
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(float16), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ +### **LinearAttention-27** + + Unified linear attention operator for autoregressive decoding (T=1) and prefill (T>1). + + The query, key, value, and (where applicable) decay/beta inputs use 3D packed format + [B, T, H*D], where heads are flattened into the last dimension; q_num_heads and + kv_num_heads are always required and are used to unpack to 4D internally for computation. + The optional past_state and present_state are 4D with shape (B, H_kv, d_k, d_v). + + Group-query attention (GQA) is supported: q_num_heads must be a positive multiple of + kv_num_heads. When q_num_heads == kv_num_heads this reduces to multi-headed linear + attention; when q_num_heads > kv_num_heads each KV head (and its recurrent state) is + shared by `q_num_heads / kv_num_heads` query heads (multi-query attention is the + special case kv_num_heads == 1). + + The update_rule attribute selects the recurrence type: + - "linear": S_t = S_{t-1} + k_t ⊗ v_t; o_t = scale * q_t^T S_t + - "gated": S_t = exp(g_t) * S_{t-1} + k_t ⊗ v_t; o_t = scale * q_t^T S_t + - "delta": S_t = S_{t-1} + β_t * k_t ⊗ (v_t - S_{t-1}^T k_t); o_t = scale * q_t^T S_t + - "gated_delta": S_t = exp(g_t) * S_{t-1} + β_t * k_t ⊗ (v_t - exp(g_t) * S_{t-1}^T k_t); o_t = scale * q_t^T S_t + + where g_t is the decay (in log-space), β_t is the update rate, and ⊗ denotes outer product. + + Semantics: Equivalent to running the recurrent update sequentially for each token, + but may be implemented using chunk-parallel algorithms for GPU efficiency. + + +#### Version + +This version of the operator has been available since version 27 of the default ONNX operator set. + +#### Attributes + +
+
chunk_size : int (default is 64)
+
Chunk size for the chunk-parallel WY decomposition during prefill (T>1). Tuning hint; does not affect output correctness.
+
kv_num_heads : int (required)
+
Number of key/value heads. Always required.
+
q_num_heads : int (required)
+
Number of query heads. Always required.
+
scale : float (default is 0.0)
+
Output scaling factor. When 0.0 (default), derives d_k = query.shape[-1] / q_num_heads and uses 1/sqrt(d_k). Set explicitly to override.
+
update_rule : string (default is gated_delta)
+
The update rule for the linear attention recurrence. One of: 'linear', 'gated', 'delta', 'gated_delta'. Default is 'gated_delta'.
+
+ +#### Inputs (3 - 6) + +
+
query (differentiable) : T
+
Query vectors with 3D packed shape (B, T, H_q * d_k). Heads are packed into the last dimension.
+
key (differentiable) : T
+
Key vectors with 3D packed shape (B, T, H_kv * d_k). Should be L2-normalized for delta/gated_delta modes.
+
value (differentiable) : T
+
Value vectors with 3D packed shape (B, T, H_kv * d_v).
+
past_state (optional, non-differentiable) : S
+
Recurrent state from previous step with shape (B, H_kv, d_k, d_v). Always 4D. If not provided, defaults to zeros.
+
decay (optional, differentiable) : T
+
Exponential decay gate in log-space. 3D packed shape: (B, T, H_kv * d_k) for per-key-dimension decay (GLA/RWKV-6), or (B, T, H_kv) for per-head scalar decay (DeltaNet/RetNet). Required for 'gated' and 'gated_delta' modes.
+
beta (optional, differentiable) : T
+
Update rate (sigmoid output). 3D packed shape: (B, T, H_kv) or (B, T, 1). Required for 'delta' and 'gated_delta' modes.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Attention output with 3D packed shape (B, T, H_q * d_v).
+
present_state (non-differentiable) : S
+
Updated recurrent state with shape (B, H_kv, d_k, d_v). Always 4D.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(bfloat16), tensor(float)
+
Constrain activation input and output types to float16, bfloat16, or float32 tensors.
+
S : tensor(float16), tensor(bfloat16), tensor(float)
+
Constrain state types to float16, bfloat16, or float32 tensors. Should be float32 or the same as T for numerical stability on long sequences.
+
+ +### **Range-27** + + Generate a tensor containing a sequence of numbers that begin at `start` and extends by increments of `delta` + up to `limit` (exclusive). + + The number of elements in the output of range is computed as below: + ``` + number_of_elements = max( ceil( (limit - start) / delta ) , 0 ) + ``` + The pseudocode determining the contents of the output is shown below: + ``` + for(int i=0; i +
stash_type : int (default is 1)
+
The data type used for intermediate computation when T is float16 or bfloat16. Defaults to 1 (float). Has no effect for other types.
+ + +#### Inputs + +
+
start : T
+
Scalar. First entry for the range of output values.
+
limit : T
+
Scalar. Exclusive upper limit for the range of output values.
+
delta : T
+
Scalar. Value to step by.
+
+ +#### Outputs + +
+
output : T
+
A 1-D tensor with same type as the inputs containing generated range of values.
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(bfloat16)
+
Constrain input types to common numeric type tensors.
+
+ +## Version 28 of the default ONNX operator set +### **Celu-28** + + Continuously Differentiable Exponential Linear Units: + Perform the linear unit element-wise on the input tensor X + using formula: + + ``` + max(0,x) + min(0,alpha*(exp(x/alpha)-1)) + ``` + +#### Version + +This version of the operator has been available since version 28 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
The Alpha value in Celu formula which control the shape of the unit. The default value is 1.0.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +# ai.onnx.preview +## Version 1 of the 'ai.onnx.preview' operator set +### **ai.onnx.preview.FlexAttention-1** + + Computes scaled dot-product attention over rank-4 (batched, multi-head) inputs, + with optional user-provided customization subgraphs at two stages: + + 1. score_mod: Modify the attention score tensor after Q·K^T + 2. prob_mod: Modify the probability tensor after Softmax + + This operator mirrors the capabilities of PyTorch's flex_attention: + https://docs.pytorch.org/docs/stable/nn.attention.flex_attention.html + + Input Shapes (MUST be rank-4 tensors): + - Q: `(batch_size, q_num_heads, q_sequence_length, head_size)` + - K: `(batch_size, kv_num_heads, kv_sequence_length, head_size)` + - V: `(batch_size, kv_num_heads, kv_sequence_length, v_head_size)` + + Output Shape: + - Y: `(batch_size, q_num_heads, q_sequence_length, v_head_size)` + + FlexAttention Computation: + ``` + Scores = (Q @ K^T) * scale + Scores = score_mod(Scores) # if 'score_mod' is provided + Probs = Softmax(Scores, axis=-1) + Probs = prob_mod(Probs) # if 'prob_mod' is provided + Y = Probs @ V + ``` + + Grouped Query Attention (GQA): + When `q_num_heads != kv_num_heads`, each K/V head is shared by a contiguous + group of query heads in head-index order. Let + `group_size = q_num_heads / kv_num_heads`; then query head `h` uses K/V head + `floor(h / group_size)`. `q_num_heads` must be a multiple of + `kv_num_heads`. + + Modifier Subgraphs (score_mod, prob_mod): + Each modifier subgraph takes exactly one rank-4 tensor input and must produce + exactly one rank-4 tensor output of the same shape and element type. + - score_mod input/output shape: `(batch_size, q_num_heads, q_sequence_length, kv_sequence_length)` + - prob_mod input/output shape: `(batch_size, q_num_heads, q_sequence_length, kv_sequence_length)` + The element type is determined by softmax_precision (defaults to float32 for + non-double inputs, otherwise double). + + Masking can be expressed in score_mod by writing masked positions as -inf (or a + large negative value appropriate for the target precision). + +#### Version + +No versioning maintained for experimental ops. +#### Attributes + +
+
prob_mod : graph
+
Optional probability modifier subgraph with 1 rank-4 tensor input and 1 rank-4 tensor output of the same shape and element type: (probs) -> probs_out. probs has softmax_precision element type and shape (B, Hq, L, S). The output must preserve the input shape.
+
scale : float
+
Scaling factor for Q*K^T. Defaults to 1/sqrt(head_size).
+
score_mod : graph
+
Optional score modifier subgraph with 1 rank-4 tensor input and 1 rank-4 tensor output of the same shape and element type: (scores) -> scores_out. scores has softmax_precision element type and shape (B, Hq, L, S). The output must preserve the input shape.
+
softmax_precision : int
+
Floating-point precision for softmax computation. Defaults to float32 for non-double inputs, otherwise uses double. Must be explicitly specified for non-float types.
+
+ +#### Inputs + +
+
Q : T1
+
Query tensor with shape `(batch_size, q_num_heads, q_seq_len, head_size)`.
+
K : T1
+
Key tensor with shape `(batch_size, kv_num_heads, kv_seq_len, head_size)`.
+
V : T1
+
Value tensor with shape `(batch_size, kv_num_heads, kv_seq_len, v_head_size)`.
+
+ +#### Outputs + +
+
Y : T1
+
Output tensor with shape `(batch_size, q_num_heads, q_seq_len, v_head_size)`.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain Q, K, V to float tensors.
+
+ +# ai.onnx.preview.training +## Version 1 of the 'ai.onnx.preview.training' operator set +### **ai.onnx.preview.training.Adagrad-1** + + Compute one iteration of ADAGRAD, a stochastic gradient based optimization + algorithm. This operator can conduct the optimization of multiple tensor variables. + + Let's define the behavior of this operator. As you can imagine, ADAGRAD requires + some parameters: + + - The initial learning-rate "R". + - The update count "T". That is, the number of training iterations conducted. + - A L2-norm regularization coefficient "norm_coefficient". + - A learning-rate decay factor "decay_factor". + - A small constant "epsilon" to avoid dividing-by-zero. + + At each ADAGRAD iteration, the optimized tensors are moved along a direction + computed based on their estimated gradient and accumulated squared gradient. Assume + that only a single tensor "X" is updated by this operator. We need the value of "X", + its gradient "G", and its accumulated squared gradient "H". Therefore, variables in + this operator's input list are sequentially "R", "T", "X", "G", and "H". Other + parameters are given as attributes because they are usually constants. Also, the + corresponding output tensors are the new value of "X" (called "X_new"), and then + the new accumulated squared gradient (called "H_new"). Those outputs are computed + from the given inputs following the pseudo code below. + + Let "+", "-", "*", and "/" are all element-wise arithmetic operations with + numpy-style broadcasting support. The pseudo code to compute those outputs is: + + // Compute a scalar learning-rate factor. At the first update of X, T is generally + // 0 (0-based update index) or 1 (1-based update index). + r = R / (1 + T * decay_factor); + + // Add gradient of 0.5 * norm_coefficient * ||X||_2^2, where ||X||_2 is the 2-norm. + G_regularized = norm_coefficient * X + G; + + // Compute new accumulated squared gradient. + H_new = H + G_regularized * G_regularized; + + // Compute the adaptive part of per-coordinate learning rate. Note that Sqrt(...) + // computes element-wise square-root. + H_adaptive = Sqrt(H_new) + epsilon + + // Compute the new value of "X". + X_new = X - r * G_regularized / H_adaptive; + + If one assign this operators to optimize multiple inputs, for example, "X_1" and "X_2", the same + pseudo code may be extended to handle all tensors jointly. More specifically, we can view "X" as a + concatenation of "X_1" and "X_2" (of course, their gradient and accumulate gradient should + be concatenated too) and then just reuse the entire pseudo code. + + Note that ADAGRAD was first proposed in http://jmlr.org/papers/volume12/duchi11a/duchi11a.pdf. + In that reference paper, this operator is a special case of the Figure 1's composite mirror + descent update. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.preview.training' operator set. + +#### Attributes + +
+
decay_factor : float (default is 0.0)
+
The decay factor of learning rate after one update.The effective learning rate is computed by r = R / (1 + T * decay_factor). Default to 0 so that increasing update counts doesn't reduce the learning rate.
+
epsilon : float (default is 0.0)
+
Small scalar to avoid dividing by zero.
+
norm_coefficient : float (default is 0.0)
+
Regularization coefficient in 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization.
+
+ +#### Inputs (3 - ∞) + +
+
R : T1
+
The initial learning rate.
+
T : T2
+
The update count of "X". It should be a scalar.
+
inputs (variadic, heterogeneous) : T3
+
The current values of optimized tensors, followed by their respective gradients, followed by their respective accumulated squared gradients.For example, if two tensor "X_1" and "X_2" are optimized, The input list would be ["X_1", "X_2", gradient of "X_1", gradient of "X_2", accumulated squared gradient of "X_1", accumulated squared gradient of "X_2"].
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : T3
+
Updated values of optimized tensors, followed by their updated values of accumulated squared gradients. For example, if two tensor "X_1" and "X_2" are optimized, the output list would be [new value of "X_1," new value of "X_2" new accumulated squared gradient of "X_1", new accumulated squared gradient of "X_2"].
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double)
+
Constrain input types to float scalars.
+
T2 : tensor(int64)
+
Constrain input types to 64-bit integer scalars.
+
T3 : tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **ai.onnx.preview.training.Adam-1** + + Compute one iteration of Adam, a stochastic gradient based optimization + algorithm. This operator can conduct the optimization of multiple tensor variables. + + Let's define the behavior of this operator. First of all, Adam requires + some parameters: + + - The learning-rate "R". + - The update count "T". That is, the number of training iterations conducted. + - A L2-norm regularization coefficient "norm_coefficient". + - A small constant "epsilon" to avoid dividing-by-zero. + - Two coefficients, "alpha" and "beta". + + At each Adam iteration, the optimized tensors are moved along a direction + computed based on their exponentially-averaged historical gradient and + exponentially-averaged historical squared gradient. Assume that only a tensor + "X" is being optimized. The rest of required information is + + - the value of "X", + - "X"'s gradient (denoted by "G"), + - "X"'s exponentially-averaged historical gradient (denoted by "V"), and + - "X"'s exponentially-averaged historical squared gradient (denoted by "H"). + + Some of those parameters are passed into this operator as input tensors and others + are stored as this operator's attributes. Specifically, this operator's input tensor + list is ["R", "T", "X", "G", "V", "H"]. That is, "R" is the first input, "T" is + the second input, and so on. Other parameters are given as attributes because they + are constants. Moreover, the corresponding output tensors are + + - the new value of "X" (called "X_new"), + - the new exponentially-averaged historical gradient (denoted by "V_new"), and + - the new exponentially-averaged historical squared gradient (denoted by "H_new"). + + Those outputs are computed following the pseudo code below. + + Let "+", "-", "*", and "/" are all element-wise arithmetic operations with + numpy-style broadcasting support. The pseudo code to compute those outputs is: + + // Add gradient of 0.5 * norm_coefficient * ||X||_2^2, where ||X||_2 is the 2-norm. + G_regularized = norm_coefficient * X + G + + // Update exponentially-averaged historical gradient. + V_new = alpha * V + (1 - alpha) * G_regularized + + // Update exponentially-averaged historical squared gradient. + H_new = beta * H + (1 - beta) * G_regularized * G_regularized + + // Compute the element-wise square-root of H_new. V_new will be element-wisely + // divided by H_sqrt for a better update direction. + H_sqrt = Sqrt(H_new) + epsilon + + // Compute learning-rate. Note that "alpha**T"/"beta**T" is alpha's/beta's T-th power. + R_adjusted = T > 0 ? R * Sqrt(1 - beta**T) / (1 - alpha**T) : R + + // Compute new value of "X". + X_new = X - R_adjusted * V_new / H_sqrt + + // Post-update regularization. + X_final = (1 - norm_coefficient_post) * X_new + + If there are multiple inputs to be optimized, the pseudo code will be applied + independently to each of them. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.preview.training' operator set. + +#### Attributes + +
+
alpha : float (default is 0.9)
+
Coefficient of previously accumulated gradient in running average. Default to 0.9.
+
beta : float (default is 0.999)
+
Coefficient of previously accumulated squared-gradient in running average. Default to 0.999.
+
epsilon : float (default is 0.0)
+
Small scalar to avoid dividing by zero.
+
norm_coefficient : float (default is 0.0)
+
Regularization coefficient of 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization.
+
norm_coefficient_post : float (default is 0.0)
+
Regularization coefficient of 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization.
+
+ +#### Inputs (3 - ∞) + +
+
R : T1
+
The initial learning rate.
+
T : T2
+
The update count of "X". It should be a scalar.
+
inputs (variadic, heterogeneous) : T3
+
The tensors to be optimized, followed by their respective gradients, followed by their respective accumulated gradients (aka momentum), followed by their respective accumulated squared gradients. For example, to optimize tensors "X_1" and "X_2,", the input list would be ["X_1", "X_2", gradient of "X_1", gradient of "X_2", accumulated gradient of "X_1", accumulated gradient of "X_2", accumulated squared gradient of "X_1", accumulated squared gradient of "X_2"].
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : T3
+
New values of optimized tensors, followed by their respective new accumulated gradients, followed by their respective new accumulated squared gradients. For example, if two tensors "X_1" and "X_2" are optimized, the outputs list would be [new value of "X_1", new value of "X_2", new accumulated gradient of "X_1", new accumulated gradient of "X_2", new accumulated squared gradient of "X_1", new accumulated squared gradient of "X_2"].
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double)
+
Constrain input types to float scalars.
+
T2 : tensor(int64)
+
Constrain input types to 64-bit integer scalars.
+
T3 : tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ +### **ai.onnx.preview.training.Gradient-1** + + Gradient operator computes the partial derivatives of a specific tensor w.r.t. + some other tensors. This operator is widely used in gradient-based training + algorithms. To illustrate its use, let's consider a computation graph, + + ``` + X -----. + | + v + W --> Conv --> H --> Gemm --> Y + ^ + | + Z + ``` + + , where W and Z are trainable tensors. Note that operators' attributes are + omitted for the sake of simplicity. Let dY/dW (dY/dZ) be the gradient of + Y with respect to W (Z). The user can compute gradient by inserting Gradient + operator to form another graph shown below. + + ``` + W --> Conv --> H --> Gemm --> Y + | ^ ^ + | | | + | X Z + | | | + | | .----------' + | | | (W/Z/X is the 1st/2nd/3rd input of Gradient as shown in + | | | "xs" followed by "zs") + | v v + '---> Gradient(xs=["W", "Z"], zs=["X"], y="Y") + | | + | '-----------------------------------> dY/dW (1st output of Gradient) + | + '---------------------------------------> dY/dZ (2nd output of Gradient) + ``` + + By definition, the tensor "y" is a function of independent variables in "xs" + and "zs". Since we only compute the gradient of "y" w.r.t. the differentiable + variables in "xs", this Gradient only outputs dY/dW and dY/dZ. Note that "H" + cannot appear in "xs" and "zs". The reason is that "H" can be determined by + tensors "W" and "X" and therefore "H" is not an independent variable. + + All outputs are optional. If needed, for example, user can assign an empty + string to the 1st output name of that Gradient to skip the generation of dY/dW. + Note that the concept of optional outputs can also be found in ONNX's RNN, GRU, + and LSTM. + + Gradient operator can compute derivative against intermediate tensors. For + example, the gradient of Y with respect to H can be done via + + ``` + W --> Conv --> H --> Gemm --> Y + ^ | ^ + | | | + X | Z + .-------' | + | .----------' + | | (H/Z is the 1st/2nd input of Gradient as shown in "xs") + v v + Gradient(xs=["H", "Z"], y="Y") + | | + | '-----------------------------------> dY/dH (1st output of Gradient) + | + '---------------------------------------> dY/dZ (2nd output of Gradient) + ``` + + It is possible to represent high-order differentiation using Gradient operators. + For example, given the following linear model: + + ``` + W --> Gemm --> Y --> Loss --> O + ^ ^ + | | + X L + ``` + + To compute the 2nd order derivative of O with respect to W (denoted by + d^2O/dW^2), one can do + + ``` + W --> Gemm --> Y --> Loss --> O + | ^ ^ + | | | + | X .------------L + | | | | + | | | v + +------+-+> Gradient(xs=["X", "W"], zs=["L"], y="O") ---> dO/dX (1st output of Gradient) + | | | | + | | | '---> dO/dW (2nd output of Gradient) + | v v + '---> Gradient(xs=["X", "W"], zs=["L"], y="dO/dW") ---> d(dO/dW)dX (1st output of + | Gradient) + | + | + '---> d^2O/dW^2 (2nd output of Gradient) + ``` + + The tensors named in attributes "xs", "zs", and "y" define the differentiated + computation graph, and the inputs to Gradient node define the values at + which the gradient is computed. We can feed different tensors to the identified + graph. For example, one can compute the gradient of Y with respect to H at + a specific value of H, H_1, by providing that value as an input to the Gradient + node. + + ``` + W --> Conv --> H --> Gemm --> Y + ^ ^ + | | + X Z + + Z_1 (2nd input of Gradient) + | + v + H_1 --> Gradient(xs=["H", "Z"], y="Y") ---> dY/dH when H = H_1 and Y = Y_1. + | + '------------------------------> dY/dZ (2nd output of Gradient) + ``` + + When the inputs of Gradient are the tensors named in "xs" and "zs", the + computation can be optimized. More specifically, intermediate variables in + forward pass can be reused if the gradient is computed via reverse-mode + auto-differentiation. + + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.preview.training' operator set. + +#### Attributes + +
+
xs : list of strings (required)
+
Input tensor names of the differentiated sub-graph. It contains only the necessary differentiated inputs of a (sub-)graph. Variables (usually called intermediate variables) that can be generated from inputs cannot be included in this attribute.
+
y : string (required)
+
The targeted tensor. It can be viewed as the output of the differentiated function. The attribute "xs" and attribute "zs" are the minimal independent variable set that determines the value of "y".
+
zs : list of strings
+
Input tensor names of the differentiated sub-graph. It contains only the necessary non-differentiated inputs of a (sub-)graph. Variables (usually called intermediate variables) that can be generated from inputs cannot be included in this attribute.
+
+ +#### Inputs (1 - ∞) + +
+
Inputs (variadic, heterogeneous) : T1
+
The values fed into graph identified by the attributes. The i-th input is the value of the i-th tensor specified in the concatenated list of the attribute "xs" and the attribute "zs". For example, if xs=["A", "B"] and zs=["C"], the first input is used as the value of symbol "A" and the 3rd input is substituted for all the occurrences of "C".
+
+ +#### Outputs (1 - ∞) + +
+
Outputs (variadic, heterogeneous) : T2
+
The gradient of the tensor specified by the attribute "y" with respect to each of tensors specified in the attribute "xs". The i-th output is the gradient of "y" with respect to the i-th tensor specified in the attribute "xs".
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Allow outputs to be any kind of tensor.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Allow inputs to be any kind of floating-point tensor.
+
+ +### **ai.onnx.preview.training.Momentum-1** + + Compute one iteration of stochastic gradient update with momentum. + This operator can conduct the optimization of multiple tensor variables. + + Let's define the behavior of this operator. As you can imagine, SG with momentum requires + several parameters: + + - The learning-rate "R". + - The update count "T". That is, the number of conducted training iterations. It should + be zero in the first training iteration. + - A L2-norm regularization coefficient "norm_coefficient". + - A decay coefficient of previous accumulated gradient (i.e., momentum) "alpha". + - The scaling coefficient of current gradient "beta". + - An attribute to choose either standard momentum or Nesterov's momentum "mode" should + be used. + + For the sake of simplicity, assume that there is only one tensor (called "X") to be optimized. + Other necessary inputs are "X"'s gradient (called "G") and "X"'s momentum (called "V"). This + Momentum operator maps all these inputs to the new value of "X" (called "X_new") and its new + momentum (called "V_new"). + + This operator supports two different momentum algorithms. Set the attribute "mode" to + "nesterov" if Nesterov's momentum is desired. Otherwise, set the attribute "model" to + "standard" to use standard momentum. Computation details are described subsequently. + + Let "+", "-", "*", and "/" are all element-wise operations with numpy-style broadcasting. + + Pseudo code for SG with standard momentum: + + // Add gradient of 0.5 * norm_coefficient * ||X||^2, where ||X|| is the sum of squared + // values of all elements in X. + G_regularized = norm_coefficient * X + G + + // In the first training iteration, beta should always be 1. + beta_adjusted = T > 0 ? beta : 1 + + // Compute the current momentum based on previous momentum and the current gradient. + V_new = alpha * V + beta_adjusted * G_regularized + + // Update X. + X_new = X - R * V_new + + Pseudo code for SG with Nesterov's momentum: + + // Add gradient of 0.5 * norm_coefficient * ||X||^2, where ||X|| is the sum of squared + // values of all elements in X. + G_regularized = norm_coefficient * X + G; + + // In the first training iteration, beta should always be 1. + beta_adjusted = T > 0 ? beta : 1 + + // Compute the current momentum based on previous momentum and the current gradient. + V_new = alpha * V + beta_adjusted * G_regularized; + + // Compute final update direction and then update X. + X_new = X - R * (G_regularized + alpha * V_new) + + If one assign this operators to optimize multiple inputs, for example, "X_1" and "X_2". The same + pseudo code would be extended to handle all tensors jointly. More specifically, we can view "X" as a + concatenation of "X_1" and "X_2" (of course, their gradient and accumulate gradient should + be concatenated too) and then our pseudo code becomes applicable. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.preview.training' operator set. + +#### Attributes + +
+
alpha : float (required)
+
The decay factor of momentum. It should be a scalar.
+
beta : float (required)
+
The coefficient of gradient in computing new momentum. It should be a scalar.
+
mode : string (required)
+
Its value should be either "nesterov" or "standard". The value "nesterov" leads to the use of Nesterov's momentum while "standard" invokes stochastic gradient method using standard momentum
+
norm_coefficient : float (required)
+
Coefficient of 0.5 * norm_coefficient * ||X||^2.
+
+ +#### Inputs (3 - ∞) + +
+
R : T1
+
The learning rate.
+
T : T2
+
Update count of "X". It should be a scalar.
+
inputs (variadic, heterogeneous) : T3
+
It sequentially contains the current values of optimized tensors, then their gradient tensors, and finally their momentum tensors. For example, if two tensors "X_1" and "X_2" are optimized, The expected input list would be ["X_1", "X_2", gradient of "X_1", gradient of "X_2", momentum of "X_1", momentum of "X_2"].
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : T3
+
It sequentially contains the new values of optimized tensors and then the new values of their momentum tensors. For example, if two tensors "X_1" and "X_2" are optimized, the output list would be [new value of "X_1," new value of "X_2" new momentum of "X_1", new momentum of "X_2"].
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double)
+
Constrain input types to float scalars.
+
T2 : tensor(int64)
+
Constrain input types to 64-bit integer scalars.
+
T3 : tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
+ diff --git a/docs/DefineDifferentiability.md b/docs/DefineDifferentiability.md new file mode 100644 index 0000000..1128f7d --- /dev/null +++ b/docs/DefineDifferentiability.md @@ -0,0 +1,126 @@ + + +# A Short Guide on the Differentiability Tag for ONNX Operators + +## Differentiability Tag + +The ONNX operator schema for each operator includes a differentiability tag for each input and output. +In this document, we explain the meaning of this tag and how to ensure the correctness of the tags. +Briefly, the tag identifies the set of differentiable inputs and differentiable outputs of an operator. +The meaning of the tag is that the partial derivative of each differentiable output is defined with respect to each differentiable output. + +## Ways to Define Differentiability Tag + +The differentiability definition of an operator consists of several aspects. + +- Differentiable inputs, which can be referenced in Gradient's `xs` attribute. +- Differentiable outputs, which can be referenced in Gradient's `y` attribute. +- The math equation to compute the Jacobian matrix (or tensor). If a variable (input or output) is differentiable or not is judged by math. If the Jacobian matrix (or tensor) exists, then the considered operator has some differentiable inputs and outputs. + +There are several strategies to implement auto-differentiation such as forward accumulation, backward accumulation, and dual variable. +Because most deep learning frameworks are backward-based, the reviewers should ensure the PR authors of tags provide enough details on that. +We present a couple of methods below to verify the differentiability for ONNX operator. + +### Method 1: Reuse Existing Deep Learning Frameworks + +The first way is to show that the considered operator's backward operation exists in an existing framework such as Pytorch or Tensorflow. In this case, the author should provide a runnable python script which computes the backward pass of the considered operator. The author should also point out how to map the Pytorch or Tensor code to ONNX format (for example, the author can call `torch.onnx.export` to save an ONNX model). The following script shows the differentiability of ONNX Reshape using Pytorch. + +```python +import torch +import torch.nn as nn + +# A single-operator model. It's literally a Pytorch Reshape. +# Note that Pytorch Reshape can be directly mapped to ONNX Reshape. +class MyModel(nn.Module): + def __init__(self): + super(MyModel, self).__init__() + + def forward(self, x): + y = torch.reshape(x, (x.numel(),)) + y.retain_grad() + return y + +model = MyModel() + +x = torch.tensor([[1., -1.], [1., 1.]], requires_grad=True) +y = model(x) +dy = torch.tensor([1., 2., 3., 4.]) + +torch.autograd.backward([y], + grad_tensors=[dy], + retain_graph=True, + create_graph=True, + grad_variables=None) + +# This example shows the input and the output in Pytorch are differentiable. +# From the exported ONNX model below, we also see that "x" is the first input +# of ONNX Reshape and "y" the output of ONNX Reshape. Therefore, we can say +# the first input and the output of ONNX Reshape are differentiable. +print(x.grad) +print(y.grad) + +with open('model.onnx', 'wb') as f: + torch.onnx.export(model, x, f) +``` + +### Method 2: Manually Do the Math + +The second way is formally proving the existence of the Jacobian matrix (or tensor) from outputs to inputs with at least two numerical examples. In this case, the reviewer should go through the math and confirm if the numerical result is correct. The author should add enough details so that any STEM graduated student can easily review it. + +For example, to show the differentiability of Add, the author may first write down its equation: + +``` +C = A + B +``` + +For the sake of simplicity, assume `A` and `B` are same-shape vector. + +``` +A = [a1, a2]^T +B = [b1, b2]^T +C = [c1, c2]^T +``` + +Here we use the symbol `^T` to denote transpose of the attached matrix or vector. +Let `X = [a1, a2, b1, b2]^T` and `Y = [c1, c2]^T` and consider Add as a function which maps `X` to `Y`. +Then, this function's Jacobian matrix is a 4-by-2 matrix, + +``` +J = [[dc1/da1, dc2/da1], + [dc1/da2, dc2/da2], + [dc1/db1, dc2/db1], + [dc1/db2, dc2/db2]] + = [[1, 0], + [0, 1], + [1, 0], + [0, 1]] +``` + +If + +``` +dL/dC = [dL/dc1, dL/dc2]^T, +``` + +then `dL/dA = [dL/da1, dL/da2]^T` and `dL/dB = [dL/db1, dL/db2]^T` can be computed from elements in + +``` + [[dL/da1], [dL/da2], [dL/db1], [dL/db2]] += J * dL/dC += [[dL/dc1], [dL/dc2], [dL/dc1], [dL/dc2]] +``` + +where `*` is standard matrix multiplication. +If `dL/dC = [0.2, 0.8]^T`, then `dL/dA = [0.2, 0.8]^T` and `dL/dB = [0.2, 0.8]^T`. +Notice that the procedure to compute `dL/dA` and `dL/dB` from `dL/dC` is usually called backward of an operator. +We can see backward operator of Add takes `dL/dC` as an input and produces two outputs `dL/dA` and `dL/dB`. +Consequently, all of `A`, `B`, and `C` are differentiable. +By flattening tensor into 1-D vector, this example can be extended to cover all tensors when shape broadcasting is not needed. +If broadcasting happens, the broadcasted element's gradient is the sum of all associated elements' gradient in its **non-broadcasting** case. +Let's consider the above example again. +If `B = [b]^T` becomes an 1-element vector, `B` may be broadcasted to `[b1, b2]^T` and `dL/dB = [dL/ db]^T = [dL/db1 + dL/db2]^T`. +For high-dimensional tensors, this is in fact a ReduceSum operation along all expanded axes. diff --git a/docs/DimensionDenotation.md b/docs/DimensionDenotation.md new file mode 100644 index 0000000..26edc2b --- /dev/null +++ b/docs/DimensionDenotation.md @@ -0,0 +1,56 @@ + + +# Dimension Denotation + +Dimension Denotation is an experimental attempt to give tensor axis semantic descriptions and thus types and perform verification steps based on them subsequently. + +## Motivation + +The motivation of such a mechanism can be illustrated via a simple example. In the linear neural network specification below, we assume a NCHW model input: + +``` +input_in_NCHW -> Transpose(input, perm=[0, 2, 1, 3]) -> AveragePool(input, ...) +``` + +In this neural network, a user mistakenly constructed a neural network that transposes an NCHW input to a weird NHCW format and pass through spatial pooling that assumes a NCHW input format. As clearly a mistake as it is, no existing infrastructure will report an error to the user. This is should be deeply unnerving to programmers who rely heavily on type checking as an integral part of program correctness guarantee. This proposal seeks to resolve this vacuum of proper type-checking inherent in the current paradigm of neural network specification. + +This proposal consists of three key components: Denotation Definition, Denotation Propagation and Denotation Verification, each of which will be discussed in detail. + +## Denotation Definition + +To begin with, we define a set of types for tensor types. Such types are defined based on the following principles: + +1. Be fine grain enough to eliminate potential pitfalls. For instance, the above example illustrated in the motivation section mandates that we distinguish between a channel dimension and a spatial feature dimension to ensure the correctness of execution of the AveragePool op. +2. Be coarse grain enough to alleviate the mental burden of users. For instance, in the above example, there is significantly less need to distinguish between a width dimension and a height dimension because operations like pooling and convolution often do not draw a distinction between various spatial dimensions. Thus, we summarize all the spatial dimensions as feature dimensions. +3. As an important corollary of 2, be model agnostic. For instance, the semantics of feature dimensions in recurrent neural networks (RNN) and the semantics of spatial dimensions in convolutional neural network (CNN) are almost indistinguishable and therefore we permit users and developers to describe either as a feature dimension. + +Specifically, in our first proposal, we define the following set of standard denotations: + +1. `DATA_BATCH` describes a batch dimension of the training data. This corresponds to the `N` dimension in the more commonly used tensor format notation `NCHW`. +2. `DATA_CHANNEL` describes a channel dimension of the training data. This corresponds to the `C` dimension. +3. `DATA_TIME` describes a time dimension. +4. `DATA_FEATURE` describes a feature dimension. This corresponds to the `H`, `W` dimension or the feature dimension in RNN. +5. `FILTER_IN_CHANNEL` describes a filter in-channel dimension. This is the dimension that is identical (in size) to the channel dimension of the input image feature maps. +6. `FILTER_OUT_CHANNEL` describes a filter out-channel dimension. This is the dimension that is identical (in size) to the channel dimension of the output image feature maps. +7. `FILTER_SPATIAL` describes a filter spatial dimension. + +## Denotation Propagation + +Denotation Propagation happens when an operation permutes, destroys or creates dimensions with respect to its input tensor. In such scenarios, we will implement customized, operation-specific functions to infer the output tensor dimension denotation based on the input tensor dimension denotation. An example operation where denotation propagation happens is Transpose operation where the pseudocode for output dimension denotation inference can be formulated as a function of the input dimension denotation: + +``` +for i, j in enumerate(perm): + out_dim_denotaion[i] = in_dim_denotation[j] +``` + +## Denotation Verification + +Denotation Verification happens when an operation expects its input to arrive in a particular format. An example operation where denotation verification happens is AveragePool operation where the input, if annotated with dimension denotation, in the 2D case should have the denotation [`DATA_BATCH`, `DATA_CHANNEL`, `DATA_FEATURE`, `DATA_FEATURE`]. If there is a mismatch between the expected dimension denotation and the actual dimension denotation, an error should be reported. + +## Type Denotation + +See the [type denotation documentation](TypeDenotation.md) for more details on how to describe images and other types. diff --git a/docs/ExternalData.md b/docs/ExternalData.md new file mode 100644 index 0000000..156ac82 --- /dev/null +++ b/docs/ExternalData.md @@ -0,0 +1,101 @@ + + +# External Data + +## Loading an ONNX Model with External Data + +* [Default] If the external data is under the same directory of the model, simply use `onnx.load()` + +```python +import onnx + +onnx_model = onnx.load("path/to/the/model.onnx") +``` + +* If the external data is under another directory, use `load_external_data_for_model()` to specify the directory path and load after using `onnx.load()` + +```python +import onnx +from onnx.external_data_helper import load_external_data_for_model + +onnx_model = onnx.load("path/to/the/model.onnx", load_external_data=False) +load_external_data_for_model(onnx_model, "data/directory/path/") +# Then the onnx_model has loaded the external data from the specific directory +``` + +## Converting an ONNX Model to External Data + +```python +import onnx +from onnx.external_data_helper import convert_model_to_external_data + +onnx_model = ... # Your model in memory as ModelProto +convert_model_to_external_data(onnx_model, all_tensors_to_one_file=True, location="filename", size_threshold=1024, convert_attribute=False) +# Must be followed by save_model to save the converted model to a specific path +onnx.save_model(onnx_model, "path/to/save/the/model.onnx") +# Then the onnx_model has converted raw data as external data and saved to specific directory +``` + +## Converting and Saving an ONNX Model to External Data + +```python +import onnx + +onnx_model = ... # Your model in memory as ModelProto +onnx.save_model(onnx_model, "path/to/save/the/model.onnx", save_as_external_data=True, all_tensors_to_one_file=True, location="filename", size_threshold=1024, convert_attribute=False) +# Then the onnx_model has converted raw data as external data and saved to specific directory +``` + +## onnx.checker for Models with External Data + +### Models with External Data (<2GB) + +Current checker supports checking models with external data. Specify either loaded onnx model or model path to the checker. + +### Large models >2GB + +However, for those models larger than 2GB, please use the model path for onnx.checker and the external data needs to be under the same directory. + +```python +import onnx + +onnx.checker.check_model("path/to/the/model.onnx") +# onnx.checker.check_model(loaded_onnx_model) will fail if given >2GB model +``` + +## TensorProto: data_location and external_data fields + +There are two fields related to the external data in TensorProto message type. + +### data_location field + +`data_location` field stores the location of data for this tensor. Value MUST be one of: + +* `DEFAULT` - data stored inside the protobuf message. Data is stored in raw_data (if set) otherwise in type-specific field. +* `EXTERNAL` - data stored in an external location as described by external_data field. + +If not set, behaves as if the value was `DEFAULT`. + +### external_data field + +`external_data` field stores key-value pairs of strings describing data location + +Recognized keys are: + +* `"location"` (required) - file path relative to the filesystem directory where the ONNX protobuf model was stored. Up-directory path components such as .. are disallowed and should be stripped when parsing. +* `"offset"` (optional) - position of byte at which stored data begins. Integer stored as string. Offset values SHOULD be multiples of the page size (usually 4kb) to enable mmap support. On Windows, offset values SHOULD be multiples of the VirtualAlloc [allocation granularity](https://learn.microsoft.com/en-us/windows/win32/api/sysinfoapi/ns-sysinfoapi-system_info) (usually 64kb) to enable [memory mapping](https://learn.microsoft.com/en-us/windows/win32/api/memoryapi/nf-memoryapi-mapviewoffile). +* `"length"` (optional) - number of bytes containing data. Integer stored as string. +* `"checksum"` (optional) - SHA1 digest of file specified in under 'location' key. + +After an ONNX file is loaded, all `external_data` fields may be updated with an additional key `("basepath")`, which stores the path to the directory from which he ONNX model file was loaded. + +### External data files + +Data stored in external data files will be in the same binary bytes string format as is used by the `raw_data` field in current ONNX implementations. + +Reference +https://github.com/onnx/onnx/pull/678 diff --git a/docs/ExternalDataSecurity.md b/docs/ExternalDataSecurity.md new file mode 100644 index 0000000..1d11456 --- /dev/null +++ b/docs/ExternalDataSecurity.md @@ -0,0 +1,163 @@ + + +# External Data Security + +This document describes the security model for loading and saving external data files in ONNX models. It is intended for maintainers working on the external data code paths. + +## Threat Model + +When an ONNX model references external data files via relative paths, an attacker who controls the model file can attempt: + +- **Symlink traversal**: A final-component symlink in the external data path pointing to a sensitive file (e.g., `/etc/shadow`), causing ONNX to read or overwrite arbitrary files. +- **Parent-directory symlink**: A symlink in a parent directory component of the external data path, bypassing a check that only inspects the final component. +- **Hardlink attacks**: A hardlink to a sensitive file appearing as a normal file, bypassing symlink-only checks while still exposing unintended data. +- **Path traversal**: Using `..` segments or absolute paths to escape the model directory. + +## Defense Layers + +We use a 4-layer defense-in-depth approach. Each entry point applies the layers appropriate to its context (see the table below). + +### Layer 1: Canonical Path Containment + +- **C++**: `std::filesystem::weakly_canonical()` resolves the path, then verifies it starts with the canonical base directory. + +This catches `..` traversal and symlinks in any path component (not just the final one). + +### Layer 2: Symlink Detection + +- **C++**: `std::filesystem::is_symlink(data_path)` rejects the final-component symlink. + +This is a belt-and-suspenders check alongside containment. It provides a clear, specific error message when the final path component is a symlink. + +### Layer 3: Secure File Open with Post-Open Verification + +Three platform-specific strategies, in order of preference: + +- **Linux 5.6+**: `openat2(RESOLVE_BENEATH | RESOLVE_NO_SYMLINKS)` — atomic kernel containment. No TOCTOU. Falls back on `ENOSYS`. +- **FreeBSD 13+ / macOS 15+**: `openat(O_RESOLVE_BENEATH | O_NOFOLLOW)` — equivalent BSD-style containment. +- **POSIX fallback**: `open(O_NOFOLLOW)` + post-open inode comparison (`fstat` fd vs `stat` canonical path). +- **Windows**: `CreateFileW(FILE_FLAG_OPEN_REPARSE_POINT)` + reparse point attribute check + inode comparison. Converted to CRT fd via `_open_osfhandle()` before returning. + +All POSIX fds are opened with `O_CLOEXEC`. All platforms return a CRT file descriptor. + +Hardlinks are detected via `fstat()` link count (POSIX) or `GetFileInformationByHandle()` (Windows), always fail closed. + +### Layer 4: Pre-Open Hardlink Count Check + +- **C++**: `std::filesystem::hard_link_count(data_path) > 1` rejects files with multiple hardlinks before opening. + +This is defense-in-depth alongside Layer 3's post-open hardlink check. It provides a clear error message at the path level before any file I/O occurs. + +## Protected Entry Points + +All Python entry points use the C++ `open_external_data()` function. In **read mode** it applies all four layers (pre-open validation via `resolve_external_data_location` for Layers 1, 2, 4, then Layer 3 post-open). In **write mode** it applies Layers 1 and 3 only — Layers 2 and 4 are skipped because the file may not yet exist. The C++ checker calls `resolve_external_data_location` directly without opening files, so Layer 3 does not apply there. + +| Entry Point | File | Mode | Layers | +|---|---|---|---| +| `resolve_external_data_location` | `onnx/checker.cc` | read | 1, 2, 4 | +| `open_external_data` | `onnx/checker.cc` | read | 1, 2, 3, 4 | +| `open_external_data` | `onnx/checker.cc` | write | 1, 3 | +| `load_external_data_for_tensor` | `onnx/external_data_helper.py` | read | 1, 2, 3, 4 (via `open_external_data`) | +| `save_external_data` | `onnx/external_data_helper.py` | write | 1, 3 (via `open_external_data`) | +| `ModelContainer._load_large_initializers` | `onnx/model_container.py` | read | 1, 2, 3, 4 (via `open_external_data`) | + +## Known Limitations + +### Windows + +- Reparse points (symlinks, junctions) are rejected via `FILE_FLAG_OPEN_REPARSE_POINT` + attribute check. +- Symlink and hardlink tests are skipped on Windows in the test suite. + +### Case-Insensitive Filesystems + +The canonical path containment checks (Layers 1 and 3) use string comparison. On case-insensitive filesystems (Windows NTFS, macOS HFS+), paths with different casing may incorrectly fail containment. This fails closed (false rejection, not a bypass). + +## Testing + +Test coverage is in: + +- **C++**: `onnx/test/cpp/checker_test.cc` — `*SymLink*` tests for symlink detection and containment, `OpenExternalData*` for secure open and verification. +- **Python**: `onnx/test/test_external_data.py`: + - `TestSaveExternalDataSymlinkProtection` — save-side symlink rejection. + - `TestLoadExternalDataSymlinkProtection` — load-side symlink rejection, parent-directory symlink, `load_external_data_for_model` rejection. + - `TestLoadExternalDataHardlinkProtection` — load-side hardlink rejection. + - `TestSaveExternalDataAbsolutePathValidation` — absolute path rejection. + +Symlink and hardlink tests are skipped on Windows (`os.name == "nt"`). + +--- + +## External Data Attribute Validation + +This section describes the security model for validating external data attributes in `ExternalDataInfo`. It covers defenses against attribute injection (CWE-915) and resource exhaustion (CWE-400) via crafted `external_data` entries in `TensorProto`. + +**Advisory:** [GHSA-538c-55jv-c5g9](https://github.com/onnx/onnx/security/advisories/GHSA-538c-55jv-c5g9) + +### Threat Model + +An attacker provides a malicious ONNX model with crafted `external_data` entries in `TensorProto`. The `external_data` field is a repeated `StringStringEntryProto` — a key-value store that accepts arbitrary strings for both key and value. + +The attack is triggered during `onnx.load()` with no explicit checker invocation required. `ExternalDataInfo.__init__` processes these key-value pairs to populate object attributes. + +Attack vectors: + +- **Arbitrary attribute injection**: Setting unknown keys (e.g. `evil_attr`) causes `setattr()` to create arbitrary attributes on the `ExternalDataInfo` object. While no current consumer iterates over attributes, injected attributes create latent risk for future code. +- **Dunder attribute injection**: Setting keys like `__class__` or `__dict__` corrupts the Python object's internal state, enabling type confusion attacks. +- **Negative offset/length**: Negative values for `offset` cause `file.seek()` to raise `OSError`. Negative `length` causes `file.read(-1)` to read the entire file to EOF, bypassing intended size limits. +- **Resource exhaustion (DoS)**: Setting `length` to a multi-petabyte value causes unbounded memory allocation when reading external data, even if the actual data file is small. + +Four Python consumers of `ExternalDataInfo` exist: `load_external_data_for_tensor`, `set_external_data` / `write_external_data_tensors`, `ModelContainer._load_large_initializers`, and `ReferenceEvaluator` (in `onnx/reference/reference_evaluator.py`). (The C++ checker validates paths but does not use the Python `ExternalDataInfo` class.) + +## Defense Layers + +We use a 3-layer defense-in-depth approach. Each layer addresses a different class of attack and operates at a different point in the processing pipeline. + +### Layer 1: Attribute Whitelist (CWE-915 Mitigation) + +`ExternalDataInfo.__init__` only accepts keys in `_ALLOWED_EXTERNAL_DATA_KEYS`: `location`, `offset`, `length`, `checksum`, `basepath`. Unknown keys are warned via `warnings.warn()` and ignored — this prevents arbitrary attribute injection. + +This also blocks dunder attribute injection (e.g. `__class__`, `__dict__`) that could cause object type confusion. + +**Rationale**: While we cannot prevent someone from constructing malicious protobuf directly, rejecting unknown keys at the Python object level is defense-in-depth that limits the attack surface. The whitelist is a `frozenset` to prevent runtime mutation. + +### Layer 2: Bounds Validation at Parse Time (CWE-400 Mitigation) + +`offset` and `length` must be non-negative integers. Non-numeric strings raise `ValueError`. This catches obviously invalid values early, before any file I/O occurs. + +**Rationale**: Negative `offset` causes `file.seek(-1)` to raise `OSError`; negative `length` causes `file.read(-1)` to read the entire file, bypassing intended size limits. Validating at parse time provides a clear error message at the point closest to the malicious input. + +### Layer 3: File-Size Validation at Consumption Time (CWE-400 Mitigation, Defense-in-Depth) + +In `load_external_data_for_tensor()` and `ModelContainer._load_large_initializers`, before reading: `offset <= file_size` and `offset + length <= file_size` are verified. A 1KB data file cannot cause a multi-petabyte memory allocation. + +**Rationale**: This is the critical safety net. It prevents memory exhaustion regardless of how the model was constructed — even via direct protobuf APIs that bypass Python-level parsing entirely. Validation happens at the point of actual file I/O, the last opportunity before harm occurs. + +## Why Layered Defense + +- **Layer 1 (whitelist)** catches the broadest class of attacks at parse time. It blocks attribute injection, dunder corruption, and any future unknown-key attack vector. +- **Layer 2 (bounds validation)** catches obviously invalid numeric values at parse time, providing clear error messages. +- **Layer 3 (file-size validation)** is the critical safety net that prevents actual harm at the I/O boundary. This layer cannot be bypassed even if an attacker crafts a model using protobuf APIs directly, because validation happens at the point of actual file read. + +## Protected Entry Points + +| Entry Point | File | Layers | +|---|---|---| +| `ExternalDataInfo.__init__` | `onnx/external_data_helper.py` | 1, 2 | +| `load_external_data_for_tensor` | `onnx/external_data_helper.py` | 1, 2, 3 | +| `set_external_data` | `onnx/external_data_helper.py` | 1 (whitelist by overwrite) | +| `ModelContainer._load_large_initializers` | `onnx/model_container.py` | 1, 2, 3 | + +## Testing + +Test coverage is in `onnx/test/test_external_data.py`: + +- `TestExternalDataInfoSecurity`: + - **CWE-915 (attribute injection):** `test_unknown_key_rejected`, `test_dunder_key_rejected`, `test_multiple_unknown_keys_all_rejected`, `test_allowed_keys_constant_is_frozen` + - **CWE-400 (bounds/DoS):** `test_negative_offset_rejected`, `test_negative_length_rejected`, `test_non_numeric_offset_raises`, `test_non_numeric_length_raises` + - **Regression guards:** `test_valid_external_data_accepted`, `test_zero_offset_and_length_accepted` +- `TestLoadExternalDataFileSizeValidation`: + - **File-size validation:** `test_offset_exceeds_file_size_raises`, `test_length_exceeds_available_data_raises`, `test_valid_offset_and_length_load_correctly` diff --git a/docs/IR.md b/docs/IR.md new file mode 100644 index 0000000..42cdd5a --- /dev/null +++ b/docs/IR.md @@ -0,0 +1,628 @@ + + +# Open Neural Network Exchange Intermediate Representation (ONNX IR) Specification + +__Purpose__ + +This document contains the normative specification of the semantics of ONNX. + +The `.proto` and `.proto3` files found under the [onnx folder](/onnx/) form the normative specification of its syntax authored in the [Protocol Buffers](https://developers.google.com/protocol-buffers) definition language. Commentary found in the `.proto` and `.proto3` files are intended to improve readability of those files, but are not normative if they conflict with this document. Such conflicts should be reported as documentation bugs. + +__Notes on model validation__ + +A [tool](../onnx/checker.py) is available to perform general validation of models against this specification. It is implemented in C++ with a Python command-line wrapper. + +__Notes on language in this and all related documents__: + +1. The use of SHOULD, MUST, MAY and so on in this document is consistent with [RFC 2119](https://www.ietf.org/rfc/rfc2119.txt). + +2. The use of 'list' shall denote an ordered collection of items, 'set' shall denote an unordered collection of unique elements, and 'bag' an unordered collection of possibly non-unique elements. + +## Components + +ONNX is an open specification that consists of the following components: + +1) A definition of an extensible computation graph model. + +2) Definitions of standard data types. + +3) Definitions of built-in operators. + +#1 and #2 together make up the ONNX Intermediate Representation, or 'IR', specification which is covered herein; the built-in operators are covered in documents listed at the end. Specifically, built-in operators are divided into a set of primitive operators and functions. A function is an operator whose semantics is formally expressed via expansion into a sub-graph (called the function body) using other operators (and functions). Functionality-wise, an ONNX compatible framework or runtime may inline a function body to execute it if it does not have corresponding implementation of the function. + +There are two official ONNX variants; the main distinction between the two is found in the default operator sets. __ONNX-ML__ extends the __ONNX__ operator set with ML algorithms that are not based on neural networks. + +Up to IR version 6, the ONNX specification and model format addressed only inference (also known as scoring). Starting from IR version 7, the ONNX specification and model format also support training. An ONNX training model is an extension of the inference-model. An inference-only runtime can consume a training model ignoring the training-related extensions. However, an inference-only model may enable a representation that is more optimal for inference purposes than a training model. + +## Runtime Agnostic + +ONNX does not pre-suppose or imply any particular method of runtime implementation. + +For example, an implementation may consist of a rich runtime which interprets the model; it may be a code generator that translates the model in its entirety to executable code for some target programming language; it may be a hardware implementation; it may be a combination of two or three of those. + +Nothing in this specification should be construed as advocating one implementation approach over any other; any comments on the inner workings of concrete implementations are to be interpreted as examples. + +## ONNX Versioning + +The IR specification, individual models, and operator sets are all versioned. Furthermore, each individual operator indicates which version of its containing operator set it was introduced or stabilized in. + +Version numbers can be used as a simple number, or used to encode [semantic versions](https://semver.org/)(AKA SemVer). If using semantic versions, the convention is to use the two most significant bytes for the major number, the next two bytes for the minor number, and the least significant four bytes for the patch/build/bugfix number. When using semantic versioning, at least one of the major/minor numbers MUST be non-zero. + +The IR specification uses simple monotonically increasing numbers for its versions. The valid IR versions are defined by the `onnx.Version` enumeration in [onnx.proto](/onnx/onnx.proto). + +Operator sets use a simple version number. Each operator set version represents a snapshot of the set of operators, and their semantics at a particular point in time. + +This specification does not provide guidance on what versioning scheme model producers should be using. + +More details on conventions and best practices for versioning of IR, operator sets, and models can be found in [Versioning](Versioning.md). + +## Extensible computation graph model + +ONNX specifies the portable, serialized format of a computation graph. It does not have to be the form a framework chooses to use internally. For example, an implementation may represent the model differently in memory if it is more efficient to manipulate during optimization passes. + +An implementation MAY extend ONNX by adding operators expressing semantics beyond the standard set of operators that all implementations MUST support. The mechanism for this is adding operator sets to the `opset_import` property in a model that depends on the extension operators. + +### Models + +The top-level ONNX construct is a ‘Model.’, and is represented in protocol buffers as the type `onnx.ModelProto` + +The main purpose of the model structure is to associate metadata with a graph which contains all the executable elements. The metadata is used when first reading the model file, giving an implementation the information it needs in order to determine whether it will be able to execute the model, generate logging messages, error reports, etc. Further, the metadata is useful to tools, such as IDEs and model galleries, which need it for informing humans about a given model’s purpose and characteristics. + +Each model has the following components: + +|Name|Type|Description| +|---|---|---| +|ir_version|int64|The ONNX version assumed by the model.| +|opset_import|OperatorSetId|A collection of operator set identifiers made available to the model. An implementation must support all operators in the set or reject the model.| +|producer_name|string|The name of the tool used to generate the model.| +|producer_version|string|The version of the generating tool.| +|domain|string|A reverse-DNS name to indicate the model namespace or domain, for example, 'org.onnx'| +|model_version|int64|The version of the model itself, encoded in an integer.| +|doc_string|string|Human-readable documentation for this model. Markdown is allowed.| +|graph|Graph|The parameterized graph that is evaluated to execute the model.| +|metadata_props|map|Named metadata values; keys should be distinct.| +|training_info|TrainingInfoProto[]|An optional extension that contains information for training.| +|functions|FunctionProto[]|An optional list of functions local to the model.| +|configuration|DeviceConfigurationProto[]|(IR version >= 11) An optional list of multi-device configurations for distributed execution.| + + Models SHOULD specify a domain and use reverse domain names based on the responsible organization's identity, the same convention that is used for [naming Java packages](https://docs.oracle.com/javase/tutorial/java/package/namingpkgs.html). + +__Note: Exploring an ONNX file__ + +You can use the `protoc` tool that is part of the Protocol Buffers distribution to examine the contents of an ONNX file, you do so like this: + +```bash +protoc --decode=onnx.ModelProto onnx.proto < yourfile.onnx +``` + +Where [onnx.proto](/onnx/onnx.proto) is the file that is part of this repository. + +Alternatively, you can use a tool like [Netron](https://github.com/lutzroeder/netron) to explore the ONNX file. + +### Model Semantics + +The semantics of an inference-model is a _stateless function_ (except possibly for the state used for random-number generation). Thus, whenever an inference-model (without random-generator operations) is used to perform inference on the same input, it is expected to produce the same output. + +The semantics of a training model is that of a _stateful object_, with the state consisting of the current values of trained-weights (and any other auxiliary state required, such as momentum, for example, used by the learning algorithm). Specifically, its semantics is captured via three methods: an initialization method (which is used to initialize or reset the values of state variables), a training step method (to train using a batch of input-output pairs), and an inference method to perform inference using the current values of the learned weights. The first two methods update the state of the object, while the third method is side-effect-free. + +### Optional Metadata + +The 'metadata_props' field in the model is available for any kind of optional metadata that a tool or model developer chooses to place there. The following are the defined “standard” optional metadata properties of a model. + +Name|Type|Format|Description +|---|---|---|---| +model_author|string|A comma-separated list of names.|The personal name of the author(s) of the model, and/or their organizations. +model_license|string|Name or URL.|The well-known name or URL of the license under which the model is made available. + +### Operator Set Identifiers + +Each operator set is uniquely identified by a (domain, version) pair. + +Name|Type|Description +|---|---|---| +domain|string|The domain of the operator set being identified. +version|int64|The version of the operator set being identified. Same as 'opset_version' in the operator set. + +The operator set version is a simple integer value that is monotonically increased as new versions of the operator set are published. + +Operator sets other than the default operator set MUST specify a domain and SHOULD use reverse domain names based on the responsible organization's identity, the same convention that is used for [naming Java packages](https://docs.oracle.com/javase/tutorial/java/package/namingpkgs.html). + +### Operator Sets + +Each model MUST explicitly name the operator sets that it relies on for its functionality. Operator sets define the available operators and their version. Each model defines the imported operator sets by their domains. All models implicitly import the default ONNX operator set. + +Each operator set SHALL be defined in a separate document, also using protobuf as the serialization format. How operator set documents are found at runtime is implementation-dependent. + +__Note: As of the publication of this document, no ONNX implementation is known to process operator set documents.__ + +The properties of an operator set are: + +Name|Type|Description +|---|---|---| +magic|string|The value ‘ONNXOPSET’ +ir_version|int32|The ONNX version corresponding to the operators. +ir_version_prerelease|string|The prerelease component of the SemVer of the IR. +ir_build_metadata|string|The build metadata of this version of the operator set. +domain|string|The domain of the operator set. Must be unique among all sets. +opset_version|int64|The version of the operator set. +doc_string|string|Human-readable documentation for this operator set. Markdown is allowed. +operator|Operator[]|The operators contained in this operator set. + +The operator set version is a simple integer value that is monotonically increased as new versions of the operator set are published. + +Operator sets other than the default operator set MUST specify a domain and SHOULD use reverse domain names based on the responsible organization's identity, the same convention that is used for [naming Java packages](https://docs.oracle.com/javase/tutorial/java/package/namingpkgs.html). + +### Operators + +Each operator used within a graph MUST be explicitly declared by one of the operator sets imported by the model. + +The properties of an operator definition are: + +Name|Type|Description +|---|---|---| +op_type|string|The name of the operator (case sensitive), as used in graph nodes. MUST be unique within the operator set’s domain. +since_version|int64|The version of the operator set when this operator was introduced. +status|OperatorStatus|One of ‘EXPERIMENTAL’ or ‘STABLE.’ +doc_string|string|A human-readable documentation string for this operator. Markdown is allowed. + +The version value MUST be the same value as the operator set version when the operator was first published. Subsequent versions of the operator set MUST NOT alter the signature or semantics of the operator once published as STABLE. + +The ‘status’ property indicates whether the syntax, semantics, or presence of the operator is in an experimental or stable stage. Once an operator is published as STABLE, it’s syntax and semantics MUST NOT change in subsequent versions of the operator set. + +There are two distinct ways to pass information to operators – inputs and attributes. Inputs represent graph inputs or values computed elsewhere in the graph, while attributes are used for values that are constants in the graph. This distinction may be highly relevant to achieving good performance for some implementations, while completely irrelevant to others. + +### Functions + +A _function_ may be thought of as an operator combined with an implementation of the operator using +other, more primitive, ops, referred to as the _function body_. The function body consists of a +topologically sorted list of nodes that form a graph. Thus, a function combines aspects of both +an operator as well a graph (described below). + +Each function contained in a Model (also referred to as a model-local function) serves +as a default or fallback implementation of the corresponding operator. A runtime, however, +may choose to use an alternative implementation of the operator (usually as an optimized kernel). +As such, the unique name of a function is significant as it is implicitly associated with a +semantic specification. + +A serialized function (a _FunctionProto_) has the following properties: + +|Name|Type|Description| +|---|---|---| +name|string|The name of the function +domain|string|The domain to which this function belongs +overload|string|Part of unique id of function (added in IR version 10) +doc_string|string|Human-readable documentation for this function. Markdown is allowed. +attribute|string[]|The attribute parameters of the function +attribute_proto|Attribute[]| (IR version 9+) The attribute parameters with default values of the function. A function attribute shall be represented either as a string attribute or an Attribute, not both. +input|string[]|The input parameters of the function +output|string[]|The output parameters of the function. +node|Node[]|A list of nodes, forming a partially ordered computation graph. It must be in topological order. +|opset_import|OperatorSetId|A collection of operator set identifiers used by the function implementation. +|value_info|ValueInfo[]| (IR version >= 10) Used to store the type and shape information of values used in the function. +|metadata_props|map|(IR version >= 10) Named metadata values; keys should be distinct. + +The name and domain serve to identify the operator uniquely in IR versions upto 9. IR version 10 adds the +field overload, and the triple (name, domain, overload) acts as a unique-id across functions stored in +a model. This is intended to support cases where distinct function-bodies are required for distinct +calls to the function within the model. +An opset version is not explicitly identified in a FunctionProto, but it is implicitly determined by +the opset version of the domain included in the model. + +The input, output, attribute, and attribute_proto (added in IR version 9) constitute the signature part of the operator. No type information +is explicitly included in the signature. The attribute_proto field describes attribute parameters of the function along with their default-value (when not specified by an call-site node), while the attribute field lists attribute parameters without a default-value. The names in these two lists must be distinct. When an attribute-parameter of the function is used in a node within the function, it is replaced by the actual parameter value specified for the attribute at a call-site node (of the function) when such a attribute is specified, and it is replaced by the default-value if the attribute has a default-value specified, and it is omitted otherwise. + +The opset_import and node fields describe the implementation of the function. + +The value_info field (added in IR version 10) allows a model to store type and shape information about the values used in a function, including its inputs and outputs. Note that this is optional, and ONNX allows functions to be polymorphic. + +### Graphs + +A graph is used to describe a side-effect-free computation (function). +A serialized graph is comprised of a set of metadata fields, a list of model parameters, and a list of computation nodes. + +Each computation dataflow graph is structured as a topologically sorted list of nodes that form a graph, which MUST be free of cycles. Each node represents a call to an operator or a model local function. Each node has zero or more inputs and one or more outputs. + +Graphs have the following properties: + +|Name|Type|Description| +|---|---|---| +name|string|The name of the model graph. +node|Node[]|A list of nodes, forming a partially ordered computation graph based on input/output data dependencies. It is in topological order. +initializer|Tensor[]|A list of named tensor values. When an initializer has the same name as a graph input, it specifies a default value for that input. When an initializer has a name different from all graph inputs, it specifies a constant value. The order of the list is unspecified. +doc_string|string|Human-readable documentation for this model. Markdown is allowed. +input|ValueInfo[]|The input parameters of the graph, possibly initialized by a default value found in ‘initializer.’ +output|ValueInfo[]|The output parameters of the graph. Once all output parameters have been written to by a graph execution, the execution is complete. +value_info|ValueInfo[]|Used to store the type and shape information of values that are not inputs or outputs. +|metadata_props|map|(IR version >= 10) Named metadata values; keys should be distinct. + +ValueInfo has the following properties: + +Name|Type|Description +|---|---|---| +name|string|The name of the value/parameter. +type|Type|The type of the value **including shape information**. +doc_string|string|Human-readable documentation for this value. Markdown is allowed. + +Each main (top-level) graph MUST define the names, types and shapes of its inputs and outputs, which are specified as ‘value info’ structures. The main graph inputs and outputs are required to have a shape, indicating the rank, even though the exact dimensions need not be specified. + +Nested subgraphs (specified as attribute values) MUST define the names of its inputs and outputs +and MAY define the types of its inputs and outputs. + +Each graph MUST specify a name. + +The graph MUST adhere to single static assignment (SSA) for all node outputs; this means that all node output names MUST be unique within a graph. + +Graphs SHOULD be populated with documentation strings, which MAY be interpreted using GitHub-style markdown syntax. HTML and other text-markup languages MAY NOT be used in documentation strings. + +### Names Within a Graph + +All names SHOULD adhere to [C90 identifier syntax rules](https://en.cppreference.com/c/language/identifier). + +Names of nodes, inputs, outputs, initializers, and attributes are organized into several namespaces. Within a namespace, each name MUST be unique for each given graph. Please see below for further clarification in the case where a graph contains nested subgraphs (as attribute values). + +The namespaces are: + +Namespace|Description +|---|---| +Attribute|The names of attributes of an operator. Unique for each operator. +Value|The names of values – node inputs & outputs, tensor values (if named), graph inputs, outputs. +Node|The names of graph nodes. +Graph|The names of graphs within a domain, unique within the model domain. +Operator|The names of operators within a domain. +Shape|The names of tensor shape variables – scoped to the value information records of a graph, which is where shape variables occur. + +### Nodes + +Computation nodes are comprised of a name, the name of an operator that it invokes, a list of named inputs, a list of named outputs, and a list of attributes. + +Input and outputs are positionally associated with operator inputs and outputs. Attributes are associated with operator attributes by name. + +They have the following properties: + +Name|Type|Description +|---|---|---| +name|string|An optional name of the node, used for diagnostic purposes only. +input|string[]|Names of the values used by the node to propagate input values to the node operator. It must refer to either a graph input, a graph initializer or a node output. +output|string[]|Names of the outputs used by the node to capture data from the operator invoked by the node. It either introduces a value in the graph or refers to a graph output. +op_type|string|The symbolic identifier of the operator to invoke. +domain|string|The domain of the operator set that contains the operator named by the op_type. +attribute|Attribute[]|Named attributes, another form of operator parameterization, used for constant values rather than propagated values. +doc_string|string|Human-readable documentation for this value. Markdown is allowed. +overload|string|Part of unique id of function (added in IR version 10) +|metadata_props|map|(IR version >= 10) Named metadata values; keys should be distinct. +|device_configurations|NodeDeviceConfigurationProto[]|(IR version >= 11) Multi-device execution configurations for this node. + +A name belonging to the Value namespace may appear in multiple places, namely as a graph input, a graph initializer, a graph output, a node input, or a node output. The occurrence of a name as a graph input, a graph initializer, or as a node output is said to be a definition and the occurrence of a name as a node input or as a graph output is said to be a use. + +A value name used in a graph must have a unique definition, with the exception that the same name MAY appear in both the graph input list and graph initializer list. (Further exceptions apply in the presence of nested subgraphs, as described later.) + +When a name appears in both the initializer list and the graph input list, a runtime MAY allow a caller to specify a value for this (input) name overriding the value specified in the initializer and a runtime MAY allow users to omit specifying a value for this (input) name, choosing the value specified in the initializer. Names of constants that are not meant to be overridden by the caller should appear only in the initializer list and not in the graph input list. In models with IR version >= 4, in nested subgraphs used as attribute values, users MUST NOT use the same name as both a subgraph initializer and subgraph input unless the corresponding op's specification explicitly allows it. In models with IR version <= 3, users MAY use the same name as both a subgraph initializer and subgraph input, but this is restricted to support constants via initializers that are not intended to correspond to any actual inputs passed from the node into the subgraph. In particular, the control-flow operator semantics determines the set of inputs supplied to the execution of the subgraph, and these input names MUST NOT appear as subgraph initializers. Subgraph initializer names must appear in the graph input list _after_ the actual inputs. This allows the actual inputs and formal inputs to be matched positionally. + +Edges in the computation graph are established by outputs of one node being referenced by name in the inputs of a subsequent node. + +The outputs of a given node introduce new names into the graph. The values of node outputs are computed by the node's operator. Node inputs MAY refer to node outputs, graph inputs, and graph initializers. When the name of a node output coincides with the name of a graph output, the graph output's value is the corresponding output value computed by that node. A node input in a nested subgraph MAY refer to names introduced in outer graphs (as node outputs, graph inputs, or graph initializers). + +The graph MUST use single static assignment for all node outputs, which means that all node output names MUST be unique within a graph. In the case of a nested subgraph, a node output name and names of inputs and initializers of the subgraph MUST be distinct from the names from the outer scopes that are visible in the nested subgraph. That is, variable shadowing is not allowed. + +Node dependencies MUST NOT create cycles in the computation graph. + +The number of inputs and outputs in a node, their types, the set of attributes specified in a node and their types MUST satisfy the constraints imposed by the signature of the node’s operator. + +The list of nodes defining the top-level computation graph MUST be ordered topologically; that is, if node K follows node N in the graph, none of the data inputs of N may refer to outputs of K. + +Node attributes are used to pass literal (static) values to operators. + +#### Input and Output Values + +The representation distinguishes between two kinds of values: attribute values, which are statically known, and input/output values. The types of values permitted in the two cases are different. + +Input and output values are found as graph inputs, outputs, and initializers, and as node inputs and outputs. Their values are determined at runtime, either by the code that initiates model execution, or by operators computing output values. + +#### Attributes + +Attribute values are only found in nodes, passed to operators by name association. Attribute values are runtime constants, in that their values are determined when a model graph is constructed and therefore not computed at runtime. A common use for attributes is to represent coefficients established during model training. + +Attributes have the following properties: + +Name|Type|Description +|---|---|---| +name|string|The name of the attribute. Must be unique among attributes, inputs, and outputs for any given operator and node. +doc_string|string|Human-readable documentation for this value. Markdown is allowed. +type|AttributeType|The type of the attribute, determining which of the remaining fields is used to hold the value of the attribute. +f|float|A 32-bit floating-point value. +i|int64|A 64-bit integer value. +s|byte[]|UTF-8 string. +t|Tensor|A tensor value. +g|Graph|A graph. +floats|float[]|A list of 32-bit floating-point values. +ints|int64[]|A list of 64-bit integer values. +strings|byte[][]|A list of UTF-8 strings. +tensors|Tensor[]|A list of tensor values. +graphs|Graph[]|A list of graphs. +ref_attr_name|string|The name of a parent function's attribute. + +The properties ‘name’ and ‘type’ are required on all attributes, and ‘doc_string’ SHOULD be used on all attributes. An attribute MUST have only one of the value-carrying properties. + +In case ‘ref_attr_name’ is set, this attribute does not contain data, and instead it's a reference to the parent function's attribute of the given name. Can only be used within the function body. + +#### Variadic Inputs and Outputs + +The last input or output of an operator MAY be marked as variadic. For example, the operator 'Max()' can be used to compute the maximum of a varying number of input values. A variadic operator has a minimum arity, which specifies the minimum number of operands that must be specified. + +For each variadic operator input, N or more node inputs must be specified where N is the minimum arity of the operator. For each variadic operator output, N or more node outputs must be specified where N is the minimum arity of the operator. + +#### Optional Inputs and Outputs + +##### Static Optional + +Some operators have inputs that are marked as optional, which means that a referring node MAY forgo providing values for such inputs. + +Some operators have outputs that are optional. When an actual output parameter of an operator is not specified, the operator implementation MAY forgo computing values for such outputs. + +There are two ways to leave an optional input or output unspecified: the first, available only for trailing inputs and outputs, is to simply not provide that input or output; the second method is to use an empty string in place of an input or output name. + +Each node referring to an operator with optional outputs MUST provide a name for each output that is computed and MUST NOT provide names for outputs that are not computed. + +Optional inputs and outputs of the above kind are referred to as _static-optional_. + +##### Dynamic Optional (since IR-8) + +**IR-8 Version** introduced a new type-constructor to represent _dynamic-optional_ inputs and outputs, +in addition to the earlier static-optional version described above. A dynamic-optional INT64 +tensor is a distinct type from an INT64 tensor type. In contrast, a static-optional INT64 +tensor does not have a distinct type, it has the same type as a INT64 tensor. +The operators `Optional` and `OptionalGetElement` MUST be explicitly used to convert between +the dynamic-optional type and the underlying non-optional type. +The dynamic-optional allows for more expressiveness than static-optional. + +#### External Tensor Data + +The raw data for large constant tensors, such as initializers, MAY be serialized in a separate file. In such a case, the tensor MUST provide the filename relative to the model file and MUST NOT use the value fields. It MAY provide a byte offset and length within that file. It MAY also specify a SHA1 digest of the file. One file MAY contain the data for multiple tensors. + +More details can be found in [External Data](ExternalData.md). + +## Standard data types + +There are two official ONNX variants; the main distinction between the two is found in the supported types and the supported operators. + +With respect to supported types, both __ONNX__ and __ONNX-ML__ definition recognize tensors, sparse tensors, sequences, maps, and optionals as input and output types. Sequences and maps were supported from the IR version 6 (ONNX 1.6.0 release). Optional type was supported from IR version 8 (ONNX 1.10.0 release). + +The following data types are supported by ONNX for inputs and outputs of graphs and nodes as well as the initializers of a graph. + +Primitive numeric, string, and Boolean types MUST be used as elements of tensors. + +### Tensor Definition + +Tensors are a generalization of vectors and matrices; whereas vectors have one dimension, and matrices two, tensors can have any number of dimensions, including zero. A zero-dimensional tensor is logically equivalent to a scalar value. + +Mathematically, a tensor can be defined as a pair of sequences/lists (V, S) where S is the shape of the tensor (a list of non-negative integers) and V is a list of values with length equal to the product of the dimensions in S. Two tensors (V, S) and (V', S') are equal if and only if V = V' and S = S'. The length of S is referred to as the rank. + +- If S has length 0, V must have length 1, since the empty product is defined to be 1. In this case, the tensor represents a scalar. +- S can contain dimensions of value 0. If any dimensions are 0, V must have length 0. +- If S has length 1, V has length equal to the single dimension in S. In this case, the tensor represents a vector. +- A tensor representing a vector of length 1 has shape [1], while a tensor representing a scalar has shape []. They both have a single element, but scalars are _not_ vectors of length 1. + +A tensor's shape S is a list but can be represented as a tensor with values S and shape [R] where R is the rank of the tensor. + +- For a tensor (V, S), the tensor representing its shape is (S, [R]). +- The shape of a scalar is []. Represented as a tensor, [] has shape [0]. + +#### Representation + +It is common to represent a tensor as a nested list. This generally works fine, but is problematic when zero dimensions are involved. A tensor of shape (5, 0) can be represented as [[], [], [], [], []], but (0, 5) is represented as [] which loses the information that the second dimension is 5. + +- A nested list is not a complete representation of a tensor with dimensions of value zero. + +### Tensor Element Types + +|Group|Types|Description| +|---|---|---| +Floating Point Types|float16, float32, float64, bfloat16, float8e4m3fn, float8e5m2, float8e4m3fnuz, float8e5m2fnuz, float4e2m1|Values adhering to the IEEE 754-2008 standard representation of floating-point data or defined in papers [FP8 Formats for Deep Learning](https://arxiv.org/abs/2209.05433), [8-bit Numerical Formats for Deep Neural Networks](https://arxiv.org/abs/2206.02915), and the [Open Compute Project](https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf) +Signed Integer Types|int2, int4, int8, int16, int32, int64|Signed integers are supported for 2-64 bit widths. +Unsigned Integer Types|uint2, uint4, uint8, uint16, uint32, uint64|Unsigned integers are supported for 2-64 bit widths. +Complex Types|complex64, complex128|A complex number with either 32- or 64-bit real and imaginary parts. +Other|string|Strings represent textual data. All strings are encoded using UTF-8. +Other|bool|Boolean values represent data with only two values, typically true and false. + +### Input / Output Data Types + +The following types are used to define the types of graph and node inputs and outputs. + +|Variant | Type | Description | +|---|---|---| +ONNX|dense tensors|Represents a Tensor. See definition above. +ONNX|sequence|Sequences are dense, ordered, collections of elements that are of homogeneous types. +ONNX|map|Maps are associative tables, defined by a key type and a value type. +ONNX|optional|Optionals are wrappers that may contain an element of tensor, sequence, or map type, or may be empty (containing none). [Details](ONNXTypes.md) + +#### Static tensor shapes + +In addition to element type, tensor types have a **static** shape. The static shape of a tensor variable is related to, but different from, the runtime (dynamic) shape of a tensor value. A static tensor shape is a list of records that indicates whether the tensor is a vector, a matrix, or a higher-dimensional value. For example, a 100x100 matrix has the shape [100,100]. + +The static shape is defined by 'TensorShapeProto': + +```proto +message TensorShapeProto { + message Dimension { + oneof value { + int64 dim_value = 1; + string dim_param = 2; + }; + }; + repeated Dimension dim = 1; +} +``` + +Which is referenced by the Tensor type message: + +```proto + message Tensor { + optional TensorProto.DataType elem_type = 1; + optional TensorShapeProto shape = 2; + } +``` + +The empty list of dimension sizes, [], is a valid tensor shape, denoting a zero-dimension (scalar) value. A zero-dimension tensor is distinct from a tensor of unknown dimensionality, which is indicated by an absent 'shape' property in the Tensor message. When the shape property is absent in the type of a value (including node input), +it indicates that the corresponding runtime value may have any shape. This sub-section describes how to interpret a missing-shape or a shape with missing dimensions etc. However, specific usage contexts may impose further constraints on a type and shape. +For example, the inputs and outputs of a model (top-level graph) are required to *have* a shape, indicating the rank of inputs and outputs, +even though the exact dimensions need not be specified. + +Each size in the list MAY be expressed as an integral value or as a "dimension variable," a string denoting that the actual size of the dimension is not statically constrained to a particular number. This is useful for declaring interfaces that care about the number of dimensions, but not the exact size of each dimension. A dimension MAY have neither dim_value nor dim_param set. Such a dimension represents an unknown dimension unrelated to other unknown dimensions. + +For example, a NxM matrix would have the shape list [N,M]. + +The name of each dimension variable SHOULD adhere to [C90 identifier syntax rules](https://en.cppreference.com/c/language/identifier). + +Currently, dimension variables are not scoped. A dimension variable "N" represents the same value across the entire graph in a model. For example, if the graph has two inputs X and Y each with shape ["N"], then at runtime the values passed in for X and Y MUST be tensors of rank 1 with the same dimension. Nested sub-graphs currently share the same scope for dimension variables as the main-graph. This allows a model to relate the dimensions of tensors inside the subgraph to the dimensions of tensors in the outer graph. + +ONNX supports types such as Sequences of Tensors. The global scoping of dimension variables means that a variable with type "Sequence" represents a sequence of tensors that *all have the same shape*. The dimension variables M or N must be omitted from the above type if that dimension does not have a fixed size across all tensors in the sequence. The entire shape must be omitted from the type if different tensors in the sequence may have different ranks. + +For example, a graph that performs matrix cross-product may be defined as taking two inputs of shape [K,M] and [M,N], and producing an output of shape [K,N]. + +Shapes MAY be defined using a combination of integers and variables. + +_Historical Notes_: The following extensions were considered early on, but were never implemented or supported. + +* The use of an empty string (as a dimension variable) to denote an unknown dimension not related to any other dimension. This was discarded in favor of using a Dimension with neither dim_value nor dim_param set. +* The use of the string "\*" (as a dimension variable) to denote a sequence of zero or more dimensions of unknown cardinality. This is not supported. In the current implementation, the number of dimensions in a shape MUST represent the rank of the tensor. A tensor of unknown rank is represented using a TypeProto::Tensor object with no shape, which is legal. +* A scoping mechanism to allow dimension variables that are local to a sub-graph (such as the body of a loop) may be useful, but is not currently supported. +* ONNX supports types such as Sequences of Tensors. A scoping mechanism for the dimension variables local to a type may be useful to distinguish between the following two types: a sequence of square matrices (of differing sizes) vs a sequence of square matrices (all of same size). This is not currently supported. + +### Attribute Types + +The type system used for attributes is related to but slightly different from that used for of inputs and outputs. Attribute values may be a dense tensor, sparse tensor, a scalar numerical value, a string, a graph, or repeated values of one of the above mentioned types. + +## Other Metadata + +The ModelProto structure, and in IR versions >= 10, various other structures (GraphProto, FunctionProto, NodeProto) +contain a metadata_props field allowing users to store other metadata in the form of key-value pairs. +It is recommended that users use key names qualified with a reverse-DNS name as prefix +(such as "ai.onnxruntime.key1") to avoid conflicts between different uses. +Unqualified names may be used in the future by the ONNX standard. + +## Training Related Information + +Training related information is described by one or more instances of _TrainingInfoProto_ contained in a model. Each TrainingInfoProto contains information describing both an initialization step and a training step. + +The initialization step is described using a Graph (TrainingInfoProto.initialization) and an initialization-binding map (TrainingInfoProto.initialization_binding). The initialization step is performed by evaluating the Graph, and assigning the outputs produced by the Graph to the _state variables_ of the training model as specified in the initialization-binding. The initialization-binding is conceptually a map, specified as a list of key-value pairs, where each key is the name of a state variable, and the value is the name of an output of the (initialization) Graph. Each name specified as a key in the binding MUST be the name of an initializer that appears in the main inference graph (i.e., in ModelProto.graph.initializer) or the name of an initializer that appears in TrainingInfoProto.algorithm.initializer. Each name specified as a value in the binding MUST be the name of an output of the TrainingInfoProto.initialization graph. Key values specified in the repeated initialization_binding field MUST be unique. + +The training step is similarly described using a Graph (TrainingInfoProto.algorithm) and an update-binding map (TrainingInfoProto.update_binding). The training step is performed by evaluating the Graph and assigning the outputs produced by the Graph to the state variables as specified in the update-binding. The constraints and description presented above for the initialization apply to the training step as well. + +Thus, the state variables of the training model consist of a subset of the initializers of the main inference graph (i.e., ModelProto.graph.initializer) and the training-algorithm graph (TrainingInfoProto.algorithm.initializer) as identified by the keys of the bindings (in TrainingInfoProto.initialization_binding and TrainingInfoProto.update_binding). Note that the state variables are not constant values in the context of training. They represent mutable variables shared by multiple graphs (implicitly declared in the top-level training model scope). This implicit declaration of shared mutable variables is used instead of an explicit declaration for purposes of backward compatibility with the inference graph representation. + +All state variables are pre-initialized to the value specified in the corresponding initializer. A subsequent call to perform the initialization step (using the appropriate API exposed by a runtime) updates the values of the state variables as described above. If the training model has more than one instance of TrainingInfoProto, the initialization step corresponding to each is performed in order. A TrainingInfoProto.initialization MAY be omitted (only if there are no initialization_bindings). For the training step, a runtime MAY allow users to invoke any one of the TrainingInfoProto.algorithm, allowing the training process to interleave the different algorithms as desired. The order in which the different TrainingProto.algorithms are called affects the training result, and it is the callers responsibility to call them in the correct order. + +## Multi-Device Configuration (IR version >= 11) + +ONNX supports multi-device execution through device configuration specifications that enable distributed inference and training. This includes support for tensor parallelism (sharding tensors across multiple devices) and pipeline parallelism (distributing different subgraphs to different devices). + +### Device Configurations + +A model MAY specify one or more multi-device configurations using _DeviceConfigurationProto_ contained in the model. Each configuration describes a specific arrangement of devices that can be used for model execution. + +The properties of a device configuration are: + +|Name|Type|Description| +|---|---|---| +|name|string|The name of the configuration. This field MUST be present for this version of the IR.| +|num_devices|int32|Number of devices in this configuration. This field MUST be present for this version of the IR.| +|device|string[]|Optional names of the devices. MUST be length of num_devices if provided.| + +### Node Device Configuration + +Individual nodes can specify device-specific execution information through _NodeDeviceConfigurationProto_. This allows fine-grained control over how computation is distributed across devices. + +The properties of a node device configuration are: + +|Name|Type|Description| +|---|---|---| +|configuration_id|string|ID of the configuration. MUST match the name of a DeviceConfigurationProto. This field MUST be present for this version of the IR.| +|sharding_spec|ShardingSpecProto[]|Sharding specifications for the node's inputs and outputs.| +|pipeline_stage|int32|Optional pipeline stage identifier for this node.| + +### Sharding Specification + +Sharding describes how tensors are partitioned or replicated across multiple devices. A _ShardingSpecProto_ defines the sharding behavior for a specific input or output tensor of a node. + +The properties of a sharding specification are: + +|Name|Type|Description| +|---|---|---| +|tensor_name|string|Identifies the input or output tensor being sharded. Must match a name in the node's input or output list. This field MUST be present for this version of the IR.| +|device|int64[]|List of devices across which the tensor is sharded or replicated.| +|index_to_device_group_map|IntIntListEntryProto[]|Optional map indicating device groups when a device ID represents multiple physical devices.| +|sharded_dim|ShardedDimProto[]|Sharding specification for each axis of the tensor.| + +### Sharded Dimension + +A _ShardedDimProto_ describes how a single axis of a tensor is sharded across devices. + +The properties of a sharded dimension are: + +|Name|Type|Description| +|---|---|---| +|axis|int64|The tensor axis being sharded. Must be in range [-r, r-1] where r is tensor rank. This field MUST be present for this version of the IR.| +|simple_sharding|SimpleShardedDimProto[]|Describes how the axis is divided into shards.| + +### Simple Sharded Dimension + +A _SimpleShardedDimProto_ specifies that N blocks are divided into M shards, where N may be symbolic but M must be constant. + +The properties of a simple sharded dimension are: + +|Name|Type|Description| +|---|---|---| +|dim_value|int64|Dimension value to be sharded (alternative to dim_param).| +|dim_param|string|Symbolic dimension parameter to be sharded (alternative to dim_value).| +|num_shards|int64|Number of shards to split the dimension into. This field MUST be present for this version of the IR.| + +### Multi-Device Execution Semantics + +The multi-device annotations are hints to execution backends and do not affect the computational semantics of the model. Backends MAY ignore these annotations if the specified configurations are not supported or available. All communication operations required for multi-device execution (such as data transfers between devices) are implicit and handled by the runtime. + +For tensor parallelism, tensors can be: + +- **Split** across devices along specified axes, distributing different portions of the data to different devices +- **Replicated** across devices, where the same tensor data is duplicated on multiple devices + +Pipeline parallelism is indicated through optional pipeline stage identifiers that suggest how to distribute subgraphs across devices for pipelined execution. + +For more detailed information about multi-device execution patterns and examples, see the [Multi-Device Proposal](proposals/0006-ONNXMultiDeviceProposal.md). + +## Other Specification Documents + +The ONNX specification is comprised of this document, which defines the semantics of the IR and the standard data types, and the following documents defining standard operator semantics and the IR syntax. The latter is specified as Protobuf v2 and v3 schema files. + +See the [metadata category documentation](MetadataProps.md) for more details. + +### Operators + +[Neural Network Operators](Operators.md) + +[Classical Machine Learning operators](Operators-ml.md) + +### Syntax + +[ONNX Models and Graphs - protobuf v2](../onnx/onnx.proto) + +[ONNX Models and Graphs - protobuf v3](../onnx/onnx.proto3) + +[ONNX-ML Models and Graphs - protobuf v2](../onnx/onnx-ml.proto) + +[ONNX-ML Models and Graphs - protobuf v3](../onnx/onnx-ml.proto3) + +[ONNX Operator Sets - protobuf v2](../onnx/onnx-operators.proto) + +[ONNX Operator Sets - protobuf v3](../onnx/onnx-operators.proto3) + +[ONNX-ML Operator Sets - protobuf v2](../onnx/onnx-operators-ml.proto) + +[ONNX-ML Operator Sets - protobuf v3](../onnx/onnx-operators-ml.proto3) + +### Versioning Conventions and Best Practices + +[Versioning](Versioning.md) diff --git a/docs/ImplementingAnOnnxBackend.md b/docs/ImplementingAnOnnxBackend.md new file mode 100644 index 0000000..c5aa5b5 --- /dev/null +++ b/docs/ImplementingAnOnnxBackend.md @@ -0,0 +1,127 @@ + + +# Implementing an ONNX backend + +## What is an ONNX backend + +An ONNX backend is a library that can run ONNX models. Since many deep learning frameworks already exist, you likely won't need to create everything from scratch. Rather, you'll likely create a converter that converts ONNX models to the corresponding framework specific representation and then delegate the execution to the framework. For example, [onnx-caffe2 (as part of caffe2)](https://github.com/pytorch/pytorch/tree/v2.3.1/caffe2/python/onnx) , [onnx-coreml](https://github.com/onnx/onnx-coreml), and [onnx-tensorflow](https://github.com/onnx/onnx-tensorflow) are all implemented as converters. + +## Unified backend interface + +ONNX has defined a unified (Python) backend interface at [onnx/backend/base.py](/onnx/backend/base.py). + +There are three core concepts in this interface: `Device`, `Backend` and `BackendRep`. + +- `Device` is a lightweight abstraction over various hardware, e.g., CPU, GPU, etc. + +- `Backend` is the entity that will take an ONNX model with inputs, perform a computation, and then return the output. + + For one-off execution, users can use `run_node` and `run_model` to obtain results quickly. + + For repeated execution, users should use `prepare`, in which the `Backend` does all of the preparation work for executing the model repeatedly (e.g., loading initializers), and returns a `BackendRep` handle. + +- `BackendRep` is the handle that a `Backend` returns after preparing to execute a model repeatedly. Users will then pass inputs to the `run` function of `BackendRep` to retrieve the corresponding results. + +Note that even though the ONNX unified backend interface is defined in Python, your backend does not need to be implemented in Python. For example, yours can be created in C++, and tools such as [pybind11](https://github.com/pybind/pybind11) or [cython](http://cython.org/) can be used to fulfill the interface. + +## ONNX backend test + +ONNX provides a standard backend test suite to assist backend implementation verification. It's strongly encouraged that each ONNX backend runs this test. + +Integrating the ONNX Backend Test suite into your CI is simple. The following are some examples demonstrating how a backend performs the integration: + +- [onnx-caffe2 onnx backend test](https://github.com/pytorch/pytorch/blob/v2.3.1/caffe2/python/onnx/tests/onnx_backend_test.py) + +- [onnx-tensorflow onnx backend test](https://github.com/onnx/onnx-tensorflow/blob/main/test/backend/test_onnx_backend.py) + +- [onnx-coreml onnx backend test](https://github.com/onnx/onnx-coreml/blob/master/tests/onnx_backend_models_test.py) + +If you have [pytest](https://docs.pytest.org/en/latest/) installed, you can get a coverage report after running the ONNX backend test to see how well your backend is doing: + +``` +---------- onnx coverage: ---------- +Operators (passed/loaded/total): 21/21/70 +------------------------------------ +╒════════════════════╤════════════════════╕ +│ Operator │ Attributes │ +│ │ (name: #values) │ +╞════════════════════╪════════════════════╡ +│ Slice │ axes: 2 │ +│ │ ends: 3 │ +│ │ starts: 3 │ +├────────────────────┼────────────────────┤ +│ Constant │ value: 1 │ +├────────────────────┼────────────────────┤ +│ Concat │ axis: 0 │ +├────────────────────┼────────────────────┤ +│ Conv │ group: 6 │ +│ │ kernel_shape: 5 │ +│ │ pads: 4 │ +│ │ strides: 3 │ +│ │ auto_pad: 0 │ +│ │ dilations: 0 │ +├────────────────────┼────────────────────┤ +│ Reshape │ shape: 9 │ +├────────────────────┼────────────────────┤ +│ BatchNormalization │ consumed_inputs: 1 │ +│ │ epsilon: 2 │ +│ │ is_test: 1 │ +│ │ momentum: 0 │ +│ │ spatial: 0 │ +├────────────────────┼────────────────────┤ +│ Dropout │ is_test: 1 │ +│ │ ratio: 2 │ +├────────────────────┼────────────────────┤ +│ MaxPool │ kernel_shape: 2 │ +│ │ pads: 3 │ +│ │ strides: 2 │ +│ │ auto_pad: 0 │ +│ │ dilations: 0 │ +├────────────────────┼────────────────────┤ +│ Transpose │ perm: 1 │ +├────────────────────┼────────────────────┤ +│ MatMul │ No attributes │ +├────────────────────┼────────────────────┤ +│ Relu │ No attributes │ +├────────────────────┼────────────────────┤ +│ LRN │ alpha: 2 │ +│ │ beta: 1 │ +│ │ bias: 2 │ +│ │ size: 1 │ +├────────────────────┼────────────────────┤ +│ Add │ axis: 1 │ +│ │ broadcast: 1 │ +├────────────────────┼────────────────────┤ +│ Abs │ No attributes │ +├────────────────────┼────────────────────┤ +│ Pad │ mode: 3 │ +│ │ paddings: 2 │ +│ │ value: 1 │ +├────────────────────┼────────────────────┤ +│ Softmax │ axis: 0 │ +├────────────────────┼────────────────────┤ +│ GlobalAveragePool │ No attributes │ +├────────────────────┼────────────────────┤ +│ Mul │ axis: 1 │ +│ │ broadcast: 1 │ +├────────────────────┼────────────────────┤ +│ Sum │ No attributes │ +├────────────────────┼────────────────────┤ +│ Gemm │ broadcast: 1 │ +│ │ transB: 1 │ +│ │ alpha: 0 │ +│ │ beta: 0 │ +│ │ transA: 0 │ +├────────────────────┼────────────────────┤ +│ AveragePool │ kernel_shape: 3 │ +│ │ pads: 3 │ +│ │ strides: 2 │ +│ │ auto_pad: 0 │ +╘════════════════════╧════════════════════╛ +``` + +The numbers in the line `Operators (passed/loaded/total): 21/21/70` indicate 21 operators covered in all test cases of your backend have passed, 21 operators were covered in all test cases of the ONNX backend test, and ONNX has a total of 70 operators. diff --git a/docs/ManagingExperimentalOps.md b/docs/ManagingExperimentalOps.md new file mode 100644 index 0000000..17b1e8e --- /dev/null +++ b/docs/ManagingExperimentalOps.md @@ -0,0 +1,42 @@ + + +# Managing Experimental Operators + +## Deprecated Experimental Operators + +The following experimental operators were deprecated and removed from ONNX. They should be removed from models, either substituted with newer superseding operators or decomposed into functionally equivalent operators: + +Old operator |New Operator +--------------------|-------------------------- +`ATen` |NA +`Affine` |`Add(Mul(X, alpha), beta)` +`ConstantFill` |`ConstantOfShape` +`Crop` |`Slice-1` +`DynamicSlice` |`Slice-10` +`GRUUnit` |NA +`GivenTensorFill` |`Const` or `ConstantOfShape` +`ImageScaler` |`Add(Mul(X, scale), Unsqueeze(bias, axes=[0, 2, 3]))` +`ParametricSoftplus`|`Mul(alpha, Softplus(Mul(beta, X)))` +`Scale` |`Mul(X, scale)` +`ScaledTanh` |`Mul(Tanh(Mul(X, beta)), alpha)` + +## Adding Experimental Operators [Deprecated - as of v1.5 experimental ops are no longer supported] + +The experimental flag in ONNX operator definitions indicates that a customer of ONNX may not be able to take a long term dependency on that op. Ops in the ONNX namespace (ai.onnx) in the _main_ branch, whether experimental or not, go through the regular review process. + +Experimental ops that are being worked on that do not have consensus yet can be managed in one of 2 ways: + +1. Use a fork or branch – what you do in the fork or branch is entirely up to you. When you are ready, you can submit a PR using the normal process. This is the recommended way. +2. If a fork/branch is not workable (for example due to complexity of mapping different branches between multiple repos), put the experimental ops in a custom namespace in the main branch. +The specific process for this is: + +* Submit an Issue with a proposal explaining the motivation and plan. It does not need to include detailed technical design. Issues will be tagged as "experimental op". +* Reviewers will generally approve by default unless the proposal directly conflicts with existing ops or somehow goes against general ONNX strategy. Approval is indicated by adding the "experiment approved" tag. +* The approval is good for 3 months, but can be renewed if needed. +* Experimental ops should be submitted in a PR in a custom namespace that is the name of the proposal, i.e. “proposal.controlflow”. The name should be descriptive rather than a company or entity name. These PRs will be approved by default as long as the parent proposal is approved and active. +* Once experimentation is done, the ops can be submitted for addition to the ONNX namespace via the regular process. The owner can also choose to end the experiment without promoting the ops. +* Either way, the custom namespace is deleted once experimentation is complete or when the approval expires. diff --git a/docs/MetadataProps.md b/docs/MetadataProps.md new file mode 100644 index 0000000..5b9fe62 --- /dev/null +++ b/docs/MetadataProps.md @@ -0,0 +1,35 @@ + + +# Metadata + +In addition to the core metadata recommendations listed in the [extensibility documentation](IR.md#optional-metadata) there is additional experimental metadata to help provide information for model inputs and outputs. + +This metadata applies to all input and output tensors of a given category. The first such category we define is: `Image`. + +## Motivation + +The motivation of such a mechanism is to allow model authors to convey to model consumers enough information for them to consume the model. + +In the case of images there are many option for providing valid image data. However a model which consumes images was trained with a particular set of these options which must +be used during inferencing. + +The goal is this proposal is to provide enough metadata that the model consumer can perform their own featurization prior to running the model and provide a compatible input or retrieve an output and know what its format is. + +## Image Category Definition + +For every tensor in this model that uses [Type Denotation](TypeDenotation.md) to declare itself an `IMAGE`, you SHOULD provide metadata to assist the model consumer. Note that any metadata provided using this mechanism is global to ALL types +with the accompanying denotation. + +Keys and values are case insensitive. + +Specifically, we define here the following set image metadata: + +|Key|Value|Description| +|-----|----|-----------| +|`Image.BitmapPixelFormat`|__string__|Specifies the format of pixel data. Each enumeration value defines a channel ordering and bit depth. Possible values:
  • `Gray8`: 1 channel image, the pixel data is 8 bpp grayscale.
  • `Rgb8`: 3 channel image, channel order is RGB, pixel data is 8bpp (No alpha)
  • `Bgr8`: 3 channel image, channel order is BGR, pixel data is 8bpp (No alpha)
  • `Rgba8`: 4 channel image, channel order is RGBA, pixel data is 8bpp (Straight alpha)
  • `Bgra8`: 4 channel image, channel order is BGRA, pixel data is 8bpp (Straight alpha)
| +|`Image.ColorSpaceGamma`|__string__|Specifies the gamma color space used. Possible values:
  • `Linear`: Linear color space, gamma == 1.0
  • `SRGB`: sRGB color space, gamma == 2.2
| +|`Image.NominalPixelRange`|__string__|Specifies the range that pixel values are stored. Possible values:
  • `NominalRange_0_255`: [0...255] for 8bpp samples
  • `Normalized_0_1`: [0...1] pixel data is stored normalized
  • `Normalized_1_1`: [-1...1] pixel data is stored normalized
  • `NominalRange_16_235`: [16...235] for 8bpp samples
| diff --git a/docs/ONNXTypes.md b/docs/ONNXTypes.md new file mode 100644 index 0000000..f59cbc6 --- /dev/null +++ b/docs/ONNXTypes.md @@ -0,0 +1,105 @@ + + +# ONNX Types + +## Optional Type + +An optional type represents a reference to either an element (could be Tensor, Sequence, Map, or Sparse Tensor) or a null value. The optional type appears in model inputs, outputs, as well as intermediate values. + +### Use-cases + +Optional type enables users to represent more dynamic typing scenarios in ONNX. Similar to Optional[X] type hint in Python typing which is equivalent to Union[None, X], Optional types in ONNX may reference a single element, or null. + +### Examples in PyTorch + +Optional type only appears in TorchScript graphs generated by jit script compiler. Scripting a model captures dynamic types where an optional value can be assigned either None or a value. + +- Example 1 + + class Model(torch.nn.Module): + def forward(self, x, y:Optional[Tensor]=None): + if y is not None: + return x + y + return x + + Corresponding TorchScript graph: + + Graph( + %self : __torch__.Model, + %x.1 : Tensor, + %y.1 : Tensor? + ): + %11 : int = prim::Constant[value=1]() + %4 : None = prim::Constant() + %5 : bool = aten::__isnot__(%y.1, %4) + %6 : Tensor = prim::If(%5) + block0(): + %y.4 : Tensor = prim::unchecked_cast(%y.1) + %12 : Tensor = aten::add(%x.1, %y.4, %11) + -> (%12) + block1(): + -> (%x.1) + return (%6) + + ONNX graph: + + Graph( + %x.1 : Float(2, 3), + %y.1 : Float(2, 3) + ): + %2 : Bool(1) = onnx::OptionalHasElement(%y.1) + %5 : Float(2, 3) = onnx::If(%2) + block0(): + %3 : Float(2, 3) = onnx::OptionalGetElement(%y.1) + %4 : Float(2, 3) = onnx::Add(%x.1, %3) + -> (%4) + block1(): + %x.2 : Float(2, 3) = onnx::Identity(%x.1) + -> (%x.2) + return (%5) + +- Example 2 + + class Model(torch.nn.Module): + def forward( + self, + src_tokens, + return_all_hiddens=torch.tensor([False]), + ): + encoder_states: Optional[Tensor] = None + if return_all_hiddens: + encoder_states = src_tokens + + return src_tokens, encoder_states + + Corresponding TorchScript graph: + + Graph( + %src_tokens.1 : Float(3, 2, 4,), + %return_all_hiddens.1 : Bool(1) + ): + %3 : None = prim::Constant() + %encoder_states : Tensor? = prim::If(%return_all_hiddens.1) + block0(): + -> (%src_tokens.1) + block1(): + -> (%3) + return (%src_tokens.1, %encoder_states) + + ONNX graph: + + Graph( + %src_tokens.1 : Float(3, 2, 4), + %return_all_hiddens.1 : Bool(1) + ): + %2 : Float(3, 2, 4) = onnx::Optional[type=tensor(float)]() + %3 : Float(3, 2, 4) = onnx::If(%return_all_hiddens.1) + block0(): + -> (%src_tokens.1) + block1(): + -> (%2) + return (%3) diff --git a/docs/OnnxBackendTest.md b/docs/OnnxBackendTest.md new file mode 100644 index 0000000..51490c4 --- /dev/null +++ b/docs/OnnxBackendTest.md @@ -0,0 +1,24 @@ + + +# ONNX Backend Test + +## What is ONNX Backend Test + +ONNX Backend Test is a test suite that each ONNX backend should run to verify whether it fulfills ONNX's standard. It serves both as a verification tool for backend implementations and one of the two ways to define each operator's expected behavior (the other way is to add it to the documentation). + +There are two types of tests in this suite – Node Tests and Model Tests: + +- **Node Tests** verify whether a backend is performing the correct computation, having the expected behavior of handling various attributes for each individual operator. In each test case, the backend will be given a node with some input, and the returned output will be compared with an expected output. +- **Model Tests** verify the backend at the model level. The test cases are similar to those of Node Tests', but instead of a node, the backend will be given an ONNX model. + +## Contributing + +As ONNX aims to become the spec of deep learning models format, it's important to ensure that there is no ambiguity in each ONNX operator's definition; adding more test cases is the only way to enforce this. + +Node Tests are created as Python/Numpy code in [onnx/backend/test/case/node](/onnx/backend/test/case/node), and then exported to protobuf files to [onnx/backend/test/data/node](/onnx/backend/test/data/node) as the source of truth by invoking the shell command `backend-test-tools generate-data`. Test cases of each operator lives in one standalone file, e.g. for the operator [Add](/docs/Operators.md#Add), its test cases are in [add.py](/onnx/backend/test/case/node/add.py), and each `expect(...)` statement in the code corresponds to one test case. The source code of all `export.*` functions will be also embedded as example code snippets in the [Operators documentation page](/docs/Operators.md). You are contributing to both the test and the documentation! + +For Model Tests, since each model protobuf file can be large in size, we don't place the file directly in the repo. Rather, we upload them to the cloud, and download them on demand when running the tests. Each test case consists of one model definition protobuf file, and several pairs of input and output files. Adding a new test case involves some manual work from admins (like uploading the files to the cloud), so if you have an ONNX model that you would like to contribute, please contact us. diff --git a/docs/OnnxReleases.md b/docs/OnnxReleases.md new file mode 100644 index 0000000..1e41117 --- /dev/null +++ b/docs/OnnxReleases.md @@ -0,0 +1,182 @@ + + +# ONNX Releases + +The ONNX project, going forward, will plan to release roughly on a four month cadence. We follow the [Semver](https://semver.org/) versioning approach and will make decisions as a community on a release by release basis on whether to do a major or minor release. + +## Preparation + +* Reach out to the SIG Arch/Infra leads to confirm whether the required status checks for the release branches are still valid and up to date, and whether any rely on outdated hardcoded runner image versions that may need updatingCheck whether the 'required checks' for the release branches are still up to date or need to be adjusted: 'Branches' -> 'Branch protection rules' +* Determine version (X.Y.Z) for the new release + * Discuss in Slack channel for Releases (https://lfaifoundation.slack.com/archives/C018VGGJUGK) + * For (v.X.Y.Z), if release is to be 1.16.0, + * X=1, Y=16, Z=0 + * The new branch will be `rel-1.16.0` + * Branch protections rules are automatically applied to branches following this format. + * The new tag will be `v1.16.0` +* Create new page for the release in [Release logistics wiki](https://github.com/onnx/onnx/wiki) (Add the release manager to the ONNX organization, if not already seen, so that the manager has write permissions in the wiki.) +* Before creating a release branch, it is highly recommended to have in mind to compile **preliminary release notes** — ideally maintained in a shared location such as the **release wiki page**. These notes should include a clear summary of the **new features**, a list of **bug fixes**, any **known issues**, and especially any **deprecations or removals**, with links to relevant tickets or documentation where applicable. Having this information ready ensures that the team can confidently and promptly create a `rc1` (release candidate 1) immediately after the branch is cut, without delays. Acting quickly at this stage also helps to **reduce the need for parallel work on both the main and release branches**, minimizing merge conflicts, duplicated effort, and coordination overhead. This practice supports a smoother, more transparent release process. + * To generate good release notes, it is helpful if pull requests have meaningful names and corresponding labels. Labels can also be added retrospectively to PRs that have already been merged. + * The labels used can be found [here](https://github.com/onnx/onnx/blob/main/.github/release.yml) + * The preliminary release notes one gets if one drafts a release on GitHub. + +## Create Release Branch + +* In `main` branch, before creating the release branch: + 1. Make sure the release version ([/VERSION_NUMBER](/VERSION_NUMBER)), IR version, ai.onnx opset version, ai.onnx.ml opset version, and ai.onnx.training opset version are correct for the new release in [ONNX proto files](/onnx/onnx.in.proto), [Versioning.md](Versioning.md), [schema.h](/onnx/defs/schema.h), [helper.py](/onnx/helper.py), and [helper_test.py](/onnx/test/helper_test.py). + +* Create a release branch + 1. Click "New branch" from [branches](https://github.com/onnx/onnx/branches) and choose `main` as Source. + 1. Make sure all tests pass on the new branch. + +* After cutting the release branch: + 1. Create PR to set [VERSION_NUMBER](/VERSION_NUMBER) file in `main` to the next future release, `X.Y+1.0`. + 1. Create PR to set `VERSION_NUMBER` file in the new release's branch to `X.Y.Zrc1`. + * For example the first release candidate for 1.16.0 would be `1.16.0rc1` + 1. Bump opset version for ai.onnx domain in `onnx/defs/operator_sets.h` and `onnx/defs/schema.h` for use by future operator additions and changes. + * For example, this [demo PR](https://github.com/onnx/onnx/pull/6001). + +## Upload release candidate to PyPI + +* Go to "Actions" -> select ["Create Releases"](https://github.com/onnx/onnx/actions/workflows/create_release.yml) -> Push the button "Run workflow" with the following config: + +make-release-config + +RC-Candidates + +* Published to https://pypi.org/ (starting with onnx 1.19.2 before it was test.pypi.org) +* Build-mode: Release + +* This button triggers the build of the different OS + +create_releases_overview_jobs + +* All artifacts of the single runs could be found associated to the job + +create_releases_artifact_overview + +* Before the final merge, it must be confirmed manually via the set up deployment environments. + +## Package verification + +**Partner Validation** + + * User should install the rc-packages with `pip install onnx=={rc version}` + * Open Issues for external repos: + * Create GitHub issues in converters' repos to provide them the package links and oppuruntity to test the release before it goes public. + * https://github.com/microsoft/onnxruntime + * Example: https://github.com/microsoft/onnxruntime/issues/19783 + * Note: [How_To_Update_ONNX_Dev_Notes](https://github.com/microsoft/onnxruntime/blob/main/docs/How_To_Update_ONNX_Dev_Notes.md) exists in their repo documenting how to pull in new ONNX releases. + * https://github.com/pytorch/pytorch + * Example: https://github.com/pytorch/pytorch/issues/121258 + * https://github.com/onnx/tensorflow-onnx + * Example: https://github.com/onnx/tensorflow-onnx/issues/2310 + * https://github.com/onnx/onnx-tensorrt + * Example: https://github.com/onnx/onnx-tensorrt/issues/956 + * https://github.com/onnx/sklearn-onnx + * Example: https://github.com/onnx/sklearn-onnx/issues/1079 + * https://github.com/microsoft/onnxconverter-common + * Example: https://github.com/microsoft/onnxconverter-common/issues/277 + * https://github.com/onnx/onnxmltools + * Example: https://github.com/onnx/onnxmltools/issues/685 + * https://github.com/Quantco/spox + * https://github.com/openvinotoolkit/openvino + * https://github.com/conda-forge/onnx-feedstock + + * If issues are found, the bugs are to be fixed in the onnx `main` branch and then cherry-picked into the release branch. + * Follow up with reporter to ensure issues are resolved (and validated in a new rc) or deferred to a new release. + +# Official Release + +Validation steps must be completed before this point! This is the point of new return. + +* git tags should not be changed once published +* Once pushed to PyPI there is no way to update the release. A new release must be made instead + +## Set final version number + +* Create PR to remove "`rcX`" suffix from `VERSION_NUMBER` file in the new release's branch. + +## Create release tag + +* [Draft a release](https://github.com/onnx/onnx/releases/new) based on the release branch: + * DO NOT click `Publish release` until you are sure no more changes are needed. + * Use `Save Draft` if need to save and update more later. + * Publishing will create the new git tag + * Tag: See top of [Preparation](#Preparation) for tag to create. + * Target: The release branch that was just cut + * Previous tag: Select the previous release. + * Write: + * Use [previous releases](https://github.com/onnx/onnx/releases) as a template + * Use information from [Release logistics wiki](https://github.com/onnx/onnx/wiki) which should have been created prior to branch cut. + * Add any additional commits that merged into the release in since wiki was written. + * .tar.gz and .zip will be auto-generated after publishing the release. + +## Upload to Official PyPI + +* Starting with the release of 1.19, the final release will also be pushed to pypi via Github “Action" -> "Create releases" (see above). Use the following config for official release: + +RunWorkflow_Final + +### NOTES: + +* Once the packages are uploaded to PyPI, **you cannot overwrite it on the same PyPI instance**. + * Please make sure everything is good on TestPyPI before uploading to PyPI** +* PyPI has separate logins, passwords, and API tokens from TestPyPI but the process is the same. An ONNX PyPI owner will need to grant access, etc. + +## After PyPI Release + +**Announce** + +* Slack: + * Post in the [onnx-release](https://lfaifoundation.slack.com/archives/C018VGGJUGK) and [onnx-general](https://lfaifoundation.slack.com/archives/C016UBNDBL2) channels. +* Notify ONNX partners via email lists: + * onnxdiscussions@service.microsoft.com + * onnxconverterteam@service.microsoft.com + * onnxruntimeteam@microsoft.com +* [ONNX News](https://onnx.ai/news.html) Post + * Update [news.json](https://github.com/onnx/onnx.github.io/blob/main/js/news.json), see [example news.json PR](https://github.com/onnx/onnx.github.io/pull/197) + +**Update conda-forge package with the new ONNX version** + +Conda builds of ONNX are done via [conda-forge/onnx-feedstock](https://github.com/conda-forge/onnx-feedstock), which runs infrastructure for building packages and uploading them to conda-forge. + +* A PR should be created automatically by `regro-cf-autotick-bot` a few hours after the release is available at https://github.com/onnx/onnx/releases. + +* If the automatic PR has build failures: + 1. Make a personal fork of conda-forge/onnx-feedstock + 1. Create a personal branch based on the automated PR branch + 1. Resolve the build issue + 1. Submit a replacement PR based on your branch + + * Example: https://github.com/conda-forge/onnx-feedstock/pull/116 + +* If the automatic PR is not created, you need to submit a PR manually + * Example: https://github.com/conda-forge/onnx-feedstock/pull/50 + * Note: Use the sha256 hash (`sha256sum onnx-X.Y.Z.tar.gz`) of the release's tar.gz file from https://github.com/onnx/onnx/releases. + +**Merge into main branch** + +* Check which changes to the release branch are also relevant for main: + * If urgent changes were made directly into the release branch, merge the release branch back into main branch. + * If all PRs merged into the release branch (after it was cut) were cherry-picks from main, the merge PR will show as empty and this step is not needed. + +**Remove old onnx-weekly packages on PyPI** + +* Remove all [onnx-weekly packages](https://pypi.org/project/onnx-weekly/#history) from PyPI for the just released version to save space. +* Steps: + * Go to [PyPI onnx-weekly/releases](https://pypi.org/manage/project/onnx-weekly/releases/) + * This is a separate project than the onnx releases so you may need to request access from an owner + * Click target package -> Options -> Delete. + +**Remove old release-candidate packages on PyPI** + +* Remove [onnx-release-candidate packages](https://test.pypi.org/project/onnx/#history) from PyPI up to at least the time specified by the previous release version to save space. +* Steps: + * Go to [PyPI onnx-weekly/releases](https://test.pypi.org/manage/project/onnx/releases/) + * This is a separate project than the onnx releases so you may need to request access from an owner + * Click target package -> Options -> Delete. diff --git a/docs/OpConventions.md b/docs/OpConventions.md new file mode 100644 index 0000000..aa721d3 --- /dev/null +++ b/docs/OpConventions.md @@ -0,0 +1,17 @@ + + +# Operator Conventions + +To maintain consistency in operator signatures, we use the following principles: + +- All attribute names should be lower case and use underscores when it helps with readability +- Any input/output represented by a single letter is capitalized (i.e. X) +- Any input/output represented by a full word or multiple words is all lower case and uses underscores when it helps with readability +- Any input/output representing a bias tensor will utilize the name "B" +- Any input/output representing a weight tensor will utilize the name “W” +- “axes” is used when an input, output or attribute is representing multiple axes +- “axis” is used when an input, output or attribute is representing a single axis diff --git a/docs/Operators-ml.md b/docs/Operators-ml.md new file mode 100644 index 0000000..abcdff0 --- /dev/null +++ b/docs/Operators-ml.md @@ -0,0 +1,1214 @@ + +## Operator Schemas +*This file is automatically generated from the + [def files](/onnx/defs) via [this script](/onnx/defs/gen_doc.py). + Do not modify directly and instead edit operator definitions.* + +For an operator input/output's differentiability, it can be differentiable, + non-differentiable, or undefined. If a variable's differentiability + is not specified, that variable has undefined differentiability. + +### ai.onnx.ml +|**Operator**|**Since version**|| +|-|-|-| +|ai.onnx.ml.ArrayFeatureExtractor|1| +|ai.onnx.ml.Binarizer|1| +|ai.onnx.ml.CastMap|1| +|ai.onnx.ml.CategoryMapper|1| +|ai.onnx.ml.DictVectorizer|1| +|ai.onnx.ml.FeatureVectorizer|1| +|ai.onnx.ml.Imputer|1| +|ai.onnx.ml.LabelEncoder|4, 2, 1| +|ai.onnx.ml.LinearClassifier|1| +|ai.onnx.ml.LinearRegressor|1| +|ai.onnx.ml.Normalizer|1| +|ai.onnx.ml.OneHotEncoder|1| +|ai.onnx.ml.SVMClassifier|1| +|ai.onnx.ml.SVMRegressor|1| +|ai.onnx.ml.Scaler|1| +|ai.onnx.ml.TreeEnsemble|5| +|ai.onnx.ml.TreeEnsembleClassifier (deprecated)|5, 3, 1| +|ai.onnx.ml.TreeEnsembleRegressor (deprecated)|5, 3, 1| +|ai.onnx.ml.ZipMap|1| + + +## ai.onnx.ml +### **ai.onnx.ml.ArrayFeatureExtractor** + + Select elements of the input tensor based on the indices passed.
+ The indices are applied to the last axes of the tensor. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Inputs + +
+
X : T
+
Data to be selected
+
Y : tensor(int64)
+
The indices, based on 0 as the first index of any dimension.
+
+ +#### Outputs + +
+
Z : T
+
Selected output data as an array
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32), tensor(string)
+
The input must be a tensor of a numeric type or string. The output will be of the same tensor type.
+
+ + +#### Examples + +
+arrayfeatureextractor + +```python +node = onnx.helper.make_node( + "ArrayFeatureExtractor", + inputs=["x", "y"], + outputs=["z"], + domain="ai.onnx.ml", +) + +x = np.arange(12).reshape((3, 4)).astype(np.float32) +y = np.array([0, 1], dtype=np.int64) +z = np.array([[0, 4, 8], [1, 5, 9]], dtype=np.float32).T +expect( + node, + inputs=[x, y], + outputs=[z], + name="test_ai_onnx_ml_array_feature_extractor", +) +``` + +
+ + +### **ai.onnx.ml.Binarizer** + + Maps the values of the input tensor to either 0 or 1, element-wise, based on the outcome of a comparison against a threshold value. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
threshold : float (default is 0.0)
+
Values greater than this are mapped to 1, others to 0.
+
+ +#### Inputs + +
+
X : T
+
Data to be binarized
+
+ +#### Outputs + +
+
Y : T
+
Binarized output data
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type. The output will be of the same tensor type.
+
+ + +#### Examples + +
+binarizer + +```python +threshold = 1.0 +node = onnx.helper.make_node( + "Binarizer", + inputs=["X"], + outputs=["Y"], + threshold=threshold, + domain="ai.onnx.ml", +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = compute_binarizer(x, threshold)[0] + +expect(node, inputs=[x], outputs=[y], name="test_ai_onnx_ml_binarizer") +``` + +
+ + +### **ai.onnx.ml.CastMap** + + Converts a map to a tensor.
The map key must be an int64 and the values will be ordered + in ascending order based on this key.
The operator supports dense packing or sparse packing. + If using sparse packing, the key cannot exceed the max_map-1 value. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
cast_to : string (default is TO_FLOAT)
+
A string indicating the desired element type of the output tensor, one of 'TO_FLOAT', 'TO_STRING', 'TO_INT64'.
+
map_form : string (default is DENSE)
+
Indicates whether to only output as many values as are in the input (dense), or position the input based on using the key of the map as the index of the output (sparse).
One of 'DENSE', 'SPARSE'.
+
max_map : int (default is 1)
+
If the value of map_form is 'SPARSE,' this attribute indicates the total length of the output tensor.
+
+ +#### Inputs + +
+
X : T1
+
The input map that is to be cast to a tensor
+
+ +#### Outputs + +
+
Y : T2
+
A tensor representing the same data as the input map, ordered by their keys
+
+ +#### Type Constraints + +
+
T1 : map(int64, string), map(int64, float)
+
The input must be an integer map to either string or float.
+
T2 : tensor(string), tensor(float), tensor(int64)
+
The output is a 1-D tensor of string, float, or integer.
+
+ + +### **ai.onnx.ml.CategoryMapper** + + Converts strings to integers and vice versa.
+ Two sequences of equal length are used to map between integers and strings, + with strings and integers at the same index detailing the mapping.
+ Each operator converts either integers to strings or strings to integers, depending + on which default value attribute is provided. Only one default value attribute + should be defined.
+ If the string default value is set, it will convert integers to strings. + If the int default value is set, it will convert strings to integers. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
cats_int64s : list of ints
+
The integers of the map. This sequence must be the same length as the 'cats_strings' sequence.
+
cats_strings : list of strings
+
The strings of the map. This sequence must be the same length as the 'cats_int64s' sequence
+
default_int64 : int (default is -1)
+
An integer to use when an input string value is not found in the map.
One and only one of the 'default_*' attributes must be defined.
+
default_string : string (default is _Unused)
+
A string to use when an input integer value is not found in the map.
One and only one of the 'default_*' attributes must be defined.
+
+ +#### Inputs + +
+
X : T1
+
Input data
+
+ +#### Outputs + +
+
Y : T2
+
Output data. If strings are input, the output values are integers, and vice versa.
+
+ +#### Type Constraints + +
+
T1 : tensor(string), tensor(int64)
+
The input must be a tensor of strings or integers, either [N,C] or [C].
+
T2 : tensor(string), tensor(int64)
+
The output is a tensor of strings or integers. Its shape will be the same as the input shape.
+
+ + +### **ai.onnx.ml.DictVectorizer** + + Uses an index mapping to convert a dictionary to an array.
+ Given a dictionary, each key is looked up in the vocabulary attribute corresponding to + the key type. The index into the vocabulary array at which the key is found is then + used to index the output 1-D tensor 'Y' and insert into it the value found in the dictionary 'X'.
+ The key type of the input map must correspond to the element type of the defined vocabulary attribute. + Therefore, the output array will be equal in length to the index mapping vector parameter. + All keys in the input dictionary must be present in the index mapping vector. + For each item in the input dictionary, insert its value in the output array. + Any keys not present in the input dictionary, will be zero in the output array.
+ For example: if the ``string_vocabulary`` parameter is set to ``["a", "c", "b", "z"]``, + then an input of ``{"a": 4, "c": 8}`` will produce an output of ``[4, 8, 0, 0]``. + + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
int64_vocabulary : list of ints
+
An integer vocabulary array.
One and only one of the vocabularies must be defined.
+
string_vocabulary : list of strings
+
A string vocabulary array.
One and only one of the vocabularies must be defined.
+
+ +#### Inputs + +
+
X : T1
+
A dictionary.
+
+ +#### Outputs + +
+
Y : T2
+
A 1-D tensor holding values from the input dictionary.
+
+ +#### Type Constraints + +
+
T1 : map(string, int64), map(int64, string), map(int64, float), map(int64, double), map(string, float), map(string, double)
+
The input must be a map from strings or integers to either strings or a numeric type. The key and value types cannot be the same.
+
T2 : tensor(int64), tensor(float), tensor(double), tensor(string)
+
The output will be a tensor of the value type of the input map. It's shape will be [1,C], where C is the length of the input dictionary.
+
+ + +### **ai.onnx.ml.FeatureVectorizer** + + Concatenates input tensors into one continuous output.
+ All input shapes are 2-D and are concatenated along the second dimension. 1-D tensors are treated as [1,C]. + Inputs are copied to the output maintaining the order of the input arguments.
+ All inputs must be integers or floats, while the output will be all floating point values. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
inputdimensions : list of ints
+
The size of each input in the input list
+
+ +#### Inputs (1 - ∞) + +
+
X (variadic) : T1
+
An ordered collection of tensors, all with the same element type.
+
+ +#### Outputs + +
+
Y : tensor(float)
+
The output array, elements ordered as the inputs.
+
+ +#### Type Constraints + +
+
T1 : tensor(int32), tensor(int64), tensor(float), tensor(double)
+
The input type must be a tensor of a numeric type.
+
+ + +### **ai.onnx.ml.Imputer** + + Replaces inputs that equal one value with another, leaving all other elements alone.
+ This operator is typically used to replace missing values in situations where they have a canonical + representation, such as -1, 0, NaN, or some extreme value.
+ One and only one of imputed_value_floats or imputed_value_int64s should be defined -- floats if the input tensor + holds floats, integers if the input tensor holds integers. The imputed values must all fit within the + width of the tensor element type. One and only one of the replaced_value_float or replaced_value_int64 should be defined, + which one depends on whether floats or integers are being processed.
+ The imputed_value attribute length can be 1 element, or it can have one element per input feature.
In other words, if the input tensor has the shape [*,F], then the length of the attribute array may be 1 or F. If it is 1, then it is broadcast along the last dimension and applied to each feature. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
imputed_value_floats : list of floats
+
Value(s) to change to
+
imputed_value_int64s : list of ints
+
Value(s) to change to.
+
replaced_value_float : float (default is 0.0)
+
A value that needs replacing.
+
replaced_value_int64 : int (default is 0)
+
A value that needs replacing.
+
+ +#### Inputs + +
+
X : T
+
Data to be processed.
+
+ +#### Outputs + +
+
Y : T
+
Imputed output data
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input type must be a tensor of a numeric type, either [N,C] or [C]. The output type will be of the same tensor type and shape.
+
+ + +### **ai.onnx.ml.LabelEncoder** + + Maps each element in the input tensor to another value.
+ The mapping is determined by the two parallel attributes, 'keys_*' and + 'values_*' attribute. The i-th value in the specified 'keys_*' attribute + would be mapped to the i-th value in the specified 'values_*' attribute. It + implies that input's element type and the element type of the specified + 'keys_*' should be identical while the output type is identical to the + specified 'values_*' attribute. Note that the 'keys_*' and 'values_*' attributes + must have the same length. If an input element can not be found in the + specified 'keys_*' attribute, the 'default_*' that matches the specified + 'values_*' attribute may be used as its output value. The type of the 'default_*' + attribute must match the 'values_*' attribute chosen.
+ Let's consider an example which maps a string tensor to an integer tensor. + Assume and 'keys_strings' is ["Amy", "Sally"], 'values_int64s' is [5, 6], + and 'default_int64' is '-1'. The input ["Dori", "Amy", "Amy", "Sally", + "Sally"] would be mapped to [-1, 5, 5, 6, 6].
+ Since this operator is an one-to-one mapping, its input and output shapes + are the same. Notice that only one of 'keys_*'/'values_*' can be set.
+ Float keys with value 'NaN' match any input 'NaN' value regardless of bit + value. If a key is repeated, the last key takes precedence. + +#### Version + +This version of the operator has been available since version 4 of the 'ai.onnx.ml' operator set. + +Other versions of this operator: 1, 2 + +#### Attributes + +
+
default_float : float (default is -0.0)
+
A float.
+
default_int64 : int (default is -1)
+
An integer.
+
default_string : string (default is _Unused)
+
A string.
+
default_tensor : tensor
+
A default tensor. {"_Unused"} if values_* has string type, {-1} if values_* has integral type, and {-0.f} if values_* has float type.
+
keys_floats : list of floats
+
A list of floats.
+
keys_int64s : list of ints
+
A list of ints.
+
keys_strings : list of strings
+
A list of strings.
+
keys_tensor : tensor
+
Keys encoded as a 1D tensor. One and only one of 'keys_*'s should be set.
+
values_floats : list of floats
+
A list of floats.
+
values_int64s : list of ints
+
A list of ints.
+
values_strings : list of strings
+
A list of strings.
+
values_tensor : tensor
+
Values encoded as a 1D tensor. One and only one of 'values_*'s should be set.
+
+ +#### Inputs + +
+
X : T1
+
Input data. It must have the same element type as the keys_* attribute set.
+
+ +#### Outputs + +
+
Y : T2
+
Output data. This tensor's element type is based on the values_* attribute set.
+
+ +#### Type Constraints + +
+
T1 : tensor(string), tensor(int64), tensor(float), tensor(int32), tensor(int16), tensor(double)
+
The input type is a tensor of any shape.
+
T2 : tensor(string), tensor(int64), tensor(float), tensor(int32), tensor(int16), tensor(double)
+
Output type is determined by the specified 'values_*' attribute.
+
+ + +#### Examples + +
+string_int_label_encoder + +```python +node = onnx.helper.make_node( + "LabelEncoder", + inputs=["X"], + outputs=["Y"], + domain="ai.onnx.ml", + keys_strings=["a", "b", "c"], + values_int64s=[0, 1, 2], + default_int64=42, +) +x = np.array(["a", "b", "d", "c", "g"]).astype(object) +y = np.array([0, 1, 42, 2, 42]).astype(np.int64) +expect( + node, + inputs=[x], + outputs=[y], + name="test_ai_onnx_ml_label_encoder_string_int", +) + +node = onnx.helper.make_node( + "LabelEncoder", + inputs=["X"], + outputs=["Y"], + domain="ai.onnx.ml", + keys_strings=["a", "b", "c"], + values_int64s=[0, 1, 2], +) +x = np.array(["a", "b", "d", "c", "g"]).astype(object) +y = np.array([0, 1, -1, 2, -1]).astype(np.int64) +expect( + node, + inputs=[x], + outputs=[y], + name="test_ai_onnx_ml_label_encoder_string_int_no_default", +) +``` + +
+ + +
+tensor_based_label_encoder + +```python +tensor_keys = make_tensor( + "keys_tensor", onnx.TensorProto.STRING, (3,), ["a", "b", "c"] +) +repeated_string_keys = ["a", "b", "c"] +x = np.array(["a", "b", "d", "c", "g"]).astype(object) +y = np.array([0, 1, 42, 2, 42]).astype(np.int16) + +node = onnx.helper.make_node( + "LabelEncoder", + inputs=["X"], + outputs=["Y"], + domain="ai.onnx.ml", + keys_tensor=tensor_keys, + values_tensor=make_tensor( + "values_tensor", onnx.TensorProto.INT16, (3,), [0, 1, 2] + ), + default_tensor=make_tensor( + "default_tensor", onnx.TensorProto.INT16, (1,), [42] + ), +) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_ai_onnx_ml_label_encoder_tensor_mapping", +) + +node = onnx.helper.make_node( + "LabelEncoder", + inputs=["X"], + outputs=["Y"], + domain="ai.onnx.ml", + keys_strings=repeated_string_keys, + values_tensor=make_tensor( + "values_tensor", onnx.TensorProto.INT16, (3,), [0, 1, 2] + ), + default_tensor=make_tensor( + "default_tensor", onnx.TensorProto.INT16, (1,), [42] + ), +) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_ai_onnx_ml_label_encoder_tensor_value_only_mapping", +) +``` + +
+ + +### **ai.onnx.ml.LinearClassifier** + + Linear classifier + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
classlabels_ints : list of ints
+
Class labels when using integer labels. One and only one 'classlabels' attribute must be defined.
+
classlabels_strings : list of strings
+
Class labels when using string labels. One and only one 'classlabels' attribute must be defined.
+
coefficients : list of floats (required)
+
A collection of weights of the model(s).
+
intercepts : list of floats
+
A collection of intercepts.
+
multi_class : int (default is 0)
+
Indicates whether to do OvR or multinomial (0=OvR is the default).
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the scores vector.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT'
+
+ +#### Inputs + +
+
X : T1
+
Data to be classified.
+
+ +#### Outputs + +
+
Y : T2
+
Classification outputs (one class per example).
+
Z : tensor(float)
+
Classification scores ([N,E] - one score for each class and example
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type, and of shape [N,C] or [C]. In the latter case, it will be treated as [1,C]
+
T2 : tensor(string), tensor(int64)
+
The output will be a tensor of strings or integers.
+
+ + +### **ai.onnx.ml.LinearRegressor** + + Generalized linear regression evaluation.
+ If targets is set to 1 (default) then univariate regression is performed.
+ If targets is set to M then M sets of coefficients must be passed in as a sequence + and M results will be output for each input n in N.
+ The coefficients array is of length n, and the coefficients for each target are contiguous. + Intercepts are optional but if provided must match the number of targets. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
coefficients : list of floats
+
Weights of the model(s).
+
intercepts : list of floats
+
Weights of the intercepts, if used.
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the regression output vector.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT'
+
targets : int (default is 1)
+
The total number of regression targets, 1 if not defined.
+
+ +#### Inputs + +
+
X : T
+
Data to be regressed.
+
+ +#### Outputs + +
+
Y : tensor(float)
+
Regression outputs (one per target, per example).
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type.
+
+ + +### **ai.onnx.ml.Normalizer** + + Normalize the input. There are three normalization modes, which have the corresponding formulas, + defined using element-wise infix operators '/' and '^' and tensor-wide functions 'max' and 'sum':
+
+ Max: Y = X / max(X)
+ L1: Y = X / sum(X)
+ L2: Y = sqrt(X^2 / sum(X^2)}
+ In all modes, if the divisor is zero, Y == X. +
+ For batches, that is, [N,C] tensors, normalization is done along the C axis. In other words, each row + of the batch is normalized independently. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
norm : string (default is MAX)
+
One of 'MAX,' 'L1,' 'L2'
+
+ +#### Inputs + +
+
X : T
+
Data to be encoded, a tensor of shape [N,C] or [C]
+
+ +#### Outputs + +
+
Y : tensor(float)
+
Encoded output data
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type.
+
+ + +### **ai.onnx.ml.OneHotEncoder** + + Replace each input element with an array of ones and zeros, where a single + one is placed at the index of the category that was passed in. The total category count + will determine the size of the extra dimension of the output array Y.
+ For example, if we pass a tensor with a single value of 4, and a category count of 8, + the output will be a tensor with ``[0,0,0,0,1,0,0,0]``.
+ This operator assumes every input feature is from the same set of categories.
+ If the input is a tensor of float, int32, or double, the data will be cast + to integers and the cats_int64s category list will be used for the lookups. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
cats_int64s : list of ints
+
List of categories, ints.
One and only one of the 'cats_*' attributes must be defined.
+
cats_strings : list of strings
+
List of categories, strings.
One and only one of the 'cats_*' attributes must be defined.
+
zeros : int (default is 1)
+
If true and category is not present, will return all zeros; if false and a category if not found, the operator will fail.
+
+ +#### Inputs + +
+
X : T
+
Data to be encoded.
+
+ +#### Outputs + +
+
Y : tensor(float)
+
Encoded output data, having one more dimension than X.
+
+ +#### Type Constraints + +
+
T : tensor(string), tensor(int64), tensor(int32), tensor(float), tensor(double)
+
The input must be a tensor of a numeric type.
+
+ + +### **ai.onnx.ml.SVMClassifier** + + Support Vector Machine classifier + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
classlabels_ints : list of ints
+
Class labels if using integer labels.
One and only one of the 'classlabels_*' attributes must be defined.
+
classlabels_strings : list of strings
+
Class labels if using string labels.
One and only one of the 'classlabels_*' attributes must be defined.
+
coefficients : list of floats
+
+
kernel_params : list of floats
+
List of 3 elements containing gamma, coef0, and degree, in that order. Zero if unused for the kernel.
+
kernel_type : string (default is LINEAR)
+
The kernel type, one of 'LINEAR,' 'POLY,' 'RBF,' 'SIGMOID'.
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the score.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT'
+
prob_a : list of floats
+
First set of probability coefficients.
+
prob_b : list of floats
+
Second set of probability coefficients. This array must be same size as prob_a.
If these are provided then output Z are probability estimates, otherwise they are raw scores.
+
rho : list of floats
+
+
support_vectors : list of floats
+
+
vectors_per_class : list of ints
+
+
+ +#### Inputs + +
+
X : T1
+
Data to be classified.
+
+ +#### Outputs + +
+
Y : T2
+
Classification outputs (one class per example).
+
Z : tensor(float)
+
Class scores (one per class per example), if prob_a and prob_b are provided they are probabilities for each class, otherwise they are raw scores.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type, either [C] or [N,C].
+
T2 : tensor(string), tensor(int64)
+
The output type will be a tensor of strings or integers, depending on which of the classlabels_* attributes is used. Its size will match the batch size of the input.
+
+ + +### **ai.onnx.ml.SVMRegressor** + + Support Vector Machine regression prediction and one-class SVM anomaly detection. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
coefficients : list of floats
+
Support vector coefficients.
+
kernel_params : list of floats
+
List of 3 elements containing gamma, coef0, and degree, in that order. Zero if unused for the kernel.
+
kernel_type : string (default is LINEAR)
+
The kernel type, one of 'LINEAR,' 'POLY,' 'RBF,' 'SIGMOID'.
+
n_supports : int (default is 0)
+
The number of support vectors.
+
one_class : int (default is 0)
+
Flag indicating whether the regression is a one-class SVM or not.
+
post_transform : string (default is NONE)
+
Indicates the transform to apply to the score.
One of 'NONE,' 'SOFTMAX,' 'LOGISTIC,' 'SOFTMAX_ZERO,' or 'PROBIT.'
+
rho : list of floats
+
+
support_vectors : list of floats
+
Chosen support vectors
+
+ +#### Inputs + +
+
X : T
+
Data to be regressed.
+
+ +#### Outputs + +
+
Y : tensor(float)
+
Regression outputs (one score per target per example).
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input type must be a tensor of a numeric type, either [C] or [N,C].
+
+ + +### **ai.onnx.ml.Scaler** + + Rescale input data, for example to standardize features by removing the mean and scaling to unit variance. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
offset : list of floats
+
First, offset by this.
Can be length of features in an [N,F] tensor or length 1, in which case it applies to all features, regardless of dimension count.
+
scale : list of floats
+
Second, multiply by this.
Can be length of features in an [N,F] tensor or length 1, in which case it applies to all features, regardless of dimension count.
Must be same length as 'offset'
+
+ +#### Inputs + +
+
X : T
+
Data to be scaled.
+
+ +#### Outputs + +
+
Y : tensor(float)
+
Scaled output data.
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int64), tensor(int32)
+
The input must be a tensor of a numeric type.
+
+ + +### **ai.onnx.ml.TreeEnsemble** + + Tree Ensemble operator. Returns the regressed values for each input in a batch. + Inputs have dimensions `[N, F]` where `N` is the input batch size and `F` is the number of input features. + Outputs have dimensions `[N, num_targets]` where `N` is the batch size and `num_targets` is the number of targets, which is a configurable attribute. + + The encoding of this attribute is split along interior nodes and the leaves of the trees. Notably, attributes with the prefix `nodes_*` are associated with interior nodes, and attributes with the prefix `leaf_*` are associated with leaves. + The attributes `nodes_*` must all have the same length and encode a sequence of tuples, as defined by taking all the `nodes_*` fields at a given position. + + All fields prefixed with `leaf_*` represent tree leaves, and similarly define tuples of leaves and must have identical length. + + This operator can be used to implement both the previous `TreeEnsembleRegressor` and `TreeEnsembleClassifier` nodes. + The `TreeEnsembleRegressor` node maps directly to this node and requires changing how the nodes are represented. + The `TreeEnsembleClassifier` node can be implemented by adding a `ArgMax` node after this node to determine the top class. + To encode class labels, a `LabelEncoder` or `GatherND` operator may be used. + +#### Version + +This version of the operator has been available since version 5 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
aggregate_function : int (default is 1)
+
Defines how to aggregate leaf values within a target.
One of 'AVERAGE' (0) 'SUM' (1) 'MIN' (2) 'MAX (3) defaults to 'SUM' (1)
+
leaf_targetids : list of ints (required)
+
The index of the target that this leaf contributes to (this must be in range `[0, n_targets)`).
+
leaf_weights : tensor (required)
+
The weight for each leaf.
+
membership_values : tensor
+
Members to test membership of for each set membership node. List all of the members to test again in the order that the 'BRANCH_MEMBER' mode appears in `node_modes`, delimited by `NaN`s. Will have the same number of sets of values as nodes with mode 'BRANCH_MEMBER'. This may be omitted if the node doesn't contain any 'BRANCH_MEMBER' nodes.
+
n_targets : int
+
The total number of targets.
+
nodes_falseleafs : list of ints (required)
+
1 if false branch is leaf for each node and 0 if an interior node. To represent a tree that is a leaf (only has one node), one can do so by having a single `nodes_*` entry with true and false branches referencing the same `leaf_*` entry
+
nodes_falsenodeids : list of ints (required)
+
If `nodes_falseleafs` is false at an entry, this represents the position of the false branch node. This position can be used to index into a `nodes_*` entry. If `nodes_falseleafs` is false, it is an index into the leaf_* attributes.
+
nodes_featureids : list of ints (required)
+
Feature id for each node.
+
nodes_hitrates : tensor
+
Popularity of each node, used for performance and may be omitted.
+
nodes_missing_value_tracks_true : list of ints
+
For each node, define whether to follow the true branch (if attribute value is 1) or false branch (if attribute value is 0) in the presence of a NaN input feature. This attribute may be left undefined and the default value is false (0) for all nodes.
+
nodes_modes : tensor (required)
+
The comparison operation performed by the node. This is encoded as an enumeration of 0 ('BRANCH_LEQ'), 1 ('BRANCH_LT'), 2 ('BRANCH_GTE'), 3 ('BRANCH_GT'), 4 ('BRANCH_EQ'), 5 ('BRANCH_NEQ'), and 6 ('BRANCH_MEMBER'). Note this is a tensor of type uint8.
+
nodes_splits : tensor (required)
+
Thresholds to do the splitting on for each node with mode that is not 'BRANCH_MEMBER'.
+
nodes_trueleafs : list of ints (required)
+
1 if true branch is leaf for each node and 0 an interior node. To represent a tree that is a leaf (only has one node), one can do so by having a single `nodes_*` entry with true and false branches referencing the same `leaf_*` entry
+
nodes_truenodeids : list of ints (required)
+
If `nodes_trueleafs` is false at an entry, this represents the position of the true branch node. This position can be used to index into a `nodes_*` entry. If `nodes_trueleafs` is false, it is an index into the leaf_* attributes.
+
post_transform : int (default is 0)
+
Indicates the transform to apply to the score.
One of 'NONE' (0), 'SOFTMAX' (1), 'LOGISTIC' (2), 'SOFTMAX_ZERO' (3) or 'PROBIT' (4), defaults to 'NONE' (0)
+
tree_roots : list of ints (required)
+
Index into `nodes_*` for the root of each tree. The tree structure is derived from the branching of each node.
+
+ +#### Inputs + +
+
X : T
+
Input of shape [Batch Size, Number of Features]
+
+ +#### Outputs + +
+
Y : T
+
Output of shape [Batch Size, Number of targets]
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(float16)
+
The input type must be a tensor of a numeric type.
+
+ + +#### Examples + +
+tree_ensemble_set_membership + +```python +node = onnx.helper.make_node( + "TreeEnsemble", + ["X"], + ["Y"], + domain="ai.onnx.ml", + n_targets=4, + aggregate_function=1, + membership_values=make_tensor( + "membership_values", + onnx.TensorProto.FLOAT, + (8,), + [1.2, 3.7, 8, 9, np.nan, 12, 7, np.nan], + ), + nodes_missing_value_tracks_true=None, + nodes_hitrates=None, + post_transform=0, + tree_roots=[0], + nodes_modes=make_tensor( + "nodes_modes", + onnx.TensorProto.UINT8, + (3,), + np.array([0, 6, 6], dtype=np.uint8), + ), + nodes_featureids=[0, 0, 0], + nodes_splits=make_tensor( + "nodes_splits", + onnx.TensorProto.FLOAT, + (3,), + np.array([11, 232344.0, np.nan], dtype=np.float32), + ), + nodes_trueleafs=[0, 1, 1], + nodes_truenodeids=[1, 0, 1], + nodes_falseleafs=[1, 0, 1], + nodes_falsenodeids=[2, 2, 3], + leaf_targetids=[0, 1, 2, 3], + leaf_weights=make_tensor( + "leaf_weights", onnx.TensorProto.FLOAT, (4,), [1, 10, 1000, 100] + ), +) + +x = np.array([1.2, 3.4, -0.12, np.nan, 12, 7], np.float32).reshape(-1, 1) +expected = np.array( + [ + [1, 0, 0, 0], + [0, 0, 0, 100], + [0, 0, 0, 100], + [0, 0, 1000, 0], + [0, 0, 1000, 0], + [0, 10, 0, 0], + ], + dtype=np.float32, +) +expect( + node, + inputs=[x], + outputs=[expected], + name="test_ai_onnx_ml_tree_ensemble_set_membership", +) +``` + +
+ + +
+tree_ensemble_single_tree + +```python +node = onnx.helper.make_node( + "TreeEnsemble", + ["X"], + ["Y"], + domain="ai.onnx.ml", + n_targets=2, + membership_values=None, + nodes_missing_value_tracks_true=None, + nodes_hitrates=None, + aggregate_function=1, + post_transform=0, + tree_roots=[0], + nodes_modes=make_tensor( + "nodes_modes", + onnx.TensorProto.UINT8, + (3,), + np.array([0, 0, 0], dtype=np.uint8), + ), + nodes_featureids=[0, 0, 0], + nodes_splits=make_tensor( + "nodes_splits", + onnx.TensorProto.DOUBLE, + (3,), + np.array([3.14, 1.2, 4.2], dtype=np.float64), + ), + nodes_truenodeids=[1, 0, 1], + nodes_trueleafs=[0, 1, 1], + nodes_falsenodeids=[2, 2, 3], + nodes_falseleafs=[0, 1, 1], + leaf_targetids=[0, 1, 0, 1], + leaf_weights=make_tensor( + "leaf_weights", + onnx.TensorProto.DOUBLE, + (4,), + np.array([5.23, 12.12, -12.23, 7.21], dtype=np.float64), + ), +) + +x = np.array([1.2, 3.4, -0.12, 1.66, 4.14, 1.77], np.float64).reshape(3, 2) +y = np.array([[5.23, 0], [5.23, 0], [0, 12.12]], dtype=np.float64) +expect( + node, + inputs=[x], + outputs=[y], + name="test_ai_onnx_ml_tree_ensemble_single_tree", +) +``` + +
+ + +### **ai.onnx.ml.TreeEnsembleClassifier** (deprecated) + + This operator is DEPRECATED. Please use TreeEnsemble with provides similar functionality. + In order to determine the top class, the ArgMax node can be applied to the output of TreeEnsemble. + To encode class labels, use a LabelEncoder operator. + Tree Ensemble classifier. Returns the top class for each of N inputs.
+ The attributes named 'nodes_X' form a sequence of tuples, associated by + index into the sequences, which must all be of equal length. These tuples + define the nodes.
+ Similarly, all fields prefixed with 'class_' are tuples of votes at the leaves. + A leaf may have multiple votes, where each vote is weighted by + the associated class_weights index.
+ One and only one of classlabels_strings or classlabels_int64s + will be defined. The class_ids are indices into this list. + All fields ending with _as_tensor can be used instead of the + same parameter without the suffix if the element type is double and not float. + +#### Version + +This version of the operator has been deprecated since version 5 of the 'ai.onnx.ml' operator set. + +Other versions of this operator: 1, 3 + + +### **ai.onnx.ml.TreeEnsembleRegressor** (deprecated) + + This operator is DEPRECATED. Please use TreeEnsemble instead which provides the same + functionality.
+ Tree Ensemble regressor. Returns the regressed values for each input in N.
+ All args with nodes_ are fields of a tuple of tree nodes, and + it is assumed they are the same length, and an index i will decode the + tuple across these inputs. Each node id can appear only once + for each tree id.
+ All fields prefixed with target_ are tuples of votes at the leaves.
+ A leaf may have multiple votes, where each vote is weighted by + the associated target_weights index.
+ All fields ending with _as_tensor can be used instead of the + same parameter without the suffix if the element type is double and not float. + All trees must have their node ids start at 0 and increment by 1.
+ Mode enum is BRANCH_LEQ, BRANCH_LT, BRANCH_GTE, BRANCH_GT, BRANCH_EQ, BRANCH_NEQ, LEAF + +#### Version + +This version of the operator has been deprecated since version 5 of the 'ai.onnx.ml' operator set. + +Other versions of this operator: 1, 3 + + +### **ai.onnx.ml.ZipMap** + + Creates a map from the input and the attributes.
+ The values are provided by the input tensor, while the keys are specified by the attributes. + Must provide keys in either classlabels_strings or classlabels_int64s (but not both).
+ The columns of the tensor correspond one-by-one to the keys specified by the attributes. There must be as many columns as keys.
+ +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.ml' operator set. + +#### Attributes + +
+
classlabels_int64s : list of ints
+
The keys when using int keys.
One and only one of the 'classlabels_*' attributes must be defined.
+
classlabels_strings : list of strings
+
The keys when using string keys.
One and only one of the 'classlabels_*' attributes must be defined.
+
+ +#### Inputs + +
+
X : tensor(float)
+
The input values
+
+ +#### Outputs + +
+
Z : T
+
The output map
+
+ +#### Type Constraints + +
+
T : seq(map(string, float)), seq(map(int64, float))
+
The output will be a sequence of string or integer maps to float.
+
+ + diff --git a/docs/Operators.md b/docs/Operators.md new file mode 100644 index 0000000..0ee4ec6 --- /dev/null +++ b/docs/Operators.md @@ -0,0 +1,44256 @@ + +## Operator Schemas +*This file is automatically generated from the + [def files](/onnx/defs) via [this script](/onnx/defs/gen_doc.py). + Do not modify directly and instead edit operator definitions.* + +For an operator input/output's differentiability, it can be differentiable, + non-differentiable, or undefined. If a variable's differentiability + is not specified, that variable has undefined differentiability. + +### ai.onnx (default) +|**Operator**|**Since version**|| +|-|-|-| +|Abs|13, 6, 1| +|Acos|22, 7| +|Acosh|22, 9| +|Add|14, 13, 7, 6, 1| +|And|7, 1| +|ArgMax|13, 12, 11, 1| +|ArgMin|13, 12, 11, 1| +|Asin|22, 7| +|Asinh|22, 9| +|Atan|22, 7| +|Atanh|22, 9| +|AveragePool|22, 19, 11, 10, 7, 1| +|BatchNormalization|15, 14, 9, 7, 6, 1| +|BitCast|26| +|BitShift|11| +|BitwiseAnd|18| +|BitwiseNot|18| +|BitwiseOr|18| +|BitwiseXor|18| +|Cast|25, 24, 23, 21, 19, 13, 9, 6, 1| +|Ceil|13, 6, 1| +|Col2Im|18| +|Compress|11, 9| +|Concat|13, 11, 4, 1| +|ConcatFromSequence|11| +|Constant|25, 24, 23, 21, 19, 13, 12, 11, 9, 1| +|ConstantOfShape|25, 24, 23, 21, 20, 9| +|Conv|22, 11, 1| +|ConvInteger|10| +|ConvTranspose|22, 11, 1| +|Cos|22, 7| +|Cosh|22, 9| +|CumProd|26| +|CumSum|14, 11| +|DFT|20, 17| +|DeformConv|22, 19| +|DepthToSpace|13, 11, 1| +|DequantizeLinear|25, 24, 23, 21, 19, 13, 10| +|Det|22, 11| +|Div|14, 13, 7, 6, 1| +|Dropout|22, 13, 12, 10, 7, 6, 1| +|Einsum|12| +|Equal|19, 13, 11, 7, 1| +|Erf|13, 9| +|Exp|13, 6, 1| +|Expand|13, 8| +|EyeLike|22, 9| +|Flatten|25, 24, 23, 21, 13, 11, 9, 1| +|Floor|13, 6, 1| +|GRU|22, 14, 7, 3, 1| +|Gather|13, 11, 1| +|GatherElements|13, 11| +|GatherND|13, 12, 11| +|Gemm|13, 11, 9, 7, 6, 1| +|GlobalAveragePool|22, 1| +|GlobalLpPool|22, 2, 1| +|GlobalMaxPool|22, 1| +|Greater|13, 9, 7, 1| +|GridSample|22, 20, 16| +|Hardmax|13, 11, 1| +|Identity|25, 24, 23, 21, 19, 16, 14, 13, 1| +|If|25, 24, 23, 21, 19, 16, 13, 11, 1| +|ImageDecoder|20| +|InstanceNormalization|22, 6, 1| +|IsInf|20, 10| +|IsNaN|20, 13, 9| +|LRN|13, 1| +|LSTM|22, 14, 7, 1| +|Less|13, 9, 7, 1| +|Log|13, 6, 1| +|Loop|25, 24, 23, 21, 19, 16, 13, 11, 1| +|LpNormalization|22, 1| +|LpPool|22, 18, 11, 2, 1| +|MatMul|13, 9, 1| +|MatMulInteger|10| +|Max|13, 12, 8, 6, 1| +|MaxPool|22, 12, 11, 10, 8, 1| +|MaxRoiPool|22, 1| +|MaxUnpool|22, 11, 9| +|Mean|13, 8, 6, 1| +|MelWeightMatrix|17| +|Min|13, 12, 8, 6, 1| +|Mod|13, 10| +|Mul|14, 13, 7, 6, 1| +|Multinomial|22, 7| +|Neg|13, 6, 1| +|NonMaxSuppression|11, 10| +|NonZero|13, 9| +|Not|1| +|OneHot|11, 9| +|Optional|15| +|OptionalGetElement|18, 15| +|OptionalHasElement|18, 15| +|Or|7, 1| +|Pad|25, 24, 23, 21, 19, 18, 13, 11, 2, 1| +|Pow|15, 13, 12, 7, 1| +|QLinearConv|10| +|QLinearMatMul|21, 10| +|QuantizeLinear|25, 24, 23, 21, 19, 13, 10| +|RNN|22, 14, 7, 1| +|RandomNormal|22, 1| +|RandomNormalLike|22, 1| +|RandomUniform|22, 1| +|RandomUniformLike|22, 1| +|Reciprocal|13, 6, 1| +|ReduceMax|20, 18, 13, 12, 11, 1| +|ReduceMean|18, 13, 11, 1| +|ReduceMin|20, 18, 13, 12, 11, 1| +|ReduceProd|18, 13, 11, 1| +|ReduceSum|13, 11, 1| +|RegexFullMatch|20| +|Reshape|25, 24, 23, 21, 19, 14, 13, 5, 1| +|Resize|19, 18, 13, 11, 10| +|ReverseSequence|10| +|RoiAlign|22, 16, 10| +|Round|22, 11| +|STFT|17| +|Scan|25, 24, 23, 21, 19, 16, 11, 9, 8| +|Scatter (deprecated)|11, 9| +|ScatterElements|18, 16, 13, 11| +|ScatterND|18, 16, 13, 11| +|SequenceAt|11| +|SequenceConstruct|11| +|SequenceEmpty|11| +|SequenceErase|11| +|SequenceInsert|11| +|SequenceLength|11| +|Shape|25, 24, 23, 21, 19, 15, 13, 1| +|Sigmoid|13, 6, 1| +|Sign|13, 9| +|Sin|22, 7| +|Sinh|22, 9| +|Size|25, 24, 23, 21, 19, 13, 1| +|Slice|13, 11, 10, 1| +|SpaceToDepth|13, 1| +|Split|18, 13, 11, 2, 1| +|SplitToSequence|24, 11| +|Sqrt|13, 6, 1| +|Squeeze|25, 24, 23, 21, 13, 11, 1| +|StringConcat|20| +|StringNormalizer|10| +|StringSplit|20| +|Sub|14, 13, 7, 6, 1| +|Sum|13, 8, 6, 1| +|Tan|22, 7| +|Tanh|13, 6, 1| +|TensorScatter|24| +|TfIdfVectorizer|9| +|Tile|13, 6, 1| +|TopK|24, 11, 10, 1| +|Transpose|25, 24, 23, 21, 13, 1| +|Trilu|14| +|Unique|11| +|Unsqueeze|25, 24, 23, 21, 13, 11, 1| +|Upsample (deprecated)|10, 9, 7| +|Where|16, 9| +|Xor|7, 1| +|**Function**|**Since version**|**Function version**| +|AffineGrid|20|20| +|Attention|24, 23|24| +|Bernoulli|22, 15|22| +|BlackmanWindow|17|17| +|CastLike|25, 24, 23, 21, 19, 15|25| +|CausalConvWithState|27|27| +|Celu|28, 12|28| +|CenterCropPad|18|18| +|Clip|13, 12, 11, 6, 1|13| +|DynamicQuantizeLinear|11|11| +|Elu|22, 6, 1|18| +|Gelu|20|20| +|GreaterOrEqual|16, 12|16| +|GroupNormalization|21, 18|21| +|HammingWindow|17|17| +|HannWindow|17|17| +|HardSigmoid|22, 6, 1|18| +|HardSwish|22, 14|22| +|LayerNormalization|17|17, 18| +|LeakyRelu|16, 6, 1|16| +|LessOrEqual|16, 12|16| +|LinearAttention|27|27| +|LogSoftmax|13, 11, 1|13, 18| +|MeanVarianceNormalization|13, 9|13, 18| +|Mish|22, 18|22| +|NegativeLogLikelihoodLoss|22, 13, 12|22| +|PRelu|16, 9, 7, 6, 1|16| +|RMSNormalization|23|23| +|Range|27, 11|27| +|ReduceL1|18, 13, 11, 1|18| +|ReduceL2|18, 13, 11, 1|18| +|ReduceLogSum|18, 13, 11, 1|18| +|ReduceLogSumExp|18, 13, 11, 1|18| +|ReduceSumSquare|18, 13, 11, 1|18| +|Relu|14, 13, 6, 1|18| +|RotaryEmbedding|23|23| +|Selu|22, 6, 1|18| +|SequenceMap|17|17| +|Shrink|9|18| +|Softmax|13, 11, 1|13, 18| +|SoftmaxCrossEntropyLoss|13, 12|13| +|Softplus|22, 1|18| +|Softsign|22, 1|18| +|Swish|24|24| +|ThresholdedRelu|22, 10|18| + +### ai.onnx.preview +|**Operator**|**Since version**|| +|-|-|-| +|**Function**|**Since version**|**Function version**| +|experimental ai.onnx.preview.FlexAttention|1|1| + +### ai.onnx.preview.training +|**Operator**|**Since version**|| +|-|-|-| +|ai.onnx.preview.training.Adagrad|1| +|ai.onnx.preview.training.Adam|1| +|ai.onnx.preview.training.Gradient|1| +|ai.onnx.preview.training.Momentum|1| + + +## ai.onnx (default) +### **Abs** + + Absolute takes one input data (Tensor) and produces one output data + (Tensor) where absolute value, y = abs(x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ + +#### Examples + +
+abs + +```python +node = onnx.helper.make_node( + "Abs", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.abs(x) + +expect(node, inputs=[x], outputs=[y], name="test_abs") +``` + +
+ + +#### Sample Implementation + +
+Abs + +```python +# SPDX-License-Identifier: Apache-2.0 +from __future__ import annotations + +import numpy as np + + +def abs(input: np.ndarray) -> np.ndarray: # noqa: A001 + return np.abs(input) # type: ignore[no-any-return] + +``` + +
+ + +### **Acos** + + Calculates the arccosine (inverse of cosine) of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 7 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The arccosine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+acos + +```python +node = onnx.helper.make_node( + "Acos", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-0.5, 0, 0.5]).astype(np.float32) +y = np.arccos(x) +expect(node, inputs=[x], outputs=[y], name="test_acos_example") + +x = np.random.rand(3, 4, 5).astype(np.float32) +y = np.arccos(x) +expect(node, inputs=[x], outputs=[y], name="test_acos") +``` + +
+ + +### **Acosh** + + Calculates the hyperbolic arccosine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic arccosine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+acosh + +```python +node = onnx.helper.make_node( + "Acosh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([10, np.e, 1]).astype(np.float32) +y = np.arccosh(x) # expected output [2.99322295, 1.65745449, 0.] +expect(node, inputs=[x], outputs=[y], name="test_acosh_example") + +x = np.random.uniform(1.0, 10.0, (3, 4, 5)).astype(np.float32) +y = np.arccosh(x) +expect(node, inputs=[x], outputs=[y], name="test_acosh") +``` + +
+ + +### **Add** + + Performs element-wise binary addition (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + + (Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 7, 13 + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ + +#### Examples + +
+add + +```python +node = onnx.helper.make_node( + "Add", + inputs=["x", "y"], + outputs=["sum"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.int8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int8) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_int8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.int16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int16) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_uint64") +``` + +
+ + +
+add_broadcast + +```python +node = onnx.helper.make_node( + "Add", + inputs=["x", "y"], + outputs=["sum"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_bcast") +``` + +
+ + +### **AffineGrid** + + Generates a 2D or 3D flow field (sampling grid), given a batch of affine matrices theta + (https://pytorch.org/docs/stable/generated/torch.nn.functional.affine_grid.html). + An affine matrix `theta` is applied to a position tensor represented in its homogeneous expression. Here is an example in 3D: + ``` + [r00, r01, r02, t0] [x] [x'] + [r10, r11, r12, t1] * [y] = [y'] + [r20, r21, r22, t2] [z] [z'] + [0, 0, 0, 1 ] [1] [1 ] + ``` + where `(x, y, z)` is the position in the original space, `(x', y', z')` is the position in the output space. + The last row is always `[0, 0, 0, 1]` and is not stored in the affine matrix. Therefore we have `theta` of shape `(N, 2, 3)` for 2D or `(N, 3, 4)` for 3D. + + Input `size` is used to define grid of positions evenly spaced in the original 2D or 3D space, with dimensions ranging from `-1` to `1`. + The output `grid` contains positions in the output space. + + When `align_corners=1`, consider `-1` and `1` to refer to the centers of the corner pixels (mark `v` in illustration). + ``` + v v v v + |-------------------|------------------| + -1 0 1 + ``` + When `align_corners=0`, consider `-1` and `1` to refer to the outer edge of the corner pixels. + ``` + v v v v + |------------------|-------------------| + -1 0 1 + ``` + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
align_corners : int (default is 0)
+
if align_corners=1, consider -1 and 1 to refer to the centers of the corner pixels. if align_corners=0, consider -1 and 1 to refer to the outer edge the corner pixels.
+
+ +#### Inputs + +
+
theta (non-differentiable) : T1
+
input batch of affine matrices with shape (N, 2, 3) for 2D or (N, 3, 4) for 3D
+
size (non-differentiable) : T2
+
the target output image size (N, C, H, W) for 2D or (N, C, D, H, W) for 3D
+
+ +#### Outputs + +
+
grid (differentiable) : T1
+
output tensor of shape (N, H, W, 2) of 2D sample coordinates or (N, D, H, W, 3) of 3D sample coordinates.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain grid types to float tensors.
+
T2 : tensor(int64)
+
Constrain size's type to int64 tensors.
+
+ + +#### Examples + +
+2d_no_reference_evaluator + +```python +theta_2d = create_theta_2d() +N, C, H, W = len(theta_2d), 3, 5, 6 +data_size = (H, W) +for align_corners in (0, 1): + node = onnx.helper.make_node( + "AffineGrid", + inputs=["theta", "size"], + outputs=["grid"], + align_corners=align_corners, + ) + + original_grid = construct_original_grid(data_size, align_corners) + grid = apply_affine_transform(theta_2d, original_grid) + + test_name = "test_affine_grid_2d" + if align_corners == 1: + test_name += "_align_corners" + expect( + node, + inputs=[theta_2d, np.array([N, C, H, W], dtype=np.int64)], + outputs=[grid], + name=test_name, + ) +``` + +
+ + +
+3d_no_reference_evaluator + +```python +theta_3d = create_theta_3d() +N, C, D, H, W = len(theta_3d), 3, 4, 5, 6 +data_size = (D, H, W) +for align_corners in (0, 1): + node = onnx.helper.make_node( + "AffineGrid", + inputs=["theta", "size"], + outputs=["grid"], + align_corners=align_corners, + ) + + original_grid = construct_original_grid(data_size, align_corners) + grid = apply_affine_transform(theta_3d, original_grid) + + test_name = "test_affine_grid_3d" + if align_corners == 1: + test_name += "_align_corners" + expect( + node, + inputs=[theta_3d, np.array([N, C, D, H, W], dtype=np.int64)], + outputs=[grid], + name=test_name, + ) +``` + +
+ + +### **And** + + Returns the tensor resulted from performing the `and` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input to boolean tensor.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ + +#### Examples + +
+and + +```python +node = onnx.helper.make_node( + "And", + inputs=["x", "y"], + outputs=["and"], +) + +# 2d +x = (np.random.randn(3, 4) > 0).astype(bool) +y = (np.random.randn(3, 4) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and2d") + +# 3d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(3, 4, 5) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and3d") + +# 4d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and4d") +``` + +
+ + +
+and_broadcast + +```python +node = onnx.helper.make_node( + "And", + inputs=["x", "y"], + outputs=["and"], +) + +# 3d vs 1d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(5) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and_bcast3v1d") + +# 3d vs 2d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(4, 5) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and_bcast3v2d") + +# 4d vs 2d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(5, 6) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and_bcast4v2d") + +# 4d vs 3d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(4, 5, 6) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and_bcast4v3d") + +# 4d vs 4d +x = (np.random.randn(1, 4, 1, 6) > 0).astype(bool) +y = (np.random.randn(3, 1, 5, 6) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and_bcast4v4d") +``` + +
+ + +### **ArgMax** + + Computes the indices of the max elements of the input tensor's element along the + provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. + If keepdims equals 0, then the resulting tensor has the reduced dimension pruned. + If select_last_index is True (default False), the index of the last occurrence of the max + is selected if the max appears more than once in the input. Otherwise the index of the + first occurrence is selected. + The type of the output tensor is integer. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 12 + +#### Attributes + +
+
axis : int (default is 0)
+
The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
select_last_index : int (default is 0)
+
Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index).
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (non-differentiable) : tensor(int64)
+
Reduced output tensor with integer data type.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ + +#### Examples + +
+default_axes_keepdims + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", inputs=["data"], outputs=["result"], keepdims=keepdims +) + +# result: [[1, 1]] +result = argmax_use_numpy(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_default_axis_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [1, 3, 4] +result = argmax_use_numpy(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_default_axis_random", +) +``` + +
+ + +
+default_axes_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", + inputs=["data"], + outputs=["result"], + keepdims=keepdims, + select_last_index=True, +) + +# result: [[1, 1]] +result = argmax_use_numpy_select_last_index(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_default_axis_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [1, 3, 4] +result = argmax_use_numpy_select_last_index(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_default_axis_random_select_last_index", +) +``` + +
+ + +
+keepdims + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# result: [[0], [1]] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmax_keepdims_example" +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 1, 4] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmax_keepdims_random" +) +``` + +
+ + +
+keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [[1], [1]] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 1, 4] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_keepdims_random_select_last_index", +) +``` + +
+ + +
+negative_axis_keepdims + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = -1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# result: [[0], [1]] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_negative_axis_keepdims_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 3, 1] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_negative_axis_keepdims_random", +) +``` + +
+ + +
+negative_axis_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = -1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [[1], [1]] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_negative_axis_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 3, 1] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_negative_axis_keepdims_random_select_last_index", +) +``` + +
+ + +
+no_keepdims + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 0 +node = onnx.helper.make_node( + "ArgMax", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# result: [0, 1] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_no_keepdims_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 4] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmax_no_keepdims_random" +) +``` + +
+ + +
+no_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 0 +node = onnx.helper.make_node( + "ArgMax", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [1, 1] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_no_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 4] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_no_keepdims_random_select_last_index", +) +``` + +
+ + +### **ArgMin** + + Computes the indices of the min elements of the input tensor's element along the + provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. + If keepdims equals 0, then the resulting tensor has the reduced dimension pruned. + If select_last_index is True (default False), the index of the last occurrence of the min + is selected if the min appears more than once in the input. Otherwise the index of the + first occurrence is selected. + The type of the output tensor is integer. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 12 + +#### Attributes + +
+
axis : int (default is 0)
+
The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data).
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
select_last_index : int (default is 0)
+
Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index).
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
reduced (non-differentiable) : tensor(int64)
+
Reduced output tensor with integer data type.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ + +#### Examples + +
+default_axes_keepdims + +```python +data = np.array([[2, 1], [3, 10]], dtype=np.float32) +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", inputs=["data"], outputs=["result"], keepdims=keepdims +) + +# The content of result is : [[0], [0]] +result = argmin_use_numpy(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_default_axis_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [1, 3, 4] +result = argmin_use_numpy(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_default_axis_random", +) +``` + +
+ + +
+default_axes_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", + inputs=["data"], + outputs=["result"], + keepdims=keepdims, + select_last_index=True, +) + +# result: [[0, 0]] +result = argmin_use_numpy_select_last_index(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_default_axis_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [1, 3, 4] +result = argmin_use_numpy_select_last_index(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_default_axis_random_select_last_index", +) +``` + +
+ + +
+keepdims + +```python +data = np.array([[2, 1], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# The content of result is : [[1], [0]] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmin_keepdims_example" +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 1, 4] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmin_keepdims_random" +) +``` + +
+ + +
+keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [[1], [0]] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 1, 4] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_keepdims_random_select_last_index", +) +``` + +
+ + +
+negative_axis_keepdims + +```python +data = np.array([[2, 1], [3, 10]], dtype=np.float32) +axis = -1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# The content of result is : [[1], [0]] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_negative_axis_keepdims_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 3, 1] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_negative_axis_keepdims_random", +) +``` + +
+ + +
+negative_axis_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = -1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [[1], [0]] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_negative_axis_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 3, 1] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_negative_axis_keepdims_random_select_last_index", +) +``` + +
+ + +
+no_keepdims + +```python +data = np.array([[2, 1], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 0 +node = onnx.helper.make_node( + "ArgMin", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# The content of result is : [[1, 0]] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_no_keepdims_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 4] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmin_no_keepdims_random" +) +``` + +
+ + +
+no_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 0 +node = onnx.helper.make_node( + "ArgMin", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [[1, 0]] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_no_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 4] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_no_keepdims_random_select_last_index", +) +``` + +
+ + +### **Asin** + + Calculates the arcsine (inverse of sine) of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 7 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The arcsine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+asin + +```python +node = onnx.helper.make_node( + "Asin", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-0.5, 0, 0.5]).astype(np.float32) +y = np.arcsin(x) +expect(node, inputs=[x], outputs=[y], name="test_asin_example") + +x = np.random.rand(3, 4, 5).astype(np.float32) +y = np.arcsin(x) +expect(node, inputs=[x], outputs=[y], name="test_asin") +``` + +
+ + +### **Asinh** + + Calculates the hyperbolic arcsine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic arcsine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+asinh + +```python +node = onnx.helper.make_node( + "Asinh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.arcsinh(x) # expected output [-0.88137358, 0., 0.88137358] +expect(node, inputs=[x], outputs=[y], name="test_asinh_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.arcsinh(x) +expect(node, inputs=[x], outputs=[y], name="test_asinh") +``` + +
+ + +### **Atan** + + Calculates the arctangent (inverse of tangent) of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 7 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The arctangent of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+atan + +```python +node = onnx.helper.make_node( + "Atan", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.arctan(x) +expect(node, inputs=[x], outputs=[y], name="test_atan_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.arctan(x) +expect(node, inputs=[x], outputs=[y], name="test_atan") +``` + +
+ + +### **Atanh** + + Calculates the hyperbolic arctangent of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic arctangent values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+atanh + +```python +node = onnx.helper.make_node( + "Atanh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-0.5, 0, 0.5]).astype(np.float32) +y = np.arctanh(x) # expected output [-0.54930615, 0., 0.54930615] +expect(node, inputs=[x], outputs=[y], name="test_atanh_example") + +x = np.random.uniform(0.0, 1.0, (3, 4, 5)).astype(np.float32) +y = np.arctanh(x) +expect(node, inputs=[x], outputs=[y], name="test_atanh") +``` + +
+ + +### **Attention** + + Computes scaled dot product attention on query, key and value tensors, using an optional attention mask if passed. + + This operator covers self and cross variants of the attention operation based on sequence lengths of K, Q and V. + + For self attention, `kv_sequence_length` equals to `q_sequence_length`. + + For cross attention, query and key might have different lengths. + + This operator also covers the 3 following variants based on the number of heads: + 1) Multi-headed Attention (MHA): Described in the paper https://arxiv.org/pdf/1706.03762, `q_num_heads = kv_num_heads`. + 2) Group-query Attention (GQA): Described in the paper https://arxiv.org/pdf/2305.13245, `q_num_heads > kv_num_heads`, `q_num_heads % kv_num_heads == 0`. + 3) Multi-query Attention (MQA): Described in the paper https://arxiv.org/pdf/1911.02150, `q_num_heads > kv_num_heads`, `kv_num_heads=1`. + + Attention bias to be added is calculated based on `attn_mask` input and `is_causal` attribute: + 1) `attn_mask`: A boolean mask where a value of `True` indicates that the element should take part in attention or a float mask of the same type as query, key, value that is added to the attention score. + 2) If `is_causal` is set to `1`, causal masking is applied with bottom-right (offset-aware) alignment: query `i` attends key `j` iff `j <= i + offset`, as illustrated below. + + ``` + 2D causal mask for Attention (PR onnx/onnx#8068) + S_q=4 queries, S_k=8 keys + Rule: query i attends key j iff j <= i + offset + offset = nonpad_kv_seqlen - S_q + + nonpad_kv_seqlen=4, offset=4-4=0 + + k0 k1 k2 k3 k4 k5 k6 k7 + +----+----+----+----+----+----+----+----+ + q0 | ## | | | | | | | | + +----+----+----+----+----+----+----+----+ + q1 | ## | ## | | | | | | | + +----+----+----+----+----+----+----+----+ + q2 | ## | ## | ## | | | | | | + +----+----+----+----+----+----+----+----+ + q3 | ## | ## | ## | ## | | | | | + +----+----+----+----+----+----+----+----+ + + + nonpad_kv_seqlen=8, offset=8-4=4 + + k0 k1 k2 k3 k4 k5 k6 k7 + +----+----+----+----+----+----+----+----+ + q0 | ## | ## | ## | ## | ## | | | | + +----+----+----+----+----+----+----+----+ + q1 | ## | ## | ## | ## | ## | ## | | | + +----+----+----+----+----+----+----+----+ + q2 | ## | ## | ## | ## | ## | ## | ## | | + +----+----+----+----+----+----+----+----+ + q3 | ## | ## | ## | ## | ## | ## | ## | ## | + +----+----+----+----+----+----+----+----+ + ``` + + With `nonpad_kv_seqlen=4` (offset=0), the mask is the standard lower-triangular. With `nonpad_kv_seqlen=8` (offset=4), the diagonal shifts right by 4, so each query sees the 4 additional valid cached keys. + + `offset` is the count of valid keys preceding the current query block: `offset = past_sequence_length` when `past_key` is provided; `offset = nonpad_kv_seqlen - q_sequence_length` (per batch) when an external cache is indicated by `nonpad_kv_seqlen` without `past_key`; `offset = 0` when neither is provided (the no-cache case, which reduces to the standard lower-triangular mask). When `offset < 0` (`nonpad_kv_seqlen < q_sequence_length`, i.e. more query tokens than cached keys) the leading query rows have an empty key set (no key satisfies `j <= i + offset`) and are fully masked. The causal frontier is computed independently of `attn_mask` and is then composed with it additively: a boolean `attn_mask` intersects the allowed set (its disallowed positions contribute `-inf` to the bias), while a float `attn_mask` is added to the attention scores rather than disabling positions. A fully-masked query row (no key attended, including the negative-offset leading rows) produces a zero output row, not `NaN`, for both `Y` and the mode-`3` `qk_matmul_output` debug output; the mode-`3` `qk_matmul_output` is emitted at the operator's output precision (`T1`). + + Errata (in-place behavioral correction, no opset bump): the reference implementation and backend tests were incorrect when `nonpad_kv_seqlen != q_sequence_length` (nonzero bottom-right offset, top-left instead of bottom-right causal alignment) and produced `NaN` for fully-masked rows; corrected in version 1.23. This fixed three behaviors described above: external-cache bottom-right causal alignment (`offset = nonpad_kv_seqlen - q_sequence_length`), zero (non-`NaN`) output for fully-masked rows including the mode-`3` `qk_matmul_output`, and the mode-`3` `qk_matmul_output` precision (`T1`). + + With respect to KV cache update, this operator allows the following two use cases: + + 1) Cache update happens inside the Attention operator. In this case, the `K` and `V` inputs contain only the incoming + tokens for the current autoregressive step, and the four optional inputs/outputs past and present key and value are + all needed. The Attention op performs a Concat operation on the past and incoming key and value to form the present + key and value, respectively. Note that this only works correctly for the special case where the past key and value + do not contain padded tokens. + 2) Cache update happens outside the Attention operator (for example, through the `TensorScatter` operator). In this + case, the `K` and `V` inputs correspond to the entire cache tensor, so the four optional inputs/outputs past and + present key and value should not be used. An additional input `nonpad_kv_seqlen` of shape (batch_size,) may be + provided to indicate the number of non-padding tokens in each sample of the batch to save unnecessary computation. + Here, the kv_sequence dimension of `attn_mask` can be shorter than `K` and `V`, but still needs to be at least as long + as the maximum value of `nonpad_kv_seqlen`. + + Both past and present state key/values are optional. They shall be used together, and not allowed to use only one of them. + The following pattern is applied to the Q, K and V inputs after appropriate reshaping of K and V inputs based on sequence lengths and num heads provided: + + ``` + The following pattern is applied by this operator: + Q K V + | | | + Q*sqrt(scale) K*sqrt(scale) | + | | | + | Transpose | + | | | + ---MatMul--- | + | | + softcap (if provided) | + | | + at_mask---Add | + | | + Softmax | + | | + -----MatMul------ + | + Y + ``` + + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +Other versions of this operator: 23 + +#### Attributes + +
+
is_causal : int (default is 0)
+
If set to `1`, causal masking is applied. For a square Q/K (no cache offset) this is a lower-triangular matrix. In general the mask is bottom-right (offset-aware): query in-block index `i` attends key `j` iff `j <= i + offset`, where `offset` is the count of valid keys preceding the query block (`past_sequence_length` for an internal `past_key` cache, or `nonpad_kv_seqlen - q_sequence_length` per batch for an external cache). When `offset = 0` this reduces to the lower-triangular (top-left) mask.
+
kv_num_heads : int
+
Number of heads of key and value. Must be used with 3D inputs of Q, K and V.
+
q_num_heads : int
+
Number of heads of query. Must be used with 3D inputs of Q, K and V.
+
qk_matmul_output_mode : int (default is 0)
+
If set to `0`, qk_matmul_output is the output of qk matmul. If set to `1`, qk_matmul_output is the output after the softcap operation (before mask addition). If set to `2`, qk_matmul_output includes the attention mask and softcap (if provided) applied to the output of qk matmul. If set to `3`, qk_matmul_output is the output after the softmax operation. In mode `3`, a fully-masked query row (every key disallowed) is a zero row, consistent with the corresponding row of the primary output `Y`: the fully-masked-row guard is applied before this output is produced. The mode-`3` output is emitted at the operator's output precision (`T1`); when `softmax_precision` differs from `T1` this is a cast of the softmax result to `T1`. Default value is 0.
+
scale : float
+
Scaling factor applied to $Q*K^T$. Default value is `1/sqrt(head_size)`. To prevent [numerical overflow](https://tinyurl.com/sudb9s96), scale `Q`, `K` by `sqrt(scale)` before matmul.
+
softcap : float (default is 0.0)
+
Softcap value for attention weights. Default value is 0.
+
softmax_precision : int
+
The floating-point precision used in softmax computation. If softmax precision is not provided, the same precision as the input of softmax (Q and K) is used.
+
+ +#### Inputs (3 - 7) + +
+
Q : T1
+
Query tensor. 4D tensor with shape `(batch_size, q_num_heads, q_sequence_length, head_size)` or 3D tensor with shape `(batch_size, q_sequence_length, q_hidden_size)`. For cases with a 3D input tensor, `q_hidden_size = q_num_heads * head_size`
+
K : T1
+
Key tensor. 4D tensor with shape `(batch_size, kv_num_heads, kv_sequence_length, head_size)` or 3D tensor with shape `(batch_size, kv_sequence_length, k_hidden_size)`. For cases with a 3D input tensor, `k_hidden_size = kv_num_heads * head_size`
+
V : T2
+
Value tensor. 4D tensor with shape `(batch_size, kv_num_heads, kv_sequence_length, v_head_size)` or 3D tensor with shape `(batch_size, kv_sequence_length, v_hidden_size)`. For cases with a 3D input tensor, `v_hidden_size = kv_num_heads * v_head_size`
+
attn_mask (optional) : U
+
Attention mask. Shape must be broadcastable to `(batch_size, q_num_heads, q_sequence_length, total_sequence_length)` where `total_sequence_length = past_sequence_length + kv_sequence_length.` The last dimension can also be shorter than `total_sequence_length` and will be padded to `total_sequence_length` with negative infinity. Two types of masks are supported: a boolean mask where a value of `True` indicates that the element should take part in attention, or a float mask of the same type as query, key, value that is added to the attention score.
+
past_key (optional) : T1
+
past state cache for key with shape `(batch_size, kv_num_heads, past_sequence_length, head_size)`
+
past_value (optional) : T2
+
past state cache for value with shape `(batch_size, kv_num_heads, past_sequence_length, v_head_size)`
+
nonpad_kv_seqlen (optional) : tensor(int64)
+
A vector of integers of shape `(batch_size,)` that indicates the number of valid (ie, non-padding) tokens in each sample. A padding mask can be derived from this. This should not be used together with `past_key` and `past_value` inputs or `present_key` and `present_value` outputs (See the KV cache use cases in the operator description).
+
+ +#### Outputs (1 - 4) + +
+
Y : T1
+
The output tensor . 4D tensor with shape `(batch_size, q_num_heads, q_sequence_length, v_head_size)` or 3D tensor with shape `(batch_size, q_sequence_length, hidden_size)`. For cases with a 3D input tensor, `hidden_size = q_num_heads * v_head_size`
+
present_key (optional) : T1
+
Updated key cache with shape `(batch_size, kv_num_heads, total_sequence_length, head_size)` where `total_sequence_length = past_sequence_length + kv_sequence_length`.
+
present_value (optional) : T2
+
Updated value cache with shape `(batch_size, kv_num_heads, total_sequence_length, v_head_size)` where `total_sequence_length = past_sequence_length + kv_sequence_length`.
+
qk_matmul_output (optional) : T1
+
The output of QK matmul. 4D tensor with shape `(batch_size, q_num_heads, q_sequence_length, total_sequence_length)` where `total_sequence_length = past_sequence_length + kv_sequence_length`.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain Q and K inputs types to float tensors.
+
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain V input types to float tensors.
+
U : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool)
+
Constrain output 'mask' types to boolean tensors and input types.
+
+ + +#### Examples + +
+attention + +```python +node = onnx.helper.make_node("Attention", inputs=["Q", "K", "V"], outputs=["Y"]) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_23_boolmask_fullymasked_row_nan_robustness + +```python +"""Opset-23 fully-masked boolean ``attn_mask`` row -> zero (not ``NaN``). + +This locks the opset-23 / ``old.cc`` function-body fully-masked-row guard +against future regressions. In opset 23 the only in-contract fully-masked +row comes from an all-``False`` boolean ``attn_mask`` row (``is_causal`` is +not set here): every key for that query is disallowed, so ``softmax`` over an +all-``-inf`` bias row is ``NaN``. The guard zeros that row's probabilities +before the ``P @ V`` contraction so the output row is exactly ``0``, while +rows with at least one allowed key are unchanged. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(4) +B, H, S, D = 1, 2, 2, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(B, H, S, D).astype(np.float32) +K = np.random.rand(B, H, S, D).astype(np.float32) +V = np.random.rand(B, H, S, D).astype(np.float32) +# Row 0: no key allowed -> fully masked (Bug-2 empty row). Row 1: both keys +# allowed -> finite, unchanged by the guard. +attn_mask = np.array([[False, False], [True, True]], dtype=np.bool_) + +Y, _, _, _ = _compute_attention(Q, K, V, attn_mask=attn_mask) + +# Fully-masked row 0 is exactly zero (not NaN); every other cell is finite. +assert np.all(np.isfinite(Y)), "non-masked rows must be finite" +assert np.array_equal(Y[:, :, 0, :], np.zeros_like(Y[:, :, 0, :])), ( + "fully-masked row must be zero (Bug-2)" +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_23_boolmask_fullymasked_row_nan_robustness", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_23_fullymasked_qk_matmul_output_mode3_zero + +```python +"""Opset-23 ``qk_matmul_output_mode=3`` fully-masked row is a zero row. + +Mode ``3`` exposes the post-softmax matrix as the optional +``qk_matmul_output``. For a fully-masked query row (all-``False`` boolean +``attn_mask`` row), the fully-masked-row guard is applied before this output +is produced, so the mode-3 row is zeroed, consistent with the primary output +``Y`` row (both are ``0``). This pins the mandated agreement between the +guarded primary output and the mode-3 output at opset 23. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(13) +B, H, S, D = 1, 2, 2, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + qk_matmul_output_mode=3, +) + +Q = np.random.rand(B, H, S, D).astype(np.float32) +K = np.random.rand(B, H, S, D).astype(np.float32) +V = np.random.rand(B, H, S, D).astype(np.float32) +# Row 0: no key allowed -> fully masked. Row 1: both keys allowed -> finite. +attn_mask = np.array([[False, False], [True, True]], dtype=np.bool_) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + qk_matmul_output_mode=3, +) + +# Primary output row 0 and the mode-3 row 0 are both guarded to zero. +assert np.array_equal(Y[:, :, 0, :], np.zeros_like(Y[:, :, 0, :])), ( + "fully-masked primary output row must be zero" +) +assert np.array_equal( + qk_matmul_output[:, :, 0, :], np.zeros_like(qk_matmul_output[:, :, 0, :]) +), "mode-3 output row for a fully-masked query must be zero (consistent with Y)" +assert np.all(np.isfinite(qk_matmul_output)), ( + "all mode-3 rows are finite (the fully-masked row is guarded to 0.0)" +) +assert np.all(np.isfinite(Y)), ( + "all Y rows are finite (the fully-masked row is guarded to 0.0)" +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_23_fullymasked_qk_matmul_output_mode3_zero", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_24_fullymasked_qk_matmul_output_mode3_zero + +```python +"""Opset-24 ``qk_matmul_output_mode=3`` fully-masked row is a zero row. + +The opset-24 twin of +``export_attention_23_fullymasked_qk_matmul_output_mode3_zero``. Mode ``3`` +exposes the post-softmax matrix as the optional ``qk_matmul_output``. For a +fully-masked query row (all-``False`` boolean ``attn_mask`` row), the +fully-masked-row guard is applied before this output is produced, so the +mode-3 row is zeroed, consistent with the primary output ``Y`` row (both are +``0``). This pins the mandated agreement between the guarded primary output +and the mode-3 output at opset 24. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(13) +B, H, S, D = 1, 2, 2, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + qk_matmul_output_mode=3, +) + +Q = np.random.rand(B, H, S, D).astype(np.float32) +K = np.random.rand(B, H, S, D).astype(np.float32) +V = np.random.rand(B, H, S, D).astype(np.float32) +# Row 0: no key allowed -> fully masked. Row 1: both keys allowed -> finite. +attn_mask = np.array([[False, False], [True, True]], dtype=np.bool_) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + qk_matmul_output_mode=3, +) + +# Primary output row 0 and the mode-3 row 0 are both guarded to zero. +assert np.array_equal(Y[:, :, 0, :], np.zeros_like(Y[:, :, 0, :])), ( + "fully-masked primary output row must be zero" +) +assert np.array_equal( + qk_matmul_output[:, :, 0, :], np.zeros_like(qk_matmul_output[:, :, 0, :]) +), "mode-3 output row for a fully-masked query must be zero (consistent with Y)" +assert np.all(np.isfinite(qk_matmul_output)), ( + "all mode-3 rows are finite (the fully-masked row is guarded to 0.0)" +) +assert np.all(np.isfinite(Y)), ( + "all Y rows are finite (the fully-masked row is guarded to 0.0)" +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_24_fullymasked_qk_matmul_output_mode3_zero", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_24_qk_matmul_output_mode3_softmax_precision + +```python +"""Mode-3 ``qk_matmul_output`` is emitted at the output precision ``T1``. + +``qk_matmul_output_mode=3`` exposes the post-softmax probabilities. When +``softmax_precision`` differs from the operator's output type ``T1`` (here +``T1 = float16`` with softmax computed in ``float32``), the mode-3 output is +cast back to ``T1`` -- matching the reference implementation, which casts the +exposed matrix to ``Q.dtype``. This locks both the dtype contract and the +fully-masked-row zeroing under a non-default ``softmax_precision``. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(17) +B, H, S, D = 1, 2, 2, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + qk_matmul_output_mode=3, + softmax_precision=int(onnx.TensorProto.FLOAT), +) + +# T1 = float16; softmax runs in float32, so the mode-3 output is cast back to +# float16 on emission. +Q = np.random.rand(B, H, S, D).astype(np.float16) +K = np.random.rand(B, H, S, D).astype(np.float16) +V = np.random.rand(B, H, S, D).astype(np.float16) +# Row 0: fully masked. Row 1: both keys allowed -> finite. +attn_mask = np.array([[False, False], [True, True]], dtype=np.bool_) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + qk_matmul_output_mode=3, + softmax_precision=int(onnx.TensorProto.FLOAT), +) + +# The mode-3 output is emitted at T1 (float16), not the float32 softmax +# precision, matching the operator's output type. +assert qk_matmul_output.dtype == np.float16, ( + "mode-3 qk_matmul_output must be emitted at the output precision T1 (float16)" +) +# The fully-masked row is still guarded to zero, consistent with Y. +assert np.array_equal( + qk_matmul_output[:, :, 0, :], np.zeros_like(qk_matmul_output[:, :, 0, :]) +), "mode-3 output row for a fully-masked query must be zero (consistent with Y)" +assert np.all(np.isfinite(qk_matmul_output)), ( + "all mode-3 rows are finite (the fully-masked row is guarded to 0.0)" +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_24_qk_matmul_output_mode3_softmax_precision", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_3d + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_attn_mask + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_3d_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_causal + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_diff_head_sizes + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_diff_heads_sizes", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_diff_head_sizes_attn_mask + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_3d_diff_heads_sizes_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_diff_head_sizes_causal + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_diff_heads_sizes_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_diff_head_sizes_scaled + +```python +scale = 1e-2 +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_diff_heads_sizes_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_diff_head_sizes_softcap + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_diff_heads_sizes_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_diff_head_sizes_with_past_and_present + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 10).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_3d_diff_heads_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_gqa + +```python +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_gqa", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_gqa_attn_mask + +```python +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_3d_gqa_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_gqa_causal + +```python +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_gqa_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_gqa_scaled + +```python +scale = 1e-2 +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_gqa_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_gqa_softcap + +```python +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_gqa_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_gqa_with_past_and_present + +```python +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_3d_gqa_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_scaled + +```python +scale = 1e-2 +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_softcap + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_transpose_verification + +```python +"""Test case to verify correct 3D to 4D transpose behavior. + +This test verifies that 3D inputs are correctly reshaped and transposed +according to the ONNX specification: +[batch_size, seq_length, hidden_size] -> +[batch_size, seq_length, num_heads, head_size] -> +[batch_size, num_heads, seq_length, head_size] +""" +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +# Test inputs that will clearly demonstrate the transpose behavior +batch_size = 1 +q_seq_length = 2 +kv_seq_length = 2 +head_size = 4 +q_hidden_size = q_num_heads * head_size # 3 * 4 = 12 +kv_hidden_size = kv_num_heads * head_size # 3 * 4 = 12 + +# Create structured inputs to verify correct transpose behavior +# Q has a pattern where each position in hidden dimension has a specific value +Q = np.zeros((batch_size, q_seq_length, q_hidden_size), dtype=np.float32) +# Fill Q with pattern: head0=[1,1,1,1], head1=[2,2,2,2], head2=[3,3,3,3] +for head in range(q_num_heads): + start_idx = head * head_size + end_idx = start_idx + head_size + Q[0, :, start_idx:end_idx] = float(head + 1) + +K = np.ones((batch_size, kv_seq_length, kv_hidden_size), dtype=np.float32) * 0.1 +V = np.ones((batch_size, kv_seq_length, kv_hidden_size), dtype=np.float32) * 0.1 + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_transpose_verification", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_with_past_and_present + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_3d_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_with_past_and_present_qk_matmul + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_3d_with_past_and_present_qk_matmul", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_with_past_and_present_qk_matmul_bias + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + qk_matmul_output_mode=2, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + qk_matmul_output_mode=2, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_3d_with_past_and_present_qk_matmul_bias", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_with_past_and_present_qk_matmul_softcap + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + softcap=2.0, + qk_matmul_output_mode=1, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + softcap=2.0, + qk_matmul_output_mode=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_3d_with_past_and_present_qk_matmul_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_3d_with_past_and_present_qk_matmul_softmax + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + qk_matmul_output_mode=3, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + qk_matmul_output_mode=3, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_3d_with_past_and_present_qk_matmul_softmax", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_4d_causal_nonpad_attn_mask_composition + +```python +"""Compose ``is_causal`` + ``nonpad_kv_seqlen`` + boolean ``attn_mask``. + +The existing nonpad tests use no ``attn_mask`` and the existing mask tests +use no ``nonpad_kv_seqlen``; this is the first to activate all three +constraints together on the external-cache path with ``batch > 1``. The +three biases are summed additively and a key is attended only if allowed by +all three. Crucially the inputs are designed so that **each constraint is +independently necessary** -- removing any one changes the golden -- to avoid a +degenerate test that a backend ignoring ``is_causal`` and/or +``nonpad_kv_seqlen`` could still pass: + +* **``is_causal`` binds.** Each batch has a key that the boolean mask allows + (``True``) but the bottom-right causal frontier disallows + (``j > i + offset``); only ``is_causal`` masks it (batch 0 row 0 key 2, + batch 1 row 0 key 3). +* **``attn_mask`` binds.** Each batch has a key the causal frontier and the + padding bound both allow but the boolean mask sets ``False`` (batch 0 row 2 + key 1, batch 1 row 2 key 2); only the mask masks it. +* **``nonpad_kv_seqlen`` binds.** ``nonpad_kv_seqlen`` sets the per-batch + causal *offset* (``offset = nonpad_kv_seqlen - q_sequence_length``), so + dropping it collapses the frontier to top-left (``offset = 0``) and shifts + which keys are attended. (Under ``is_causal=1`` the causal frontier already + subsumes the ``j < nonpad`` padding bound, so ``nonpad_kv_seqlen`` binds + through the offset it induces rather than through a redundant padding cut.) + +The mask is chosen to leave at least one allowed key on every query row, so +this exercises the *intersection* of the three constraints with finite outputs +(the fully-masked-row guard is covered by +``test_attention_4d_causal_nonpad_negative_offset_structural_empty`` and +``test_attention_24_fullymasked_qk_matmul_output_mode3_zero``). + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(11) +B, H, L, D = 2, 2, 6, 8 +S_q = 3 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, L, D).astype(np.float32) +V = np.random.rand(B, H, L, D).astype(np.float32) +nonpad_kv_seqlen = np.array([4, 5], dtype=np.int64) # offsets [1, 2] +# Per-batch (B, 1, S_q, L) bool mask. Each batch is laid out so all three +# constraints uniquely bind (see the docstring): a causal-only-masked key +# (mask True, j > i + offset), a mask-only-masked key (mask False, causal + +# nonpad allow it), and >=1 allowed key per row. +attn_mask = np.array( + [ + [ + [ + [True, True, True, False, False, False], + [True, True, True, False, False, False], + [True, False, True, True, False, False], + ] + ], + [ + [ + [True, True, True, True, False, False], + [True, True, True, True, False, False], + [True, True, False, True, True, False], + ] + ], + ], + dtype=np.bool_, +) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +# The chosen mask leaves >=1 allowed key per row, so the composition stays +# finite (no fully-masked row in this case). +assert np.all(np.isfinite(Y)), "composed-constraint output must be finite" + +# Non-degeneracy: each constraint is independently necessary. Removing any one +# of the three (is_causal, attn_mask, nonpad_kv_seqlen) must change the result, +# so a backend that ignores is_causal or nonpad_kv_seqlen cannot reproduce the +# golden by applying only the most restrictive mask. +y_no_causal, _, _, _ = _compute_attention( + Q, K, V, attn_mask=attn_mask, nonpad_kv_seqlen=nonpad_kv_seqlen, is_causal=0 +) +y_no_mask, _, _, _ = _compute_attention( + Q, K, V, nonpad_kv_seqlen=nonpad_kv_seqlen, is_causal=1 +) +y_no_nonpad, _, _, _ = _compute_attention( + Q, K, V, attn_mask=attn_mask, is_causal=1 +) +assert not np.allclose(Y, y_no_causal, equal_nan=True), ( + "is_causal must bind: dropping it changes the result" +) +assert not np.allclose(Y, y_no_mask, equal_nan=True), ( + "attn_mask must bind: dropping it changes the result" +) +assert not np.allclose(Y, y_no_nonpad, equal_nan=True), ( + "nonpad_kv_seqlen must bind (via the causal offset): dropping it changes the result" +) + +expect( + node, + inputs=[Q, K, V, attn_mask, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_causal_nonpad_attn_mask_composition", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_4d_causal_nonpad_batch_prefill + +```python +"""Batch>1 continued prefill with distinct per-batch bottom-right offsets. + +The batched generalization of the ``batch == 1`` continued-prefill case: with +``nonpad_kv_seqlen = [4, 5, 6]`` and ``S_q = 2`` the per-batch bottom-right +offsets are ``[2, 3, 4]`` (all ``>= 0``), so each batch realigns its causal +frontier to its own valid-key prefix. This pins that the per-batch offset is +applied independently across the batch dimension. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(12) +B, H, L, D = 3, 2, 6, 8 +S_q = 2 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, L, D).astype(np.float32) +V = np.random.rand(B, H, L, D).astype(np.float32) +nonpad_kv_seqlen = np.array([4, 5, 6], dtype=np.int64) # offsets [2, 3, 4] + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +assert np.all(np.isfinite(Y)), "per-batch prefill output must be finite" + +expect( + node, + inputs=[Q, K, V, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_causal_nonpad_batch_prefill", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_4d_causal_nonpad_continued_prefill + +```python +"""Continued / chunked prefill (S_q=2) into a partially-filled static cache. + +With ``nonpad_kv_seqlen = [4]`` and ``S_q = 2`` the bottom-right offset is +``4 - 2 = 2``: query 0 attends keys ``{0,1,2}`` and query 1 attends +``{0,1,2,3}``. The old top-left alignment would mask everything past the +diagonal (``{0}`` and ``{0,1}``), so this test fails pre-fix. +""" +np.random.seed(1) +B, H, L, D = 1, 2, 4, 8 +S_q = 2 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, L, D).astype(np.float32) +V = np.random.rand(B, H, L, D).astype(np.float32) +nonpad_kv_seqlen = np.array([4], dtype=np.int64) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_causal_nonpad_continued_prefill", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_4d_causal_nonpad_negative_offset_structural_empty + +```python +"""Negative bottom-right offset: structurally-empty early query rows -> zero. + +This is the onnx-node twin of the ORT gtest +``Attention_Causal_NonPadKVSeqLen_StructuralEmptyRow_Zero`` / +``StructuralEmptyRows_Zero_CUDA``. With ``nonpad_kv_seqlen = [2]`` and +``S_q = 4`` the bottom-right offset is ``2 - 4 = -2``: query row ``sq`` +attends keys ``0..(sq - 2)``, so rows 0 and 1 have an empty key set. Their +``softmax`` over an all-``-inf`` bias row is ``NaN``; the fully-masked-row +guard zeros those rows before the ``P @ V`` contraction so the output rows are +exactly ``0``, while rows 2 and 3 (attending keys ``{0}`` and ``{0,1}``) stay finite +and nonzero. A ``nonpad_kv_seqlen[b] < q_sequence_length`` input is out of +the contract's intended use, but its result is still well-defined (zeroed +rows) rather than ``NaN``; this test pins that defined behavior. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(7) +B, H, L, D = 1, 2, 4, 8 +S_q = 4 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, L, D).astype(np.float32) +V = np.random.rand(B, H, L, D).astype(np.float32) +# offset = nonpad - S_q = 2 - 4 = -2 -> rows 0,1 structurally empty. +nonpad_kv_seqlen = np.array([2], dtype=np.int64) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +# Structurally-empty early rows are exactly zero (not NaN); later rows finite. +assert np.all(np.isfinite(Y)), "all output rows must be finite" +assert np.array_equal(Y[:, :, 0, :], np.zeros_like(Y[:, :, 0, :])), ( + "structurally-empty row 0 must be zero" +) +assert np.array_equal(Y[:, :, 1, :], np.zeros_like(Y[:, :, 1, :])), ( + "structurally-empty row 1 must be zero" +) +assert np.any(Y[:, :, 2, :] != 0) and np.any(Y[:, :, 3, :] != 0), ( + "rows with a non-empty key set must be nonzero" +) + +expect( + node, + inputs=[Q, K, V, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_causal_nonpad_negative_offset_structural_empty", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_4d_causal_with_past_and_present + +```python +"""Regression guard: internal (past_key) cache + is_causal. + +This exercises the unchanged scalar bottom-right path (offset = +past_sequence_length). Its golden output must remain identical to the +pre-fix behavior, proving the external-cache change does not touch the +past_key path. +""" +np.random.seed(2) +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], + is_causal=1, +) + +past_sequence_length = 3 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 4, 8).astype(np.float32) +V = np.random.rand(2, 3, 4, 8).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + past_key=past_key, + past_value=past_value, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_causal_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_4d_diff_heads_mask4d_padded_kv + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 4).astype(np.float32) +nonpad_kv_seqlen = np.array([3, 4], dtype=np.int64) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + nonpad_kv_seqlen=nonpad_kv_seqlen, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_diff_heads_mask4d_padded_kv", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_4d_gqa_causal_nonpad_decode + +```python +"""External/static-cache decode (S_q=1) with per-batch valid lengths. + +K/V are the full static cache buffer; ``nonpad_kv_seqlen`` marks how many +leading keys are valid per batch. With bottom-right (offset-aware) causal +masking the single decode query attends keys ``0..nonpad[b]-1``. Under the +old top-left alignment it would attend only key 0, so this test fails +pre-fix and passes post-fix. +""" +np.random.seed(0) +B, H_q, H_kv, L, D = 2, 4, 2, 8, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H_q, 1, D).astype(np.float32) +K = np.random.rand(B, H_kv, L, D).astype(np.float32) +V = np.random.rand(B, H_kv, L, D).astype(np.float32) +# Batch 0 has all 8 keys valid, batch 1 only the first 5. +nonpad_kv_seqlen = np.array([8, 5], dtype=np.int64) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_gqa_causal_nonpad_decode", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_4d_gqa_causal_nonpad_decode_fp16 + +```python +"""fp16 variant of the external-cache decode case (locks -inf dtype handling).""" +np.random.seed(0) +B, H_q, H_kv, L, D = 2, 4, 2, 8, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H_q, 1, D).astype(np.float16) +K = np.random.rand(B, H_kv, L, D).astype(np.float16) +V = np.random.rand(B, H_kv, L, D).astype(np.float16) +nonpad_kv_seqlen = np.array([8, 5], dtype=np.int64) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_gqa_causal_nonpad_decode_fp16", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_attn_3d_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 1, 4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_3d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_attn_3d_mask_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 1, 4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_3d_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_attn_4d_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_4d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_attn_4d_mask_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_4d_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_attn_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_attn_mask_bool + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(bool) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_bool", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_attn_mask_bool_4d + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6).astype(bool) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_bool_4d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, is_causal=1) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_causal_boolmask_nan_robustness + +```python +"""Composed ``is_causal`` + boolean ``attn_mask`` NaN-robustness. + +The causal frontier (lower-triangular here, offset 0) and the boolean +``attn_mask`` are intersected: a key is attended only if allowed by both. +This exercises two pre-fix NaN sources on the same forward pass: + +* **Bug-1 (allowed cells stay finite).** Query 0 is allowed key 0 by both + the causal frontier (``{0}``) and the mask (``True`` at key 0). The old + ``(1 - attn_mask) * -inf`` conversion computes ``0 * -inf = NaN`` at that + allowed cell, poisoning the row. The select conversion + ``where(attn_mask, 0, -inf)`` keeps it finite. +* **Bug-2 (fully-masked row -> 0).** Query 1 is allowed keys ``{0, 1}`` by + the causal frontier but the mask is ``False`` at both, so the combined + constraint allows no key. ``softmax`` of an all-``-inf`` row is ``NaN``; + the fully-masked-row guard zeros it before the ``P @ V`` contraction so + the output row is ``0``. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(3) +B, H, S, D = 1, 2, 2, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H, S, D).astype(np.float32) +K = np.random.rand(B, H, S, D).astype(np.float32) +V = np.random.rand(B, H, S, D).astype(np.float32) +# Row 0: key 0 allowed (Bug-1 allowed cell). Row 1: no key allowed -> fully +# masked once intersected with the causal frontier (Bug-2 empty row). +attn_mask = np.array([[True, False], [False, False]], dtype=np.bool_) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + is_causal=1, +) + +# Bug-1: allowed cells are finite (no NaN anywhere). Bug-2: the fully-masked +# query row is exactly zero, not NaN. +assert np.all(np.isfinite(Y)), "allowed cells must be finite (Bug-1)" +assert np.array_equal(Y[:, :, 1, :], np.zeros_like(Y[:, :, 1, :])), ( + "fully-masked row must be zero (Bug-2)" +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_causal_boolmask_nan_robustness", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +
+attention_diff_head_sizes + +```python +node = onnx.helper.make_node("Attention", inputs=["Q", "K", "V"], outputs=["Y"]) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_diff_heads_sizes", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_diff_head_sizes_attn_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_diff_heads_sizes_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_diff_head_sizes_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_diff_heads_sizes_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_diff_head_sizes_scaled + +```python +scale = 1e-2 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, scale=scale) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_diff_heads_sizes_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_diff_head_sizes_softcap + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=2.0, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + softcap=2.0, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_diff_heads_sizes_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_diff_head_sizes_with_past_and_present + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 10).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_diff_heads_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_diff_head_sizes_with_past_and_present_mask3D + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) +attn_mask = np.random.rand(2, 1, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 10).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_diff_heads_with_past_and_present_mask3d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_diff_head_sizes_with_past_and_present_mask4D + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 10).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_diff_heads_with_past_and_present_mask4d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_fp16 + +```python +node = onnx.helper.make_node("Attention", inputs=["Q", "K", "V"], outputs=["Y"]) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float16) +K = np.random.rand(2, 3, 6, 8).astype(np.float16) +V = np.random.rand(2, 3, 6, 8).astype(np.float16) + +Y, _, _, _ = _compute_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_fp16", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_gqa + +```python +node = onnx.helper.make_node("Attention", inputs=["Q", "K", "V"], outputs=["Y"]) + +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_gqa", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_gqa_attn_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_gqa_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_gqa_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, is_causal=1) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_gqa_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_gqa_scaled + +```python +scale = 1e-2 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, +) + +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, scale=scale) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_gqa_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_gqa_softcap + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=2.0, +) + +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, softcap=2.0) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_gqa_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_gqa_with_past_and_present + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_gqa_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_gqa_with_past_and_present_fp16 + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 9, 4, 8).astype(np.float16) +K = np.random.rand(2, 3, 6, 8).astype(np.float16) +V = np.random.rand(2, 3, 6, 8).astype(np.float16) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float16) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float16) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float16) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_gqa_with_past_and_present_fp16", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_scaled + +```python +scale = 1e-2 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, scale=scale) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_softcap + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=2.0, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, softcap=2.0) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_softcap_with_neginf_mask + +```python +"""Softcap + -inf mask: verifies softcap is applied BEFORE mask/bias. + +If ordering were wrong (mask then softcap), tanh(-inf/softcap) = -1, +so softcap * tanh(-inf/softcap) = -softcap (finite). That leaks +probability to masked positions. With correct ordering (softcap then +mask), the -inf mask values survive to softmax and yield zero weight. +""" +np.random.seed(42) +B, H, S_q, S_kv, D = 1, 1, 4, 6, 8 + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, S_kv, D).astype(np.float32) +V = np.random.rand(B, H, S_kv, D).astype(np.float32) + +# All Q positions are blocked from KV positions 4 and 5. +attn_mask = np.zeros((S_q, S_kv), dtype=np.float32) +attn_mask[:, 4:] = -np.inf + +softcap = 0.5 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + softcap=softcap, +) + +Y, _, _, _ = _compute_attention(Q, K, V, attn_mask=attn_mask, softcap=softcap) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_softcap_neginf_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_softcap_with_neginf_mask_poison + +```python +"""Softcap + -inf mask + poison values at masked KV positions. + +V has value 1000 at the masked positions (4 and 5). With correct +ordering the output stays in [0, 1] because the mask zeros out those +positions. With wrong ordering the output explodes (> 50), making +the failure obvious even with loose tolerances. +""" +np.random.seed(42) +B, H, S_q, S_kv, D = 1, 1, 4, 6, 8 + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, S_kv, D).astype(np.float32) +V = np.random.rand(B, H, S_kv, D).astype(np.float32) + +# Block all Q positions from KV positions 4 and 5. +attn_mask = np.zeros((S_q, S_kv), dtype=np.float32) +attn_mask[:, 4:] = -np.inf + +# Poison: if attention leaks to masked positions, output >> 1. +V[:, :, 4:, :] = 1000.0 + +softcap = 0.5 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + softcap=softcap, +) + +Y, _, _, _ = _compute_attention(Q, K, V, attn_mask=attn_mask, softcap=softcap) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_softcap_neginf_mask_poison", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_past_and_present + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_past_and_present_qk_matmul + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_past_and_present_qk_matmul_bias + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + qk_matmul_output_mode=2, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + qk_matmul_output_mode=2, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul_bias", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_past_and_present_qk_matmul_bias_3d_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + qk_matmul_output_mode=2, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 1, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + qk_matmul_output_mode=2, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_past_and_present_qk_matmul_bias_3d_mask_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + qk_matmul_output_mode=2, + is_causal=1, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 1, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + qk_matmul_output_mode=2, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_past_and_present_qk_matmul_bias_4d_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + qk_matmul_output_mode=2, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + qk_matmul_output_mode=2, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_past_and_present_qk_matmul_bias_4d_mask_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + qk_matmul_output_mode=2, + is_causal=1, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + qk_matmul_output_mode=2, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_qk_matmul + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y", "", "", "qk_matmul_output"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, qk_matmul_output = _compute_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y, qk_matmul_output], + name="test_attention_4d_with_qk_matmul", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_qk_matmul_bias + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + qk_matmul_output_mode=2, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + qk_matmul_output_mode=2, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_4d_with_qk_matmul_bias", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_qk_matmul_softcap + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + softcap=2.0, + qk_matmul_output_mode=1, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + softcap=2.0, + qk_matmul_output_mode=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_4d_with_qk_matmul_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +
+attention_with_qk_matmul_softmax + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + qk_matmul_output_mode=3, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + qk_matmul_output_mode=3, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_4d_with_qk_matmul_softmax", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +### **AveragePool** + + AveragePool consumes an input tensor X and applies average pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + average pooling consisting of computing the average on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape is calculated differently + depending on whether explicit padding is used, where pads is employed, or auto padding is used, where auto_pad is utilized. + With explicit padding (https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html?highlight=maxpool#torch.nn.MaxPool2d): + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled. `pad_shape[i]` is the sum of pads along axis `i`. Sliding windows that would start in the right padded region are ignored. + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following when ceil_mode is enabled: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + or when ceil_mode is disabled (https://www.tensorflow.org/api_docs/python/tf/keras/layers/AveragePooling2D): + ``` + VALID: output_spatial_shape[i] = floor((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i]) + 1 + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = floor((input_spatial_shape[i] - 1) / strides_spatial_shape[i]) + 1 + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i] + ``` + The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero). + + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 7, 10, 11, 19 + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
count_include_pad : int (default is 0)
+
Whether include pad pixels when calculating values for the edges. Default is 0, doesn't count include pad.
+
dilations : list of ints
+
Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+averagepool_1d_default + +```python +"""input_shape: [1, 3, 32] +output_shape: [1, 3, 31] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2], +) +x = np.random.randn(1, 3, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = [2] +strides = [1] +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "AVG") + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_1d_default") +``` + +
+ + +
+averagepool_2d_ceil + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + strides=[2, 2], + ceil_mode=True, +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[6, 7.5], [12, 13.5]]]]).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_ceil") +``` + +
+ + +
+averagepool_2d_ceil_last_window_starts_on_pad + +```python +"""input_shape: [1, 3, 2, 2] +output_shape: [1, 3, 1, 1] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + strides=[3, 3], + pads=[1, 1, 1, 1], + ceil_mode=True, + count_include_pad=1, +) +x = np.array( + [ + [ + [[0.8580, 0.0786], [0.2692, 0.1537]], + [[0.8816, 0.4353], [0.5772, 0.6623]], + [[0.9067, 0.9483], [0.5970, 0.7630]], + ] + ] +).astype(np.float32) +y = np.array([[[[0.1511]], [[0.2841]], [[0.3572]]]]).astype(np.float32) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_averagepool_2d_ceil_last_window_starts_on_pad", +) +``` + +
+ + +
+averagepool_2d_default + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 31, 31] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = (2, 2) +strides = (1, 1) +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "AVG") + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_default") +``` + +
+ + +
+averagepool_2d_dilations + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[1, 1], + dilations=[2, 2], + ceil_mode=True, +) + +# input shape: [1, 1, 4, 4] +x = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ] +).astype(np.float32) + +y = np.array([[[[6, 7], [10, 11]]]]).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_dilations") +``` + +
+ + +
+averagepool_2d_pads + +```python +"""input_shape: [1, 3, 28, 28] +output_shape: [1, 3, 30, 30] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[2, 2, 2, 2], +) +x = np.random.randn(1, 3, 28, 28).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (3, 3) +strides = (1, 1) +pad_bottom = 2 +pad_top = 2 +pad_right = 2 +pad_left = 2 +pads = [pad_top, pad_left, pad_bottom, pad_right] +out_shape, extra_pads = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides, ceil_mode=False +) +padded = np.pad( + x, + ( + (0, 0), + (0, 0), + (extra_pads[0], extra_pads[2]), + (extra_pads[1], extra_pads[3]), + ), + mode="constant", + constant_values=np.nan, +) +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=extra_pads, + pads=pads, +) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_pads") +``` + +
+ + +
+averagepool_2d_pads_count_include_pad + +```python +"""input_shape: [1, 3, 28, 28] +output_shape: [1, 3, 30, 30] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[2, 2, 2, 2], + count_include_pad=1, +) +x = np.random.randn(1, 3, 28, 28).astype(np.float32) +x_shape = np.shape(x) +dilations = (1, 1) +kernel_shape = (3, 3) +strides = (1, 1) +pad_bottom = 2 +pad_top = 2 +pad_right = 2 +pad_left = 2 +pads = [pad_top, pad_left, pad_bottom, pad_right] +out_shape, extra_pads = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides, dilations, ceil_mode=False +) +padded = np.pad( + x, + ( + (0, 0), + (0, 0), + (extra_pads[0], extra_pads[2]), + (extra_pads[1], extra_pads[3]), + ), + mode="constant", + constant_values=0, +) +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=extra_pads, + pads=pads, + count_include_pad=1, +) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_averagepool_2d_pads_count_include_pad", +) +``` + +
+ + +
+averagepool_2d_precomputed_pads + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 5, 5] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + pads=[2, 2, 2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array( + [ + [ + [ + [7, 7.5, 8, 8.5, 9], + [9.5, 10, 10.5, 11, 11.5], + [12, 12.5, 13, 13.5, 14], + [14.5, 15, 15.5, 16, 16.5], + [17, 17.5, 18, 18.5, 19], + ] + ] + ] +).astype(np.float32) + +expect( + node, inputs=[x], outputs=[y], name="test_averagepool_2d_precomputed_pads" +) +``` + +
+ + +
+averagepool_2d_precomputed_pads_count_include_pad + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 5, 5] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + pads=[2, 2, 2, 2], + count_include_pad=1, +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array( + [ + [ + [ + [2.5200, 3.6000, 4.8000, 4.0800, 3.2400], + [4.5600, 6.4000, 8.4000, 7.0400, 5.5200], + [7.2000, 10.0000, 13.0000, 10.8000, 8.4000], + [6.9600, 9.6000, 12.4000, 10.2400, 7.9200], + [6.1200, 8.4000, 10.8000, 8.8800, 6.8400], + ] + ] + ] +).astype(np.float32) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_averagepool_2d_precomputed_pads_count_include_pad", +) +``` + +
+ + +
+averagepool_2d_precomputed_same_upper + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 3, 3] +pad_shape: [2, 2] -> [1, 1, 1, 1] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + strides=[2, 2], + auto_pad="SAME_UPPER", +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[4, 5.5, 7], [11.5, 13, 14.5], [19, 20.5, 22]]]]).astype( + np.float32 +) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_averagepool_2d_precomputed_same_upper", +) +``` + +
+ + +
+averagepool_2d_precomputed_strides + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[4, 6], [14, 16]]]]).astype(np.float32) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_averagepool_2d_precomputed_strides", +) +``` + +
+ + +
+averagepool_2d_same_lower + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [1, 0, 1, 0] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_LOWER", +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_LOWER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_LOWER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_bottom = pad_shape[0] // 2 +pad_top = pad_shape[0] - pad_bottom +pad_right = pad_shape[1] // 2 +pad_left = pad_shape[1] - pad_right +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=np.nan, +) +pads = (pad_top, pad_left, pad_bottom, pad_right) +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=pads, + pads=pads, +) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_same_lower") +``` + +
+ + +
+averagepool_2d_same_upper + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [0, 1, 0, 1] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_UPPER", +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_UPPER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_UPPER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_top = pad_shape[0] // 2 +pad_bottom = pad_shape[0] - pad_top +pad_left = pad_shape[1] // 2 +pad_right = pad_shape[1] - pad_left +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=np.nan, +) +pads = (pad_top, pad_left, pad_bottom, pad_right) +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=pads, + pads=pads, +) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_same_upper") +``` + +
+ + +
+averagepool_2d_strides + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 10, 10] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + strides=[3, 3], +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (5, 5) +strides = (3, 3) +out_shape, pads = get_output_shape_explicit_padding( + None, x_shape[2:], kernel_shape, strides, ceil_mode=False +) +padded = x +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=pads, + pads=None, +) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_strides") +``` + +
+ + +
+averagepool_3d_default + +```python +"""input_shape: [1, 3, 32, 32, 32] +output_shape: [1, 3, 31, 31, 31] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], +) +x = np.random.randn(1, 3, 32, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = [2, 2, 2] +strides = [1, 1, 1] +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "AVG") + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_3d_default") +``` + +
+ + +
+averagepool_3d_dilations + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], + strides=[1, 1, 1], + dilations=[2, 2, 2], + ceil_mode=True, +) + +# input shape: [1, 1, 4, 4, 4] +x = np.array( + [ + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + ] + ] + ] +).astype(np.float32) + +y = np.array([[[[[6, 7], [10, 11]], [[6, 7], [10, 11]]]]]).astype(np.float32) + +expect( + node, inputs=[x], outputs=[y], name="test_averagepool_3d_dilations_small" +) +``` + +
+ + +
+averagepool_3d_dilations_large + +```python +x_shape = (32, 32, 32) +dilations = (2, 2, 2) +kernel_shape = (5, 5, 5) +strides = (3, 3, 3) +count_include_pad = 0 + +for count_include_pad in (0, 1): + for ceil_mode in (True, False): + node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=kernel_shape, + strides=strides, + dilations=dilations, + count_include_pad=count_include_pad, + ceil_mode=ceil_mode, + ) + + x = np.random.randn(1, 1, *x_shape).astype(np.float32) + out_shape, extra_pads = get_output_shape_explicit_padding( + None, + x_shape, + kernel_shape, + strides, + dilations=dilations, + ceil_mode=ceil_mode, + ) + padded = np.pad( + x, + ( + (0, 0), + (0, 0), + (extra_pads[0], extra_pads[3]), + (extra_pads[1], extra_pads[4]), + (extra_pads[2], extra_pads[5]), + ), + mode="constant", + constant_values=0 if count_include_pad == 1 else np.nan, + ) + y = pool( + padded, + (1, 1, *x_shape), + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=extra_pads, + pads=None, + dilations=dilations, + count_include_pad=count_include_pad, + ) + + test_name = f"test_averagepool_3d_dilations_large_count_include_pad_is_{count_include_pad}_ceil_mode_is_{ceil_mode}" + expect(node, inputs=[x], outputs=[y], name=test_name) +``` + +
+ + +### **BatchNormalization** + + Carries out batch normalization as described in the paper + https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, + There are five required inputs 'X', 'scale', 'B', 'input_mean' and + 'input_var'. + Note that 'input_mean' and 'input_var' are expected to be the estimated + statistics in inference mode (training_mode=False, default), + and the running statistics in training mode (training_mode=True). + There are multiple cases for the number of outputs, which we list below: + + * Output case #1: Y, running_mean, running_var (training_mode=True) + * Output case #2: Y (training_mode=False) + + When training_mode=False, extra outputs are invalid. + The outputs are updated as follows when training_mode=True: + ``` + running_mean = input_mean * momentum + current_mean * (1 - momentum) + running_var = input_var * momentum + current_var * (1 - momentum) + + Y = (X - current_mean) / sqrt(current_var + epsilon) * scale + B + ``` + where: + ``` + current_mean = ReduceMean(X, axis=all_except_channel_index) + current_var = ReduceVar(X, axis=all_except_channel_index) + ``` + Notice that `ReduceVar` refers to the population variance, and it equals to + `sum(sqrd(x_i - x_avg)) / N` + where `N` is the population size (this formula does not use sample size `N - 1`). + + The computation of ReduceMean and ReduceVar uses float to avoid overflow for float16 inputs. + + When training_mode=False: + ``` + Y = (X - input_mean) / sqrt(input_var + epsilon) * scale + B + ``` + + For previous (depreciated) non-spatial cases, implementors are suggested + to flatten the input shape to (N x C * D1 * D2 * ... * Dn) before a BatchNormalization Op. + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 7, 9, 14 + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
momentum : float (default is 0.9)
+
Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum).
+
training_mode : int (default is 0)
+
If set to true, it indicates BatchNormalization is being used for training, and outputs 1 and 2 are to be computed.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size, C is the number of channels. Statistics are computed for every channel of C over N and D1 to Dn dimensions. For image data, input dimensions become (N x C x H x W). The op also accepts single dimension input of size N in which case C is assumed to be 1
+
scale (differentiable) : T1
+
Scale tensor of shape (C).
+
B (differentiable) : T1
+
Bias tensor of shape (C).
+
input_mean (differentiable) : T2
+
running (training) or estimated (testing) mean tensor of shape (C).
+
input_var (differentiable) : T2
+
running (training) or estimated (testing) variance tensor of shape (C).
+
+ +#### Outputs (1 - 3) + +
+
Y (differentiable) : T
+
The output tensor of the same shape as X
+
running_mean (optional, non-differentiable) : T2
+
The running mean after the BatchNormalization operator.
+
running_var (optional, non-differentiable) : T2
+
The running variance after the BatchNormalization operator. This op uses the population size (N) for calculating variance, and not the sample size N-1.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
T1 : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain scale and bias types to float tensors.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain mean and variance types to float tensors.
+
+ + +#### Examples + +
+batchnormalization + +```python +# input size: (2, 3, 4, 5) +x = np.random.randn(2, 3, 4, 5).astype(np.float32) +s = np.random.randn(3).astype(np.float32) +bias = np.random.randn(3).astype(np.float32) +mean = np.random.randn(3).astype(np.float32) +var = np.random.rand(3).astype(np.float32) +y = _batchnorm_test_mode(x, s, bias, mean, var).astype(np.float32) + +node = onnx.helper.make_node( + "BatchNormalization", + inputs=["x", "s", "bias", "mean", "var"], + outputs=["y"], +) + +# output size: (2, 3, 4, 5) +expect( + node, + inputs=[x, s, bias, mean, var], + outputs=[y], + name="test_batchnorm_example", +) + +# input size: (2, 3, 4, 5) +x = np.random.randn(2, 3, 4, 5).astype(np.float32) +s = np.random.randn(3).astype(np.float32) +bias = np.random.randn(3).astype(np.float32) +mean = np.random.randn(3).astype(np.float32) +var = np.random.rand(3).astype(np.float32) +epsilon = 1e-2 +y = _batchnorm_test_mode(x, s, bias, mean, var, epsilon).astype(np.float32) + +node = onnx.helper.make_node( + "BatchNormalization", + inputs=["x", "s", "bias", "mean", "var"], + outputs=["y"], + epsilon=epsilon, +) + +# output size: (2, 3, 4, 5) +expect( + node, + inputs=[x, s, bias, mean, var], + outputs=[y], + name="test_batchnorm_epsilon", +) +``` + +
+ + +
+train + +```python +# input size: (2, 3, 4, 5) +x = np.random.randn(2, 3, 4, 5).astype(np.float32) +s = np.random.randn(3).astype(np.float32) +bias = np.random.randn(3).astype(np.float32) +mean = np.random.randn(3).astype(np.float32) +var = np.random.rand(3).astype(np.float32) +# using np.bool(1) while generating test data with "'bool' object has no attribute 'dtype'" +# working around by using np.byte(1).astype(bool) +training_mode = 1 +y, output_mean, output_var = _batchnorm_training_mode(x, s, bias, mean, var) + +node = onnx.helper.make_node( + "BatchNormalization", + inputs=["x", "s", "bias", "mean", "var"], + outputs=["y", "output_mean", "output_var"], + training_mode=training_mode, +) + +# output size: (2, 3, 4, 5) +expect( + node, + inputs=[x, s, bias, mean, var], + outputs=[y, output_mean, output_var], + name="test_batchnorm_example_training_mode", +) + +# input size: (2, 3, 4, 5) +x = np.random.randn(2, 3, 4, 5).astype(np.float32) +s = np.random.randn(3).astype(np.float32) +bias = np.random.randn(3).astype(np.float32) +mean = np.random.randn(3).astype(np.float32) +var = np.random.rand(3).astype(np.float32) +training_mode = 1 +momentum = 0.9 +epsilon = 1e-2 +y, output_mean, output_var = _batchnorm_training_mode( + x, s, bias, mean, var, momentum, epsilon +) + +node = onnx.helper.make_node( + "BatchNormalization", + inputs=["x", "s", "bias", "mean", "var"], + outputs=["y", "output_mean", "output_var"], + epsilon=epsilon, + training_mode=training_mode, +) + +# output size: (2, 3, 4, 5) +expect( + node, + inputs=[x, s, bias, mean, var], + outputs=[y, output_mean, output_var], + name="test_batchnorm_epsilon_training_mode", +) +``` + +
+ + +### **Bernoulli** + + Draws binary random numbers (0 or 1) from a Bernoulli distribution. The input tensor should be a tensor + containing probabilities p (a value in the range [0,1]) to be used for drawing the binary random number, + where an output of 1 is produced with probability p and an output of 0 is produced with probability (1-p). + + This operator is non-deterministic and may not produce the same values in different + implementations (even if a seed is specified). + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 15 + +#### Attributes + +
+
dtype : int
+
The data type for the elements of the output tensor. if not specified, we will use the data type of the input tensor.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
All values in input have to be in the range:[0, 1].
+
+ +#### Outputs + +
+
output : T2
+
The returned output tensor only has values 0 or 1, same shape as input tensor.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool)
+
Constrain output types to all numeric tensors and bool tensors.
+
+ + +#### Examples + +
+bernoulli_with_dtype + +```python +node = onnx.helper.make_node( + "Bernoulli", + inputs=["x"], + outputs=["y"], + dtype=onnx.TensorProto.DOUBLE, +) + +x = np.random.uniform(0.0, 1.0, 10).astype(np.float32) +y = bernoulli_reference_implementation(x, float) +expect(node, inputs=[x], outputs=[y], name="test_bernoulli_double") +``` + +
+ + +
+bernoulli_with_seed + +```python +seed = float(0) +node = onnx.helper.make_node( + "Bernoulli", + inputs=["x"], + outputs=["y"], + seed=seed, +) + +x = np.random.uniform(0.0, 1.0, 10).astype(np.float32) +y = bernoulli_reference_implementation(x, np.float32) +expect(node, inputs=[x], outputs=[y], name="test_bernoulli_seed") +``` + +
+ + +
+bernoulli_without_dtype + +```python +node = onnx.helper.make_node( + "Bernoulli", + inputs=["x"], + outputs=["y"], +) + +x = np.random.uniform(0.0, 1.0, 10).astype(float) +y = bernoulli_reference_implementation(x, float) +expect(node, inputs=[x], outputs=[y], name="test_bernoulli") +``` + +
+ + +### **BitCast** + + Reinterprets the binary representation of a tensor as a different data type, + specified by the 'to' attribute. Unlike Cast, BitCast preserves the exact bit + pattern without any value conversion. + + The target data type must have the same bit-width as the input data type. + The output tensor has the same shape as the input tensor. + All types except string are supported. Implementations must treat the + underlying bytes as little endian. + +#### Version + +This version of the operator has been available since version 26 of the default ONNX operator set. + +#### Attributes + +
+
to : int (required)
+
The data type to which the input tensor is bitwise reinterpreted. Must be one of the non-string types from DataType enum in TensorProto. The target type must have the same bit-width as the input type.
+
+ +#### Inputs + +
+
input (non-differentiable) : T1
+
Input tensor to be bitcast.
+
+ +#### Outputs + +
+
output (non-differentiable) : T2
+
Output tensor with the same shape as the input.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input types. Bitcasting from string is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain output types. Bitcasting to string is not supported.
+
+ + +#### Examples + +
+bitcast_2d_float32_to_int32 + +```python +"""Test bitcasting 2D array from float32 to int32.""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT32, +) +x = np.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]], dtype=np.float32) +y = x.view(np.int32) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_2d_float32_to_int32") +``` + +
+ + +
+bitcast_bool_to_uint8 + +```python +"""Test bitcasting from bool to uint8 (same size).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.UINT8, +) +x = np.array([True, False, True, False], dtype=np.bool_) +y = x.view(np.uint8) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_bool_to_uint8") +``` + +
+ + +
+bitcast_float32_to_int32 + +```python +"""Test bitcasting from float32 to int32 (same size).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT32, +) +x = np.array([1.0, -2.5, 3.75], dtype=np.float32) +y = x.view(np.int32) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_float32_to_int32") +``` + +
+ + +
+bitcast_float64_to_int64 + +```python +"""Test bitcasting from float64 to int64 (same size).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT64, +) +x = np.array([1.0, -2.5, 3.75], dtype=np.float64) +y = x.view(np.int64) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_float64_to_int64") +``` + +
+ + +
+bitcast_int32_to_float32 + +```python +"""Test bitcasting from int32 to float32 (same size).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.FLOAT, +) +x = np.array([1065353216, -1071644672, 1081081856], dtype=np.int32) +y = x.view(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_int32_to_float32") +``` + +
+ + +
+bitcast_int64_to_float64 + +```python +"""Test bitcasting from int64 to float64 (same size).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.DOUBLE, +) +x = np.array( + [4607182418800017408, -4611686018427387904, 4614256656552045184], + dtype=np.int64, +) +y = x.view(np.float64) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_int64_to_float64") +``` + +
+ + +
+bitcast_int8_to_uint8 + +```python +"""Test bitcasting from int8 to uint8 (same size, different signedness).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.UINT8, +) +x = np.array([-1, -128, 127, 0], dtype=np.int8) +y = x.view(np.uint8) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_int8_to_uint8") +``` + +
+ + +
+bitcast_scalar_float32_to_int32 + +```python +"""Test bitcasting scalar from float32 to int32.""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT32, +) +x = np.array(1.0, dtype=np.float32) +y = x.view(np.int32) +expect( + node, inputs=[x], outputs=[y], name="test_bitcast_scalar_float32_to_int32" +) +``` + +
+ + +
+bitcast_uint16_to_int16 + +```python +"""Test bitcasting from uint16 to int16 (same size, different signedness).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT16, +) +x = np.array([1, 32768, 65535], dtype=np.uint16) +y = x.view(np.int16) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_uint16_to_int16") +``` + +
+ + +
+bitcast_uint32_to_int32 + +```python +"""Test bitcasting from uint32 to int32 (same size, different signedness).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT32, +) +x = np.array([4294967295, 2147483648, 2147483647], dtype=np.uint32) +y = x.view(np.int32) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_uint32_to_int32") +``` + +
+ + +### **BitShift** + + Bitwise shift operator performs element-wise operation. For each input element, if the + attribute "direction" is "RIGHT", this operator moves its binary representation toward + the right side so that the input value is effectively decreased. If the attribute "direction" + is "LEFT", bits of binary representation moves toward the left side, which results the + increase of its actual value. The input X is the tensor to be shifted and another input + Y specifies the amounts of shifting. For example, if "direction" is "Right", X is [1, 4], + and S is [1, 1], the corresponding output Z would be [0, 2]. If "direction" is "LEFT" with + X=[1, 2] and S=[1, 2], the corresponding output Y would be [2, 8]. + + Because this operator supports Numpy-style broadcasting, X's and Y's shapes are + not necessarily identical. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
direction : string (required)
+
Direction of moving bits. It can be either "RIGHT" (for right shift) or "LEFT" (for left shift).
+
+ +#### Inputs + +
+
X (non-differentiable) : T
+
First operand, input to be shifted.
+
Y (non-differentiable) : T
+
Second operand, amounts of shift.
+
+ +#### Outputs + +
+
Z (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64)
+
Constrain input and output types to integer tensors.
+
+ + +#### Examples + +
+left_unit16 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="LEFT" +) + +x = np.array([16, 4, 1]).astype(np.uint16) +y = np.array([1, 2, 3]).astype(np.uint16) +z = x << y # expected output [32, 16, 8] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_left_uint16") +``` + +
+ + +
+left_unit32 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="LEFT" +) + +x = np.array([16, 4, 1]).astype(np.uint32) +y = np.array([1, 2, 3]).astype(np.uint32) +z = x << y # expected output [32, 16, 8] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_left_uint32") +``` + +
+ + +
+left_unit64 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="LEFT" +) + +x = np.array([16, 4, 1]).astype(np.uint64) +y = np.array([1, 2, 3]).astype(np.uint64) +z = x << y # expected output [32, 16, 8] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_left_uint64") +``` + +
+ + +
+left_unit8 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="LEFT" +) + +x = np.array([16, 4, 1]).astype(np.uint8) +y = np.array([1, 2, 3]).astype(np.uint8) +z = x << y # expected output [32, 16, 8] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_left_uint8") +``` + +
+ + +
+right_unit16 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="RIGHT" +) + +x = np.array([16, 4, 1]).astype(np.uint16) +y = np.array([1, 2, 3]).astype(np.uint16) +z = x >> y # expected output [8, 1, 0] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_right_uint16") +``` + +
+ + +
+right_unit32 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="RIGHT" +) + +x = np.array([16, 4, 1]).astype(np.uint32) +y = np.array([1, 2, 3]).astype(np.uint32) +z = x >> y # expected output [8, 1, 0] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_right_uint32") +``` + +
+ + +
+right_unit64 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="RIGHT" +) + +x = np.array([16, 4, 1]).astype(np.uint64) +y = np.array([1, 2, 3]).astype(np.uint64) +z = x >> y # expected output [8, 1, 0] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_right_uint64") +``` + +
+ + +
+right_unit8 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="RIGHT" +) + +x = np.array([16, 4, 1]).astype(np.uint8) +y = np.array([1, 2, 3]).astype(np.uint8) +z = x >> y # expected output [8, 1, 0] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_right_uint8") +``` + +
+ + +### **BitwiseAnd** + + Returns the tensor resulting from performing the bitwise `and` operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the bitwise operator.
+
B (non-differentiable) : T
+
Second input operand for the bitwise operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64)
+
Constrain input to integer tensors.
+
+ + +#### Examples + +
+bitwiseand + +```python +node = onnx.helper.make_node( + "BitwiseAnd", + inputs=["x", "y"], + outputs=["bitwiseand"], +) + +# 2d +x = create_random_int((3, 4), np.int32) +y = create_random_int((3, 4), np.int32) +z = np.bitwise_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_and_i32_2d") + +# 3d +x = create_random_int((3, 4, 5), np.int16) +y = create_random_int((3, 4, 5), np.int16) +z = np.bitwise_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_and_i16_3d") +``` + +
+ + +
+bitwiseand_broadcast + +```python +node = onnx.helper.make_node( + "BitwiseAnd", + inputs=["x", "y"], + outputs=["bitwiseand"], +) + +# 3d vs 1d +x = create_random_int((3, 4, 5), np.uint64) +y = create_random_int((5,), np.uint64) +z = np.bitwise_and(x, y) +expect( + node, inputs=[x, y], outputs=[z], name="test_bitwise_and_ui64_bcast_3v1d" +) + +# 4d vs 3d +x = create_random_int((3, 4, 5, 6), np.uint8) +y = create_random_int((4, 5, 6), np.uint8) +z = np.bitwise_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_and_ui8_bcast_4v3d") +``` + +
+ + +### **BitwiseNot** + + Returns the bitwise not of the input tensor element-wise. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64)
+
Constrain input/output to integer tensors.
+
+ + +#### Examples + +
+bitwisenot + +```python +node = onnx.helper.make_node( + "BitwiseNot", + inputs=["x"], + outputs=["bitwise_not"], +) + +# 2d +x = create_random_int((3, 4), np.int32) +y = np.bitwise_not(x) +expect(node, inputs=[x], outputs=[y], name="test_bitwise_not_2d") + +# 3d +x = create_random_int((3, 4, 5), np.uint16) +y = np.bitwise_not(x) +expect(node, inputs=[x], outputs=[y], name="test_bitwise_not_3d") + +# 4d +x = create_random_int((3, 4, 5, 6), np.uint8) +y = np.bitwise_not(x) +expect(node, inputs=[x], outputs=[y], name="test_bitwise_not_4d") +``` + +
+ + +### **BitwiseOr** + + Returns the tensor resulting from performing the bitwise `or` operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the bitwise operator.
+
B (non-differentiable) : T
+
Second input operand for the bitwise operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64)
+
Constrain input to integer tensors.
+
+ + +#### Examples + +
+bitwiseor + +```python +node = onnx.helper.make_node( + "BitwiseOr", + inputs=["x", "y"], + outputs=["bitwiseor"], +) +# 2d +x = create_random_int((3, 4), np.int32) +y = create_random_int((3, 4), np.int32) +z = np.bitwise_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_or_i32_2d") + +# 4d +x = create_random_int((3, 4, 5, 6), np.int8) +y = create_random_int((3, 4, 5, 6), np.int8) +z = np.bitwise_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_or_i16_4d") +``` + +
+ + +
+bitwiseor_broadcast + +```python +node = onnx.helper.make_node( + "BitwiseOr", + inputs=["x", "y"], + outputs=["bitwiseor"], +) + +# 3d vs 1d +x = create_random_int((3, 4, 5), np.uint64) +y = create_random_int((5,), np.uint64) +z = np.bitwise_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_or_ui64_bcast_3v1d") + +# 4d vs 3d +x = create_random_int((3, 4, 5, 6), np.uint8) +y = create_random_int((4, 5, 6), np.uint8) +z = np.bitwise_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_or_ui8_bcast_4v3d") +``` + +
+ + +### **BitwiseXor** + + Returns the tensor resulting from performing the bitwise `xor` operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the bitwise operator.
+
B (non-differentiable) : T
+
Second input operand for the bitwise operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64)
+
Constrain input to integer tensors.
+
+ + +#### Examples + +
+bitwiseor_broadcast + +```python +node = onnx.helper.make_node( + "BitwiseXor", + inputs=["x", "y"], + outputs=["bitwisexor"], +) + +# 3d vs 1d +x = create_random_int((3, 4, 5), np.uint64) +y = create_random_int((5,), np.uint64) +z = np.bitwise_xor(x, y) +expect( + node, inputs=[x, y], outputs=[z], name="test_bitwise_xor_ui64_bcast_3v1d" +) + +# 4d vs 3d +x = create_random_int((3, 4, 5, 6), np.uint8) +y = create_random_int((4, 5, 6), np.uint8) +z = np.bitwise_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_xor_ui8_bcast_4v3d") +``` + +
+ + +
+bitwisexor + +```python +node = onnx.helper.make_node( + "BitwiseXor", + inputs=["x", "y"], + outputs=["bitwisexor"], +) + +# 2d +x = create_random_int((3, 4), np.int32) +y = create_random_int((3, 4), np.int32) +z = np.bitwise_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_xor_i32_2d") + +# 3d +x = create_random_int((3, 4, 5), np.int16) +y = create_random_int((3, 4, 5), np.int16) +z = np.bitwise_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_xor_i16_3d") +``` + +
+ + +### **BlackmanWindow** + + Generates a Blackman window as described in the paper https://ieeexplore.ieee.org/document/1455106. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
output_datatype : int (default is 1)
+
The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT.
+
periodic : int (default is 1)
+
If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When 'periodic' is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1.
+
+ +#### Inputs + +
+
size (non-differentiable) : T1
+
A scalar value indicating the length of the window.
+
+ +#### Outputs + +
+
output (non-differentiable) : T2
+
A Blackman window with length: size. The output has the shape: [size].
+
+ +#### Type Constraints + +
+
T1 : tensor(int32), tensor(int64)
+
Constrain the input size to int32_t or int64_t.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain output types to numeric tensors.
+
+ + +#### Examples + +
+blackmanwindow + +```python +# Test periodic window +node = onnx.helper.make_node( + "BlackmanWindow", + inputs=["x"], + outputs=["y"], +) +size = np.int32(10) +a0 = 0.42 +a1 = -0.5 +a2 = 0.08 +y = a0 +y += a1 * np.cos(2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / size) +y += a2 * np.cos(4 * np.pi * np.arange(0, size, 1, dtype=np.float32) / size) +expect( + node, + inputs=[size], + outputs=[y.astype(np.float32)], + name="test_blackmanwindow", +) + +# Test symmetric window +node = onnx.helper.make_node( + "BlackmanWindow", inputs=["x"], outputs=["y"], periodic=0 +) +size = np.int32(10) +a0 = 0.42 +a1 = -0.5 +a2 = 0.08 +y = a0 +y += a1 * np.cos( + 2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / (size - 1) +) +y += a2 * np.cos( + 4 * np.pi * np.arange(0, size, 1, dtype=np.float32) / (size - 1) +) +expect( + node, + inputs=[size], + outputs=[y.astype(np.float32)], + name="test_blackmanwindow_symmetric", +) +``` + +
+ + +### **Cast** + + The operator casts the elements of a given input tensor to a data type + specified by the 'to' argument and returns an output tensor of the same size in + the converted type. The 'to' argument must be one of the data types specified + in the 'DataType' enum field in the TensorProto message. + + Casting from string tensor in plain (e.g., "3.14" and "1000") and scientific numeric representations + (e.g., "1e-5" and "1E8") to float types is supported. For example, converting string "100.5" to an integer may + yield result 100. There are some string literals reserved for special floating-point values; + "+INF" (and "INF"), "-INF", and "NaN" are positive infinity, negative infinity, and not-a-number, respectively. + Any string which can exactly match "+INF" in a case-insensitive way would be mapped to positive infinite. Similarly, + this case-insensitive rule is applied to "INF" and "NaN". When casting from numeric tensors + to string tensors, plain floating-point representation (such as "314.15926") would be used. + Converting non-numerical-literal string such as "Hello World!" is an undefined behavior. Cases + of converting string representing floating-point arithmetic value, such as "2.718", to INT is an undefined behavior. + + Conversion from a numerical type to any numerical type is always allowed. + User must be aware of precision loss and value change caused by range difference between two types. + For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting + an integer 36 to Boolean may produce 1 because we truncate bits which can't be stored in the targeted type. + + In more detail, the conversion among numerical types should follow these rules + if the destination type is not a float 8 type. + + * Casting from floating point to: + * floating point: +/- infinity if OOR (out of range). + * fixed point: undefined if OOR. + * bool: +/- 0.0 to False; all else to True. + * Casting from fixed point to: + * floating point: +/- infinity if OOR. (+ infinity in the case of uint) + * fixed point: when OOR, discard higher bits and reinterpret (with respect to two's complement representation for + signed types). For example, 200 (int16) -> -56 (int8). + * bool: zero to False; nonzero to True. + * Casting from bool to: + * floating point: `{1.0, 0.0}`. + * fixed point: `{1, 0}`. + * bool: no change. + + Float 8 types (E4M3FN, E4M3FNUZ, E5M2, E5M2FNUZ) were introduced to speed up the training of + deep models. By default the conversion of a float *x* obeys + to the following rules. `[x]` means the value rounded to + the target mantissa width. + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | -------- | -------- | -------- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | Inf | FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | + | -Inf | -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | + | \[x\] > FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | FLT_MAX | + | \[x\] \< -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | -FLT_MAX | + | else | RNE | RNE | RNE | RNE | + + The behavior changes if the parameter 'saturate' is set to False. + The rules then become: + + | x | E4M3FN | E4M3FNUZ | E5M2 | E5M2FNUZ | + | ----------------- | ------ | -------- | ---- | -------- | + | 0 | 0 | 0 | 0 | 0 | + | -0 | -0 | 0 | -0 | 0 | + | NaN | NaN | NaN | NaN | NaN | + | -NaN | -NaN | NaN | -NaN | NaN | + | Inf | NaN | NaN | Inf | NaN | + | -Inf | -NaN | NaN | -Inf | NaN | + | \[x\] > FLT_MAX | NaN | NaN | Inf | NaN | + | \[x\] \< -FLT_MAX | NaN | NaN | -Inf | NaN | + | else | RNE | RNE | RNE | RNE | + + FLOAT8E8M0 type was introduced to enable [Microscaling (MX) formats](https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf). + When casting to FLOAT8E8M0, the rounding behavior can be specified using the `round_mode` and `saturate` attributes. + The current CUDA behavior is to round up and saturate. Casting negative values to FLOAT8E8M0 gives undefined behavior. + The following table describes the casting behavior of special values to FLOAT8E8M0 in the two most common cases. + + | x | saturate + up | non-saturate + nearest | + | ----------------- | ------------- | --------------------- | + | 0 | 0 | NaN | + | -0 | Unspecified | Unspecified | + | NaN | NaN | NaN | + | Inf | E8M0_MAX | NaN | + | x > E8M0_MAX | E8M0_MAX | NaN | + | x < E8M0_MIN | E8M0_MIN | NaN | + | x < 0 | Unspecified | Unspecified | + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 9, 13, 19, 21, 23, 24 + +#### Attributes + +
+
round_mode : string (default is up)
+
Rounding mode for conversion to float8e8m0. It only applies to casting to float8e8m0 and is `up` by default. `up`: round to nearest value away from zero, `down`: round to nearest value towards zero, `nearest`: round to nearest value and ties round up.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz, float8e8m0). It is true by default. All cases are fully described in the tables inserted in the operator description.
+
to : int (required)
+
The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor with the same shape as input with type specified by the 'to' argument
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain output types. Casting to complex is not supported.
+
+ + +#### Examples + +
+cast + +```python +test_cases = [ + ("FLOAT", "FLOAT16"), + ("FLOAT", "DOUBLE"), + ("FLOAT16", "FLOAT"), + ("FLOAT16", "DOUBLE"), + ("DOUBLE", "FLOAT"), + ("DOUBLE", "FLOAT16"), + ("FLOAT", "BFLOAT16"), + ("BFLOAT16", "FLOAT"), + ("FLOAT", "FLOAT8E4M3FN"), + ("FLOAT16", "FLOAT8E4M3FN"), + ("FLOAT", "FLOAT8E4M3FNUZ"), + ("FLOAT16", "FLOAT8E4M3FNUZ"), + ("FLOAT8E4M3FN", "FLOAT"), + ("FLOAT8E4M3FN", "FLOAT16"), + ("FLOAT8E4M3FNUZ", "FLOAT"), + ("FLOAT8E4M3FNUZ", "FLOAT16"), + ("FLOAT", "FLOAT8E5M2"), + ("FLOAT16", "FLOAT8E5M2"), + ("FLOAT", "FLOAT8E5M2FNUZ"), + ("FLOAT16", "FLOAT8E5M2FNUZ"), + ("FLOAT8E5M2", "FLOAT"), + ("FLOAT8E5M2", "FLOAT16"), + ("FLOAT8E5M2FNUZ", "FLOAT"), + ("FLOAT8E5M2FNUZ", "FLOAT16"), + ("FLOAT", "UINT4"), + ("FLOAT16", "UINT4"), + ("FLOAT", "INT4"), + ("FLOAT16", "INT4"), + ("UINT4", "FLOAT"), + ("UINT4", "FLOAT16"), + ("UINT4", "UINT8"), + ("INT4", "FLOAT"), + ("INT4", "FLOAT16"), + ("INT4", "INT8"), + ("FLOAT4E2M1", "FLOAT"), + ("FLOAT4E2M1", "FLOAT16"), + ("FLOAT", "FLOAT4E2M1"), + ("FLOAT16", "FLOAT4E2M1"), + ("FLOAT", "UINT2"), + ("FLOAT16", "UINT2"), + ("FLOAT", "INT2"), + ("FLOAT16", "INT2"), + ("UINT2", "FLOAT"), + ("UINT2", "FLOAT16"), + ("UINT2", "UINT8"), + ("INT2", "FLOAT"), + ("INT2", "FLOAT16"), + ("INT2", "INT8"), +] + +for from_type, to_type in test_cases: + if from_type == to_type: + # Skip cases where from_type and to_type are the same + continue + from_dtype = getattr(TensorProto, from_type) + to_dtype = getattr(TensorProto, to_type) + from_np_dtype = tensor_dtype_to_np_dtype(from_dtype) + to_np_dtype = tensor_dtype_to_np_dtype(to_dtype) + + if from_type == "BFLOAT16" or to_type == "BFLOAT16": + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.816468", + "0.21087195", + "0.7229038", + "NaN", + "INF", + "+INF", + "-INF", + ], + dtype=np.float32, + ) + input_shape = (3, 4) + + elif from_type in F8_TYPES or to_type in F8_TYPES: + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.7229038", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-0.0000001", + "0.0000001", + "-1000000", + ], + dtype=np.float32, + ) + input_shape = (3, 5) + elif from_type in ("UINT4", "INT4") or to_type in ("UINT4", "INT4"): + np_fp32 = np.arange(-9, 16).astype(np.float32) + input_shape = (5, 5) + elif from_type in ("UINT2", "INT2") or to_type in ("UINT2", "INT2"): + np_fp32 = np.arange(-3, 4).astype(np.float32) + input_shape = (7, 1) + elif from_type == "FLOAT4E2M1" or to_type == "FLOAT4E2M1": + np_fp32 = np.array( + [ + "0.48", + "0.25", + "1.05", + "-3.5", + "-8", + "9", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-4", + "0.01", + "-0.0", + ], + dtype=np.float32, + ) + input_shape = (3, 5) + + else: + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.816468", + "0.21087195", + "0.7229038", + "NaN", + "INF", + "+INF", + "-INF", + ], + dtype=np.float32, + ).reshape([3, 4]) + input_shape = (3, 4) + + if from_type in F8_TYPES: + np_from = onnx.numpy_helper.saturate_cast(np_fp32, from_np_dtype) + input = make_tensor( + "input", + from_dtype, + input_shape, + vals=np_from, + raw=True, + ) + elif from_type in FOUR_BIT_TYPES: + np_from = np_fp32.astype(from_np_dtype) + packed = onnx.numpy_helper._pack_4bitx2(np_from) + # No byteswap needed on big-endian machines as _pack_4bitx2() + # returns a numpy array with uint8 datatype. + input = make_tensor( + "input", from_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + elif from_type in TWO_BIT_TYPES: + np_from = np_fp32.astype(from_np_dtype) + packed = onnx.numpy_helper._pack_2bitx4(np_from) + input = make_tensor( + "input", from_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + else: + np_from = np_fp32.astype(from_np_dtype) + input = make_tensor( + "input", from_dtype, input_shape, vals=np_from, raw=True + ) + + if to_type in F8_TYPES: + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=onnx.numpy_helper.saturate_cast(np_from, to_np_dtype), + raw=True, + ) + elif to_type in FOUR_BIT_TYPES: + packed = onnx.numpy_helper._pack_4bitx2(np_from.astype(to_np_dtype)) + # No byteswap needed on big-endian machines as _pack_4bitx2() + # returns a numpy array with uint8 datatype. + output = make_tensor( + "output", to_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + elif to_type in TWO_BIT_TYPES: + packed = onnx.numpy_helper._pack_2bitx4(np_from.astype(to_np_dtype)) + output = make_tensor( + "output", to_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + else: + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=np_from.astype(to_np_dtype), + raw=True, + ) + + node = onnx.helper.make_node( + "Cast", + inputs=["input"], + outputs=["output"], + to=to_dtype, + ) + expect( + node, + inputs=[input], + outputs=[output], + name="test_cast_" + from_type + "_to_" + to_type, + ) +``` + +
+ + +
+e8m0 + +```python +np_fp32 = np.array( + [ + "0.0", + "0.124", + "0.25", + "0.5", + "1.1", + "2.0", + "4.0", + "8.0", + ], + dtype=np.float32, +) +test_cases = [ + ("FLOAT", "FLOAT8E8M0"), + ("FLOAT16", "FLOAT8E8M0"), + ("FLOAT8E8M0", "FLOAT"), + ("FLOAT8E8M0", "FLOAT16"), +] +for from_type, to_type in test_cases: + if from_type == "FLOAT": + input_np = np_fp32 + output_np = to_float8e8m0(np_fp32) + elif from_type == "FLOAT16": + input_np = np_fp32.astype(np.float16) + output_np = to_float8e8m0(input_np) + elif from_type == "FLOAT8E8M0": + input_np = to_float8e8m0(np_fp32) + if to_type == "FLOAT": + output_np = input_np.astype(np.float32) + elif to_type == "FLOAT16": + output_np = input_np.astype(np.float16) + else: + raise ValueError( + f"Conversion from {from_type} to {to_type} is not tested." + ) + else: + raise ValueError( + f"Conversion from {from_type} to {to_type} is not tested." + ) + input = make_tensor( + "input", + getattr(TensorProto, from_type), + [2, 4], + input_np, + raw=True, + ) + output = make_tensor( + "output", + getattr(TensorProto, to_type), + [2, 4], + output_np, + raw=True, + ) + if to_type == "FLOAT8E8M0": + node = onnx.helper.make_node( + "Cast", + inputs=["input"], + outputs=["output"], + to=getattr(TensorProto, to_type), + saturate=1, + round_mode="up", + ) + else: + node = onnx.helper.make_node( + "Cast", + inputs=["input"], + outputs=["output"], + to=getattr(TensorProto, to_type), + ) + + expect( + node, + inputs=[input], + outputs=[output], + name="test_cast_e8m0_" + from_type + "_to_" + to_type, + ) +``` + +
+ + +
+saturate_false + +```python +test_cases = itertools.product( + [ + "FLOAT", + "FLOAT16", + ], + [ + "FLOAT8E4M3FN", + "FLOAT8E4M3FNUZ", + "FLOAT8E5M2", + "FLOAT8E5M2FNUZ", + ], +) +input_shape = (3, 5) +for from_type, to_type in test_cases: + from_dtype = getattr(TensorProto, from_type) + to_dtype = getattr(TensorProto, to_type) + from_np_dtype = tensor_dtype_to_np_dtype(from_dtype) + to_np_dtype = tensor_dtype_to_np_dtype(to_dtype) + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.7229038", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-0.0000001", + "0.0000001", + "-1000000", + ], + dtype=np.float32, + ) + + input = make_tensor( + "input", + from_dtype, + input_shape, + vals=np_fp32.astype(from_np_dtype), + raw=True, + ) + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=np_fp32.astype(from_np_dtype).astype(to_np_dtype), + raw=True, + ) + + node = onnx.helper.make_node( + "Cast", + inputs=["input"], + outputs=["output"], + to=to_dtype, + saturate=0, + ) + expect( + node, + inputs=[input], + outputs=[output], + name="test_cast_no_saturate_" + from_type + "_to_" + to_type, + ) +``` + +
+ + +### **CastLike** + + The operator casts the elements of a given input tensor (the first input) to + the same data type as the elements of the second input tensor. + See documentation of the Cast operator for further details. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 15, 19, 21, 23, 24 + +#### Attributes + +
+
round_mode : string (default is up)
+
Rounding mode for conversion to float8e8m0. It only applies to casting to float8e8m0 and is `up` by default. `up`: round to nearest value away from zero, `down`: round to nearest value towards zero, `nearest`: round to nearest value and ties round up. Please refer to operator Cast description for further details.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz, float8e8m0). It is true by default. Please refer to operator Cast description for further details.
+
+ +#### Inputs + +
+
input (differentiable) : T1
+
Input tensor to be cast.
+
target_type (non-differentiable) : T2
+
The (first) input tensor will be cast to produce a tensor of the same type as this (second input) tensor.
+
+ +#### Outputs + +
+
output (differentiable) : T2
+
Output tensor produced by casting the first input tensor to have the same type as the second input tensor.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input types. Casting from complex is not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain output types. Casting to complex is not supported.
+
+ + +#### Examples + +
+castlike + +```python +test_cases = [ + ("FLOAT", "FLOAT16"), + ("FLOAT", "DOUBLE"), + ("FLOAT16", "FLOAT"), + ("FLOAT16", "DOUBLE"), + ("DOUBLE", "FLOAT"), + ("DOUBLE", "FLOAT16"), + ("FLOAT", "BFLOAT16"), + ("BFLOAT16", "FLOAT"), + ("FLOAT", "FLOAT8E4M3FN"), + ("FLOAT16", "FLOAT8E4M3FN"), + ("FLOAT", "FLOAT8E4M3FNUZ"), + ("FLOAT16", "FLOAT8E4M3FNUZ"), + ("FLOAT8E4M3FN", "FLOAT"), + ("FLOAT8E4M3FN", "FLOAT16"), + ("FLOAT8E4M3FNUZ", "FLOAT"), + ("FLOAT8E4M3FNUZ", "FLOAT16"), + ("FLOAT", "FLOAT8E5M2"), + ("FLOAT16", "FLOAT8E5M2"), + ("FLOAT", "FLOAT8E5M2FNUZ"), + ("FLOAT16", "FLOAT8E5M2FNUZ"), + ("FLOAT8E5M2", "FLOAT"), + ("FLOAT8E5M2", "FLOAT16"), + ("FLOAT8E5M2FNUZ", "FLOAT"), + ("FLOAT8E5M2FNUZ", "FLOAT16"), + ("FLOAT", "UINT4"), + ("FLOAT16", "UINT4"), + ("FLOAT", "INT4"), + ("FLOAT16", "INT4"), + ("UINT4", "FLOAT"), + ("UINT4", "FLOAT16"), + ("UINT4", "UINT8"), + ("INT4", "FLOAT"), + ("INT4", "FLOAT16"), + ("INT4", "INT8"), + ("FLOAT4E2M1", "FLOAT"), + ("FLOAT4E2M1", "FLOAT16"), + ("FLOAT", "FLOAT4E2M1"), + ("FLOAT16", "FLOAT4E2M1"), + ("FLOAT", "UINT2"), + ("FLOAT16", "UINT2"), + ("FLOAT", "INT2"), + ("FLOAT16", "INT2"), + ("UINT2", "FLOAT"), + ("UINT2", "FLOAT16"), + ("UINT2", "UINT8"), + ("INT2", "FLOAT"), + ("INT2", "FLOAT16"), + ("INT2", "INT8"), +] + +f8_types = {"FLOAT8E4M3FN", "FLOAT8E4M3FNUZ", "FLOAT8E5M2", "FLOAT8E5M2FNUZ"} + +for from_type, to_type in test_cases: + if from_type == to_type: + # Skip cases where from_type and to_type are the same + continue + from_dtype = getattr(TensorProto, from_type) + to_dtype = getattr(TensorProto, to_type) + from_np_dtype = tensor_dtype_to_np_dtype(from_dtype) + to_np_dtype = tensor_dtype_to_np_dtype(to_dtype) + + if from_type == "BFLOAT16" or to_type == "BFLOAT16": + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.816468", + "0.21087195", + "0.7229038", + "NaN", + "INF", + "+INF", + "-INF", + ], + dtype=np.float32, + ) + input_shape = (3, 4) + + elif from_type in f8_types or to_type in f8_types: + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.7229038", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-0.0000001", + "0.0000001", + "-1000000", + ], + dtype=np.float32, + ) + input_shape = (3, 5) + elif from_type in ("UINT4", "INT4") or to_type in ("UINT4", "INT4"): + np_fp32 = np.arange(-9, 16).astype(np.float32) + input_shape = (5, 5) + elif from_type in ("UINT2", "INT2") or to_type in ("UINT2", "INT2"): + np_fp32 = np.arange(-3, 4).astype(np.float32) + input_shape = (7, 1) + elif from_type == "FLOAT4E2M1" or to_type == "FLOAT4E2M1": + np_fp32 = np.array( + [ + "0.48", + "0.25", + "1.05", + "-3.5", + "-8", + "9", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-4", + "0.01", + "-0.0", + ], + dtype=np.float32, + ) + input_shape = (3, 5) + + else: + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.816468", + "0.21087195", + "0.7229038", + "NaN", + "INF", + "+INF", + "-INF", + ], + dtype=np.float32, + ).reshape([3, 4]) + input_shape = (3, 4) + + if from_type in F8_TYPES: + np_from = onnx.numpy_helper.saturate_cast(np_fp32, from_np_dtype) + input = make_tensor( + "input", + from_dtype, + input_shape, + vals=np_from, + raw=True, + ) + elif from_type in FOUR_BIT_TYPES: + np_from = np_fp32.astype(from_np_dtype) + packed = onnx.numpy_helper._pack_4bitx2(np_from) + # No byteswap needed on big-endian machines as _pack_4bitx2() + # returns a numpy array with uint8 datatype. + input = make_tensor( + "input", from_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + elif from_type in TWO_BIT_TYPES: + np_from = np_fp32.astype(from_np_dtype) + packed = onnx.numpy_helper._pack_2bitx4(np_from) + # No byteswap needed on big-endian machines as _pack_2bitx4() + # returns a numpy array with uint8 datatype. + input = make_tensor( + "input", from_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + else: + np_from = np_fp32.astype(from_np_dtype) + input = make_tensor( + "input", from_dtype, input_shape, vals=np_from, raw=True + ) + + if to_type in F8_TYPES: + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=onnx.numpy_helper.saturate_cast(np_from, to_np_dtype), + raw=True, + ) + elif to_type in FOUR_BIT_TYPES: + packed = onnx.numpy_helper._pack_4bitx2(np_from.astype(to_np_dtype)) + # No byteswap needed on big-endian machines as _pack_4bitx2() + # returns a numpy array with uint8 datatype. + output = make_tensor( + "output", to_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + elif to_type in TWO_BIT_TYPES: + packed = onnx.numpy_helper._pack_2bitx4(np_from.astype(to_np_dtype)) + # No byteswap needed on big-endian machines as _pack_2bitx4() + # returns a numpy array with uint8 datatype. + output = make_tensor( + "output", to_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + else: + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=np_from.astype(to_np_dtype), + raw=True, + ) + + like = make_tensor("like", to_dtype, (0,), vals=[]) + + node = onnx.helper.make_node( + "CastLike", + inputs=["input", "like"], + outputs=["output"], + ) + + expect( + node, + inputs=[input, like], + outputs=[output], + name="test_castlike_" + from_type + "_to_" + to_type, + ) +``` + +
+ + +
+saturate_false + +```python +test_cases = itertools.product( + [ + "FLOAT", + "FLOAT16", + ], + [ + "FLOAT8E4M3FN", + "FLOAT8E4M3FNUZ", + "FLOAT8E5M2", + "FLOAT8E5M2FNUZ", + ], +) +input_shape = (3, 5) +for from_type, to_type in test_cases: + from_dtype = getattr(TensorProto, from_type) + to_dtype = getattr(TensorProto, to_type) + from_np_dtype = tensor_dtype_to_np_dtype(from_dtype) + to_np_dtype = tensor_dtype_to_np_dtype(to_dtype) + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.7229038", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-0.0000001", + "0.0000001", + "-1000000", + ], + dtype=np.float32, + ) + + input = make_tensor( + "input", + from_dtype, + input_shape, + vals=np_fp32.astype(from_np_dtype), + raw=True, + ) + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=np_fp32.astype(from_np_dtype).astype(to_np_dtype), + raw=True, + ) + + like = make_tensor("like", to_dtype, (0,), vals=[]) + + node = onnx.helper.make_node( + "CastLike", + inputs=["input", "like"], + outputs=["output"], + saturate=0, + ) + + expect( + node, + inputs=[input, like], + outputs=[output], + name="test_castlike_no_saturate_" + from_type + "_to_" + to_type, + ) +``` + +
+ + +### **CausalConvWithState** + + Stateful causal 1D depthwise convolution. + + Used by Gated DeltaNet (Qwen3.5) and Mamba (Jamba, FalconMamba) as a preprocessing step. + Replaces the 3-op pattern (Concat + Conv + Slice) with a single fused operation. + + The convolution is causal (looks only at current and past positions) and depthwise + (each channel is convolved independently with its own kernel). + + The input, weight, past_state, output, and present_state tensors are rank-3 with + shape (batch_size, channels, length). The optional bias input is rank-1 with + shape (channels). For higher-dimensional data, use Reshape nodes before and + after this operator to pack extra dimensions into the batch or channel axis. + + Weight layout: (channels, 1, k) for depthwise convolution. + The carry state stores the last (k-1) positions for incremental decode. + + The optional activation attribute supports fused SiLU/Swish activation. + + +#### Version + +This version of the operator has been available since version 27 of the default ONNX operator set. + +#### Attributes + +
+
activation : string (default is none)
+
Fused activation function. One of: 'silu', 'swish', 'none'. Default is 'none'.
+
+ +#### Inputs (2 - 4) + +
+
input (differentiable) : T
+
Input tensor with shape (batch_size, channels, length). Channels-first layout.
+
weight (differentiable) : T
+
Depthwise convolution kernel with shape (channels, 1, k) where k is the kernel size. The middle dim of size 1 follows the ONNX `Conv` weight layout `(M, C/group, k1, ..., kn)`: since this op is always depthwise, `group = channels`, so `C/group = 1`. Keeping this layout makes the weight tensor a drop-in for a depthwise `Conv(group=channels)` weight, so `Conv` <-> `CausalConvWithState` rewrites require no reshape.
+
bias (optional, differentiable) : T
+
Optional per-channel bias with shape (channels).
+
past_state (optional, non-differentiable) : T
+
Carry state from previous step with shape (batch_size, channels, k - 1). If not provided, padding is zero.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Convolution output with same shape as input.
+
present_state (non-differentiable) : T
+
Updated carry state with shape (batch_size, channels, k - 1). Contains the last (k - 1) values of the effective padded/concatenated sequence along the causal axis, including any values from past_state or zero-padding when the current input is shorter than k - 1.
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(float16), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+b1_c1_degenerate + +```python +# Mamba/GDN inner-head edge case: B=1, C=1. +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 1, 1, 6, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight) + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_b1_c1_degenerate", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+basic + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight) + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_basic", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+decode_step + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight", "bias", "past_state"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 1, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) +bias = np.random.randn(channels).astype(np.float32) +past_state = np.random.randn(batch_size, channels, k - 1).astype(np.float32) + +output, present_state = _compute( + input_, weight, bias=bias, past_state=past_state +) + +expect( + node, + inputs=[input_, weight, bias, past_state], + outputs=[output, present_state], + name="test_causal_conv_with_state_decode_step", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+fp16 + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.rand(batch_size, channels, length).astype(np.float16) +weight = np.random.rand(channels, 1, k).astype(np.float16) + +output, present_state = _compute(input_, weight) + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_fp16", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+kernel_size_one + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 1 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight) + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_kernel_size_one", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+short_input_no_past_state + +```python +# L < k-1 with no past_state: zero-pad is wider than the input. +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 2, 5 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight) + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_short_input_no_past_state", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+silu + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], + activation="silu", +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight, activation="silu") + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_silu", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+silu_fp16 + +```python +# fp16 + SiLU: the reference upcasts Sigmoid/Mul to float32, so the +# function-body expansion must do the same to stay numerically faithful. +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], + activation="silu", +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.rand(batch_size, channels, length).astype(np.float16) +weight = np.random.rand(channels, 1, k).astype(np.float16) + +output, present_state = _compute(input_, weight, activation="silu") + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_silu_fp16", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+silu_with_past_state + +```python +# Fused activation combined with concat-from-past variant of PaddedInput. +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight", "", "past_state"], + outputs=["output", "present_state"], + activation="silu", +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) +past_state = np.random.randn(batch_size, channels, k - 1).astype(np.float32) + +output, present_state = _compute( + input_, weight, past_state=past_state, activation="silu" +) + +expect( + node, + inputs=[input_, weight, past_state], + outputs=[output, present_state], + name="test_causal_conv_with_state_silu_with_past_state", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+swish_alias + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], + activation="swish", +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight, activation="swish") + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_swish_alias", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+with_bias + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight", "bias"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) +bias = np.random.randn(channels).astype(np.float32) + +output, present_state = _compute(input_, weight, bias=bias) + +expect( + node, + inputs=[input_, weight, bias], + outputs=[output, present_state], + name="test_causal_conv_with_state_with_bias", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+with_bias_and_past_state + +```python +# Multi-token (T>1) path through Concat(past, input) -> Conv(+bias). +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight", "bias", "past_state"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) +bias = np.random.randn(channels).astype(np.float32) +past_state = np.random.randn(batch_size, channels, k - 1).astype(np.float32) + +output, present_state = _compute( + input_, weight, bias=bias, past_state=past_state +) + +expect( + node, + inputs=[input_, weight, bias, past_state], + outputs=[output, present_state], + name="test_causal_conv_with_state_with_bias_and_past_state", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +
+with_past_state + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight", "", "past_state"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) +past_state = np.random.randn(batch_size, channels, k - 1).astype(np.float32) + +output, present_state = _compute(input_, weight, past_state=past_state) + +expect( + node, + inputs=[input_, weight, past_state], + outputs=[output, present_state], + name="test_causal_conv_with_state_with_past_state", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +### **Ceil** + + Ceil takes one input data (Tensor) and produces one output data + (Tensor) where the ceil is, y = ceil(x), is applied to + the tensor elementwise. If x is integral, +0, -0, NaN, or infinite, x itself is returned. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+ceil + +```python +node = onnx.helper.make_node( + "Ceil", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.5, 1.2]).astype(np.float32) +y = np.ceil(x) # expected output [-1., 2.] +expect(node, inputs=[x], outputs=[y], name="test_ceil_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.ceil(x) +expect(node, inputs=[x], outputs=[y], name="test_ceil") +``` + +
+ + +### **Celu** + + Continuously Differentiable Exponential Linear Units: + Perform the linear unit element-wise on the input tensor X + using formula: + + ``` + max(0,x) + min(0,alpha*(exp(x/alpha)-1)) + ``` + +#### Version + +This version of the operator has been available since version 28 of the default ONNX operator set. + +Other versions of this operator: 12 + +#### Attributes + +
+
alpha : float (default is 1.0)
+
The Alpha value in Celu formula which control the shape of the unit. The default value is 1.0.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+celu + +```python +alpha = 2.0 +node = onnx.helper.make_node( + "Celu", + inputs=["X"], + outputs=["Y"], + alpha=alpha, +) + +input_data = np.array( + [ + [ + [[0.8439683], [0.5665144], [0.05836735]], + [[0.02916367], [0.12964272], [0.5060197]], + [[0.79538304], [0.9411346], [0.9546573]], + ], + [ + [[0.17730942], [0.46192095], [0.26480448]], + [[0.6746842], [0.01665257], [0.62473077]], + [[0.9240844], [0.9722341], [0.11965699]], + ], + [ + [[0.41356155], [0.9129373], [0.59330076]], + [[0.81929934], [0.7862604], [0.11799799]], + [[0.69248444], [0.54119414], [0.07513223]], + ], + ], + dtype=np.float32, +) + +# Calculate expected output data +positive_input = np.maximum(0, input_data) +negative_input = np.minimum(0, alpha * (np.exp(input_data / alpha) - 1)) +expected_output = positive_input + negative_input + +expect(node, inputs=[input_data], outputs=[expected_output], name="test_celu") +``` + +
+ + +
+celu_bfloat16 + +```python +alpha = 2.0 +node = onnx.helper.make_node( + "Celu", + inputs=["X"], + outputs=["Y"], + alpha=alpha, +) + +input_data = np.array([-3.0, -0.5, 0.0, 0.5, 3.0], dtype=ml_dtypes.bfloat16) + +positive_input = np.maximum(0, input_data) +negative_input = np.minimum(0, alpha * (np.exp(input_data / alpha) - 1)) +expected_output = (positive_input + negative_input).astype(ml_dtypes.bfloat16) + +expect( + node, + inputs=[input_data], + outputs=[expected_output], + name="test_celu_bfloat16", +) +``` + +
+ + +
+celu_float16 + +```python +alpha = 2.0 +node = onnx.helper.make_node( + "Celu", + inputs=["X"], + outputs=["Y"], + alpha=alpha, +) + +input_data = np.array([-3.0, -0.5, 0.0, 0.5, 3.0], dtype=np.float16) + +positive_input = np.maximum(0, input_data) +negative_input = np.minimum(0, alpha * (np.exp(input_data / alpha) - 1)) +expected_output = (positive_input + negative_input).astype(np.float16) + +expect( + node, + inputs=[input_data], + outputs=[expected_output], + name="test_celu_float16", +) +``` + +
+ + +### **CenterCropPad** + + Center crop or pad an input to given dimensions. + + The crop/pad dimensions can be specified for a subset of the `axes`; unspecified dimensions will remain unchanged. + + If the input dimensions are larger than the target crop dimensions, a centered cropping window will be extracted + from the input. The starting value for the cropping window is rounded down, which means that if the difference + between the input shape and the crop shape is odd, the cropping window will be shifted half a pixel to the left + of the input center. + + If the input dimensions are smaller than the target crop dimensions, the input will be padded equally on both sides + to center it in the output. In cases where the total number of padding pixels is odd, an additional pixel will be + added to the right side. + + The padding value used is zero. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
axes : list of ints
+
If provided, it specifies a subset of axes that 'shape' refer to. If not provided, all axes are assumed [0, 1, ..., r-1], where r = rank(data). Negative value means counting dimensions from the back. Accepted range is [-r, r-1], where r = rank(data). Behavior is undefined if an axis is repeated.
+
+ +#### Inputs + +
+
input_data (differentiable) : T
+
Input to extract the centered crop from.
+
shape (non-differentiable) : Tind
+
1-D tensor representing the cropping window dimensions.
+
+ +#### Outputs + +
+
output_data (differentiable) : T
+
Output data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ + +#### Examples + +
+center_crop_pad_crop + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], +) + +# First dim is even diff, second is uneven +x = np.random.randn(20, 10, 3).astype(np.float32) +shape = np.array([10, 7, 3], dtype=np.int64) +y = x[5:15, 1:8, :] + +expect(node, inputs=[x, shape], outputs=[y], name="test_center_crop_pad_crop") +``` + +
+ + +
+center_crop_pad_crop_and_pad + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], +) + +# Cropping on first dim, padding on second, third stays the same +x = np.random.randn(20, 8, 3).astype(np.float32) +shape = np.array([10, 10, 3], dtype=np.int64) +y = np.zeros([10, 10, 3], dtype=np.float32) +y[:, 1:9, :] = x[5:15, :, :] + +expect( + node, + inputs=[x, shape], + outputs=[y], + name="test_center_crop_pad_crop_and_pad", +) +``` + +
+ + +
+center_crop_pad_crop_axes_chw + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], + axes=[1, 2], +) + +# Cropping on second dim, padding on third, first stays the same +x = np.random.randn(3, 20, 8).astype(np.float32) +shape = np.array([10, 9], dtype=np.int64) +y = np.zeros([3, 10, 9], dtype=np.float32) +y[:, :, :8] = x[:, 5:15, :] + +expect( + node, + inputs=[x, shape], + outputs=[y], + name="test_center_crop_pad_crop_axes_chw", +) +``` + +
+ + +
+center_crop_pad_crop_axes_hwc + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], + axes=[0, 1], +) + +# Cropping on first dim, padding on second, third stays the same +x = np.random.randn(20, 8, 3).astype(np.float32) +shape = np.array([10, 9], dtype=np.int64) +y = np.zeros([10, 9, 3], dtype=np.float32) +y[:, :8, :] = x[5:15, :, :] + +expect( + node, + inputs=[x, shape], + outputs=[y], + name="test_center_crop_pad_crop_axes_hwc", +) +``` + +
+ + +
+center_crop_pad_crop_negative_axes_hwc + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], + axes=[-3, -2], +) + +# Cropping on first dim, padding on second, third stays the same +x = np.random.randn(20, 8, 3).astype(np.float32) +shape = np.array([10, 9], dtype=np.int64) +y = np.zeros([10, 9, 3], dtype=np.float32) +y[:, :8, :] = x[5:15, :, :] + +expect( + node, + inputs=[x, shape], + outputs=[y], + name="test_center_crop_pad_crop_negative_axes_hwc", +) +``` + +
+ + +
+center_crop_pad_pad + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], +) + +# First dim is even diff, second is uneven +x = np.random.randn(10, 7, 3).astype(np.float32) +shape = np.array([20, 10, 3], dtype=np.int64) +y = np.zeros([20, 10, 3], dtype=np.float32) +y[5:15, 1:8, :] = x + +expect(node, inputs=[x, shape], outputs=[y], name="test_center_crop_pad_pad") +``` + +
+ + +### **Clip** + + Clip operator limits the given input within an interval. The interval is + specified by the inputs 'min' and 'max'. They default to + numeric_limits::lowest() and numeric_limits::max(), respectively. + When 'min' is greater than 'max', the clip operator sets all the 'input' values to + the value of 'max'. Thus, this is equivalent to 'Min(max, Max(input, min))'. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 11, 12 + +#### Inputs (1 - 3) + +
+
input (differentiable) : T
+
Input tensor whose elements to be clipped
+
min (optional, non-differentiable) : T
+
Minimum value, under which element is replaced by min. It must be a scalar(tensor of empty shape).
+
max (optional, non-differentiable) : T
+
Maximum value, above which element is replaced by max. It must be a scalar(tensor of empty shape).
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor with clipped input elements
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ + +#### Examples + +
+clip + +```python +node = onnx.helper.make_node( + "Clip", + inputs=["x", "min", "max"], + outputs=["y"], +) + +x = np.array([-2, 0, 2]).astype(np.float32) +min_val = np.float32(-1) +max_val = np.float32(1) +y = np.clip(x, min_val, max_val) # expected output [-1., 0., 1.] +expect( + node, inputs=[x, min_val, max_val], outputs=[y], name="test_clip_example" +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, min_val, max_val) +expect(node, inputs=[x, min_val, max_val], outputs=[y], name="test_clip") +node = onnx.helper.make_node( + "Clip", + inputs=["x", "min", "max"], + outputs=["y"], +) + +min_val = np.float32(-5) +max_val = np.float32(5) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.array([-1, 0, 1]).astype(np.float32) +expect( + node, inputs=[x, min_val, max_val], outputs=[y], name="test_clip_inbounds" +) + +x = np.array([-6, 0, 6]).astype(np.float32) +y = np.array([-5, 0, 5]).astype(np.float32) +expect( + node, inputs=[x, min_val, max_val], outputs=[y], name="test_clip_outbounds" +) + +x = np.array([-1, 0, 6]).astype(np.float32) +y = np.array([-1, 0, 5]).astype(np.float32) +expect( + node, + inputs=[x, min_val, max_val], + outputs=[y], + name="test_clip_splitbounds", +) + +x = np.array([-2, 0, 6]).astype(np.float32) +y = np.array([1, 1, 1]).astype(np.float32) +min_val = np.float32(2) +max_val = np.float32(1) +expect( + node, + inputs=[x, min_val, max_val], + outputs=[y], + name="test_clip_min_greater_than_max", +) +``` + +
+ + +
+clip_default + +```python +node = onnx.helper.make_node( + "Clip", + inputs=["x", "min"], + outputs=["y"], +) +min_val = np.float32(0) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, min_val, np.inf) +expect(node, inputs=[x, min_val], outputs=[y], name="test_clip_default_min") + +no_min = "" # optional input, not supplied +node = onnx.helper.make_node( + "Clip", + inputs=["x", no_min, "max"], + outputs=["y"], +) +max_val = np.float32(0) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, -np.inf, max_val) +expect(node, inputs=[x, max_val], outputs=[y], name="test_clip_default_max") + +no_max = "" # optional input, not supplied +node = onnx.helper.make_node( + "Clip", + inputs=["x", no_min, no_max], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.array([-1, 0, 1]).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_clip_default_inbounds") +``` + +
+ + +
+clip_default_int8 + +```python +node = onnx.helper.make_node( + "Clip", + inputs=["x", "min"], + outputs=["y"], +) +min_val = np.int8(0) +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.clip(x, min_val, np.iinfo(np.int8).max) +expect( + node, inputs=[x, min_val], outputs=[y], name="test_clip_default_int8_min" +) + +no_min = "" # optional input, not supplied +node = onnx.helper.make_node( + "Clip", + inputs=["x", no_min, "max"], + outputs=["y"], +) +max_val = np.int8(0) +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.clip(x, np.iinfo(np.int8).min, max_val) +expect( + node, inputs=[x, max_val], outputs=[y], name="test_clip_default_int8_max" +) + +no_max = "" # optional input, not supplied +node = onnx.helper.make_node( + "Clip", + inputs=["x", no_min, no_max], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.int8) +y = np.array([-1, 0, 1]).astype(np.int8) +expect(node, inputs=[x], outputs=[y], name="test_clip_default_int8_inbounds") +``` + +
+ + +### **Col2Im** + + The operator rearranges column blocks back into a multidimensional image + + Col2Im behaves similarly to PyTorch's fold https://pytorch.org/docs/stable/generated/torch.nn.Fold.html, + but it only supports *batched* multi-dimensional image tensors. + Another implementation in Python with N-dimension support can be found at https://github.com/f-dangel/unfoldNd/. + + NOTE: + Although specifying image_shape looks redundant because it could be calculated from + convolution formulas, it is required as input for more advanced scenarios as explained + at PyTorch's implementation (https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/Col2Im.cpp#L10) + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +#### Attributes + +
+
dilations : list of ints
+
1-dimensional tensor with dilation value along each spatial axis of the image. If not present, the dilation defaults to 1 along each spatial axis of the image.
+
pads : list of ints
+
1-dimensional tensor with padding value for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin is the number of pixels added at the beginning of axis `i` and xi_end is the number of pixels added at the end of axis `i`. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
1-dimensional tensor with stride value along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input data tensor to be rearranged from column blocks back into an image. This is a 3-dimensional tensor containing [N, C * n-ary-product(block_shape), L], where N is batch dimension, C is image channel dimension and L is number of blocks.The blocks are enumerated in increasing lexicographic-order of their indices.For example, with an image-size 10*20 and block-size 9*18, there would be 2*3 blocks, enumerated in the order block(0, 0), block(0, 1), block(0, 2), block(1, 0), block(1, 1), block(1, 2).
+
image_shape (non-differentiable) : tensor(int64)
+
The shape of the spatial dimensions of the image after rearranging the column blocks.This is a 1-dimensional tensor with size of at least 2, containing the value [H_img, W_img] for a 2-D image or [dim_i1, dim_i2, ..., dim_iN] for a N-D image.
+
block_shape (non-differentiable) : tensor(int64)
+
The shape of the block to apply on the input.This is a 1-dimensional tensor of size of at least 2, containing the value [H_block, W_block] for a 2-D image or [dim_b1, dim_b2, ..., dim_bN] for a N-D block.This is the block-shape before dilation is applied to it.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor produced by rearranging blocks into an image.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all numeric tensor types.
+
+ + +#### Examples + +
+col2im + +```python +input = np.array( + [ + [ + [1.0, 6.0, 11.0, 16.0, 21.0], # (1, 5, 5) + [2.0, 7.0, 12.0, 17.0, 22.0], + [3.0, 8.0, 13.0, 18.0, 23.0], + [4.0, 9.0, 14.0, 19.0, 24.0], + [5.0, 0.0, 15.0, 20.0, 25.0], + ] + ] +).astype(np.float32) + +image_shape = np.array([5, 5]).astype(np.int64) +block_shape = np.array([1, 5]).astype(np.int64) +node = onnx.helper.make_node( + "Col2Im", ["input", "image_shape", "block_shape"], ["output"] +) + +output = np.array( + [ + [ + [ + [1.0, 2.0, 3.0, 4.0, 5.0], # (1, 1, 5, 5) + [6.0, 7.0, 8.0, 9.0, 0.0], + [11.0, 12.0, 13.0, 14.0, 15.0], + [16.0, 17.0, 18.0, 19.0, 20.0], + [21.0, 22.0, 23.0, 24.0, 25.0], + ] + ] + ] +).astype(np.float32) + +expect( + node, + inputs=[input, image_shape, block_shape], + outputs=[output], + name="test_col2im", +) +``` + +
+ + +
+col2im_5d + +```python +input = np.array( + [ + [ + [1, 6, 11, 16, 21, 26, 31, 36, 41, 46, 51, 56], # (1, 10, 12) + [2, 7, 12, 17, 22, 27, 32, 37, 42, 47, 52, 57], + [3, 8, 13, 18, 23, 28, 33, 38, 43, 48, 53, 58], + [4, 9, 14, 19, 24, 29, 34, 39, 44, 49, 54, 59], + [5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60], + [61, 66, 71, 76, 81, 86, 91, 96, 101, 106, 111, 116], + [62, 67, 72, 77, 82, 87, 92, 97, 102, 107, 112, 117], + [63, 68, 73, 78, 83, 88, 93, 98, 103, 108, 113, 118], + [64, 69, 74, 79, 84, 89, 94, 99, 104, 109, 114, 119], + [65, 70, 75, 80, 85, 90, 95, 100, 105, 110, 115, 120], + ] + ] +).astype(np.float32) +image_shape = np.array([3, 4, 5]).astype(np.int64) +block_shape = np.array([1, 1, 5]).astype(np.int64) + +output = np.array( + [ + [ + [ + [ + [1, 2, 3, 4, 5], # (1, 2, 3, 4, 5) + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + ], + [ + [21, 22, 23, 24, 25], + [26, 27, 28, 29, 30], + [31, 32, 33, 34, 35], + [36, 37, 38, 39, 40], + ], + [ + [41, 42, 43, 44, 45], + [46, 47, 48, 49, 50], + [51, 52, 53, 54, 55], + [56, 57, 58, 59, 60], + ], + ], + [ + [ + [61, 62, 63, 64, 65], + [66, 67, 68, 69, 70], + [71, 72, 73, 74, 75], + [76, 77, 78, 79, 80], + ], + [ + [81, 82, 83, 84, 85], + [86, 87, 88, 89, 90], + [91, 92, 93, 94, 95], + [96, 97, 98, 99, 100], + ], + [ + [101, 102, 103, 104, 105], + [106, 107, 108, 109, 110], + [111, 112, 113, 114, 115], + [116, 117, 118, 119, 120], + ], + ], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "Col2Im", ["input", "image_shape", "block_shape"], ["output"] +) +expect( + node, + inputs=[input, image_shape, block_shape], + outputs=[output], + name="test_col2im_5d", +) +``` + +
+ + +
+col2im_dilations + +```python +input = np.array( + [ + [ + [1.0, 5.0, 9.0, 13.0, 17], # (1, 4, 5) + [2.0, 6.0, 10.0, 14.0, 18], + [3.0, 7.0, 11.0, 15.0, 19], + [4.0, 8.0, 12.0, 16.0, 20], + ] + ] +).astype(np.float32) +image_shape = np.array([6, 6]).astype(np.int64) +block_shape = np.array([2, 2]).astype(np.int64) + +output = np.array( + [ + [ + [ + [1.0, 0.0, 0.0, 0.0, 0.0, 2.0], # (1, 1, 6, 6) + [8.0, 0.0, 0.0, 0.0, 0.0, 10.0], + [16.0, 0.0, 0.0, 0.0, 0.0, 18.0], + [24.0, 0.0, 0.0, 0.0, 0.0, 26.0], + [32.0, 0.0, 0.0, 0.0, 0.0, 34.0], + [19.0, 0.0, 0.0, 0.0, 0.0, 20.0], + ] + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "Col2Im", + ["input", "image_shape", "block_shape"], + ["output"], + dilations=[1, 5], +) +expect( + node, + inputs=[input, image_shape, block_shape], + outputs=[output], + name="test_col2im_dilations", +) +``` + +
+ + +
+col2im_pads + +```python +input = np.array( + [ + [ + [ + 1.0, + 6.0, + 11.0, + 16.0, + 21.0, + 26, + 31, + 36, + 41, + 46, + 51, + 56, + 61, + 66, + 71, + ], # (1, 5, 15) + [ + 2.0, + 7.0, + 12.0, + 17.0, + 22.0, + 27, + 32, + 37, + 42, + 47, + 52, + 57, + 62, + 67, + 72, + ], + [ + 3.0, + 8.0, + 13.0, + 18.0, + 23.0, + 28, + 33, + 38, + 43, + 48, + 53, + 58, + 63, + 68, + 73, + ], + [ + 4.0, + 9.0, + 14.0, + 19.0, + 24.0, + 29, + 34, + 39, + 44, + 49, + 54, + 59, + 64, + 69, + 74, + ], + [ + 5.0, + 10.0, + 15.0, + 20.0, + 25.0, + 30, + 35, + 40, + 45, + 50, + 55, + 60, + 65, + 70, + 75, + ], + ] + ] +).astype(np.float32) +image_shape = np.array([5, 5]).astype(np.int64) +block_shape = np.array([1, 5]).astype(np.int64) + +output = np.array( + [ + [ + [ + [8.0, 21.0, 24.0, 27.0, 24.0], # (1, 1, 5, 5) + [38.0, 66.0, 69.0, 72.0, 54.0], + [68.0, 111.0, 114.0, 117.0, 84.0], + [98.0, 156.0, 159.0, 162.0, 114.0], + [128.0, 201.0, 204.0, 207.0, 144.0], + ] + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "Col2Im", + ["input", "image_shape", "block_shape"], + ["output"], + pads=[0, 1, 0, 1], +) +expect( + node, + inputs=[input, image_shape, block_shape], + outputs=[output], + name="test_col2im_pads", +) +``` + +
+ + +
+col2im_strides + +```python +input = np.array( + [ + [ + [0.0, 0.0, 0.0, 0.0], # (1, 9, 4) + [1.0, 1.0, 1.0, 1.0], + [1.0, 1.0, 1.0, 1.0], + [1.0, 1.0, 1.0, 1.0], + [0.0, 0.0, 0.0, 0.0], + [0.0, 0.0, 0.0, 0.0], + [0.0, 0.0, 0.0, 0.0], + [1.0, 1.0, 1.0, 1.0], + [0.0, 0.0, 0.0, 0.0], + ] + ] +).astype(np.float32) +image_shape = np.array([5, 5]).astype(np.int64) +block_shape = np.array([3, 3]).astype(np.int64) + +output = np.array( + [ + [ + [ + [0.0, 1.0, 1.0, 1.0, 1.0], # (1, 1, 5, 5) + [1.0, 0.0, 1.0, 0.0, 0.0], + [0.0, 2.0, 1.0, 2.0, 1.0], + [1.0, 0.0, 1.0, 0.0, 0.0], + [0.0, 1.0, 0.0, 1.0, 0.0], + ] + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "Col2Im", + ["input", "image_shape", "block_shape"], + ["output"], + strides=[2, 2], +) +expect( + node, + inputs=[input, image_shape, block_shape], + outputs=[output], + name="test_col2im_strides", +) +``` + +
+ + +### **Compress** + + Selects slices from an input tensor along a given axis where condition evaluates to True for each axis index. + In case axis is not provided, input is flattened before elements are selected. + Compress behaves like numpy.compress: https://docs.scipy.org/doc/numpy/reference/generated/numpy.compress.html + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Attributes + +
+
axis : int
+
(Optional) Axis along which to take slices. If not specified, input is flattened before elements being selected. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Tensor of rank r >= 1.
+
condition (non-differentiable) : T1
+
Rank 1 tensor of booleans to indicate which slices or data elements to be selected. Its length can be less than the input length along the axis or the flattened input size if axis is not specified. In such cases data slices or elements exceeding the condition length are discarded.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank r if axis is specified. Otherwise output is a Tensor of rank 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
T1 : tensor(bool)
+
Constrain to boolean tensors.
+
+ + +#### Examples + +
+compress_0 + +```python +node = onnx.helper.make_node( + "Compress", + inputs=["input", "condition"], + outputs=["output"], + axis=0, +) +input = np.array([[1, 2], [3, 4], [5, 6]]).astype(np.float32) +condition = np.array([0, 1, 1]) +output = np.compress(condition, input, axis=0) +# print(output) +# [[ 3. 4.] +# [ 5. 6.]] + +expect( + node, + inputs=[input, condition.astype(bool)], + outputs=[output], + name="test_compress_0", +) +``` + +
+ + +
+compress_1 + +```python +node = onnx.helper.make_node( + "Compress", + inputs=["input", "condition"], + outputs=["output"], + axis=1, +) +input = np.array([[1, 2], [3, 4], [5, 6]]).astype(np.float32) +condition = np.array([0, 1]) +output = np.compress(condition, input, axis=1) +# print(output) +# [[ 2.] +# [ 4.] +# [ 6.]] + +expect( + node, + inputs=[input, condition.astype(bool)], + outputs=[output], + name="test_compress_1", +) +``` + +
+ + +
+compress_default_axis + +```python +node = onnx.helper.make_node( + "Compress", + inputs=["input", "condition"], + outputs=["output"], +) +input = np.array([[1, 2], [3, 4], [5, 6]]).astype(np.float32) +condition = np.array([0, 1, 0, 0, 1]) +output = np.compress(condition, input) +# print(output) +# [ 2., 5.] + +expect( + node, + inputs=[input, condition.astype(bool)], + outputs=[output], + name="test_compress_default_axis", +) +``` + +
+ + +
+compress_negative_axis + +```python +node = onnx.helper.make_node( + "Compress", + inputs=["input", "condition"], + outputs=["output"], + axis=-1, +) +input = np.array([[1, 2], [3, 4], [5, 6]]).astype(np.float32) +condition = np.array([0, 1]) +output = np.compress(condition, input, axis=-1) +# print(output) +# [[ 2.] +# [ 4.] +# [ 6.]] +expect( + node, + inputs=[input, condition.astype(bool)], + outputs=[output], + name="test_compress_negative_axis", +) +``` + +
+ + +### **Concat** + + Concatenate a list of tensors into a single tensor. All input tensors must have the same shape, except for the dimension size of the axis to concatenate on. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 4, 11 + +#### Attributes + +
+
axis : int (required)
+
Which axis to concat on. A negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(inputs)..
+
+ +#### Inputs (1 - ∞) + +
+
inputs (variadic, differentiable) : T
+
List of tensors for concatenation
+
+ +#### Outputs + +
+
concat_result (differentiable) : T
+
Concatenated tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain output types to any tensor type.
+
+ + +#### Examples + +
+concat + +```python +test_cases: dict[str, Sequence[Any]] = { + "1d": ([1, 2], [3, 4]), + "2d": ([[1, 2], [3, 4]], [[5, 6], [7, 8]]), + "3d": ( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]]], + [[[9, 10], [11, 12]], [[13, 14], [15, 16]]], + ), +} + +for test_case, values_ in test_cases.items(): + values = [np.asarray(v, dtype=np.float32) for v in values_] + for i in range(len(values[0].shape)): + in_args = ["value" + str(k) for k in range(len(values))] + node = onnx.helper.make_node( + "Concat", inputs=list(in_args), outputs=["output"], axis=i + ) + output = np.concatenate(values, i) + expect( + node, + inputs=list(values), + outputs=[output], + name="test_concat_" + test_case + "_axis_" + str(i), + ) + + for i in range(-len(values[0].shape), 0): + in_args = ["value" + str(k) for k in range(len(values))] + node = onnx.helper.make_node( + "Concat", inputs=list(in_args), outputs=["output"], axis=i + ) + output = np.concatenate(values, i) + expect( + node, + inputs=list(values), + outputs=[output], + name="test_concat_" + test_case + "_axis_negative_" + str(abs(i)), + ) +``` + +
+ + +### **ConcatFromSequence** + + Concatenate a sequence of tensors into a single tensor. + All input tensors must have the same shape, except for the dimension size of the axis to concatenate on. + By default 'new_axis' is 0, the behavior is similar to numpy.concatenate. + When 'new_axis' is 1, the behavior is similar to numpy.stack. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (required)
+
Which axis to concat on. Accepted range in `[-r, r - 1]`, where `r` is the rank of input tensors. When `new_axis` is 1, accepted range is `[-r - 1, r]`.
+
new_axis : int (default is 0)
+
Insert and concatenate on a new axis or not, default 0 means do not insert new axis.
+
+ +#### Inputs + +
+
input_sequence : S
+
Sequence of tensors for concatenation
+
+ +#### Outputs + +
+
concat_result : T
+
Concatenated tensor
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input types to any tensor type.
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain output types to any tensor type.
+
+ + +### **Constant** + + This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, + or value_* must be specified. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 9, 11, 12, 13, 19, 21, 23, 24 + +#### Attributes + +
+
sparse_value : sparse_tensor
+
The value for the elements of the output tensor in sparse format.
+
value : tensor
+
The value for the elements of the output tensor.
+
value_float : float
+
The value for the sole element for the scalar, float32, output tensor.
+
value_floats : list of floats
+
The values for the elements for the 1D, float32, output tensor.
+
value_int : int
+
The value for the sole element for the scalar, int64, output tensor.
+
value_ints : list of ints
+
The values for the elements for the 1D, int64, output tensor.
+
value_string : string
+
The value for the sole element for the scalar, UTF-8 string, output tensor.
+
value_strings : list of strings
+
The values for the elements for the 1D, UTF-8 string, output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor containing the same value of the provided tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types.
+
+ + +#### Examples + +
+constant + +```python +values = np.random.randn(5, 5).astype(np.float32) +node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["values"], + value=onnx.helper.make_tensor( + name="const_tensor", + data_type=onnx.TensorProto.FLOAT, + dims=values.shape, + vals=values.flatten().astype(float), + ), +) + +expect(node, inputs=[], outputs=[values], name="test_constant") +``` + +
+ + +### **ConstantOfShape** + + Generate a tensor with given value and shape. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 9, 20, 21, 23, 24 + +#### Attributes + +
+
value : tensor
+
(Optional) The value of the output elements.Should be a one-element tensor. If not specified, it defaults to a tensor of value 0 and datatype float32
+
+ +#### Inputs + +
+
input : T1
+
1D tensor. The shape of the expected output tensor. If empty tensor is given, the output would be a scalar. All values must be >= 0.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of shape specified by 'input'.If attribute 'value' is specified, the value and datatype of the output tensor is taken from 'value'.If attribute 'value' is not specified, the value in the output defaults to 0, and the datatype defaults to float32.
+
+ +#### Type Constraints + +
+
T1 : tensor(int64)
+
Constrain input types.
+
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint4), tensor(int4), tensor(bool), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain output types to be numerics or boolean.
+
+ + +#### Examples + +
+float_ones + +```python +x = np.array([4, 3, 2]).astype(np.int64) +tensor_value = onnx.helper.make_tensor( + "value", onnx.TensorProto.FLOAT, [1], [1] +) +node = onnx.helper.make_node( + "ConstantOfShape", + inputs=["x"], + outputs=["y"], + value=tensor_value, +) + +y = np.ones(x, dtype=np.float32) +expect(node, inputs=[x], outputs=[y], name="test_constantofshape_float_ones") +``` + +
+ + +
+int32_shape_zero + +```python +x = np.array( + [ + 0, + ] +).astype(np.int64) +tensor_value = onnx.helper.make_tensor( + "value", onnx.TensorProto.INT32, [1], [0] +) +node = onnx.helper.make_node( + "ConstantOfShape", + inputs=["x"], + outputs=["y"], + value=tensor_value, +) +y = np.zeros(x, dtype=np.int32) +expect( + node, inputs=[x], outputs=[y], name="test_constantofshape_int_shape_zero" +) +``` + +
+ + +
+int32_zeros + +```python +x = np.array([10, 6]).astype(np.int64) +tensor_value = onnx.helper.make_tensor( + "value", onnx.TensorProto.INT32, [1], [0] +) +node = onnx.helper.make_node( + "ConstantOfShape", + inputs=["x"], + outputs=["y"], + value=tensor_value, +) +y = np.zeros(x, dtype=np.int32) +expect(node, inputs=[x], outputs=[y], name="test_constantofshape_int_zeros") +``` + +
+ + +### **Conv** + + The convolution operator consumes an input tensor and a filter, and + computes the output. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 11 + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input W.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults is 1 along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X (differentiable) : T
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
W (differentiable) : T
+
The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.
+
B (optional, differentiable) : T
+
Optional 1D bias to be added to the convolution, has size of M.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+conv + +```python +x = np.array( + [ + [ + [ + [0.0, 1.0, 2.0, 3.0, 4.0], # (1, 1, 5, 5) input tensor + [5.0, 6.0, 7.0, 8.0, 9.0], + [10.0, 11.0, 12.0, 13.0, 14.0], + [15.0, 16.0, 17.0, 18.0, 19.0], + [20.0, 21.0, 22.0, 23.0, 24.0], + ] + ] + ] +).astype(np.float32) +W = np.array( + [ + [ + [ + [1.0, 1.0, 1.0], # (1, 1, 3, 3) tensor for convolution weights + [1.0, 1.0, 1.0], + [1.0, 1.0, 1.0], + ] + ] + ] +).astype(np.float32) + +# Convolution with padding +node_with_padding = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + kernel_shape=[3, 3], + # Default values for other attributes: strides=[1, 1], dilations=[1, 1], groups=1 + pads=[1, 1, 1, 1], +) +y_with_padding = np.array( + [ + [ + [ + [12.0, 21.0, 27.0, 33.0, 24.0], # (1, 1, 5, 5) output tensor + [33.0, 54.0, 63.0, 72.0, 51.0], + [63.0, 99.0, 108.0, 117.0, 81.0], + [93.0, 144.0, 153.0, 162.0, 111.0], + [72.0, 111.0, 117.0, 123.0, 84.0], + ] + ] + ] +).astype(np.float32) +expect( + node_with_padding, + inputs=[x, W], + outputs=[y_with_padding], + name="test_basic_conv_with_padding", +) + +# Convolution without padding +node_without_padding = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + kernel_shape=[3, 3], + # Default values for other attributes: strides=[1, 1], dilations=[1, 1], groups=1 + pads=[0, 0, 0, 0], +) +y_without_padding = np.array( + [ + [ + [ + [54.0, 63.0, 72.0], # (1, 1, 3, 3) output tensor + [99.0, 108.0, 117.0], + [144.0, 153.0, 162.0], + ] + ] + ] +).astype(np.float32) +expect( + node_without_padding, + inputs=[x, W], + outputs=[y_without_padding], + name="test_basic_conv_without_padding", +) +``` + +
+ + +
+conv_with_autopad_same + +```python +x = np.array( + [ + [ + [ + [0.0, 1.0, 2.0, 3.0, 4.0], # (1, 1, 5, 5) input tensor + [5.0, 6.0, 7.0, 8.0, 9.0], + [10.0, 11.0, 12.0, 13.0, 14.0], + [15.0, 16.0, 17.0, 18.0, 19.0], + [20.0, 21.0, 22.0, 23.0, 24.0], + ] + ] + ] +).astype(np.float32) +W = np.array( + [ + [ + [ + [1.0, 1.0, 1.0], # (1, 1, 3, 3) tensor for convolution weights + [1.0, 1.0, 1.0], + [1.0, 1.0, 1.0], + ] + ] + ] +).astype(np.float32) + +# Convolution with auto_pad='SAME_LOWER' and strides=2 +node = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + auto_pad="SAME_LOWER", + kernel_shape=[3, 3], + strides=[2, 2], +) +y = np.array( + [[[[12.0, 27.0, 24.0], [63.0, 108.0, 81.0], [72.0, 117.0, 84.0]]]] +).astype(np.float32) +expect(node, inputs=[x, W], outputs=[y], name="test_conv_with_autopad_same") +``` + +
+ + +
+conv_with_strides + +```python +x = np.array( + [ + [ + [ + [0.0, 1.0, 2.0, 3.0, 4.0], # (1, 1, 7, 5) input tensor + [5.0, 6.0, 7.0, 8.0, 9.0], + [10.0, 11.0, 12.0, 13.0, 14.0], + [15.0, 16.0, 17.0, 18.0, 19.0], + [20.0, 21.0, 22.0, 23.0, 24.0], + [25.0, 26.0, 27.0, 28.0, 29.0], + [30.0, 31.0, 32.0, 33.0, 34.0], + ] + ] + ] +).astype(np.float32) +W = np.array( + [ + [ + [ + [1.0, 1.0, 1.0], # (1, 1, 3, 3) tensor for convolution weights + [1.0, 1.0, 1.0], + [1.0, 1.0, 1.0], + ] + ] + ] +).astype(np.float32) + +# Convolution with strides=2 and padding +node_with_padding = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[1, 1, 1, 1], + strides=[ + 2, + 2, + ], # Default values for other attributes: dilations=[1, 1], groups=1 +) +y_with_padding = np.array( + [ + [ + [ + [12.0, 27.0, 24.0], # (1, 1, 4, 3) output tensor + [63.0, 108.0, 81.0], + [123.0, 198.0, 141.0], + [112.0, 177.0, 124.0], + ] + ] + ] +).astype(np.float32) +expect( + node_with_padding, + inputs=[x, W], + outputs=[y_with_padding], + name="test_conv_with_strides_padding", +) + +# Convolution with strides=2 and no padding +node_without_padding = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[0, 0, 0, 0], + strides=[ + 2, + 2, + ], # Default values for other attributes: dilations=[1, 1], groups=1 +) +y_without_padding = np.array( + [ + [ + [ + [54.0, 72.0], # (1, 1, 3, 2) output tensor + [144.0, 162.0], + [234.0, 252.0], + ] + ] + ] +).astype(np.float32) +expect( + node_without_padding, + inputs=[x, W], + outputs=[y_without_padding], + name="test_conv_with_strides_no_padding", +) + +# Convolution with strides=2 and padding only along one dimension (the H dimension in NxCxHxW tensor) +node_with_asymmetric_padding = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[1, 0, 1, 0], + strides=[ + 2, + 2, + ], # Default values for other attributes: dilations=[1, 1], groups=1 +) +y_with_asymmetric_padding = np.array( + [ + [ + [ + [21.0, 33.0], # (1, 1, 4, 2) output tensor + [99.0, 117.0], + [189.0, 207.0], + [171.0, 183.0], + ] + ] + ] +).astype(np.float32) +expect( + node_with_asymmetric_padding, + inputs=[x, W], + outputs=[y_with_asymmetric_padding], + name="test_conv_with_strides_and_asymmetric_padding", +) +``` + +
+ + +### **ConvInteger** + + The integer convolution operator consumes an input tensor, its zero-point, a filter, and its zero-point, + and computes the output. The production MUST never overflow. The accumulation may overflow if and only if in 32 bits. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each axis.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into. default is 1.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input 'w'.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0.The value represent the number of pixels added to the beginning and end part of the corresponding axis.`pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number ofpixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`.This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaultsto 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each axis.
+
+ +#### Inputs (2 - 4) + +
+
x : T1
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
w : T2
+
The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
+
x_zero_point (optional) : T1
+
Zero point tensor for input 'x'. It's optional and default value is 0. It's a scalar, which means a per-tensor/layer quantization.
+
w_zero_point (optional) : T2
+
Zero point tensor for input 'w'. It's optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it's a 1-D tensor, its number of elements should be equal to the number of output channels (M)
+
+ +#### Outputs + +
+
y : T3
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8)
+
Constrain input x and its zero point data type to 8-bit integer tensor.
+
T2 : tensor(int8), tensor(uint8)
+
Constrain input w and its zero point data type to 8-bit integer tensor.
+
T3 : tensor(int32)
+
Constrain output y data type to 32-bit integer tensor.
+
+ + +#### Examples + +
+with_padding + +```python +x = ( + np.array([2, 3, 4, 5, 6, 7, 8, 9, 10]) + .astype(np.uint8) + .reshape((1, 1, 3, 3)) +) +x_zero_point = np.uint8(1) +w_zero_points = np.array([0, 1], dtype=np.uint8) +w = np.array([1, 1, 1, 1, 1, 1, 1, 1]).astype(np.uint8).reshape((2, 1, 2, 2)) + +y = ( + np.array( + [ + 1, + 3, + 5, + 3, + 5, + 12, + 16, + 9, + 11, + 24, + 28, + 15, + 7, + 15, + 17, + 9, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + ] + ) + .astype(np.int32) + .reshape((1, 2, 4, 4)) +) + +# ConvInteger with padding +convinteger_node_with_padding = onnx.helper.make_node( + "ConvInteger", + inputs=["x", "w", "x_zero_point", "w_zero_points"], + outputs=["y"], + pads=[1, 1, 1, 1], +) + +expect( + convinteger_node_with_padding, + inputs=[x, w, x_zero_point, w_zero_points], + outputs=[y], + name="test_convinteger_with_padding", +) +``` + +
+ + +
+without_padding + +```python +x = ( + np.array([2, 3, 4, 5, 6, 7, 8, 9, 10]) + .astype(np.uint8) + .reshape((1, 1, 3, 3)) +) +x_zero_point = np.uint8(1) +w = np.array([1, 1, 1, 1]).astype(np.uint8).reshape((1, 1, 2, 2)) + +y = np.array([12, 16, 24, 28]).astype(np.int32).reshape(1, 1, 2, 2) + +# ConvInteger without padding +convinteger_node = onnx.helper.make_node( + "ConvInteger", inputs=["x", "w", "x_zero_point"], outputs=["y"] +) + +expect( + convinteger_node, + inputs=[x, w, x_zero_point], + outputs=[y], + name="test_convinteger_without_padding", +) +``` + +
+ + +### **ConvTranspose** + + The convolution transpose operator consumes an input tensor and a filter, + and computes the output. + + If the pads parameter is provided the shape of the output is calculated via the following equation: + + output_shape[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - pads[start_i] - pads[end_i] + + output_shape can also be explicitly specified in which case pads values are auto generated using these equations: + + total_padding[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - output_shape[i] + If (auto_pads == SAME_UPPER): pads[start_i] = total_padding[i]/2; pads[end_i] = total_padding[i] - (total_padding[i]/2) + Else: pads[start_i] = total_padding[i] - (total_padding[i]/2); pads[end_i] = (total_padding[i]/2). + + + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 11 + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = input_shape[i] * strides[i]` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each spatial axis.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input W.
+
output_padding : list of ints
+
Additional elements added to the side with higher coordinate indices in the output. Each padding value in "output_padding" must be less than the corresponding stride/dilation dimension. By default, this attribute is a zero vector. Note that this attribute doesn't directly affect the computed output values. It only controls the selection of the computed values, so changing this attribute only adds or removes output elements. If "output_shape" is explicitly provided, "output_padding" does not contribute additional size to "output_shape" but participates in the computation of the needed padding amount. This is also called adjs or adjustment in some frameworks.
+
output_shape : list of ints
+
The shape of the output can be explicitly set which will cause pads values to be auto generated. If output_shape is specified pads values are ignored. See doc for details for equations to generate pads. Note that the output_shape attribute value should not include dimensions for batch size and channels, which are automatically inferred.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X (differentiable) : T
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn)
+
W (differentiable) : T
+
The weight tensor that will be used in the convolutions; has size (C x M/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the weight shape will be (C x M/group x k1 x k2 x ... x kn), where (k1 x k2 x ... x kn) is the dimension of the kernel. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
+
B (optional, differentiable) : T
+
Optional 1D bias to be added to the convolution, has size of M.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, pad lengths and group count. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+convtranspose + +```python +x = np.array( + [[[[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]]]] # (1, 1, 3, 3) +).astype(np.float32) + +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], # (1, 2, 3, 3) + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node("ConvTranspose", ["X", "W"], ["Y"]) + +y = np.array( + [ + [ + [ + [0.0, 1.0, 3.0, 3.0, 2.0], # (1, 2, 5, 5) + [3.0, 8.0, 15.0, 12.0, 7.0], + [9.0, 21.0, 36.0, 27.0, 15.0], + [9.0, 20.0, 33.0, 24.0, 13.0], + [6.0, 13.0, 21.0, 15.0, 8.0], + ], + [ + [0.0, 1.0, 3.0, 3.0, 2.0], + [3.0, 8.0, 15.0, 12.0, 7.0], + [9.0, 21.0, 36.0, 27.0, 15.0], + [9.0, 20.0, 33.0, 24.0, 13.0], + [6.0, 13.0, 21.0, 15.0, 8.0], + ], + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose") +``` + +
+ + +
+convtranspose_1d + +```python +x = np.array([[[0.0, 1.0, 2.0]]]).astype(np.float32) # (1, 1, 3) + +W = np.array([[[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]]]).astype( # (1, 2, 3) + np.float32 +) + +node = onnx.helper.make_node("ConvTranspose", ["X", "W"], ["Y"]) + +y = np.array( + [[[0.0, 1.0, 3.0, 3.0, 2.0], [0.0, 1.0, 3.0, 3.0, 2.0]]] # (1, 2, 5) +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_1d") +``` + +
+ + +
+convtranspose_3d + +```python +x = np.array( + [ + [ + [ + [ + [0.0, 1.0, 2.0, 3.0, 4.0], # (1, 1, 3, 4, 5) + [5.0, 6.0, 7.0, 8.0, 9.0], + [10.0, 11.0, 12.0, 13.0, 14.0], + [15.0, 16.0, 17.0, 18.0, 19.0], + ], + [ + [20.0, 21.0, 22.0, 23.0, 24.0], + [25.0, 26.0, 27.0, 28.0, 29.0], + [30.0, 31.0, 32.0, 33.0, 34.0], + [35.0, 36.0, 37.0, 38.0, 39.0], + ], + [ + [40.0, 41.0, 42.0, 43.0, 44.0], + [45.0, 46.0, 47.0, 48.0, 49.0], + [50.0, 51.0, 52.0, 53.0, 54.0], + [55.0, 56.0, 57.0, 58.0, 59.0], + ], + ] + ] + ] +).astype(np.float32) + +W = np.array( + [ + [ + [ + [ + [1.0, 1.0, 1.0], # (1, 2, 3, 3, 3) + [1.0, 1.0, 1.0], + [1.0, 1.0, 1.0], + ], + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node("ConvTranspose", ["X", "W"], ["Y"]) + +y = np.array( + [ + [ + [ + [ + [0.0, 1.0, 3.0, 6.0, 9.0, 7.0, 4.0], # (1, 2, 5, 6, 7) + [5.0, 12.0, 21.0, 27.0, 33.0, 24.0, 13.0], + [15.0, 33.0, 54.0, 63.0, 72.0, 51.0, 27.0], + [30.0, 63.0, 99.0, 108.0, 117.0, 81.0, 42.0], + [25.0, 52.0, 81.0, 87.0, 93.0, 64.0, 33.0], + [15.0, 31.0, 48.0, 51.0, 54.0, 37.0, 19.0], + ], + [ + [20.0, 42.0, 66.0, 72.0, 78.0, 54.0, 28.0], + [50.0, 104.0, 162.0, 174.0, 186.0, 128.0, 66.0], + [90.0, 186.0, 288.0, 306.0, 324.0, 222.0, 114.0], + [120.0, 246.0, 378.0, 396.0, 414.0, 282.0, 144.0], + [90.0, 184.0, 282.0, 294.0, 306.0, 208.0, 106.0], + [50.0, 102.0, 156.0, 162.0, 168.0, 114.0, 58.0], + ], + [ + [60.0, 123.0, 189.0, 198.0, 207.0, 141.0, 72.0], + [135.0, 276.0, 423.0, 441.0, 459.0, 312.0, 159.0], + [225.0, 459.0, 702.0, 729.0, 756.0, 513.0, 261.0], + [270.0, 549.0, 837.0, 864.0, 891.0, 603.0, 306.0], + [195.0, 396.0, 603.0, 621.0, 639.0, 432.0, 219.0], + [105.0, 213.0, 324.0, 333.0, 342.0, 231.0, 117.0], + ], + [ + [60.0, 122.0, 186.0, 192.0, 198.0, 134.0, 68.0], + [130.0, 264.0, 402.0, 414.0, 426.0, 288.0, 146.0], + [210.0, 426.0, 648.0, 666.0, 684.0, 462.0, 234.0], + [240.0, 486.0, 738.0, 756.0, 774.0, 522.0, 264.0], + [170.0, 344.0, 522.0, 534.0, 546.0, 368.0, 186.0], + [90.0, 182.0, 276.0, 282.0, 288.0, 194.0, 98.0], + ], + [ + [40.0, 81.0, 123.0, 126.0, 129.0, 87.0, 44.0], + [85.0, 172.0, 261.0, 267.0, 273.0, 184.0, 93.0], + [135.0, 273.0, 414.0, 423.0, 432.0, 291.0, 147.0], + [150.0, 303.0, 459.0, 468.0, 477.0, 321.0, 162.0], + [105.0, 212.0, 321.0, 327.0, 333.0, 224.0, 113.0], + [55.0, 111.0, 168.0, 171.0, 174.0, 117.0, 59.0], + ], + ], + [ + [ + [0.0, 1.0, 3.0, 6.0, 9.0, 7.0, 4.0], + [5.0, 12.0, 21.0, 27.0, 33.0, 24.0, 13.0], + [15.0, 33.0, 54.0, 63.0, 72.0, 51.0, 27.0], + [30.0, 63.0, 99.0, 108.0, 117.0, 81.0, 42.0], + [25.0, 52.0, 81.0, 87.0, 93.0, 64.0, 33.0], + [15.0, 31.0, 48.0, 51.0, 54.0, 37.0, 19.0], + ], + [ + [20.0, 42.0, 66.0, 72.0, 78.0, 54.0, 28.0], + [50.0, 104.0, 162.0, 174.0, 186.0, 128.0, 66.0], + [90.0, 186.0, 288.0, 306.0, 324.0, 222.0, 114.0], + [120.0, 246.0, 378.0, 396.0, 414.0, 282.0, 144.0], + [90.0, 184.0, 282.0, 294.0, 306.0, 208.0, 106.0], + [50.0, 102.0, 156.0, 162.0, 168.0, 114.0, 58.0], + ], + [ + [60.0, 123.0, 189.0, 198.0, 207.0, 141.0, 72.0], + [135.0, 276.0, 423.0, 441.0, 459.0, 312.0, 159.0], + [225.0, 459.0, 702.0, 729.0, 756.0, 513.0, 261.0], + [270.0, 549.0, 837.0, 864.0, 891.0, 603.0, 306.0], + [195.0, 396.0, 603.0, 621.0, 639.0, 432.0, 219.0], + [105.0, 213.0, 324.0, 333.0, 342.0, 231.0, 117.0], + ], + [ + [60.0, 122.0, 186.0, 192.0, 198.0, 134.0, 68.0], + [130.0, 264.0, 402.0, 414.0, 426.0, 288.0, 146.0], + [210.0, 426.0, 648.0, 666.0, 684.0, 462.0, 234.0], + [240.0, 486.0, 738.0, 756.0, 774.0, 522.0, 264.0], + [170.0, 344.0, 522.0, 534.0, 546.0, 368.0, 186.0], + [90.0, 182.0, 276.0, 282.0, 288.0, 194.0, 98.0], + ], + [ + [40.0, 81.0, 123.0, 126.0, 129.0, 87.0, 44.0], + [85.0, 172.0, 261.0, 267.0, 273.0, 184.0, 93.0], + [135.0, 273.0, 414.0, 423.0, 432.0, 291.0, 147.0], + [150.0, 303.0, 459.0, 468.0, 477.0, 321.0, 162.0], + [105.0, 212.0, 321.0, 327.0, 333.0, 224.0, 113.0], + [55.0, 111.0, 168.0, 171.0, 174.0, 117.0, 59.0], + ], + ], + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_3d") +``` + +
+ + +
+convtranspose_attributes + +```python +x = np.array( + [[[[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]]]] # (1, 1, 3, 3) +).astype(np.float32) + +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], # (1, 2, 3, 3) + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ] + ] +).astype(np.float32) + +y = np.array( + [ + [ + [ + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], # (1, 2, 10, 8) + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], + ], + [ + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], + ], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "ConvTranspose", ["X", "W"], ["Y"], strides=[3, 2], output_shape=[10, 8] +) +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_output_shape") + +node = onnx.helper.make_node( + "ConvTranspose", ["X", "W"], ["Y"], strides=[3, 2], output_padding=[1, 1] +) +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_pad") + +node = onnx.helper.make_node( + "ConvTranspose", + ["X", "W"], + ["Y"], + name="test", + strides=[3, 2], + output_shape=[10, 8], + kernel_shape=[3, 3], + output_padding=[1, 1], +) +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_kernel_shape") +``` + +
+ + +
+convtranspose_autopad_same + +```python +x = np.array( + [[[[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]]]] # (1, 1, 3, 3) +).astype(np.float32) + +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], # (1, 2, 3, 3) + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "ConvTranspose", ["X", "W"], ["Y"], auto_pad="SAME_UPPER", strides=[2, 2] +) + +y = np.array( + [ + [ + [ + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0], + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0], + [3.0, 3.0, 8.0, 5.0, 12.0, 7.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0], + [9.0, 9.0, 20.0, 11.0, 24.0, 13.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0], + ], + [ + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0], + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0], + [3.0, 3.0, 8.0, 5.0, 12.0, 7.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0], + [9.0, 9.0, 20.0, 11.0, 24.0, 13.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0], + ], + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_autopad_same") +``` + +
+ + +
+convtranspose_dilations + +```python +x = np.array( + [[[[3.0, 8.0, 1.0], [9.0, 5.0, 7.0], [3.0, 2.0, 6.0]]]] # (1, 1, 3, 3) +).astype(np.float32) +W = np.array([[[[7.0, 2.0], [1.0, 9.0]]]]).astype(np.float32) # (1, 1, 2, 2) + +node = onnx.helper.make_node( + "ConvTranspose", ["X", "W"], ["Y"], dilations=[2, 2] +) + +y = np.array( + [ + [ + [ + [21.0, 56.0, 13.0, 16.0, 2.0], # [1, 1, 5, 5] + [63.0, 35.0, 67.0, 10.0, 14.0], + [24.0, 22.0, 76.0, 76.0, 21.0], + [9.0, 5.0, 88.0, 45.0, 63.0], + [3.0, 2.0, 33.0, 18.0, 54.0], + ] + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_dilations") +``` + +
+ + +
+convtranspose_group_2 + +```python +x = np.array( + [ + [ + [[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0], [15.0, 16.0, 17.0]], + ] + ] +).astype(np.float32) +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + ] +).astype(np.float32) + +node = onnx.helper.make_node("ConvTranspose", ["X", "W"], ["Y"], group=2) + +y = np.array( + [ + [ + [ + [0.0, 1.0, 3.0, 3.0, 2.0], + [3.0, 8.0, 15.0, 12.0, 7.0], + [9.0, 21.0, 36.0, 27.0, 15.0], + [9.0, 20.0, 33.0, 24.0, 13.0], + [6.0, 13.0, 21.0, 15.0, 8.0], + ], + [ + [9.0, 19.0, 30.0, 21.0, 11.0], + [21.0, 44.0, 69.0, 48.0, 25.0], + [36.0, 75.0, 117.0, 81.0, 42.0], + [27.0, 56.0, 87.0, 60.0, 31.0], + [15.0, 31.0, 48.0, 33.0, 17.0], + ], + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_group_2") +``` + +
+ + +
+convtranspose_group_2_image_3 + +```python +x = np.array( + [ + [ + [[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0], [15.0, 16.0, 17.0]], + ], + [ + [[18.0, 19.0, 20.0], [21.0, 22.0, 23.0], [24.0, 25.0, 26.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0], [15.0, 16.0, 17.0]], + ], + [ + [[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0], [15.0, 16.0, 17.0]], + ], + ] +).astype(np.float32) +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + ] +).astype(np.float32) + +node = onnx.helper.make_node("ConvTranspose", ["X", "W"], ["Y"], group=2) + +y = np.array( + [ + [ + [ + [0.0, 1.0, 3.0, 3.0, 2.0], + [3.0, 8.0, 15.0, 12.0, 7.0], + [9.0, 21.0, 36.0, 27.0, 15.0], + [9.0, 20.0, 33.0, 24.0, 13.0], + [6.0, 13.0, 21.0, 15.0, 8.0], + ], + [ + [9.0, 19.0, 30.0, 21.0, 11.0], + [21.0, 44.0, 69.0, 48.0, 25.0], + [36.0, 75.0, 117.0, 81.0, 42.0], + [27.0, 56.0, 87.0, 60.0, 31.0], + [15.0, 31.0, 48.0, 33.0, 17.0], + ], + ], + [ + [ + [18.0, 37.0, 57.0, 39.0, 20.0], + [39.0, 80.0, 123.0, 84.0, 43.0], + [63.0, 129.0, 198.0, 135.0, 69.0], + [45.0, 92.0, 141.0, 96.0, 49.0], + [24.0, 49.0, 75.0, 51.0, 26.0], + ], + [ + [9.0, 19.0, 30.0, 21.0, 11.0], + [21.0, 44.0, 69.0, 48.0, 25.0], + [36.0, 75.0, 117.0, 81.0, 42.0], + [27.0, 56.0, 87.0, 60.0, 31.0], + [15.0, 31.0, 48.0, 33.0, 17.0], + ], + ], + [ + [ + [0.0, 1.0, 3.0, 3.0, 2.0], + [3.0, 8.0, 15.0, 12.0, 7.0], + [9.0, 21.0, 36.0, 27.0, 15.0], + [9.0, 20.0, 33.0, 24.0, 13.0], + [6.0, 13.0, 21.0, 15.0, 8.0], + ], + [ + [9.0, 19.0, 30.0, 21.0, 11.0], + [21.0, 44.0, 69.0, 48.0, 25.0], + [36.0, 75.0, 117.0, 81.0, 42.0], + [27.0, 56.0, 87.0, 60.0, 31.0], + [15.0, 31.0, 48.0, 33.0, 17.0], + ], + ], + ] +).astype(np.float32) + +expect( + node, inputs=[x, W], outputs=[y], name="test_convtranspose_group_2_image_3" +) +``` + +
+ + +
+convtranspose_pads + +```python +x = np.array( + [[[[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]]]] # (1, 1, 3, 3) +).astype(np.float32) + +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], # (1, 2, 3, 3) + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "ConvTranspose", ["X", "W"], ["Y"], strides=[3, 2], pads=[1, 2, 1, 2] +) + +y = np.array( + [ + [ + [ + [1.0, 1.0, 3.0], # (1, 2, 7, 3) + [1.0, 1.0, 3.0], + [7.0, 4.0, 9.0], + [7.0, 4.0, 9.0], + [7.0, 4.0, 9.0], + [13.0, 7.0, 15.0], + [13.0, 7.0, 15.0], + ], + [ + [1.0, 1.0, 3.0], + [1.0, 1.0, 3.0], + [7.0, 4.0, 9.0], + [7.0, 4.0, 9.0], + [7.0, 4.0, 9.0], + [13.0, 7.0, 15.0], + [13.0, 7.0, 15.0], + ], + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_pads") +``` + +
+ + +### **Cos** + + Calculates the cosine of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 7 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The cosine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+cos + +```python +node = onnx.helper.make_node( + "Cos", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.cos(x) +expect(node, inputs=[x], outputs=[y], name="test_cos_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.cos(x) +expect(node, inputs=[x], outputs=[y], name="test_cos") +``` + +
+ + +### **Cosh** + + Calculates the hyperbolic cosine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic cosine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+cosh + +```python +node = onnx.helper.make_node( + "Cosh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.cosh(x) # expected output [1.54308069, 1., 1.54308069] +expect(node, inputs=[x], outputs=[y], name="test_cosh_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.cosh(x) +expect(node, inputs=[x], outputs=[y], name="test_cosh") +``` + +
+ + +### **CumProd** + + Performs cumulative product of the input elements along the given axis. + By default, it will do the product inclusively meaning the first element is copied as is. + Through an `exclusive` attribute, this behavior can change to exclude the first element. + It can also perform product in the opposite direction of the axis. For that, set `reverse` attribute to 1. + + Example: + ``` + input_x = [1, 2, 3] + axis=0 + output = [1, 2, 6] + exclusive=1 + output = [1, 1, 2] + exclusive=0 + reverse=1 + output = [6, 6, 3] + exclusive=1 + reverse=1 + output = [6, 3, 1] + ``` + + +#### Version + +This version of the operator has been available since version 26 of the default ONNX operator set. + +#### Attributes + +
+
exclusive : int (default is 0)
+
If set to 1 will return exclusive product in which the top element is not included. In other terms, if set to 1, the j-th output element would be the product of the first (j-1) elements. Otherwise, it would be the product of the first j elements.
+
reverse : int (default is 0)
+
If set to 1 will perform the products in reverse direction.
+
+ +#### Inputs + +
+
x (differentiable) : T
+
An input tensor that is to be processed.
+
axis (non-differentiable) : T2
+
A 0-D tensor. Must be in the range [-rank(x), rank(x)-1]. Negative value means counting dimensions from the back.
+
+ +#### Outputs + +
+
y (differentiable) : T
+
Output tensor of the same type as 'x' with cumulative products of the x's elements
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
T2 : tensor(int32), tensor(int64)
+
axis tensor can be int32 or int64 only
+
+ + +#### Examples + +
+cumprod_1d + +```python +node = onnx.helper.make_node("CumProd", inputs=["x", "axis"], outputs=["y"]) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.array(0, dtype=np.int32) +y = np.array([1.0, 2.0, 6.0, 24.0, 120.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_1d") +``` + +
+ + +
+cumprod_1d_exclusive + +```python +node = onnx.helper.make_node( + "CumProd", inputs=["x", "axis"], outputs=["y"], exclusive=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.array(0, dtype=np.int32) +y = np.array([1.0, 1.0, 2.0, 6.0, 24.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_1d_exclusive") +``` + +
+ + +
+cumprod_1d_int32_exclusive + +```python +node = onnx.helper.make_node( + "CumProd", inputs=["x", "axis"], outputs=["y"], exclusive=1 +) +x = np.array([1, 2, 3, 4, 5]).astype(np.int32) +axis = np.array(0, dtype=np.int32) +y = np.array([1, 1, 2, 6, 24]).astype(np.int32) +expect( + node, inputs=[x, axis], outputs=[y], name="test_cumprod_1d_int32_exclusive" +) +``` + +
+ + +
+cumprod_1d_reverse + +```python +node = onnx.helper.make_node( + "CumProd", inputs=["x", "axis"], outputs=["y"], reverse=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.array(0, dtype=np.int32) +y = np.array([120.0, 120.0, 60.0, 20.0, 5.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_1d_reverse") +``` + +
+ + +
+cumprod_1d_reverse_exclusive + +```python +node = onnx.helper.make_node( + "CumProd", inputs=["x", "axis"], outputs=["y"], reverse=1, exclusive=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.array(0, dtype=np.int32) +y = np.array([120.0, 60.0, 20.0, 5.0, 1.0]).astype(np.float64) +expect( + node, + inputs=[x, axis], + outputs=[y], + name="test_cumprod_1d_reverse_exclusive", +) +``` + +
+ + +
+cumprod_2d_axis_0 + +```python +node = onnx.helper.make_node( + "CumProd", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.array(0, dtype=np.int32) +y = ( + np.array([1.0, 2.0, 3.0, 4.0, 10.0, 18.0]) + .astype(np.float64) + .reshape((2, 3)) +) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_2d_axis_0") +``` + +
+ + +
+cumprod_2d_axis_1 + +```python +node = onnx.helper.make_node( + "CumProd", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.array(1, dtype=np.int32) +y = ( + np.array([1.0, 2.0, 6.0, 4.0, 20.0, 120.0]) + .astype(np.float64) + .reshape((2, 3)) +) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_2d_axis_1") +``` + +
+ + +
+cumprod_2d_int32 + +```python +node = onnx.helper.make_node( + "CumProd", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1, 2, 3, 4, 5, 6]).astype(np.int32).reshape((2, 3)) +axis = np.array(0, dtype=np.int32) +y = np.array([1, 2, 3, 4, 10, 18]).astype(np.int32).reshape((2, 3)) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_2d_int32") +``` + +
+ + +
+cumprod_2d_negative_axis + +```python +node = onnx.helper.make_node( + "CumProd", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.array(-1, dtype=np.int32) +y = ( + np.array([1.0, 2.0, 6.0, 4.0, 20.0, 120.0]) + .astype(np.float64) + .reshape((2, 3)) +) +expect( + node, inputs=[x, axis], outputs=[y], name="test_cumprod_2d_negative_axis" +) +``` + +
+ + +### **CumSum** + + Performs cumulative sum of the input elements along the given axis. + By default, it will do the sum inclusively meaning the first element is copied as is. + Through an `exclusive` attribute, this behavior can change to exclude the first element. + It can also perform summation in the opposite direction of the axis. For that, set `reverse` attribute to 1. + + Example: + ``` + input_x = [1, 2, 3] + axis=0 + output = [1, 3, 6] + exclusive=1 + output = [0, 1, 3] + exclusive=0 + reverse=1 + output = [6, 5, 3] + exclusive=1 + reverse=1 + output = [5, 3, 0] + ``` + + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +Other versions of this operator: 11 + +#### Attributes + +
+
exclusive : int (default is 0)
+
If set to 1 will return exclusive sum in which the top element is not included. In other terms, if set to 1, the j-th output element would be the sum of the first (j-1) elements. Otherwise, it would be the sum of the first j elements.
+
reverse : int (default is 0)
+
If set to 1 will perform the sums in reverse direction.
+
+ +#### Inputs + +
+
x (differentiable) : T
+
An input tensor that is to be processed.
+
axis (non-differentiable) : T2
+
A 0-D tensor. Must be in the range [-rank(x), rank(x)-1]. Negative value means counting dimensions from the back.
+
+ +#### Outputs + +
+
y (differentiable) : T
+
Output tensor of the same type as 'x' with cumulative sums of the x's elements
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
T2 : tensor(int32), tensor(int64)
+
axis tensor can be int32 or int64 only
+
+ + +#### Examples + +
+cumsum_1d + +```python +node = onnx.helper.make_node("CumSum", inputs=["x", "axis"], outputs=["y"]) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.int32(0) +y = np.array([1.0, 3.0, 6.0, 10.0, 15.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_1d") +``` + +
+ + +
+cumsum_1d_exclusive + +```python +node = onnx.helper.make_node( + "CumSum", inputs=["x", "axis"], outputs=["y"], exclusive=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.int32(0) +y = np.array([0.0, 1.0, 3.0, 6.0, 10.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_1d_exclusive") +``` + +
+ + +
+cumsum_1d_int32_exclusive + +```python +node = onnx.helper.make_node( + "CumSum", inputs=["x", "axis"], outputs=["y"], exclusive=1 +) +x = np.array([1, 2, 3, 4, 5]).astype(np.int32) +axis = np.int32(0) +y = np.array([0, 1, 3, 6, 10]).astype(np.int32) +expect( + node, inputs=[x, axis], outputs=[y], name="test_cumsum_1d_int32_exclusive" +) +``` + +
+ + +
+cumsum_1d_reverse + +```python +node = onnx.helper.make_node( + "CumSum", inputs=["x", "axis"], outputs=["y"], reverse=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.int32(0) +y = np.array([15.0, 14.0, 12.0, 9.0, 5.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_1d_reverse") +``` + +
+ + +
+cumsum_1d_reverse_exclusive + +```python +node = onnx.helper.make_node( + "CumSum", inputs=["x", "axis"], outputs=["y"], reverse=1, exclusive=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.int32(0) +y = np.array([14.0, 12.0, 9.0, 5.0, 0.0]).astype(np.float64) +expect( + node, inputs=[x, axis], outputs=[y], name="test_cumsum_1d_reverse_exclusive" +) +``` + +
+ + +
+cumsum_2d_axis_0 + +```python +node = onnx.helper.make_node( + "CumSum", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.int32(0) +y = np.array([1.0, 2.0, 3.0, 5.0, 7.0, 9.0]).astype(np.float64).reshape((2, 3)) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_2d_axis_0") +``` + +
+ + +
+cumsum_2d_axis_1 + +```python +node = onnx.helper.make_node( + "CumSum", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.int32(1) +y = np.array([1.0, 3.0, 6.0, 4.0, 9.0, 15.0]).astype(np.float64).reshape((2, 3)) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_2d_axis_1") +``` + +
+ + +
+cumsum_2d_int32 + +```python +node = onnx.helper.make_node( + "CumSum", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1, 2, 3, 4, 5, 6]).astype(np.int32).reshape((2, 3)) +axis = np.int32(0) +y = np.array([1, 2, 3, 5, 7, 9]).astype(np.int32).reshape((2, 3)) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_2d_int32") +``` + +
+ + +
+cumsum_2d_negative_axis + +```python +node = onnx.helper.make_node( + "CumSum", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.int32(-1) +y = np.array([1.0, 3.0, 6.0, 4.0, 9.0, 15.0]).astype(np.float64).reshape((2, 3)) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_2d_negative_axis") +``` + +
+ + +### **DFT** + + Computes the discrete Fourier Transform (DFT) of the input. + + Assuming the input has shape `[M, N]`, where `N` is the dimension over which the + DFT is computed and `M` denotes the conceptual "all other dimensions," + the DFT `y[m, k]` of shape `[M, N]` is defined as + + $$y[m, k] = \sum_{n=0}^{N-1} e^{-2 \pi j \frac{k n}{N} } x[m, n] ,$$ + + and the inverse transform is defined as + + $$x[m, n] = \frac{1}{N} \sum_{k=0}^{N-1} e^{2 \pi j \frac{k n}{N} } y[m, k] ,$$ + + where $j$ is the imaginary unit. + + The actual shape of the output is specified in the "output" section. + + Reference: https://docs.scipy.org/doc/scipy/tutorial/fft.html + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +Other versions of this operator: 17 + +#### Attributes + +
+
inverse : int (default is 0)
+
Whether to perform the inverse discrete Fourier Transform. Default is 0, which corresponds to `false`.
+
onesided : int (default is 0)
+
If `onesided` is `1`, only values for `k` in `[0, 1, 2, ..., floor(n_fft/2) + 1]` are used or returned because the real-to-complex Fourier transform satisfies the conjugate symmetry, i.e., `X[m, k] = X[m, n_fft-k]*`, where `m` denotes "all other dimensions" DFT was not applied on. When `onesided=1` and `inverse=0` (forward DFT), only real input is supported and a one-sided complex spectrum is returned (RFFT). When `onesided=1` and `inverse=1` (inverse DFT), only complex input is supported and a full real signal is returned (IRFFT). Value can be `0` or `1`. Default is `0`.
+
+ +#### Inputs (1 - 3) + +
+
input (non-differentiable) : T1
+
For real input, the following shape is expected: `[signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][1]`. For complex input, the following shape is expected: `[signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][2]`. The final dimension represents the real and imaginary parts of the value in that order.
+
dft_length (optional, non-differentiable) : T2
+
The length of the signal as a scalar. If greater than the axis dimension, the signal will be zero-padded up to `dft_length`. If less than the axis dimension, only the first `dft_length` values will be used as the signal. If not provided, the default `dft_length = signal_dim_axis`, except for the IRFFT case (`onesided=1`, `inverse=1`), in which case the default dft_length is `2 * (signal_dim_axis - 1)`.
+
axis (optional, non-differentiable) : tensor(int64)
+
The axis as a scalar on which to perform the DFT. Default is `-2` (last signal axis). Negative value means counting dimensions from the back. Accepted range is $[-r, -2] \cup [0, r-2]$ where `r = rank(input)`. The last dimension is for representing complex numbers and thus is an invalid axis.
+
+ +#### Outputs + +
+
output : T1
+
The Fourier Transform of the input vector. For standard DFT (`onesided=0`), the output shape is: `[signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][2]` (complex), with `signal_dim_axis = dft_length`. For RFFT (`onesided=1`, `inverse=0`), the output shape is: `[signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][2]` (one-sided complex), with `signal_dim_axis = floor(dft_length/2) + 1`. For IRFFT (`onesided=1`, `inverse=1`), the output shape is: `[signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][1]` (real), where `signal_dim_axis = dft_length`.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T2 : tensor(int32), tensor(int64)
+
Constrain scalar length types to integers.
+
+ + +#### Examples + +
+dft + +```python +node = onnx.helper.make_node("DFT", inputs=["x", "", "axis"], outputs=["y"]) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +axis = np.array(1, dtype=np.int64) +y = np.fft.fft(x, axis=0) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect(node, inputs=[x, axis], outputs=[y], name="test_dft") + +node = onnx.helper.make_node("DFT", inputs=["x", "", "axis"], outputs=["y"]) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +axis = np.array(2, dtype=np.int64) +y = np.fft.fft(x, axis=1) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect(node, inputs=[x, axis], outputs=[y], name="test_dft_axis") + +node = onnx.helper.make_node( + "DFT", inputs=["x", "", "axis"], outputs=["y"], inverse=1 +) +x = np.arange(0, 100, dtype=np.complex64).reshape(10, 10) +axis = np.array(1, dtype=np.int64) +y = np.fft.ifft(x, axis=0) + +x = np.stack((x.real, x.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect(node, inputs=[x, axis], outputs=[y], name="test_dft_inverse") + +# Test RFFT (Real FFT): real input -> one-sided complex output +node = onnx.helper.make_node( + "DFT", inputs=["x", "", "axis"], outputs=["y"], onesided=1 +) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +axis = np.array(1, dtype=np.int64) +y = np.fft.rfft(x, axis=0) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 6, 10, 2) +expect(node, inputs=[x, axis], outputs=[y], name="test_dft_rfft") + +# Test IRFFT (Inverse Real FFT): one-sided complex input -> real output +node = onnx.helper.make_node( + "DFT", inputs=["x", "", "axis"], outputs=["y"], onesided=1, inverse=1 +) +# Create one-sided complex input (6 bins for signal length 10) +x = np.fft.rfft(np.arange(0, 100).reshape(10, 10), axis=0).astype(np.complex64) +axis = np.array(1, dtype=np.int64) +y = np.fft.irfft(x, n=10, axis=0) + +x = np.stack((x.real, x.imag), axis=2).astype(np.float32).reshape(1, 6, 10, 2) +y = y.reshape(1, 10, 10, 1).astype(np.float32) +expect(node, inputs=[x, axis], outputs=[y], name="test_dft_irfft") +``` + +
+ + +
+opset19 + +```python +node = onnx.helper.make_node("DFT", inputs=["x"], outputs=["y"], axis=1) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +y = np.fft.fft(x, axis=0) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect( + node, + inputs=[x], + outputs=[y], + name="test_dft_opset19", + opset_imports=[onnx.helper.make_opsetid("", 19)], +) + +node = onnx.helper.make_node("DFT", inputs=["x"], outputs=["y"], axis=2) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +y = np.fft.fft(x, axis=1) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect( + node, + inputs=[x], + outputs=[y], + name="test_dft_axis_opset19", + opset_imports=[onnx.helper.make_opsetid("", 19)], +) + +node = onnx.helper.make_node( + "DFT", inputs=["x"], outputs=["y"], inverse=1, axis=1 +) +x = np.arange(0, 100, dtype=np.complex64).reshape( + 10, + 10, +) +y = np.fft.ifft(x, axis=0) + +x = np.stack((x.real, x.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect( + node, + inputs=[x], + outputs=[y], + name="test_dft_inverse_opset19", + opset_imports=[onnx.helper.make_opsetid("", 19)], +) + +# Test RFFT (Real FFT): real input -> one-sided complex output +node = onnx.helper.make_node( + "DFT", inputs=["x"], outputs=["y"], onesided=1, axis=1 +) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +y = np.fft.rfft(x, axis=0) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 6, 10, 2) +expect( + node, + inputs=[x], + outputs=[y], + name="test_dft_rfft_opset19", + opset_imports=[onnx.helper.make_opsetid("", 19)], +) + +# Test IRFFT (Inverse Real FFT): one-sided complex input -> real output +node = onnx.helper.make_node( + "DFT", inputs=["x"], outputs=["y"], onesided=1, inverse=1, axis=1 +) +# Create one-sided complex input (6 bins for signal length 10) +x = np.fft.rfft(np.arange(0, 100).reshape(10, 10), axis=0).astype(np.complex64) +y = np.fft.irfft(x, n=10, axis=0) + +x = np.stack((x.real, x.imag), axis=2).astype(np.float32).reshape(1, 6, 10, 2) +y = y.reshape(1, 10, 10, 1).astype(np.float32) +expect( + node, + inputs=[x], + outputs=[y], + name="test_dft_irfft_opset19", + opset_imports=[onnx.helper.make_opsetid("", 19)], +) +``` + +
+ + +### **DeformConv** + + Performs deformable convolution as described in https://arxiv.org/abs/1703.06211 and https://arxiv.org/abs/1811.11168. + This operator specification supports the general N-D case. Note that most common use cases have 2D or 3D data. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 19 + +#### Attributes + +
+
dilations : list of ints
+
Dilation value along each spatial axis of the kernel. Default is 1 along each axis.
+
group : int (default is 1)
+
Number of groups the input and output channels, C and oC, are divided into. C and oC must both be divisible by group. Default is 1.
+
kernel_shape : list of ints
+
Shape of the convolution kernel. If not present, it is inferred from the shape of input W.
+
offset_group : int (default is 1)
+
Number of groups of offset. C must be divisible by offset_group. Default is 1.
+
pads : list of ints
+
Padding for the beginning and end along each spatial axis. The values represent the number of pixels added to the beginning and end of the corresponding axis and can take any nonnegative value. The format should be as follows: [x1_begin, x2_begin, ..., x1_end, x2_end, ...], where xi_begin is the number of pixels added at the beginning of axis `i` and xi_end is the number of pixels added at the end of axis `i`. Default is 0 along each axis.
+
strides : list of ints
+
Stride along each spatial axis. Default is 1 along each axis.
+
+ +#### Inputs (3 - 5) + +
+
X : T
+
Input data tensor. For 2D image data, it has shape (N, C, H, W) where N is the batch size, C is the number of input channels, and H and W are the height and width. In general, the shape is (N, C, D1, D2, ... , Dn) for n-dimensional data, where D1 to Dn are the spatial dimension sizes. Most common use cases have n = 2 or 3.
+
W : T
+
Weight tensor that will be used in the convolutions. It has shape (oC, C/group, kH, kW), where oC is the number of output channels and kH and kW are the kernel height and width. For more than 2 dimensions, it has shape (oC, C/group, k1, k2, ... , kn).
+
offset : T
+
Offset tensor denoting the offset for the sampling locations in the convolution kernel. It has shape (N, offset_group * kH * kW * 2, oH, oW) for 2D data or (N, offset_group * k1 * k2 * ... * kn * n, o1, o2, ... , on) for nD data. Use linear interpolationfor fractional offset values. Sampling locations outside of the padded input tensor gives zero.
+
B (optional) : T
+
Optional 1D bias of length oC to be added to the convolution. Default is a tensor of zeros.
+
mask (optional) : T
+
The mask tensor to be applied to each position in the convolution kernel. It has shape (N, offset_group * kH * kW, oH, oW) for 2D data or (N, offset_group * k1 * k2 * ... * kn * n, o1, o2, ... , on) for nD data. Default is a tensor of ones.
+
+ +#### Outputs + +
+
Y : T
+
Output data tensor that contains the result of convolution. It has shape (N, oC, oH, oW) for 2D data or (N, oC, o1, o2, ..., on) for nD data
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+deformconv + +```python +X = np.arange(9).astype(np.float32) +X.shape = (1, 1, 3, 3) +W = np.ones((1, 1, 2, 2), dtype=np.float32) + +# Convolution with padding +offset_with_padding = np.zeros((1, 8, 4, 4), dtype=np.float32) +# h-coord of [0, 0] element of kernel, at output position [0, 0] +offset_with_padding[0, 0, 0, 0] = 0.5 +# w-coord of [1, 0] element of kernel, at output position [1, 2] +offset_with_padding[0, 5, 1, 2] = -0.1 + +node_with_padding = onnx.helper.make_node( + "DeformConv", + inputs=["X", "W", "offset_with_padding"], + outputs=["Y_with_padding"], + kernel_shape=[2, 2], + pads=[1, 1, 1, 1], +) +Y_with_padding = np.array( + [ + [ + [ + [0.0, 1.0, 3.0, 2.0], # (1, 1, 4, 4) output tensor + [3.0, 8.0, 11.9, 7.0], + [9.0, 20.0, 24.0, 13.0], + [6.0, 13.0, 15.0, 8.0], + ] + ] + ] +).astype(np.float32) +expect( + node_with_padding, + inputs=[X, W, offset_with_padding], + outputs=[Y_with_padding], + name="test_basic_deform_conv_with_padding", +) + +# Convolution without padding +offset_without_padding = np.zeros((1, 8, 2, 2), dtype=np.float32) +# h-coord of [0, 0] element of kernel, at output position [0, 0] +offset_without_padding[0, 0, 0, 0] = 0.5 +# w-coord of [1, 0] element of kernel, at output position [0, 1] +offset_without_padding[0, 5, 0, 1] = -0.1 + +node_without_padding = onnx.helper.make_node( + "DeformConv", + inputs=["X", "W", "offset_without_padding"], + outputs=["Y_without_padding"], + kernel_shape=[2, 2], + pads=[0, 0, 0, 0], +) +Y_without_padding = np.array( + [ + [ + [ + [9.5, 11.9], # (1, 1, 2, 2) output tensor + [20.0, 24.0], + ] + ] + ] +).astype(np.float32) +expect( + node_without_padding, + inputs=[X, W, offset_without_padding], + outputs=[Y_without_padding], + name="test_basic_deform_conv_without_padding", +) +``` + +
+ + +
+deformconv_with_mask_bias + +```python +X = np.arange(9).astype(np.float32) +X.shape = (1, 1, 3, 3) +W = np.ones((1, 1, 2, 2), dtype=np.float32) +B = np.ones((1,), dtype=np.float32) + +offset = np.zeros((1, 8, 2, 2), dtype=np.float32) +# h-coord of [0, 0] element of kernel, at output position [0, 0] +offset[0, 0, 0, 0] = 0.5 +# w-coord of [1, 0] element of kernel, at output position [0, 1] +offset[0, 5, 0, 1] = -0.1 + +mask = np.ones((1, 4, 2, 2), dtype=np.float32) +mask[0, 2, 1, 1] = 0.2 # [1, 0] element of kernel at output position [1, 1] + +node = onnx.helper.make_node( + "DeformConv", + inputs=["X", "W", "offset", "B", "mask"], + outputs=["Y"], + kernel_shape=[2, 2], + pads=[0, 0, 0, 0], +) +Y = np.array( + [ + [ + [ + [10.5, 12.9], # (1, 1, 2, 2) output tensor + [21.0, 19.4], + ] + ] + ] +).astype(np.float32) +expect( + node, + inputs=[X, W, offset, B, mask], + outputs=[Y], + name="test_deform_conv_with_mask_bias", +) +``` + +
+ + +
+deformconv_with_multiple_offset_groups + +```python +X = np.zeros((1, 2, 3, 3), dtype=np.float32) +X[0, 0] = np.reshape(np.arange(9).astype(np.float32), (3, 3)) +X[0, 1] = np.reshape(np.arange(8, -1, -1).astype(np.float32), (3, 3)) +X.shape = (1, 2, 3, 3) +W = np.ones((1, 2, 2, 2), dtype=np.float32) + +offset = np.zeros((1, 16, 2, 2), dtype=np.float32) +# h-coord of [0, 0] element of kernel in channel 0, at output position [0, 0] +offset[0, 0, 0, 0] = 0.5 +# w-coord of [1, 0] element of kernel in channel 1, at output position [0, 1] +offset[0, 13, 0, 1] = -0.1 + +node = onnx.helper.make_node( + "DeformConv", + inputs=["X", "W", "offset"], + outputs=["Y"], + kernel_shape=[2, 2], + pads=[0, 0, 0, 0], + offset_group=2, +) +Y = np.array( + [ + [ + [ + [33.5, 32.1], # (1, 1, 2, 2) output tensor + [32.0, 32.0], + ] + ] + ] +).astype(np.float32) +expect( + node, + inputs=[X, W, offset], + outputs=[Y], + name="test_deform_conv_with_multiple_offset_groups", +) +``` + +
+ + +### **DepthToSpace** + + DepthToSpace rearranges (permutes) data from depth into blocks of spatial data. + This is the reverse transformation of SpaceToDepth. More specifically, this op outputs a copy of + the input tensor where values from the depth dimension are moved in spatial blocks to the height + and width dimensions. By default, `mode` = `DCR`. + In the DCR mode, elements along the depth dimension from the input tensor are rearranged in the + following order: depth, column, and then row. The output y is computed from the input x as below: + + ``` + b, c, h, w = x.shape + tmp = np.reshape(x, [b, blocksize, blocksize, c // (blocksize**2), h, w]) + tmp = np.transpose(tmp, [0, 3, 4, 1, 5, 2]) + y = np.reshape(tmp, [b, c // (blocksize**2), h * blocksize, w * blocksize]) + ``` + + In the CRD mode, elements along the depth dimension from the input tensor are rearranged in the + following order: column, row, and the depth. The output y is computed from the input x as below: + + ``` + b, c, h, w = x.shape + tmp = np.reshape(x, [b, c // (blocksize ** 2), blocksize, blocksize, h, w]) + tmp = np.transpose(tmp, [0, 1, 4, 2, 5, 3]) + y = np.reshape(tmp, [b, c // (blocksize ** 2), h * blocksize, w * blocksize]) + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 11 + +#### Attributes + +
+
blocksize : int (required)
+
Blocks of [blocksize, blocksize] are moved.
+
mode : string (default is DCR)
+
DCR (default) for depth-column-row order re-arrangement. Use CRD for column-row-depth order.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor of [N, C/(blocksize * blocksize), H * blocksize, W * blocksize].
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ + +#### Examples + +
+crd_mode_example + +```python +node = onnx.helper.make_node( + "DepthToSpace", inputs=["x"], outputs=["y"], blocksize=2, mode="CRD" +) + +# (1, 8, 2, 3) input tensor +x = np.array( + [ + [ + [[0.0, 1.0, 2.0], [3.0, 4.0, 5.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0]], + [[18.0, 19.0, 20.0], [21.0, 22.0, 23.0]], + [[27.0, 28.0, 29.0], [30.0, 31.0, 32.0]], + [[36.0, 37.0, 38.0], [39.0, 40.0, 41.0]], + [[45.0, 46.0, 47.0], [48.0, 49.0, 50.0]], + [[54.0, 55.0, 56.0], [57.0, 58.0, 59.0]], + [[63.0, 64.0, 65.0], [66.0, 67.0, 68.0]], + ] + ] +).astype(np.float32) + +# (1, 2, 4, 6) output tensor +y = np.array( + [ + [ + [ + [0.0, 9.0, 1.0, 10.0, 2.0, 11.0], + [18.0, 27.0, 19.0, 28.0, 20.0, 29.0], + [3.0, 12.0, 4.0, 13.0, 5.0, 14.0], + [21.0, 30.0, 22.0, 31.0, 23.0, 32.0], + ], + [ + [36.0, 45.0, 37.0, 46.0, 38.0, 47.0], + [54.0, 63.0, 55.0, 64.0, 56.0, 65.0], + [39.0, 48.0, 40.0, 49.0, 41.0, 50.0], + [57.0, 66.0, 58.0, 67.0, 59.0, 68.0], + ], + ] + ] +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_depthtospace_crd_mode_example") +``` + +
+ + +
+default_mode_example + +```python +node = onnx.helper.make_node( + "DepthToSpace", inputs=["x"], outputs=["y"], blocksize=2, mode="DCR" +) + +# (1, 8, 2, 3) input tensor +x = np.array( + [ + [ + [[0.0, 1.0, 2.0], [3.0, 4.0, 5.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0]], + [[18.0, 19.0, 20.0], [21.0, 22.0, 23.0]], + [[27.0, 28.0, 29.0], [30.0, 31.0, 32.0]], + [[36.0, 37.0, 38.0], [39.0, 40.0, 41.0]], + [[45.0, 46.0, 47.0], [48.0, 49.0, 50.0]], + [[54.0, 55.0, 56.0], [57.0, 58.0, 59.0]], + [[63.0, 64.0, 65.0], [66.0, 67.0, 68.0]], + ] + ] +).astype(np.float32) + +# (1, 2, 4, 6) output tensor +y = np.array( + [ + [ + [ + [0.0, 18.0, 1.0, 19.0, 2.0, 20.0], + [36.0, 54.0, 37.0, 55.0, 38.0, 56.0], + [3.0, 21.0, 4.0, 22.0, 5.0, 23.0], + [39.0, 57.0, 40.0, 58.0, 41.0, 59.0], + ], + [ + [9.0, 27.0, 10.0, 28.0, 11.0, 29.0], + [45.0, 63.0, 46.0, 64.0, 47.0, 65.0], + [12.0, 30.0, 13.0, 31.0, 14.0, 32.0], + [48.0, 66.0, 49.0, 67.0, 50.0, 68.0], + ], + ] + ] +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_depthtospace_example") +``` + +
+ + +### **DequantizeLinear** + + The linear dequantization operator. It consumes a quantized tensor, a scale, and a zero point to compute the + full-precision tensor. The dequantization formula is `y = (x - x_zero_point) * x_scale`. `x_scale` and `x_zero_point` + must have the same shape, determining the quantization's granularity: a scalar for per-tensor/per-layer quantization, + a 1-D tensor for per-axis quantization, or have a rank identical to the input for blocked quantization. + See QuantizeLinear for details on quantization granularity. + + `x_zero_point` and `x` must have the same type. `x` and `y` must have the same shape. In the case of dequantizing + `int32`, there's no zero point (zero point is supposed to be 0). + `zero-point` is usually not used in the case of float8 and 4-bit types quantization, but the dequantization formula remains the same + for consistency. The output type is determined by the attribute `output_dtype`. If `output_dtype` is not supplied then the output type + is the same as `x_scale`. The output type also determines the precision of the multiplication operation. + + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 10, 13, 19, 21, 23, 24 + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`.
+
block_size : int (default is 0)
+
(Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given `x` shape `(D0, ..., Di, ..., Dn)`, `y_scale` shape `(S0, ... Si, ...Sn)` and `axis=i`, the accepted range is `[ceil(Di/Si), ceil(Di/(Si-1))-1]`
+
output_dtype : int (default is 0)
+
(Optional) The output data type. If not supplied, the output data type is inferred from `x_scale` data type (`T2`)
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D quantized input tensor to be de-quantized.
+
x_scale : T2
+
Scale for input `x`. For per-tensor/layer dequantization the scale is a scalar, for per per-axis dequantization it is a 1-D Tensor and for blocked dequantization it has the same shape as the input, except for one dimension in which blocking is performed.
+
x_zero_point (optional) : T1
+
Zero point for input `x`. Shape must match x_scale. It's optional. Zero point is 0 when it's not specified.
+
+ +#### Outputs + +
+
y : T3
+
N-D full precision output tensor. It has the same shape as input `x`. The data type is specified by the `output_dtype` attribute or, in its absence, the type of `x_scale`.
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(int32), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(uint2), tensor(int2)
+
The type of the inputs 'x_zero_point' and 'x'.
+
T2 : tensor(float), tensor(float16), tensor(bfloat16), tensor(float8e8m0)
+
The type of the input 'x_scale'.
+
T3 : tensor(float), tensor(float16), tensor(bfloat16)
+
The type of the output 'y'.
+
+ + +#### Examples + +
+axis + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], +) + +# 1-D tensor zero point and scale of size equal to axis 1 of the input tensor +x = np.array( + [ + [ + [[3, 89], [34, 200], [74, 59]], + [[5, 24], [24, 87], [32, 13]], + [[245, 99], [4, 142], [121, 102]], + ], + ], + dtype=np.uint8, +) +x_scale = np.array([2, 4, 5], dtype=np.float32) +x_zero_point = np.array([84, 24, 196], dtype=np.uint8) +y = ( + x.astype(np.float32) - x_zero_point.reshape(1, 3, 1, 1).astype(np.float32) +) * x_scale.reshape(1, 3, 1, 1) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_axis", +) +``` + +
+ + +
+blocked + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=1, + block_size=2, +) + +x = np.array( + [ + [ + [[3, 89], [34, 200], [74, 59]], + [[5, 24], [24, 87], [32, 13]], + [[5, 12], [12, 33], [65, 42]], + [[245, 99], [4, 142], [121, 102]], + ], + ], + dtype=np.uint8, +) + +x_scale = np.array( + [ + [ + [[3.0, 2.0], [4.0, 1.0], [2.0, 2.0]], + [[5.0, 2.0], [4.0, 3.0], [5.0, 2.0]], + ], + ], + dtype=np.float32, +) +x_zero_point = np.array( + [ + [ + [[1, 0], [0, 1], [2, 20]], + [[3, 2], [4, 3], [15, 2]], + ], + ], + dtype=np.uint8, +) + +# x.shape = (1, 4, 3, 2) +# x_scale.shape = (1, 2, 3, 2) +assert x_scale.shape == x_zero_point.shape +block_axis = 1 +# The block shape is [x.shape[i] // x_scale.shape[i] for i in range(len(x.shape))] = (1, 2, 1, 1) +assert all( + x.shape[i] == x_scale.shape[i] + for i in range(len(x.shape)) + if i != block_axis +) +assert x.shape[block_axis] % x_scale.shape[block_axis] == 0 +repeats = x.shape[block_axis] // x_scale.shape[block_axis] + +# Create element-wise scale and zero point +x_scale_elementwise = np.repeat(x_scale, repeats=repeats, axis=block_axis) +x_zero_point_elementwise = np.repeat( + x_zero_point, repeats=repeats, axis=block_axis +) + +y = ( + x.astype(np.float32) - x_zero_point_elementwise.astype(np.float32) +) * x_scale_elementwise + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_blocked", +) +``` + +
+ + +
+dequantizelinear + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], +) + +# scalar zero point and scale +x = np.array([0, 3, 128, 255]).astype(np.uint8) +x_scale = np.float32(2) +x_zero_point = np.uint8(128) +y = np.array([-256, -250, 0, 254], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear", +) +``` + +
+ + +
+e4m3fn + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.FLOAT8E4M3FN, [5], [0, 0.5, 1, 448, -104]) +x_scale = np.float32(2) +y = np.array([0.0, 1.0, 2.0, 896.0, -208.0], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale], + outputs=[y], + name="test_dequantizelinear_e4m3fn", +) +``` + +
+ + +
+e4m3fn_float16 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.FLOAT8E4M3FN, [5], [0, 0.5, 1, 448, -104]) +x_scale = np.float16(2) +y = np.array([0.0, 1.0, 2.0, 896.0, -208.0], dtype=np.float16) + +expect( + node, + inputs=[x, x_scale], + outputs=[y], + name="test_dequantizelinear_e4m3fn_float16", +) +``` + +
+ + +
+e4m3fn_zero_point + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.FLOAT8E4M3FN, [5], [0, 0.5, 1, 448, -104]) +zero_point = make_tensor("zero_point", TensorProto.FLOAT8E4M3FN, [1], [0]) +x_scale = np.float32(2) +y = np.array([0.0, 1.0, 2.0, 896.0, -208.0], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, zero_point], + outputs=[y], + name="test_dequantizelinear_e4m3fn_zero_point", +) +``` + +
+ + +
+e5m2 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.FLOAT8E5M2, [5], [0, 0.5, 1, 49152, -96]) +x_scale = np.float32(2) +y = np.array([0.0, 1.0, 2.0, 98304.0, -192.0], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale], + outputs=[y], + name="test_dequantizelinear_e5m2", +) +``` + +
+ + +
+float4e2m1 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.FLOAT4E2M1, [5], [0, 1, -1, 1.5, -4]) +x_scale = np.float32(2) +x_zero_point = make_tensor("x_zero_point", TensorProto.FLOAT4E2M1, (1,), [0]) +y = np.array([0, 2, -2, 3, -8], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_float4e2m1", +) +``` + +
+ + +
+int16 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], +) + +x = np.array([-300, -30, -1025, 1270]).astype(np.int16) +x_scale = np.float32(2) +x_zero_point = np.int16(-1024) +y = np.array([1448.0, 1988.0, -2.0, 4588.0], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_int16", +) +``` + +
+ + +
+int2 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.INT2, [4], [0, 1, -1, -2]) +x_scale = np.float32(2) +x_zero_point = make_tensor("x_zero_point", TensorProto.INT2, (1,), [1]) +y = np.array([-2, 0, -4, -6], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_int2", +) +``` + +
+ + +
+int4 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.INT4, [5], [0, 1, 7, -4, -8]) +x_scale = np.float32(2) +x_zero_point = make_tensor("x_zero_point", TensorProto.INT4, (1,), [1]) +y = np.array([-2, 0, 12, -10, -18], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_int4", +) +``` + +
+ + +
+uint16 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], +) + +x = np.array([30000, 31000, 32768, 33000]).astype(np.uint16) +x_scale = np.float32(2) +x_zero_point = np.uint16(32767) +y = np.array([-5534.0, -3534.0, 2.0, 466.0], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_uint16", +) +``` + +
+ + +
+uint2 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.UINT2, [4], [0, 1, 2, 3]) +x_scale = np.float32(2) +x_zero_point = make_tensor("x_zero_point", TensorProto.UINT2, (1,), [1]) +y = np.array([-2, 0, 2, 4], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_uint2", +) +``` + +
+ + +
+uint4 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.UINT4, [5], [0, 1, 7, 10, 15]) +x_scale = np.float32(2) +x_zero_point = make_tensor("x_zero_point", TensorProto.UINT4, (1,), [1]) +y = np.array([-2, 0, 12, 18, 28], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_uint4", +) +``` + +
+ + +### **Det** + + Det calculates determinant of a square matrix or batches of square matrices. + Det takes one input tensor of shape `[*, M, M]`, where `*` is zero or more batch dimensions, + and the inner-most 2 dimensions form square matrices. + The output is a tensor of shape `[*]`, containing the determinants of all input submatrices. + e.g., When the input is 2-D, the output is a scalar(shape is empty: `[]`). + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 11 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to floating-point tensors.
+
+ + +#### Examples + +
+2d + +```python +node = onnx.helper.make_node( + "Det", + inputs=["x"], + outputs=["y"], +) + +x = np.arange(4).reshape(2, 2).astype(np.float32) +y = np.linalg.det(x) # expect -2 +expect(node, inputs=[x], outputs=[y], name="test_det_2d") +``` + +
+ + +
+nd + +```python +node = onnx.helper.make_node( + "Det", + inputs=["x"], + outputs=["y"], +) + +x = np.array([[[1, 2], [3, 4]], [[1, 2], [2, 1]], [[1, 3], [3, 1]]]).astype( + np.float32 +) +y = np.linalg.det(x) # expect array([-2., -3., -8.]) +expect(node, inputs=[x], outputs=[y], name="test_det_nd") +``` + +
+ + +### **Div** + + Performs element-wise binary division (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + + For integer inputs, the result is computed using truncating division (rounding toward zero). + (Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 7, 13 + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ + +#### Examples + +
+div + +```python +node = onnx.helper.make_node( + "Div", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([3, 4]).astype(np.float32) +y = np.array([1, 2]).astype(np.float32) +z = x / y # expected output [3., 2.] +expect(node, inputs=[x, y], outputs=[z], name="test_div_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.rand(3, 4, 5).astype(np.float32) + 1.0 +z = x / y +expect(node, inputs=[x, y], outputs=[z], name="test_div") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.int8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int8) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_int8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.int16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int16) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_int16") + +x = np.array([-3, 3, -3, 3], dtype=np.int32) +y = np.array([2, 2, -2, -2], dtype=np.int32) +z = np.array([-1, 1, 1, -1], dtype=np.int32) +expect(node, inputs=[x, y], outputs=[z], name="test_div_int32_trunc") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_uint64") +``` + +
+ + +
+div_broadcast + +```python +node = onnx.helper.make_node( + "Div", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.rand(5).astype(np.float32) + 1.0 +z = x / y +expect(node, inputs=[x, y], outputs=[z], name="test_div_bcast") +``` + +
+ + +### **Dropout** + + Dropout takes an input floating-point tensor, an optional input ratio (floating-point scalar) and an optional input training_mode (boolean scalar). It produces two tensor outputs, + output (floating-point tensor) and mask (optional `Tensor`). If `training_mode` is true then the output Y will be a random dropout; + Note that this Dropout scales the masked input data by the following equation, so to convert the trained model into inference mode, + the user can simply not pass `training_mode` input or set it to false. + ``` + output = scale * data * mask, + ``` + where + ``` + scale = 1. / (1. - ratio). + ``` + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 7, 10, 12, 13 + +#### Attributes + +
+
seed : int
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs (1 - 3) + +
+
data (differentiable) : T
+
The input data as Tensor.
+
ratio (optional, non-differentiable) : T1
+
The ratio of random dropout, with value in [0, 1). If set to 0, the output would be a simple copy of the input. If it's non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
+
training_mode (optional, non-differentiable) : T2
+
If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
+
+ +#### Outputs (1 - 2) + +
+
output (differentiable) : T
+
The output.
+
mask (optional, non-differentiable) : T2
+
The output mask.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input and output types to float tensors.
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input 'ratio' types to float tensors.
+
T2 : tensor(bool)
+
Constrain output 'mask' types to boolean tensors.
+
+ + +#### Examples + +
+default + +```python +seed = np.int64(0) +node = onnx.helper.make_node("Dropout", inputs=["x"], outputs=["y"], seed=seed) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = dropout(x) +expect(node, inputs=[x], outputs=[y], name="test_dropout_default") +``` + +
+ + +
+default_mask + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x"], outputs=["y", "z"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y, z = dropout(x, return_mask=True) +expect(node, inputs=[x], outputs=[y, z], name="test_dropout_default_mask") +``` + +
+ + +
+default_mask_ratio + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r"], outputs=["y", "z"], seed=seed +) + +r = np.float32(0.1) +x = np.random.randn(3, 4, 5).astype(np.float32) +y, z = dropout(x, r, return_mask=True) +expect( + node, inputs=[x, r], outputs=[y, z], name="test_dropout_default_mask_ratio" +) +``` + +
+ + +
+default_old + +```python +node = onnx.helper.make_node( + "Dropout", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = x +expect( + node, + inputs=[x], + outputs=[y], + name="test_dropout_default_old", + opset_imports=[helper.make_opsetid("", 11)], +) +``` + +
+ + +
+default_ratio + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r"], outputs=["y"], seed=seed +) + +r = np.float32(0.1) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = dropout(x, r) +expect(node, inputs=[x, r], outputs=[y], name="test_dropout_default_ratio") +``` + +
+ + +
+random_old + +```python +node = onnx.helper.make_node( + "Dropout", + inputs=["x"], + outputs=["y"], + ratio=0.2, +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = x +expect( + node, + inputs=[x], + outputs=[y], + name="test_dropout_random_old", + opset_imports=[helper.make_opsetid("", 11)], +) +``` + +
+ + +
+training + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.75) +t = np.bool_(True) +y = dropout(x, r, training_mode=t) +expect(node, inputs=[x, r, t], outputs=[y], name="test_training_dropout") +``` + +
+ + +
+training_default + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.5) +t = np.bool_(True) +y = dropout(x, r, training_mode=t) +expect( + node, inputs=[x, r, t], outputs=[y], name="test_training_dropout_default" +) +``` + +
+ + +
+training_default_ratio_mask + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y", "z"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.5) +t = np.bool_(True) +y, z = dropout(x, r, training_mode=t, return_mask=True) +expect( + node, + inputs=[x, r, t], + outputs=[y, z], + name="test_training_dropout_default_mask", +) +``` + +
+ + +
+training_default_zero_ratio + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.0) +t = np.bool_(True) +y = dropout(x, r, training_mode=t) +expect( + node, inputs=[x, r, t], outputs=[y], name="test_training_dropout_zero_ratio" +) +``` + +
+ + +
+training_default_zero_ratio_mask + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y", "z"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.0) +t = np.bool_(True) +y, z = dropout(x, r, training_mode=t, return_mask=True) +expect( + node, + inputs=[x, r, t], + outputs=[y, z], + name="test_training_dropout_zero_ratio_mask", +) +``` + +
+ + +
+training_ratio_mask + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y", "z"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.75) +t = np.bool_(True) +y, z = dropout(x, r, training_mode=t, return_mask=True) +expect( + node, inputs=[x, r, t], outputs=[y, z], name="test_training_dropout_mask" +) +``` + +
+ + +### **DynamicQuantizeLinear** + + A Function to fuse calculation for Scale, Zero Point and FP32->8Bit conversion of FP32 Input data. + Outputs Scale, ZeroPoint and Quantized Input for a given FP32 Input. + Scale is calculated as: + ``` + y_scale = (maximum(0, max(x)) - minimum(0, min(x))) / (qmax - qmin) + ``` + + * where qmax and qmin are max and min values for quantization range i.e. [0, 255] in case of uint8 + * data range is adjusted to include 0. + + Zero point is calculated as: + ``` + intermediate_zero_point = qmin - min(x)/y_scale + y_zero_point = cast(round(saturate(intermediate_zero_point))) + ``` + + * where qmax and qmin are max and min values for quantization range .i.e [0, 255] in case of uint8 + * for saturation, it saturates to [0, 255] if it's uint8, or [-127, 127] if it's int8. Right now only uint8 is supported. + * rounding to nearest ties to even. + + Data quantization formula is: + ``` + y = saturate (round (x / y_scale) + y_zero_point) + ``` + + * for saturation, it saturates to [0, 255] if it's uint8, or [-127, 127] if it's int8. Right now only uint8 is supported. + * rounding to nearest ties to even. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
x : T1
+
Input tensor
+
+ +#### Outputs + +
+
y : T2
+
Quantized output tensor
+
y_scale : tensor(float)
+
Output scale. It's a scalar, which means a per-tensor/layer quantization.
+
y_zero_point : T2
+
Output zero point. It's a scalar, which means a per-tensor/layer quantization.
+
+ +#### Type Constraints + +
+
T1 : tensor(float)
+
Constrain 'x' to float tensor.
+
T2 : tensor(uint8)
+
Constrain 'y_zero_point' and 'y' to 8-bit unsigned integer tensor.
+
+ + +#### Examples + +
+dynamicquantizelinear + +```python +node = onnx.helper.make_node( + "DynamicQuantizeLinear", + inputs=["x"], + outputs=["y", "y_scale", "y_zero_point"], +) + +# expected scale 0.0196078438 and zero point 153 +X = np.array([0, 2, -3, -2.5, 1.34, 0.5]).astype(np.float32) +x_min = np.minimum(0, np.min(X)) +x_max = np.maximum(0, np.max(X)) +Y_Scale = np.float32((x_max - x_min) / (255 - 0)) # uint8 -> [0, 255] +Y_ZeroPoint = np.clip(round((0 - x_min) / Y_Scale), 0, 255).astype(np.uint8) +Y = np.clip(np.round(X / Y_Scale) + Y_ZeroPoint, 0, 255).astype(np.uint8) + +expect( + node, + inputs=[X], + outputs=[Y, Y_Scale, Y_ZeroPoint], + name="test_dynamicquantizelinear", +) + +# expected scale 0.0156862754 and zero point 255 +X = np.array([-1.0, -2.1, -1.3, -2.5, -3.34, -4.0]).astype(np.float32) +x_min = np.minimum(0, np.min(X)) +x_max = np.maximum(0, np.max(X)) +Y_Scale = np.float32((x_max - x_min) / (255 - 0)) # uint8 -> [0, 255] +Y_ZeroPoint = np.clip(round((0 - x_min) / Y_Scale), 0, 255).astype(np.uint8) +Y = np.clip(np.round(X / Y_Scale) + Y_ZeroPoint, 0, 255).astype(np.uint8) + +expect( + node, + inputs=[X], + outputs=[Y, Y_Scale, Y_ZeroPoint], + name="test_dynamicquantizelinear_max_adjusted", +) + +X = ( + np.array([1, 2.1, 1.3, 2.5, 3.34, 4.0, 1.5, 2.6, 3.9, 4.0, 3.0, 2.345]) + .astype(np.float32) + .reshape((3, 4)) +) + +# expected scale 0.0156862754 and zero point 0 +x_min = np.minimum(0, np.min(X)) +x_max = np.maximum(0, np.max(X)) +Y_Scale = np.float32((x_max - x_min) / (255 - 0)) # uint8 -> [0, 255] +Y_ZeroPoint = np.clip(round((0 - x_min) / Y_Scale), 0, 255).astype(np.uint8) +Y = np.clip(np.round(X / Y_Scale) + Y_ZeroPoint, 0, 255).astype(np.uint8) + +expect( + node, + inputs=[X], + outputs=[Y, Y_Scale, Y_ZeroPoint], + name="test_dynamicquantizelinear_min_adjusted", +) +``` + +
+ + +### **Einsum** + + An einsum of the form `term1, term2 -> output-term` produces an output tensor using the following equation + + ``` + output[output-term] = reduce-sum( input1[term1] * input2[term2] ) + ``` + + where the reduce-sum performs a summation over all the indices occurring in the input terms (term1, term2) + that do not occur in the output-term. + + The Einsum operator evaluates algebraic tensor operations on a sequence of tensors, using the Einstein summation + convention. The equation string contains a comma-separated sequence of lower case letters. Each term corresponds to + an operand tensor, and the characters within the terms correspond to operands dimensions. + + This sequence may be followed by "->" to separate the left and right hand side of the equation. + If the equation contains "->" followed by the right-hand side, the explicit (not classical) form of the Einstein + summation is performed, and the right-hand side indices indicate output tensor dimensions. In other cases, + output indices are (implicitly) set to the alphabetically sorted sequence of indices appearing exactly once in the + equation. + + When a dimension character is repeated in the left-hand side, it represents summation along the dimension. + + The equation may contain ellipsis ("...") to enable broadcasting. Ellipsis must indicate a fixed number of dimensions. + Specifically, every occurrence of ellipsis in the equation must represent the same number of dimensions. + The right-hand side may contain exactly one ellipsis. In implicit mode, the ellipsis dimensions are set to the + beginning of the output. The equation string may contain space (U+0020) character. + +#### Version + +This version of the operator has been available since version 12 of the default ONNX operator set. + +#### Attributes + +
+
equation : string (required)
+
Einsum expression string.
+
+ +#### Inputs (1 - ∞) + +
+
Inputs (variadic, differentiable) : T
+
Operands
+
+ +#### Outputs + +
+
Output (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to all numerical tensor types.
+
+ + +#### Examples + +
+einsum_batch_diagonal + +```python +Eqn = "...ii ->...i" +node = onnx.helper.make_node( + "Einsum", inputs=["x"], outputs=["y"], equation=Eqn +) + +X = np.random.randn(3, 5, 5) +Z = einsum_reference_implementation(Eqn, (X,)) + +expect(node, inputs=[X], outputs=[Z], name="test_einsum_batch_diagonal") +``` + +
+ + +
+einsum_batch_matmul + +```python +Eqn = "bij, bjk -> bik" +node = onnx.helper.make_node( + "Einsum", inputs=["x", "y"], outputs=["z"], equation=Eqn +) + +X = np.random.randn(5, 2, 3) +Y = np.random.randn(5, 3, 4) +Z = einsum_reference_implementation(Eqn, (X, Y)) + +expect(node, inputs=[X, Y], outputs=[Z], name="test_einsum_batch_matmul") +``` + +
+ + +
+einsum_inner_prod + +```python +Eqn = "i,i" +node = onnx.helper.make_node( + "Einsum", inputs=["x", "y"], outputs=["z"], equation=Eqn +) + +X = np.random.randn(5) +Y = np.random.randn(5) +Z = einsum_reference_implementation(Eqn, (X, Y)) + +expect(node, inputs=[X, Y], outputs=[Z], name="test_einsum_inner_prod") +``` + +
+ + +
+einsum_scalar + +```python +Eqn = "->" +node = onnx.helper.make_node( + "Einsum", inputs=["x"], outputs=["y"], equation=Eqn +) + +X = np.array(5.0) # scalar input +Z = einsum_reference_implementation(Eqn, (X,)) + +expect(node, inputs=[X], outputs=[Z], name="test_einsum_scalar") +``` + +
+ + +
+einsum_sum + +```python +Eqn = "ij->i" +node = onnx.helper.make_node( + "Einsum", inputs=["x"], outputs=["y"], equation=Eqn +) + +X = np.random.randn(3, 4) +Z = einsum_reference_implementation(Eqn, (X,)) + +expect(node, inputs=[X], outputs=[Z], name="test_einsum_sum") +``` + +
+ + +
+einsum_transpose + +```python +Eqn = "ij->ji" +node = onnx.helper.make_node( + "Einsum", inputs=["x"], outputs=["y"], equation=Eqn +) + +X = np.random.randn(3, 4) +Y = einsum_reference_implementation(Eqn, (X,)) + +expect(node, inputs=[X], outputs=[Y], name="test_einsum_transpose") +``` + +
+ + +### **Elu** + + Elu takes one input data (Tensor) and produces one output data + (Tensor) where the function `f(x) = alpha * (exp(x) - 1.) for x < + 0`, `f(x) = x for x >= 0`., is applied to the tensor elementwise. + + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Coefficient of ELU.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+elu + +```python +node = onnx.helper.make_node("Elu", inputs=["x"], outputs=["y"], alpha=2.0) + +x = np.array([-1, 0, 1]).astype(np.float32) +# expected output [-1.2642411, 0., 1.] +y = np.clip(x, 0, np.inf) + (np.exp(np.clip(x, -np.inf, 0)) - 1) * 2.0 +expect(node, inputs=[x], outputs=[y], name="test_elu_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + (np.exp(np.clip(x, -np.inf, 0)) - 1) * 2.0 +expect(node, inputs=[x], outputs=[y], name="test_elu") +``` + +
+ + +
+elu_default + +```python +default_alpha = 1.0 +node = onnx.helper.make_node( + "Elu", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + (np.exp(np.clip(x, -np.inf, 0)) - 1) * default_alpha +expect(node, inputs=[x], outputs=[y], name="test_elu_default") +``` + +
+ + +### **Equal** + + Returns the tensor resulted from performing the `equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +Other versions of this operator: 1, 7, 11, 13 + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(string)
+
Constrain input types to all (non-complex) tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ + +#### Examples + +
+equal + +```python +node = onnx.helper.make_node( + "Equal", + inputs=["x", "y"], + outputs=["z"], +) + +x = (np.random.randn(3, 4, 5) * 10).astype(np.int32) +y = (np.random.randn(3, 4, 5) * 10).astype(np.int32) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal") + +x = (np.random.randn(3, 4, 5) * 10).astype(np.int8) +y = (np.random.randn(3, 4, 5) * 10).astype(np.int8) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_int8") + +x = (np.random.randn(3, 4, 5) * 10).astype(np.int16) +y = (np.random.randn(3, 4, 5) * 10).astype(np.int16) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_uint64") +``` + +
+ + +
+equal_broadcast + +```python +node = onnx.helper.make_node( + "Equal", + inputs=["x", "y"], + outputs=["z"], +) + +x = (np.random.randn(3, 4, 5) * 10).astype(np.int32) +y = (np.random.randn(5) * 10).astype(np.int32) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_bcast") +``` + +
+ + +
+equal_string + +```python +node = onnx.helper.make_node( + "Equal", + inputs=["x", "y"], + outputs=["z"], +) +x = np.array(["string1", "string2"], dtype=np.dtype(object)) +y = np.array(["string1", "string3"], dtype=np.dtype(object)) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_string") +``` + +
+ + +
+equal_string_broadcast + +```python +node = onnx.helper.make_node( + "Equal", + inputs=["x", "y"], + outputs=["z"], +) +x = np.array(["string1", "string2"], dtype=np.dtype(object)) +y = np.array(["string1"], dtype=np.dtype(object)) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_string_broadcast") +``` + +
+ + +### **Erf** + + Computes the error function of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The error function of the input tensor computed element-wise. It has the same shape and type of the input.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+erf + +```python +node = onnx.helper.make_node( + "Erf", + inputs=["x"], + outputs=["y"], +) + +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +y = np.vectorize(math.erf)(x).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_erf") +``` + +
+ + +### **Exp** + + Calculates the exponential of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The exponential of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+exp + +```python +node = onnx.helper.make_node( + "Exp", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.exp(x) # expected output [0.36787945, 1., 2.71828175] +expect(node, inputs=[x], outputs=[y], name="test_exp_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.exp(x) +expect(node, inputs=[x], outputs=[y], name="test_exp") +``` + +
+ + +### **Expand** + + Broadcast the input tensor following the given shape and the broadcast rule. + The broadcast rule is similar to numpy.array(input) * numpy.ones(shape): + Dimensions are right alignment; + Two corresponding dimensions must have the same value, or one of them is equal to 1. + Also, this operator is similar to numpy.broadcast_to(input, shape), + but the major difference is numpy.broadcast_to() does not allow shape to be smaller than input.size(). + It is possible that the output.shape is not equal to shape, when some dimensions in shape is equal to 1, + or the shape.ndim < input.shape.ndim. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 8 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
shape (non-differentiable) : tensor(int64)
+
A 1-D tensor indicates the shape you want to expand to, following the broadcast rule
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensors.
+
+ + +#### Examples + +
+dim_changed + +```python +node = onnx.helper.make_node( + "Expand", + inputs=["data", "new_shape"], + outputs=["expanded"], +) +shape = [3, 1] +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[1.], [2.], [3.]] +new_shape = [2, 1, 6] +expanded = data * np.ones(new_shape, dtype=np.float32) +# print(expanded) +# [[[1., 1., 1., 1., 1., 1.], +# [2., 2., 2., 2., 2., 2.], +# [3., 3., 3., 3., 3., 3.]], +# +# [[1., 1., 1., 1., 1., 1.], +# [2., 2., 2., 2., 2., 2.], +# [3., 3., 3., 3., 3., 3.]]] +new_shape = np.array(new_shape, dtype=np.int64) +expect( + node, + inputs=[data, new_shape], + outputs=[expanded], + name="test_expand_dim_changed", +) +``` + +
+ + +
+dim_unchanged + +```python +node = onnx.helper.make_node( + "Expand", + inputs=["data", "new_shape"], + outputs=["expanded"], +) +shape = [3, 1] +new_shape = [3, 4] +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[1.], [2.], [3.]] +expanded = np.tile(data, 4) +# print(expanded) +# [[1., 1., 1., 1.], +# [2., 2., 2., 2.], +# [3., 3., 3., 3.]] +new_shape = np.array(new_shape, dtype=np.int64) +expect( + node, + inputs=[data, new_shape], + outputs=[expanded], + name="test_expand_dim_unchanged", +) +``` + +
+ + +### **EyeLike** + + Generate a 2D tensor (matrix) with ones on the diagonal and zeros everywhere else. Only 2D + tensors are supported, i.e. input T1 must be of rank 2. The shape of the output tensor is the + same as the input tensor. The data type can be specified by the 'dtype' argument. If + 'dtype' is not specified, then the type of input tensor is used. By default, the main diagonal + is populated with ones, but attribute 'k' can be used to populate upper or lower diagonals. + The 'dtype' argument must be one of the data types specified in the 'DataType' enum field in the + TensorProto message and be valid as an output type. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Attributes + +
+
dtype : int
+
(Optional) The data type for the elements of the output tensor. If not specified, the data type of the input tensor T1 is used.
+
k : int (default is 0)
+
(Optional) Index of the diagonal to be populated with ones. Default is 0. If T2 is the output, this op sets T2[i, i+k] = 1. k = 0 populates the main diagonal, k > 0 populates an upper diagonal, and k < 0 populates a lower diagonal.
+
+ +#### Inputs + +
+
input : T1
+
2D input tensor to copy shape, and optionally, type information from.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor, same shape as input tensor T1.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool)
+
Constrain input types. Strings and complex are not supported.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool)
+
Constrain output types. Strings and complex are not supported.
+
+ + +#### Examples + +
+populate_off_main_diagonal + +```python +shape = (4, 5) +off_diagonal_offset = 1 +node = onnx.helper.make_node( + "EyeLike", + inputs=["x"], + outputs=["y"], + k=off_diagonal_offset, + dtype=onnx.TensorProto.FLOAT, +) + +x = np.random.randint(0, 100, size=shape, dtype=np.int32) +y = np.eye(shape[0], shape[1], k=off_diagonal_offset, dtype=np.float32) +expect( + node, + inputs=[x], + outputs=[y], + name="test_eyelike_populate_off_main_diagonal", +) +``` + +
+ + +
+with_dtype + +```python +shape = (3, 4) +node = onnx.helper.make_node( + "EyeLike", + inputs=["x"], + outputs=["y"], + dtype=onnx.TensorProto.DOUBLE, +) + +x = np.random.randint(0, 100, size=shape, dtype=np.int32) +y = np.eye(shape[0], shape[1], dtype=np.float64) +expect(node, inputs=[x], outputs=[y], name="test_eyelike_with_dtype") +``` + +
+ + +
+without_dtype + +```python +shape = (4, 4) +node = onnx.helper.make_node( + "EyeLike", + inputs=["x"], + outputs=["y"], +) + +x = np.random.randint(0, 100, size=shape, dtype=np.int32) +y = np.eye(shape[0], shape[1], dtype=np.int32) +expect(node, inputs=[x], outputs=[y], name="test_eyelike_without_dtype") +``` + +
+ + +### **Flatten** + + Flattens the input tensor into a 2D matrix. If input tensor has shape + (d_0, d_1, ... d_n) then the output will have shape + (d_0 X d_1 ... d_(axis-1), d_axis X d_(axis+1) ... X dn). + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 9, 11, 13, 21, 23, 24 + +#### Attributes + +
+
axis : int (default is 1)
+
Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 ... d_n), where the shape of the input tensor is (d_0, d_1, ... d_n).
+
+ +#### Inputs + +
+
input (differentiable) : T
+
A tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output to all tensor types up to IRv13.
+
+ + +#### Examples + +
+flatten + +```python +shape = (2, 3, 4, 5) +a = np.random.random_sample(shape).astype(np.float32) + +for i in range(len(shape)): + node = onnx.helper.make_node( + "Flatten", + inputs=["a"], + outputs=["b"], + axis=i, + ) + + new_shape = (1, -1) if i == 0 else (np.prod(shape[0:i]).astype(int), -1) + b = np.reshape(a, new_shape) + expect(node, inputs=[a], outputs=[b], name="test_flatten_axis" + str(i)) +``` + +
+ + +
+flatten_negative_axis + +```python +shape = (2, 3, 4, 5) +a = np.random.random_sample(shape).astype(np.float32) + +for i in range(-len(shape), 0): + node = onnx.helper.make_node( + "Flatten", + inputs=["a"], + outputs=["b"], + axis=i, + ) + + new_shape = (np.prod(shape[0:i]).astype(int), -1) + b = np.reshape(a, new_shape) + expect( + node, + inputs=[a], + outputs=[b], + name="test_flatten_negative_axis" + str(abs(i)), + ) +``` + +
+ + +
+flatten_with_default_axis + +```python +node = onnx.helper.make_node( + "Flatten", + inputs=["a"], + outputs=["b"], # Default value for axis: axis=1 +) + +shape = (5, 4, 3, 2) +a = np.random.random_sample(shape).astype(np.float32) +new_shape = (5, 24) +b = np.reshape(a, new_shape) +expect(node, inputs=[a], outputs=[b], name="test_flatten_default_axis") +``` + +
+ + +### **Floor** + + Floor takes one input data (Tensor) and produces one output data + (Tensor) where the floor is, y = floor(x), is applied to + the tensor elementwise. If x is integral, +0, -0, NaN, or infinite, x itself is returned. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+floor + +```python +node = onnx.helper.make_node( + "Floor", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.5, 1.2, 2]).astype(np.float32) +y = np.floor(x) # expected output [-2., 1., 2.] +expect(node, inputs=[x], outputs=[y], name="test_floor_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.floor(x) +expect(node, inputs=[x], outputs=[y], name="test_floor") +``` + +
+ + +### **GRU** + + Computes an one-layer GRU. This operator is usually supported via some custom + implementation such as CuDNN. + + Notations: + + * `X` - input tensor + * `z` - update gate + * `r` - reset gate + * `h` - hidden gate + * `t` - time step (t-1 means previous time step) + * `W[zrh]` - W parameter weight matrix for update, reset, and hidden gates + * `R[zrh]` - R recurrence weight matrix for update, reset, and hidden gates + * `Wb[zrh]` - W bias vectors for update, reset, and hidden gates + * `Rb[zrh]` - R bias vectors for update, reset, and hidden gates + * `WB[zrh]` - W parameter weight matrix for backward update, reset, and hidden gates + * `RB[zrh]` - R recurrence weight matrix for backward update, reset, and hidden gates + * `WBb[zrh]` - W bias vectors for backward update, reset, and hidden gates + * `RBb[zrh]` - R bias vectors for backward update, reset, and hidden gates + * `H` - Hidden state + * `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + * Relu(x) - max(0, x) + * Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + * Sigmoid(x) - 1/(1 + e^{-x}) + + NOTE: + Below are optional + + * Affine(x) - alpha * x + beta + * LeakyRelu(x) - x if x >= 0 else alpha * x + * ThresholdedRelu(x) - x if x >= alpha else 0 + * ScaledTanh(x) - alpha * Tanh(beta * x) + * HardSigmoid(x) - min(max(alpha * x + beta, 0), 1) + * Elu(x) - x if x >= 0 else alpha * (e^x - 1) + * Softsign(x) - x/(1 + |x|) + * Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh): + + * zt = f(Xt*(Wz^T) + Ht-1*(Rz^T) + Wbz + Rbz) + * rt = f(Xt*(Wr^T) + Ht-1*(Rr^T) + Wbr + Rbr) + * ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*(Rh^T) + Rbh + Wbh) # default, when linear_before_reset = 0 + * ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*(Rh^T) + Rbh)) + Wbh) # when linear_before_reset != 0 + * Ht = (1 - zt) (.) ht + zt (.) Ht-1 + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 3, 7, 14 + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings
+
A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
layout : int (default is 0)
+
The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size].
+
linear_before_reset : int (default is 0)
+
When computing the output of the hidden gate, apply the linear transformation before multiplying by the output of the reset gate.
+
+ +#### Inputs (3 - 6) + +
+
X (differentiable) : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W (differentiable) : T
+
The weight tensor for the gates. Concatenation of `W[zrh]` and `WB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, input_size]`.
+
R (differentiable) : T
+
The recurrence weight tensor. Concatenation of `R[zrh]` and `RB[zrh]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 3*hidden_size, hidden_size]`.
+
B (optional, differentiable) : T
+
The bias tensor for the gates. Concatenation of `[Wb[zrh], Rb[zrh]]` and `[WBb[zrh], RBb[zrh]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 6*hidden_size]`. Optional: If not specified - assumed to be 0
+
sequence_lens (optional, non-differentiable) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional, non-differentiable) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs (0 - 2) + +
+
Y (optional, differentiable) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional, differentiable) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ + +#### Examples + +
+batchwise + +```python +input = np.array([[[1.0, 2.0]], [[3.0, 4.0]], [[5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 6 +number_of_gates = 3 +weight_scale = 0.2 +layout = 1 + +node = onnx.helper.make_node( + "GRU", + inputs=["X", "W", "R"], + outputs=["Y", "Y_h"], + hidden_size=hidden_size, + layout=layout, +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +gru = GRUHelper(X=input, W=W, R=R, layout=layout) +Y, Y_h = gru.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y.astype(np.float32), Y_h.astype(np.float32)], + name="test_gru_batchwise", +) +``` + +
+ + +
+defaults + +```python +input = np.array([[[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 5 +weight_scale = 0.1 +number_of_gates = 3 + +node = onnx.helper.make_node( + "GRU", inputs=["X", "W", "R"], outputs=["", "Y_h"], hidden_size=hidden_size +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +gru = GRUHelper(X=input, W=W, R=R) +_, Y_h = gru.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y_h.astype(np.float32)], + name="test_gru_defaults", +) +``` + +
+ + +
+initial_bias + +```python +input = np.array([[[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]]]).astype( + np.float32 +) + +input_size = 3 +hidden_size = 3 +weight_scale = 0.1 +custom_bias = 0.1 +number_of_gates = 3 + +node = onnx.helper.make_node( + "GRU", + inputs=["X", "W", "R", "B"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +# Adding custom bias +W_B = custom_bias * np.ones((1, number_of_gates * hidden_size)).astype( + np.float32 +) +R_B = np.zeros((1, number_of_gates * hidden_size)).astype(np.float32) +B = np.concatenate((W_B, R_B), axis=1) + +gru = GRUHelper(X=input, W=W, R=R, B=B) +_, Y_h = gru.step() +expect( + node, + inputs=[input, W, R, B], + outputs=[Y_h.astype(np.float32)], + name="test_gru_with_initial_bias", +) +``` + +
+ + +
+seq_length + +```python +input = np.array( + [ + [[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]], + [[10.0, 11.0, 12.0], [13.0, 14.0, 15.0], [16.0, 17.0, 18.0]], + ] +).astype(np.float32) + +input_size = 3 +hidden_size = 5 +number_of_gates = 3 + +node = onnx.helper.make_node( + "GRU", + inputs=["X", "W", "R", "B"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +W = np.random.randn(1, number_of_gates * hidden_size, input_size).astype( + np.float32 +) +R = np.random.randn(1, number_of_gates * hidden_size, hidden_size).astype( + np.float32 +) + +# Adding custom bias +W_B = np.random.randn(1, number_of_gates * hidden_size).astype(np.float32) +R_B = np.random.randn(1, number_of_gates * hidden_size).astype(np.float32) +B = np.concatenate((W_B, R_B), axis=1) + +gru = GRUHelper(X=input, W=W, R=R, B=B) +_, Y_h = gru.step() +expect( + node, + inputs=[input, W, R, B], + outputs=[Y_h.astype(np.float32)], + name="test_gru_seq_length", +) +``` + +
+ + +### **Gather** + + Given `data` tensor of rank r >= 1, and `indices` tensor of rank q, gather + entries of the axis dimension of `data` (by default outer-most one as axis=0) indexed by `indices`, and concatenates + them in an output tensor of rank q + (r - 1). + + It is an indexing operation that indexes into the input `data` along a single (specified) axis. + Each entry in `indices` produces a `r-1` dimensional slice of the input tensor. + The entire operation produces, conceptually, a `q`-dimensional tensor of `r-1` dimensional slices, + which is arranged into a `q + (r-1)`-dimensional tensor, with the `q` dimensions taking the + place of the original `axis` that is being indexed into. + + The following few examples illustrate how `Gather` works for specific shapes of `data`, + `indices`, and given value of `axis`: + | data shape | indices shape | axis | output shape | output equation | + | --- | --- | --- | --- | --- | + | (P, Q) | ( ) (a scalar) | 0 | (Q) | output[q] = data[indices, q] | + | (P, Q, R) | ( ) (a scalar) | 1 | (P, R) | output[p, r] = data[p, indices, r] | + | (P, Q) | (R, S) | 0 | (R, S, Q) | output[r, s, q] = data[ [indices[r, s], q] | + | (P, Q) | (R, S) | 1 | (P, R, S) | output[p, r, s] = data[ p, indices[r, s]] | + + More generally, if `axis = 0`, let `k = indices[i_{0}, ..., i_{q-1}]` + then `output[i_{0}, ..., i_{q-1}, j_{0}, ..., j_{r-2}] = input[k , j_{0}, ..., j_{r-2}]`: + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + indices = [ + [0, 1], + [1, 2], + ] + output = [ + [ + [1.0, 1.2], + [2.3, 3.4], + ], + [ + [2.3, 3.4], + [4.5, 5.7], + ], + ] + ``` + + If `axis = 1`, let `k = indices[i_{0}, ..., i_{q-1}]` + then `output[j_{0}, i_{0}, ..., i_{q-1}, j_{1}, ..., j_{r-2}] = input[j_{0}, k, j_{1}, ..., j_{r-2}]`: + + ``` + data = [ + [1.0, 1.2, 1.9], + [2.3, 3.4, 3.9], + [4.5, 5.7, 5.9], + ] + indices = [ + [0, 2], + ] + axis = 1, + output = [ + [[1.0, 1.9]], + [[2.3, 3.9]], + [[4.5, 5.9]], + ] + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 11 + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : Tind
+
Tensor of int32/int64 indices, of any rank q. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank q + (r - 1).
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ + +#### Examples + +
+gather_0 + +```python +node = onnx.helper.make_node( + "Gather", + inputs=["data", "indices"], + outputs=["y"], + axis=0, +) +data = np.random.randn(5, 4, 3, 2).astype(np.float32) +indices = np.array([0, 1, 3]) +y = np.take(data, indices, axis=0) + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_0", +) +``` + +
+ + +
+gather_1 + +```python +node = onnx.helper.make_node( + "Gather", + inputs=["data", "indices"], + outputs=["y"], + axis=1, +) +data = np.random.randn(5, 4, 3, 2).astype(np.float32) +indices = np.array([0, 1, 3]) +y = np.take(data, indices, axis=1) + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_1", +) +``` + +
+ + +
+gather_2d_indices + +```python +node = onnx.helper.make_node( + "Gather", + inputs=["data", "indices"], + outputs=["y"], + axis=1, +) +data = np.random.randn(3, 3).astype(np.float32) +indices = np.array([[0, 2]]) +y = np.take(data, indices, axis=1) + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_2d_indices", +) +``` + +
+ + +
+gather_negative_indices + +```python +node = onnx.helper.make_node( + "Gather", + inputs=["data", "indices"], + outputs=["y"], + axis=0, +) +data = np.arange(10).astype(np.float32) +indices = np.array([0, -9, -10]) +y = np.take(data, indices, axis=0) + +# print(y) +# [0. 1. 0.] + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_negative_indices", +) +``` + +
+ + +### **GatherElements** + + GatherElements takes two inputs `data` and `indices` of the same rank r >= 1 + and an optional attribute `axis` that identifies an axis of `data` + (by default, the outer-most axis, that is axis 0). It is an indexing operation + that produces its output by indexing into the input data tensor at index + positions determined by elements of the `indices` tensor. + Its output shape is the same as the shape of `indices` and consists of one value + (gathered from the `data`) for each element in `indices`. + + For instance, in the 3-D case (r = 3), the output produced is determined + by the following equations: + ``` + out[i][j][k] = input[index[i][j][k]][j][k] if axis = 0, + out[i][j][k] = input[i][index[i][j][k]][k] if axis = 1, + out[i][j][k] = input[i][j][index[i][j][k]] if axis = 2, + ``` + + This operator is also the inverse of ScatterElements. It is similar to Torch's gather operation. + + Example 1: + ``` + data = [ + [1, 2], + [3, 4], + ] + indices = [ + [0, 0], + [1, 0], + ] + axis = 1 + output = [ + [1, 1], + [4, 3], + ] + ``` + Example 2: + ``` + data = [ + [1, 2, 3], + [4, 5, 6], + [7, 8, 9], + ] + indices = [ + [1, 2, 0], + [2, 0, 0], + ] + axis = 0 + output = [ + [4, 8, 3], + [7, 2, 3], + ] + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 11 + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : Tind
+
Tensor of int32/int64 indices, with the same rank r as the input. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of the same shape as indices.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ + +#### Examples + +
+gather_elements_0 + +```python +axis = 1 +node = onnx.helper.make_node( + "GatherElements", + inputs=["data", "indices"], + outputs=["y"], + axis=axis, +) +data = np.array([[1, 2], [3, 4]], dtype=np.float32) +indices = np.array([[0, 0], [1, 0]], dtype=np.int32) + +y = gather_elements(data, indices, axis) +# print(y) produces +# [[1, 1], +# [4, 3]] + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_elements_0", +) +``` + +
+ + +
+gather_elements_1 + +```python +axis = 0 +node = onnx.helper.make_node( + "GatherElements", + inputs=["data", "indices"], + outputs=["y"], + axis=axis, +) +data = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype=np.float32) +indices = np.array([[1, 2, 0], [2, 0, 0]], dtype=np.int32) + +y = gather_elements(data, indices, axis) +# print(y) produces +# [[4, 8, 3], +# [7, 2, 3]] + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_elements_1", +) +``` + +
+ + +
+gather_elements_negative_indices + +```python +axis = 0 +node = onnx.helper.make_node( + "GatherElements", + inputs=["data", "indices"], + outputs=["y"], + axis=axis, +) +data = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype=np.float32) +indices = np.array([[-1, -2, 0], [-2, 0, 0]], dtype=np.int32) + +y = gather_elements(data, indices, axis) +# print(y) produces +# [[7, 5, 3], +# [4, 2, 3]] + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_elements_negative_indices", +) +``` + +
+ + +### **GatherND** + + Given `data` tensor of rank `r` >= 1, `indices` tensor of rank `q` >= 1, and `batch_dims` integer `b`, this operator gathers + slices of `data` into an output tensor of rank `q + r - indices_shape[-1] - 1 - b`. + + `indices` is an q-dimensional integer tensor, best thought of as a `(q-1)`-dimensional tensor of index-tuples into `data`, + where each element defines a slice of `data` + + `batch_dims` (denoted as `b`) is an integer indicating the number of batch dimensions, i.e the leading `b` number of dimensions of + `data` tensor and `indices` are representing the batches, and the gather starts from the `b+1` dimension. + + Some salient points about the inputs' rank and shape: + + 1) r >= 1 and q >= 1 are to be honored. There is no dependency condition to be met between ranks `r` and `q` + + 2) The first `b` dimensions of the shape of `indices` tensor and `data` tensor must be equal. + + 3) b < min(q, r) is to be honored. + + 4) The `indices_shape[-1]` should have a value between 1 (inclusive) and rank `r-b` (inclusive) + + 5) All values in `indices` are expected to be within bounds [-s, s-1] along axis of size `s` (i.e.) `-data_shape[i] <= indices[...,i] <= data_shape[i] - 1`. + It is an error if any of the index values are out of bounds. + + The output is computed as follows: + + The output tensor is obtained by mapping each index-tuple in the `indices` tensor to the corresponding slice of the input `data`. + + 1) If `indices_shape[-1] > r-b` => error condition + + 2) If `indices_shape[-1] == r-b`, since the rank of `indices` is `q`, `indices` can be thought of as `N` `(q-b-1)`-dimensional tensors + containing 1-D tensors of dimension `r-b`, where `N` is an integer equals to the product of 1 and all the elements in the batch dimensions + of the indices_shape. Let us think of each such `r-b` ranked tensor as `indices_slice`. Each *scalar value* corresponding to `data[0:b-1,indices_slice]` + is filled into the corresponding location of the `(q-b-1)`-dimensional tensor to form the `output` tensor (Example 1 below) + + 3) If `indices_shape[-1] < r-b`, since the rank of `indices` is `q`, `indices` can be thought of as `N` `(q-b-1)`-dimensional tensor + containing 1-D tensors of dimension `< r-b`. Let us think of each such tensors as `indices_slice`. Each *tensor slice* corresponding + to `data[0:b-1, indices_slice , :]` is filled into the corresponding location of the `(q-b-1)`-dimensional tensor + to form the `output` tensor (Examples 2, 3, 4 and 5 below) + + This operator is the inverse of `ScatterND`. + + **Example 1** + + ``` + batch_dims = 0 + data = [[0,1],[2,3]] # data_shape = [2, 2] + indices = [[0,0],[1,1]] # indices_shape = [2, 2] + output = [0,3] # output_shape = [2] + ``` + + **Example 2** + + ``` + batch_dims = 0 + data = [[0,1],[2,3]] # data_shape = [2, 2] + indices = [[1],[0]] # indices_shape = [2, 1] + output = [[2,3],[0,1]] # output_shape = [2, 2] + ``` + + **Example 3** + + ``` + batch_dims = 0 + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + indices = [[0,1],[1,0]] # indices_shape = [2, 2] + output = [[2,3],[4,5]] # output_shape = [2, 2] + ``` + + **Example 4** + + ``` + batch_dims = 0 + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + indices = [[[0,1]],[[1,0]]] # indices_shape = [2, 1, 2] + output = [[[2,3]],[[4,5]]] # output_shape = [2, 1, 2] + ``` + + **Example 5** + + ``` + batch_dims = 1 + data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2] + indices = [[1],[0]] # indices_shape = [2, 1] + output = [[2,3],[4,5]] # output_shape = [2, 2] + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 11, 12 + +#### Attributes + +
+
batch_dims : int (default is 0)
+
The number of batch dimensions. The gather of indexing starts from dimension of data[batch_dims:]
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : tensor(int64)
+
Tensor of rank q >= 1. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank q + r - indices_shape[-1] - 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
+ + +#### Examples + +
+float32 + +```python +node = onnx.helper.make_node( + "GatherND", + inputs=["data", "indices"], + outputs=["output"], +) + +data = np.array([[[0, 1], [2, 3]], [[4, 5], [6, 7]]], dtype=np.float32) +indices = np.array([[[0, 1]], [[1, 0]]], dtype=np.int64) +output = gather_nd_impl(data, indices, 0) +expected_output = np.array([[[2, 3]], [[4, 5]]], dtype=np.float32) +assert np.array_equal(output, expected_output) +expect( + node, + inputs=[data, indices], + outputs=[output], + name="test_gathernd_example_float32", +) +``` + +
+ + +
+int32 + +```python +node = onnx.helper.make_node( + "GatherND", + inputs=["data", "indices"], + outputs=["output"], +) + +data = np.array([[0, 1], [2, 3]], dtype=np.int32) +indices = np.array([[0, 0], [1, 1]], dtype=np.int64) +output = gather_nd_impl(data, indices, 0) +expected_output = np.array([0, 3], dtype=np.int32) +assert np.array_equal(output, expected_output) +expect( + node, + inputs=[data, indices], + outputs=[output], + name="test_gathernd_example_int32", +) +``` + +
+ + +
+int32_batchdim_1 + +```python +node = onnx.helper.make_node( + "GatherND", + inputs=["data", "indices"], + outputs=["output"], + batch_dims=1, +) + +data = np.array([[[0, 1], [2, 3]], [[4, 5], [6, 7]]], dtype=np.int32) +indices = np.array([[1], [0]], dtype=np.int64) +output = gather_nd_impl(data, indices, 1) +expected_output = np.array([[2, 3], [4, 5]], dtype=np.int32) +assert np.array_equal(output, expected_output) +expect( + node, + inputs=[data, indices], + outputs=[output], + name="test_gathernd_example_int32_batch_dim1", +) +``` + +
+ + +### **Gelu** + + Gelu takes one input data (Tensor) and produces one + output data (Tensor) where the gaussian error linear units function, + $y = 0.5 * x * (1 + erf(x/sqrt(2)))$ is applied to the tensor elementwise. + If the attribute "approximate" is set to "tanh", the function estimation, + $y = 0.5 * x * (1 + Tanh(sqrt(2/\pi) * (x + 0.044715 * x^3)))$ is used and applied + to the tensor elementwise. + + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
approximate : string (default is none)
+
Gelu approximation algorithm: `"tanh"`, `"none"`(default).`"none"`: do not use approximation.`"tanh"`: use tanh approximation.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+gelu_default + +```python +node = onnx.helper.make_node("Gelu", inputs=["x"], outputs=["y"]) + +x = np.array([-1, 0, 1]).astype(np.float32) +# expected output [-0.15865526, 0., 0.84134474] +y = (0.5 * x * (1 + np.vectorize(math.erf)(x / np.sqrt(2)))).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_gelu_default_1") + +x = np.random.randn(3, 4, 5).astype(np.float32) +# expected output [2.99595031, 3.99987331, 4.99999857] +y = (0.5 * x * (1 + np.vectorize(math.erf)(x / np.sqrt(2)))).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_gelu_default_2") +``` + +
+ + +
+gelu_tanh + +```python +node = onnx.helper.make_node( + "Gelu", inputs=["x"], outputs=["y"], approximate="tanh" +) + +x = np.array([-1, 0, 1]).astype(np.float32) +# expected output [-0.158808, 0., 0.841192] +y = ( + 0.5 + * x + * (1 + np.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * np.power(x, 3)))) +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_gelu_tanh_1") + +x = np.random.randn(3, 4, 5).astype(np.float32) +# expected output [2.9963627, 3.99993, 4.9999995] +y = ( + 0.5 + * x + * (1 + np.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * np.power(x, 3)))) +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_gelu_tanh_2") +``` + +
+ + +### **Gemm** + + General Matrix multiplication: + https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3 + + * A' = transpose(A) if transA else A + * B' = transpose(B) if transB else B + + Compute Y = alpha * A' * B' + beta * C, where input tensor A has shape (M, K) or (K, M), + input tensor B has shape (K, N) or (N, K), input tensor C is broadcastable to shape (M, N), + and output tensor Y has shape (M, N). A will be transposed before doing the + computation if attribute transA is non-zero, same for B and transB. + This operator supports **unidirectional broadcasting** (tensor C should be unidirectional broadcastable to tensor A * B); for more details please check [the doc](Broadcasting.md). + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 7, 9, 11 + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Scalar multiplier for the product of input tensors A * B.
+
beta : float (default is 1.0)
+
Scalar multiplier for input tensor C.
+
transA : int (default is 0)
+
Whether A should be transposed
+
transB : int (default is 0)
+
Whether B should be transposed
+
+ +#### Inputs (2 - 3) + +
+
A (differentiable) : T
+
Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
+
B (differentiable) : T
+
Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
+
C (optional, differentiable) : T
+
Optional input tensor C. If not specified, the computation is done as if C is a scalar 0. The shape of C should be unidirectional broadcastable to (M, N).
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor of shape (M, N).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(bfloat16)
+
Constrain input and output types to float/int tensors.
+
+ + +#### Examples + +
+all_attributes + +```python +node = onnx.helper.make_node( + "Gemm", + inputs=["a", "b", "c"], + outputs=["y"], + alpha=0.25, + beta=0.35, + transA=1, + transB=1, +) +a = np.random.ranf([4, 3]).astype(np.float32) +b = np.random.ranf([5, 4]).astype(np.float32) +c = np.random.ranf([1, 5]).astype(np.float32) +y = gemm_reference_implementation( + a, b, c, transA=1, transB=1, alpha=0.25, beta=0.35 +) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_all_attributes") +``` + +
+ + +
+alpha + +```python +node = onnx.helper.make_node( + "Gemm", inputs=["a", "b", "c"], outputs=["y"], alpha=0.5 +) +a = np.random.ranf([3, 5]).astype(np.float32) +b = np.random.ranf([5, 4]).astype(np.float32) +c = np.zeros([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c, alpha=0.5) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_alpha") +``` + +
+ + +
+beta + +```python +node = onnx.helper.make_node( + "Gemm", inputs=["a", "b", "c"], outputs=["y"], beta=0.5 +) +a = np.random.ranf([2, 7]).astype(np.float32) +b = np.random.ranf([7, 4]).astype(np.float32) +c = np.random.ranf([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c, beta=0.5) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_beta") +``` + +
+ + +
+default_matrix_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b", "c"], outputs=["y"]) +a = np.random.ranf([3, 6]).astype(np.float32) +b = np.random.ranf([6, 4]).astype(np.float32) +c = np.random.ranf([3, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c) +expect( + node, inputs=[a, b, c], outputs=[y], name="test_gemm_default_matrix_bias" +) +``` + +
+ + +
+default_no_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b"], outputs=["y"]) +a = np.random.ranf([2, 10]).astype(np.float32) +b = np.random.ranf([10, 3]).astype(np.float32) +y = gemm_reference_implementation(a, b) +expect(node, inputs=[a, b], outputs=[y], name="test_gemm_default_no_bias") +``` + +
+ + +
+default_scalar_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b", "c"], outputs=["y"]) +a = np.random.ranf([2, 3]).astype(np.float32) +b = np.random.ranf([3, 4]).astype(np.float32) +c = np.array(3.14).astype(np.float32) +y = gemm_reference_implementation(a, b, c) +expect( + node, inputs=[a, b, c], outputs=[y], name="test_gemm_default_scalar_bias" +) +``` + +
+ + +
+default_single_elem_vector_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b", "c"], outputs=["y"]) +a = np.random.ranf([3, 7]).astype(np.float32) +b = np.random.ranf([7, 3]).astype(np.float32) +c = np.random.ranf([1]).astype(np.float32) +y = gemm_reference_implementation(a, b, c) +expect( + node, + inputs=[a, b, c], + outputs=[y], + name="test_gemm_default_single_elem_vector_bias", +) +``` + +
+ + +
+default_vector_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b", "c"], outputs=["y"]) +a = np.random.ranf([2, 7]).astype(np.float32) +b = np.random.ranf([7, 4]).astype(np.float32) +c = np.random.ranf([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c) +expect( + node, inputs=[a, b, c], outputs=[y], name="test_gemm_default_vector_bias" +) +``` + +
+ + +
+default_zero_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b", "c"], outputs=["y"]) +a = np.random.ranf([3, 5]).astype(np.float32) +b = np.random.ranf([5, 4]).astype(np.float32) +c = np.zeros([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_default_zero_bias") +``` + +
+ + +
+transposeA + +```python +node = onnx.helper.make_node( + "Gemm", inputs=["a", "b", "c"], outputs=["y"], transA=1 +) +a = np.random.ranf([6, 3]).astype(np.float32) +b = np.random.ranf([6, 4]).astype(np.float32) +c = np.zeros([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c, transA=1) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_transposeA") +``` + +
+ + +
+transposeB + +```python +node = onnx.helper.make_node( + "Gemm", inputs=["a", "b", "c"], outputs=["y"], transB=1 +) +a = np.random.ranf([3, 6]).astype(np.float32) +b = np.random.ranf([4, 6]).astype(np.float32) +c = np.zeros([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c, transB=1) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_transposeB") +``` + +
+ + +### **GlobalAveragePool** + + GlobalAveragePool consumes an input tensor X and applies average pooling across + the values in the same channel. This is equivalent to AveragePool with kernel size + equal to the spatial dimension of input tensor. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+globalaveragepool + +```python +node = onnx.helper.make_node( + "GlobalAveragePool", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(1, 3, 5, 5).astype(np.float32) +y = np.mean(x, axis=tuple(range(2, np.ndim(x))), keepdims=True) +expect(node, inputs=[x], outputs=[y], name="test_globalaveragepool") +``` + +
+ + +
+globalaveragepool_precomputed + +```python +node = onnx.helper.make_node( + "GlobalAveragePool", + inputs=["x"], + outputs=["y"], +) +x = np.array( + [ + [ + [ + [1, 2, 3], + [4, 5, 6], + [7, 8, 9], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[5]]]]).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_globalaveragepool_precomputed") +``` + +
+ + +### **GlobalLpPool** + + GlobalLpPool consumes an input tensor X and applies lp pool pooling across + the values in the same channel. This is equivalent to LpPool with kernel size + equal to the spatial dimension of input tensor. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 2 + +#### Attributes + +
+
p : int (default is 2)
+
p value of the Lp norm used to pool over the input data.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +### **GlobalMaxPool** + + GlobalMaxPool consumes an input tensor X and applies max pooling across + the values in the same channel. This is equivalent to MaxPool with kernel size + equal to the spatial dimension of input tensor. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+globalmaxpool + +```python +node = onnx.helper.make_node( + "GlobalMaxPool", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(1, 3, 5, 5).astype(np.float32) +y = np.max(x, axis=tuple(range(2, np.ndim(x))), keepdims=True) +expect(node, inputs=[x], outputs=[y], name="test_globalmaxpool") +``` + +
+ + +
+globalmaxpool_precomputed + +```python +node = onnx.helper.make_node( + "GlobalMaxPool", + inputs=["x"], + outputs=["y"], +) +x = np.array( + [ + [ + [ + [1, 2, 3], + [4, 5, 6], + [7, 8, 9], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[9]]]]).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_globalmaxpool_precomputed") +``` + +
+ + +### **Greater** + + Returns the tensor resulted from performing the `greater` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 7, 9 + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ + +#### Examples + +
+greater + +```python +node = onnx.helper.make_node( + "Greater", + inputs=["x", "y"], + outputs=["greater"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater") + +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.random.randn(3, 4, 5).astype(np.int8) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_int8") + +x = np.random.randn(3, 4, 5).astype(np.int16) +y = np.random.randn(3, 4, 5).astype(np.int16) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_uint64") +``` + +
+ + +
+greater + +```python +node = onnx.helper.make_node( + "GreaterOrEqual", + inputs=["x", "y"], + outputs=["greater_equal"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal") + +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.random.randn(3, 4, 5).astype(np.int8) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_int8") + +x = np.random.randn(3, 4, 5).astype(np.int16) +y = np.random.randn(3, 4, 5).astype(np.int16) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_uint64") +``` + +
+ + +
+greater_broadcast + +```python +node = onnx.helper.make_node( + "Greater", + inputs=["x", "y"], + outputs=["greater"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_bcast") +``` + +
+ + +
+greater_broadcast + +```python +node = onnx.helper.make_node( + "GreaterOrEqual", + inputs=["x", "y"], + outputs=["greater_equal"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_bcast") +``` + +
+ + +### **GreaterOrEqual** + + Returns the tensor resulted from performing the `greater_equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +Other versions of this operator: 12 + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ + +### **GridSample** + + Given an input `X` and a flow-field `grid`, computes the output `Y` using `X` values and pixel locations from the `grid`. + For spatial input `X` with shape (N, C, H, W), the `grid` will have shape (N, H_out, W_out, 2), + the output `Y` will have shape (N, C, H_out, W_out). For volumetric input `X` with shape (N, C, D, H, W), + the `grid` will have shape (N, D_out, H_out, W_out, 3), the output `Y` will have shape (N, C, D_out, H_out, W_out). + More generally, for an input `X` of rank r+2 with shape (N, C, d1, d2, ..., dr), + the `grid` will have shape (N, D1_out, D2_out, ..., Dr_out, r), the output `Y` will have shape (N, C, D1_out, D2_out, ..., Dr_out). + + The tensor `X` contains values at centers of square pixels (voxels, etc) locations such as (n, c, d1_in, d2_in, ..., dr_in). + The (n, d1_out, d2_out, ..., dr_out, :) values from the tensor `grid` are the normalized positions for interpolating the values + at the (n, c, d1_out, d2_out, ..., dr_out) locations from the output tensor `Y` using a specified interpolation method (the mode) + and a padding mode (for `grid` positions falling outside the 2-dimensional image). + + For example, the values in `grid[n, h_out, w_out, :]` are size-2 vectors specifying normalized positions in the 2-dimensional space of `X`. + They are used to interpolate output values of `Y[n, c, h_out, w_out]`. + + The GridSample operator is often used in doing grid generator and sampler in the + [Spatial Transformer Networks](https://arxiv.org/abs/1506.02025). + See also in [torch.nn.functional.grid_sample](https://pytorch.org/docs/stable/generated/torch.nn.functional.grid_sample.html). + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 16, 20 + +#### Attributes + +
+
align_corners : int (default is 0)
+
If align_corners=1, the extrema (-1 and 1) are considered as referring to the center points of the input's corner pixels (voxels, etc.). If align_corners=0, they are instead considered as referring to the corner points of the input's corner pixels (voxels, etc.), making the sampling more resolution agnostic.
+
mode : string (default is linear)
+
Three interpolation modes: linear (default), nearest and cubic. The "linear" mode includes linear and N-linear interpolation modes depending on the number of spatial dimensions of the input tensor (i.e. linear for 1 spatial dimension, bilinear for 2 spatial dimensions, etc.). The "cubic" mode also includes N-cubic interpolation modes following the same rules. The "nearest" mode rounds to the nearest even index when the sampling point falls halfway between two indices.
+
padding_mode : string (default is zeros)
+
Support padding modes for outside grid values: `zeros`(default), `border`, `reflection`. zeros: use 0 for out-of-bound grid locations, border: use border values for out-of-bound grid locations, reflection: use values at locations reflected by the border for out-of-bound grid locations. If index 0 represents the margin pixel, the reflected value at index -1 will be the same as the value at index 1. For location far away from the border, it will keep being reflected until becoming in bound. If pixel location x = -3.5 reflects by border -1 and becomes x' = 1.5, then reflects by border 1 and becomes x'' = 0.5.
+
+ +#### Inputs + +
+
X (differentiable) : T1
+
Input tensor of rank r+2 that has shape (N, C, D1, D2, ..., Dr), where N is the batch size, C is the number of channels, D1, D2, ..., Dr are the spatial dimensions.
+
grid (non-differentiable) : T2
+
Input offset of shape (N, D1_out, D2_out, ..., Dr_out, r), where D1_out, D2_out, ..., Dr_out are the spatial dimensions of the grid and output, and r is the number of spatial dimensions. Grid specifies the sampling locations normalized by the input spatial dimensions. Therefore, it should have most values in the range of [-1, 1]. If the grid has values outside the range of [-1, 1], the corresponding outputs will be handled as defined by padding_mode. Following computer vision convention, the coordinates in the length-r location vector are listed from the innermost tensor dimension to the outermost, the opposite of regular tensor indexing.
+
+ +#### Outputs + +
+
Y (differentiable) : T1
+
Output tensor of rank r+2 that has shape (N, C, D1_out, D2_out, ..., Dr_out) of the sampled values. For integer input types, intermediate values are computed as floating point and cast to integer at the end.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input `X` and output `Y` types to all tensor types.
+
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain grid types to float tensors.
+
+ + +#### Examples + +
+gridsample + +```python +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + padding_mode="zeros", + align_corners=0, +) +# X shape, [N, C, H, W] - [1, 1, 4, 4] +X = np.array( + [ + [ + [ + [0.0, 1.0, 2.0, 3.0], + [4.0, 5.0, 6.0, 7.0], + [8.0, 9.0, 10.0, 11.0], + [12.0, 13.0, 14.0, 15.0], + ] + ] + ], + dtype=np.float32, +) +# Grid shape, [N, H_out, W_out, 2] - [1, 6, 6, 2] +Grid = np.array( + [ + [ + [ + [-1.0000, -1.0000], + [-0.6000, -1.0000], + [-0.2000, -1.0000], + [0.2000, -1.0000], + [0.6000, -1.0000], + [1.0000, -1.0000], + ], + [ + [-1.0000, -0.6000], + [-0.6000, -0.6000], + [-0.2000, -0.6000], + [0.2000, -0.6000], + [0.6000, -0.6000], + [1.0000, -0.6000], + ], + [ + [-1.0000, -0.2000], + [-0.6000, -0.2000], + [-0.2000, -0.2000], + [0.2000, -0.2000], + [0.6000, -0.2000], + [1.0000, -0.2000], + ], + [ + [-1.0000, 0.2000], + [-0.6000, 0.2000], + [-0.2000, 0.2000], + [0.2000, 0.2000], + [0.6000, 0.2000], + [1.0000, 0.2000], + ], + [ + [-1.0000, 0.6000], + [-0.6000, 0.6000], + [-0.2000, 0.6000], + [0.2000, 0.6000], + [0.6000, 0.6000], + [1.0000, 0.6000], + ], + [ + [-1.0000, 1.0000], + [-0.6000, 1.0000], + [-0.2000, 1.0000], + [0.2000, 1.0000], + [0.6000, 1.0000], + [1.0000, 1.0000], + ], + ] + ], + dtype=np.float32, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 6, 6] +Y = np.array( + [ + [ + [ + [0.0000, 0.1500, 0.5500, 0.9500, 1.3500, 0.7500], + [0.6000, 1.5000, 2.3000, 3.1000, 3.9000, 2.1000], + [2.2000, 4.7000, 5.5000, 6.3000, 7.1000, 3.7000], + [3.8000, 7.9000, 8.7000, 9.5000, 10.3000, 5.3000], + [5.4000, 11.1000, 11.9000, 12.7000, 13.5000, 6.9000], + [3.0000, 6.1500, 6.5500, 6.9500, 7.3500, 3.7500], + ] + ] + ], + dtype=np.float32, +) +expect(node, inputs=[X, Grid], outputs=[Y], name="test_gridsample") +``` + +
+ + +
+gridsample_mode_aligncorners + +```python +# X shape, [N, C, H, W] - [1, 1, 3, 2] +X = np.array( + [[[[0.0, 1.0], [2.0, 3.0], [4.0, 5.0]]]], + dtype=np.float32, +) +# Grid shape, [N, H_out, W_out, 2] - [1, 2, 4, 2] +Grid = np.array( + [ + [ + [ + [-1.0000, -1.0000], + [-0.5000, -0.5000], + [-0.2000, -0.2000], + [0.0000, 0.0000], + ], + [ + [0.0000, 0.0000], + [-0.2000, -0.2000], + [0.5000, 0.5000], + [1.0000, 1.0000], + ], + ] + ], + dtype=np.float32, +) + +# setting mode = 'bilinear', default align_corners = 0 +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bilinear = np.array( + [[[[0.0000, 0.5000, 1.7000, 2.5000], [2.5000, 1.7000, 4.5000, 1.2500]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bilinear], + name="test_gridsample_bilinear", +) + +# setting mode = 'bilinear', align_corners = 1 +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_align_corners = np.array( + [[[[0.0000, 1.2500, 2.0000, 2.5000], [2.5000, 2.0000, 3.7500, 5.0000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_align_corners], + name="test_gridsample_aligncorners_true", +) + +# setting mode = 'nearest' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="nearest", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_nearest = np.array( + [[[[0.0, 0.0, 2.0, 2.0], [2.0, 2.0, 5.0, 0.0]]]], + dtype=np.float32, +) + +expect( + node, inputs=[X, Grid], outputs=[Y_nearest], name="test_gridsample_nearest" +) + +# setting mode = 'bicubic' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="cubic", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bicubic = np.array( + [[[[-0.1406, 0.3828, 1.7556, 2.9688], [2.9688, 1.7556, 5.1445, 1.3906]]]], + dtype=np.float32, +) + +expect( + node, inputs=[X, Grid], outputs=[Y_bicubic], name="test_gridsample_bicubic" +) + +# ============================================================================ +# Additional tests +# The reference output tensors were generated using PyTorch 2.0. +Grid = np.array( + [ + [ + [[-1.0, -0.8], [-0.6, -0.5], [-0.1, -0.2], [0.7, 0.0]], + [[0.0, 0.4], [0.2, -0.2], [-0.3, 0.5], [-1.0, 1.0]], + ] + ], + dtype=np.float32, +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="nearest", + align_corners=0, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_nearest = np.array( + [[[[0.0, 0.0, 2.0, 3.0], [4.0, 3.0, 4.0, 4.0]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_nearest], + name="test_gridsample_nearest_align_corners_0_additional_1", +) + +# setting mode = 'nearest' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="nearest", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_nearest = np.array( + [[[[0.0, 0.0, 2.0, 3.0], [2.0, 3.0, 4.0, 4.0]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_nearest], + name="test_gridsample_nearest_align_corners_1_additional_1", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + align_corners=0, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bilinear = np.array( + [[[[0.0000, 0.4500, 1.8000, 2.4000], [3.7000, 2.1000, 3.7000, 1.0000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bilinear], + name="test_gridsample_bilinear_align_corners_0_additional_1", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bilinear = np.array( + [[[[0.4000, 1.2000, 2.0500, 2.8500], [3.3000, 2.2000, 3.3500, 4.0000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bilinear], + name="test_gridsample_bilinear_align_corners_1_additional_1", +) + +# These two new bicubic tests produces slightly higher error ~5e-5 +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="cubic", + align_corners=0, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bicubic = np.array( + [ + [ + [ + [-0.173250, 0.284265, 1.923106, 2.568000], + [5.170375, 2.284414, 4.744844, 1.046875], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bicubic], + name="test_gridsample_bicubic_align_corners_0_additional_1", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="cubic", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bicubic = np.array( + [ + [ + [ + [0.304001, 1.128750, 2.266270, 3.144844], + [4.531500, 2.455360, 4.599819, 4.000000], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bicubic], + name="test_gridsample_bicubic_align_corners_1_additional_1", +) +``` + +
+ + +
+gridsample_paddingmode + +```python +# X shape, [N, C, H, W] - [1, 1, 3, 2] +X = np.array( + [[[[0.0, 1.0], [2.0, 3.0], [4.0, 5.0]]]], + dtype=np.float32, +) +# Grid shape, [N, H_out, W_out, 2] - [1, 2, 4, 2] +Grid = np.array( + [ + [ + [ + [-10.0000, -10.0000], + [-5.0000, -5.0000], + [-0.2000, -0.2000], + [10.0000, 10.0000], + ], + [ + [10.0000, 10.0000], + [-0.2000, -0.2000], + [5.0000, 5.0000], + [10.0000, 10.0000], + ], + ] + ], + dtype=np.float32, +) + +# setting padding_mode = 'zeros' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + padding_mode="zeros", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_zeros = np.array( + [[[[0.0000, 0.0000, 1.7000, 0.0000], [0.0000, 1.7000, 0.0000, 0.0000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_zeros], + name="test_gridsample_zeros_padding", +) + +# setting padding_mode = 'border' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + padding_mode="border", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_border = np.array( + [[[[0.0000, 0.0000, 1.7000, 5.0000], [5.0000, 1.7000, 5.0000, 5.0000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_border], + name="test_gridsample_border_padding", +) + +# setting padding_mode = 'reflection' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + padding_mode="reflection", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_reflection = np.array( + [[[[2.5000, 0.0000, 1.7000, 2.5000], [2.5000, 1.7000, 5.0000, 2.5000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_reflection], + name="test_gridsample_reflection_padding", +) +``` + +
+ + +
+volumeetric_gridsample_mode_aligncorners + +```python +X = np.array( + [ + [ + [ + [[1.0, 2.0], [3.0, 4.0]], + [[5.0, 6.0], [7.0, 8.0]], + [[9.0, 10.0], [11.0, 12.0]], + ] + ] + ], + dtype=np.float32, +) + +Grid = np.array( + [ + [ + [ + [[-1.0, -1.0, -1.0], [-1.0, -0.5, 0.3]], + [[-0.5, -0.5, -0.5], [1.0, -0.6, -1.0]], + [[-0.2, -0.2, -0.2], [0.4, 0.2, 0.6]], + [[0.0, 0.0, 0.0], [-1.0, 0.0, 0.0]], + ], + [ + [[0.0, 0.0, 0.0], [-1.0, 1.0, 0.0]], + [[-0.2, -0.2, -0.2], [1.0, 0.4, -0.2]], + [[0.5, 0.5, 0.5], [-1.0, -0.8, 0.8]], + [[1.0, 1.0, 1.0], [0.4, 0.6, -0.3]], + ], + ] + ], + dtype=np.float32, +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="nearest", + align_corners=0, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_nearest = np.array( + [ + [ + [ + [[1.0, 5.0], [1.0, 0.0], [5.0, 12.0], [5.0, 5.0]], + [[5.0, 0.0], [5.0, 0.0], [12.0, 9.0], [0.0, 8.0]], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_nearest], + name="test_gridsample_volumetric_nearest_align_corners_0", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="nearest", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_nearest = np.array( + [ + [ + [ + [[1.0, 5.0], [1.0, 2.0], [5.0, 12.0], [5.0, 5.0]], + [[5.0, 7.0], [5.0, 8.0], [12.0, 9.0], [12.0, 8.0]], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_nearest], + name="test_gridsample_volumetric_nearest_align_corners_1", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + align_corners=0, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bilinear = np.array( + [ + [ + [ + [ + [0.1250, 3.4000], + [2.0000, 0.4500], + [4.7000, 10.9000], + [6.5000, 3.0000], + ], + [ + [6.5000, 1.7500], + [4.7000, 3.3000], + [11.0000, 2.5200], + [1.5000, 5.4900], + ], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bilinear], + name="test_gridsample_volumetric_bilinear_align_corners_0", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bilinear = np.array( + [ + [ + [ + [ + [1.0000, 6.7000], + [3.7500, 2.4000], + [5.4000, 9.3000], + [6.5000, 6.0000], + ], + [ + [6.5000, 7.0000], + [5.4000, 6.6000], + [9.2500, 8.4000], + [12.0000, 6.1000], + ], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bilinear], + name="test_gridsample_volumetric_bilinear_align_corners_1", +) +``` + +
+ + +### **GroupNormalization** + + A GroupNormalization function. Carries out group normalization as described in + the paper https://arxiv.org/abs/1803.08494 + + This operator transforms input according to + ``` + y = scale * (x - mean) / sqrt(variance + epsilon) + bias, + ``` + where the mean and variance are computed per instance per group of channels, and + `scale` and `bias` should be specified for each channel. The number of + groups `num_groups` should be divisible by the number of channels so that there are + an equal number of channels per group. + + The overall computation has two stages: the first stage normalizes the elements to + have zero mean and unit variance for each instance in each group, and the second + stage scales and shifts the results of the first stage. The floating-point precision + used in the first stage is determined by the `stash_type` attribute. For example, + if `stash_type` is 1, the operator casts all input variables to 32-bit float, + performs the computation, and finally casts the normalized results back to the + original type of `X`. The second stage does not depend on `stash_type`. + + When the number of groups is the same as the number of channels, this operator is + equivalent to InstanceNormalization. When there is only one group, this operator + is equivalent to LayerNormalization. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +Other versions of this operator: 18 + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
num_groups : int (required)
+
The number of groups of channels. It should be a divisor of the number of channels `C`.
+
stash_type : int (default is 1)
+
The floating-point precision used in stage one of the computation.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor. Dimensions for image cases are `(N x C x H x W)`, where `N` is the batch size, `C` is the number of channels, and `H` and `W` are the height and width of the data. Statistics are computed for every group of channels over `C`, `H`, and `W`. For non-image cases, the dimensions are in the form of `(N x C x D1 x D2 ... Dn)`.
+
scale (differentiable) : T
+
Scale tensor of shape `(C)`.
+
bias (differentiable) : T
+
Bias tensor of shape `(C)`.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
The output tensor of the same shape as `X`.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+epsilon + +```python +c = 4 +num_groups = 2 +x = np.random.randn(3, c, 2, 2).astype(np.float32) +scale = np.random.randn(c).astype(np.float32) +bias = np.random.randn(c).astype(np.float32) +epsilon = 1e-2 +y = _group_normalization(x, num_groups, scale, bias, epsilon).astype(np.float32) + +node = onnx.helper.make_node( + "GroupNormalization", + inputs=["x", "scale", "bias"], + outputs=["y"], + epsilon=epsilon, + num_groups=num_groups, +) + +expect( + node, + inputs=[x, scale, bias], + outputs=[y], + name="test_group_normalization_epsilon", +) +``` + +
+ + +
+groupnormalization + +```python +c = 4 +num_groups = 2 +x = np.random.randn(3, c, 2, 2).astype(np.float32) +scale = np.random.randn(c).astype(np.float32) +bias = np.random.randn(c).astype(np.float32) +y = _group_normalization(x, num_groups, scale, bias).astype(np.float32) + +node = onnx.helper.make_node( + "GroupNormalization", + inputs=["x", "scale", "bias"], + outputs=["y"], + num_groups=num_groups, +) + +expect( + node, + inputs=[x, scale, bias], + outputs=[y], + name="test_group_normalization_example", +) +``` + +
+ + +### **HammingWindow** + + Generates a Hamming window as described in the paper https://ieeexplore.ieee.org/document/1455106. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
output_datatype : int (default is 1)
+
The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT.
+
periodic : int (default is 1)
+
If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When 'periodic' is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1.
+
+ +#### Inputs + +
+
size (non-differentiable) : T1
+
A scalar value indicating the length of the window.
+
+ +#### Outputs + +
+
output (non-differentiable) : T2
+
A Hamming window with length: size. The output has the shape: [size].
+
+ +#### Type Constraints + +
+
T1 : tensor(int32), tensor(int64)
+
Constrain the input size to int32_t or int64_t.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain output types to numeric tensors.
+
+ + +#### Examples + +
+hammingwindow + +```python +# Test periodic window +node = onnx.helper.make_node( + "HammingWindow", + inputs=["x"], + outputs=["y"], +) +size = np.int32(10) +a0 = 25 / 46 +a1 = 1 - a0 +y = a0 - a1 * np.cos(2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / size) +expect( + node, + inputs=[size], + outputs=[y.astype(np.float32)], + name="test_hammingwindow", +) + +# Test symmetric window +node = onnx.helper.make_node( + "HammingWindow", inputs=["x"], outputs=["y"], periodic=0 +) +size = np.int32(10) +a0 = 25 / 46 +a1 = 1 - a0 +y = a0 - a1 * np.cos( + 2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / (size - 1) +) +expect( + node, + inputs=[size], + outputs=[y.astype(np.float32)], + name="test_hammingwindow_symmetric", +) +``` + +
+ + +### **HannWindow** + + Generates a Hann window as described in the paper https://ieeexplore.ieee.org/document/1455106. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
output_datatype : int (default is 1)
+
The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT.
+
periodic : int (default is 1)
+
If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When 'periodic' is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1.
+
+ +#### Inputs + +
+
size (non-differentiable) : T1
+
A scalar value indicating the length of the window.
+
+ +#### Outputs + +
+
output (non-differentiable) : T2
+
A Hann window with length: size. The output has the shape: [size].
+
+ +#### Type Constraints + +
+
T1 : tensor(int32), tensor(int64)
+
Constrain the input size to int32_t or int64_t.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain output types to numeric tensors.
+
+ + +#### Examples + +
+hannwindow + +```python +# Test periodic window +node = onnx.helper.make_node( + "HannWindow", + inputs=["x"], + outputs=["y"], +) +size = np.int32(10) +a0 = 0.5 +a1 = 0.5 +y = a0 - a1 * np.cos(2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / size) +expect( + node, inputs=[size], outputs=[y.astype(np.float32)], name="test_hannwindow" +) + +# Test symmetric window +node = onnx.helper.make_node( + "HannWindow", inputs=["x"], outputs=["y"], periodic=0 +) +size = np.int32(10) +a0 = 0.5 +a1 = 0.5 +y = a0 - a1 * np.cos( + 2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / (size - 1) +) +expect( + node, + inputs=[size], + outputs=[y.astype(np.float32)], + name="test_hannwindow_symmetric", +) +``` + +
+ + +### **HardSigmoid** + + HardSigmoid takes one input data (Tensor) and produces one output data + (Tensor) where the HardSigmoid function, y = max(0, min(1, alpha * x + beta)), + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Attributes + +
+
alpha : float (default is 0.2)
+
Value of alpha.
+
beta : float (default is 0.5)
+
Value of beta.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+hardsigmoid + +```python +node = onnx.helper.make_node( + "HardSigmoid", inputs=["x"], outputs=["y"], alpha=0.5, beta=0.6 +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.clip(x * 0.5 + 0.6, 0, 1) # expected output [0.1, 0.6, 1.] +expect(node, inputs=[x], outputs=[y], name="test_hardsigmoid_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x * 0.5 + 0.6, 0, 1) +expect(node, inputs=[x], outputs=[y], name="test_hardsigmoid") +``` + +
+ + +
+hardsigmoid_default + +```python +default_alpha = 0.2 +default_beta = 0.5 +node = onnx.helper.make_node( + "HardSigmoid", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x * default_alpha + default_beta, 0, 1) +expect(node, inputs=[x], outputs=[y], name="test_hardsigmoid_default") +``` + +
+ + +### **HardSwish** + + HardSwish takes one input data (Tensor) and produces one output data (Tensor) where + the HardSwish function, y = x * max(0, min(1, alpha * x + beta)) = x * HardSigmoid(x), + where alpha = 1/6 and beta = 0.5, is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 14 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+hardswish + +```python +node = onnx.helper.make_node( + "HardSwish", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = hardswish(x) + +expect(node, inputs=[x], outputs=[y], name="test_hardswish") +``` + +
+ + +### **Hardmax** + + The operator computes the hardmax values for the given input: + + Hardmax(element in input, axis) = 1 if the element is the first maximum value along the specified axis, 0 otherwise + + The "axis" attribute indicates the dimension along which Hardmax + will be performed. The output tensor has the same shape + and contains the Hardmax values of the corresponding input. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 11 + +#### Attributes + +
+
axis : int (default is -1)
+
+Describes the dimension Hardmax will be performed on. +Negative value means counting dimensions +from the back. Accepted range is [-r, r-1] where r = rank(input). +
+
+ +#### Inputs + +
+
input (differentiable) : T
+
The input tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output values with the same shape as the input tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+hardmax + +```python +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], +) + +x = np.array([[3, 0, 1, 2], [2, 5, 1, 0], [0, 1, 3, 2], [0, 1, 2, 3]]).astype( + np.float32 +) +# expect result: +# [[1. 0. 0. 0.] +# [0. 1. 0. 0.] +# [0. 0. 1. 0.] +# [0. 0. 0. 1.]] +y = hardmax(x) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_example") + +# For multiple occurrences of the maximal values, the first occurrence is selected for one-hot output +x = np.array([[3, 3, 3, 1]]).astype(np.float32) +# expect result: +# [[1, 0, 0, 0]] +y = hardmax(x) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_one_hot") +``` + +
+ + +
+hardmax_axis + +```python +x = np.random.randn(3, 4, 5).astype(np.float32) +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], + axis=0, +) +y = hardmax(x, axis=0) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_axis_0") + +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], + axis=1, +) +y = hardmax(x, axis=1) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_axis_1") + +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], + axis=2, +) +y = hardmax(x, axis=2) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_axis_2") + +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], + axis=-1, +) +y = hardmax(x, axis=-1) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_negative_axis") + +# default axis is -1 +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], +) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_default_axis") +``` + +
+ + +### **Identity** + + Identity operator + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 13, 14, 16, 19, 21, 23, 24 + +#### Inputs + +
+
input (differentiable) : V
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : V
+
Tensor to copy input into.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain input and output types to all tensor, sequence, and optional types.
+
+ + +#### Examples + +
+identity + +```python +node = onnx.helper.make_node( + "Identity", + inputs=["x"], + outputs=["y"], +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +expect(node, inputs=[data], outputs=[data], name="test_identity") +``` + +
+ + +
+identity_opt + +```python +ten_in_tp = onnx.helper.make_tensor_type_proto( + onnx.TensorProto.FLOAT, shape=[5] +) +seq_in_tp = onnx.helper.make_sequence_type_proto(ten_in_tp) +opt_in_tp = onnx.helper.make_optional_type_proto(seq_in_tp) + +identity_node = onnx.helper.make_node( + "Identity", inputs=["opt_in"], outputs=["opt_out"] +) + +x = [np.array([1, 2, 3, 4, 5]).astype(np.float32)] + +expect( + identity_node, + inputs=[x], + outputs=[x], + name="test_identity_opt", + opset_imports=[onnx.helper.make_opsetid("", 16)], + input_type_protos=[opt_in_tp], + output_type_protos=[opt_in_tp], +) +``` + +
+ + +
+sequence + +```python +node = onnx.helper.make_node( + "Identity", + inputs=["x"], + outputs=["y"], +) + +data = [ + np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, + ), + np.array( + [ + [ + [ + [2, 3], + [1, 5], + ] + ] + ], + dtype=np.float32, + ), +] + +expect(node, inputs=[data], outputs=[data], name="test_identity_sequence") +``` + +
+ + +### **If** + + If conditional + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13, 16, 19, 21, 23, 24 + +#### Attributes + +
+
else_branch : graph (required)
+
Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
+
then_branch : graph (required)
+
Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
+
+ +#### Inputs + +
+
cond : B
+
Condition for the if. The tensor must contain a single element.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : V
+
Values that are live-out to the enclosing scope. The return values in the `then_branch` and `else_branch` must be of the same data type. The `then_branch` and `else_branch` may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if-node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of `then_branch` is typed float tensor with shape [2] and the first output of `else_branch` is another float tensor with shape [3], If's first output should have (a) no shape set, or (b) a shape of rank 1 with neither `dim_value` nor `dim_param` set, or (c) a shape of rank 1 with a unique `dim_param`. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), seq(tensor(float4e2m1)), seq(tensor(float8e8m0)), seq(tensor(uint2)), seq(tensor(int2)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4)), optional(tensor(float4e2m1)), optional(tensor(float8e8m0)), optional(tensor(uint2)), optional(tensor(int2))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv13.
+
B : tensor(bool)
+
Only bool
+
+ + +#### Examples + +
+if + +```python +# Given a bool scalar input cond. +# return constant tensor x if cond is True, otherwise return constant tensor y. + +then_out = onnx.helper.make_tensor_value_info( + "then_out", onnx.TensorProto.FLOAT, [5] +) +else_out = onnx.helper.make_tensor_value_info( + "else_out", onnx.TensorProto.FLOAT, [5] +) + +x = np.array([1, 2, 3, 4, 5]).astype(np.float32) +y = np.array([5, 4, 3, 2, 1]).astype(np.float32) + +then_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["then_out"], + value=onnx.numpy_helper.from_array(x), +) + +else_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["else_out"], + value=onnx.numpy_helper.from_array(y), +) + +then_body = onnx.helper.make_graph( + [then_const_node], "then_body", [], [then_out] +) + +else_body = onnx.helper.make_graph( + [else_const_node], "else_body", [], [else_out] +) + +if_node = onnx.helper.make_node( + "If", + inputs=["cond"], + outputs=["res"], + then_branch=then_body, + else_branch=else_body, +) + +cond = np.array(1).astype(bool) +res = x if cond else y +expect( + if_node, + inputs=[cond], + outputs=[res], + name="test_if", + opset_imports=[onnx.helper.make_opsetid("", 11)], +) +``` + +
+ + +
+if_optional + +```python +# Given a bool scalar input cond, return an empty optional sequence of +# tensor if True, return an optional sequence with value x +# (the input optional sequence) otherwise. + +ten_in_tp = onnx.helper.make_tensor_type_proto( + onnx.TensorProto.FLOAT, shape=[5] +) +seq_in_tp = onnx.helper.make_sequence_type_proto(ten_in_tp) + +then_out_tensor_tp = onnx.helper.make_tensor_type_proto( + onnx.TensorProto.FLOAT, shape=[5] +) +then_out_seq_tp = onnx.helper.make_sequence_type_proto(then_out_tensor_tp) +then_out_opt_tp = onnx.helper.make_optional_type_proto(then_out_seq_tp) +then_out = onnx.helper.make_value_info("optional_empty", then_out_opt_tp) + +else_out_tensor_tp = onnx.helper.make_tensor_type_proto( + onnx.TensorProto.FLOAT, shape=[5] +) +else_out_seq_tp = onnx.helper.make_sequence_type_proto(else_out_tensor_tp) +else_out_opt_tp = onnx.helper.make_optional_type_proto(else_out_seq_tp) +else_out = onnx.helper.make_value_info("else_opt", else_out_opt_tp) + +x = [np.array([1, 2, 3, 4, 5]).astype(np.float32)] +cond = np.array(0).astype(bool) +res = compute_if_outputs(x, cond) + +opt_empty_in = onnx.helper.make_node( + "Optional", inputs=[], outputs=["optional_empty"], type=seq_in_tp +) + +then_body = onnx.helper.make_graph([opt_empty_in], "then_body", [], [then_out]) + +else_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["x"], + value=onnx.numpy_helper.from_array(x[0]), +) + +else_seq_node = onnx.helper.make_node( + "SequenceConstruct", inputs=["x"], outputs=["else_seq"] +) + +else_optional_seq_node = onnx.helper.make_node( + "Optional", inputs=["else_seq"], outputs=["else_opt"] +) + +else_body = onnx.helper.make_graph( + [else_const_node, else_seq_node, else_optional_seq_node], + "else_body", + [], + [else_out], +) + +if_node = onnx.helper.make_node( + "If", + inputs=["cond"], + outputs=["sequence"], + then_branch=then_body, + else_branch=else_body, +) + +expect( + if_node, + inputs=[cond], + outputs=[res], + name="test_if_opt", + output_type_protos=[else_out_opt_tp], + opset_imports=[onnx.helper.make_opsetid("", 16)], +) +``` + +
+ + +
+if_seq + +```python +# Given a bool scalar input cond. +# return constant sequence x if cond is True, otherwise return constant sequence y. + +then_out = onnx.helper.make_tensor_sequence_value_info( + "then_out", onnx.TensorProto.FLOAT, shape=[5] +) +else_out = onnx.helper.make_tensor_sequence_value_info( + "else_out", onnx.TensorProto.FLOAT, shape=[5] +) + +x = [np.array([1, 2, 3, 4, 5]).astype(np.float32)] +y = [np.array([5, 4, 3, 2, 1]).astype(np.float32)] + +then_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["x"], + value=onnx.numpy_helper.from_array(x[0]), +) + +then_seq_node = onnx.helper.make_node( + "SequenceConstruct", inputs=["x"], outputs=["then_out"] +) + +else_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["y"], + value=onnx.numpy_helper.from_array(y[0]), +) + +else_seq_node = onnx.helper.make_node( + "SequenceConstruct", inputs=["y"], outputs=["else_out"] +) + +then_body = onnx.helper.make_graph( + [then_const_node, then_seq_node], "then_body", [], [then_out] +) + +else_body = onnx.helper.make_graph( + [else_const_node, else_seq_node], "else_body", [], [else_out] +) + +if_node = onnx.helper.make_node( + "If", + inputs=["cond"], + outputs=["res"], + then_branch=then_body, + else_branch=else_body, +) + +cond = np.array(1).astype(bool) +res = x if cond else y +expect( + if_node, + inputs=[cond], + outputs=[res], + name="test_if_seq", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) +``` + +
+ + +### **ImageDecoder** + + Loads and decodes and image from a file. If it can't decode for any reason (e.g. corrupted encoded + stream, invalid format, it will return an empty matrix). + The following image formats are supported: + * BMP + * JPEG (note: Lossless JPEG support is optional) + * JPEG2000 + * TIFF + * PNG + * WebP + * Portable image format (PBM, PGM, PPM, PXM, PNM) + Decoded images follow a channel-last layout: (Height, Width, Channels). + **JPEG chroma upsampling method:** + When upsampling the chroma components by a factor of 2, the pixels are linearly interpolated so that the + centers of the output pixels are 1/4 and 3/4 of the way between input pixel centers. + When rounding, 0.5 is rounded down and up at alternative pixels locations to prevent bias towards + larger values (ordered dither pattern). + Considering adjacent input pixels A, B, and C, B is upsampled to pixels B0 and B1 so that + ``` + B0 = round_half_down((1/4) * A + (3/4) * B) + B1 = round_half_up((3/4) * B + (1/4) * C) + ``` + This method, is the default chroma upsampling method in the well-established libjpeg-turbo library, + also referred as "smooth" or "fancy" upsampling. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
pixel_format : string (default is RGB)
+
Pixel format. Can be one of "RGB", "BGR", or "Grayscale".
+
+ +#### Inputs + +
+
encoded_stream (non-differentiable) : T1
+
Encoded stream
+
+ +#### Outputs + +
+
image (non-differentiable) : T2
+
Decoded image
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8)
+
Constrain input types to 8-bit unsigned integer tensor.
+
T2 : tensor(uint8)
+
Constrain output types to 8-bit unsigned integer tensor.
+
+ + +#### Examples + +
+image_decoder_decode_bmp_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "bmp", _image_decoder_data.image_decoder_decode_bmp_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_bmp_rgb", +) +``` + +
+ + +
+image_decoder_decode_jpeg2k_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "jpeg2000", _image_decoder_data.image_decoder_decode_jpeg2k_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_jpeg2k_rgb", +) +``` + +
+ + +
+image_decoder_decode_jpeg_bgr + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="BGR", +) + +data, output = _generate_test_data( + "jpeg", _image_decoder_data.image_decoder_decode_jpeg_bgr, "BGR" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_jpeg_bgr", +) +``` + +
+ + +
+image_decoder_decode_jpeg_grayscale + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="Grayscale", +) + +data, output = _generate_test_data( + "jpeg", _image_decoder_data.image_decoder_decode_jpeg_grayscale, "Grayscale" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_jpeg_grayscale", +) +``` + +
+ + +
+image_decoder_decode_jpeg_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "jpeg", _image_decoder_data.image_decoder_decode_jpeg_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_jpeg_rgb", +) +``` + +
+ + +
+image_decoder_decode_png_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "png", _image_decoder_data.image_decoder_decode_png_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_png_rgb", +) +``` + +
+ + +
+image_decoder_decode_pnm_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "ppm", _image_decoder_data.image_decoder_decode_pnm_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_pnm_rgb", +) +``` + +
+ + +
+image_decoder_decode_tiff_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "tiff", _image_decoder_data.image_decoder_decode_tiff_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_tiff_rgb", +) +``` + +
+ + +
+image_decoder_decode_webp_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "webp", _image_decoder_data.image_decoder_decode_webp_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_webp_rgb", +) +``` + +
+ + +### **InstanceNormalization** + + Carries out instance normalization as described in the paper + https://arxiv.org/abs/1607.08022. + + y = scale * (x - mean) / sqrt(variance + epsilon) + B, + where mean and variance are computed per instance per channel. + + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Attributes + +
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
scale (differentiable) : T
+
The input 1-dimensional scale tensor of size C.
+
B (differentiable) : T
+
The input 1-dimensional bias tensor of size C.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output tensor of the same shape as input.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+instancenormalization + +```python +def _instancenorm_test_mode( + x: np.ndarray, s: np.ndarray, bias: np.ndarray, epsilon: float = 1e-5 +) -> np.ndarray: + dims_x = len(x.shape) + axis = tuple(range(2, dims_x)) + mean = np.mean(x, axis=axis, keepdims=True) + var = np.var(x, axis=axis, keepdims=True) + dim_ones = (1,) * (dims_x - 2) + s = s.reshape(-1, *dim_ones) + bias = bias.reshape(-1, *dim_ones) + return s * (x - mean) / np.sqrt(var + epsilon) + bias + +# input size: (1, 2, 1, 3) +x = np.array([[[[-1, 0, 1]], [[2, 3, 4]]]]).astype(np.float32) +s = np.array([1.0, 1.5]).astype(np.float32) +bias = np.array([0, 1]).astype(np.float32) +y = _instancenorm_test_mode(x, s, bias).astype(np.float32) + +node = onnx.helper.make_node( + "InstanceNormalization", + inputs=["x", "s", "bias"], + outputs=["y"], +) + +# output size: (1, 2, 1, 3) +expect(node, inputs=[x, s, bias], outputs=[y], name="test_instancenorm_example") + +# input size: (2, 3, 4, 5) +x = np.random.randn(2, 3, 4, 5).astype(np.float32) +s = np.random.randn(3).astype(np.float32) +bias = np.random.randn(3).astype(np.float32) +epsilon = 1e-2 +y = _instancenorm_test_mode(x, s, bias, epsilon).astype(np.float32) + +node = onnx.helper.make_node( + "InstanceNormalization", + inputs=["x", "s", "bias"], + outputs=["y"], + epsilon=epsilon, +) + +# output size: (2, 3, 4, 5) +expect(node, inputs=[x, s, bias], outputs=[y], name="test_instancenorm_epsilon") +``` + +
+ + +### **IsInf** + + Map infinity to true and other values to false. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +Other versions of this operator: 10 + +#### Attributes + +
+
detect_negative : int (default is 1)
+
(Optional) Whether map negative infinity to true. Default to 1 so that negative infinity induces true. Set this attribute to 0 if negative infinity should be mapped to false.
+
detect_positive : int (default is 1)
+
(Optional) Whether map positive infinity to true. Default to 1 so that positive infinity induces true. Set this attribute to 0 if positive infinity should be mapped to false.
+
+ +#### Inputs + +
+
X (non-differentiable) : T1
+
input
+
+ +#### Outputs + +
+
Y (non-differentiable) : T2
+
output
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input types to float tensors.
+
T2 : tensor(bool)
+
Constrain output types to boolean tensors.
+
+ + +#### Examples + +
+infinity + +```python +node = onnx.helper.make_node( + "IsInf", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.2, np.nan, np.inf, 2.8, -np.inf, np.inf], dtype=np.float32) +y = np.isinf(x) +expect(node, inputs=[x], outputs=[y], name="test_isinf") +``` + +
+ + +
+infinity_float16 + +```python +node = onnx.helper.make_node( + "IsInf", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.2, np.nan, np.inf, 2.8, -np.inf, np.inf], dtype=np.float16) +y = np.isinf(x) +expect(node, inputs=[x], outputs=[y], name="test_isinf_float16") +``` + +
+ + +
+negative_infinity_only + +```python +node = onnx.helper.make_node( + "IsInf", inputs=["x"], outputs=["y"], detect_positive=0 +) + +x = np.array([-1.7, np.nan, np.inf, -3.6, -np.inf, np.inf], dtype=np.float32) +y = np.isneginf(x) +expect(node, inputs=[x], outputs=[y], name="test_isinf_negative") +``` + +
+ + +
+positive_infinity_only + +```python +node = onnx.helper.make_node( + "IsInf", inputs=["x"], outputs=["y"], detect_negative=0 +) + +x = np.array([-1.7, np.nan, np.inf, 3.6, -np.inf, np.inf], dtype=np.float32) +y = np.isposinf(x) +expect(node, inputs=[x], outputs=[y], name="test_isinf_positive") +``` + +
+ + +### **IsNaN** + + Returns which elements of the input are NaN. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +Other versions of this operator: 9, 13 + +#### Inputs + +
+
X (non-differentiable) : T1
+
input
+
+ +#### Outputs + +
+
Y (non-differentiable) : T2
+
output
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
Constrain input types to float tensors.
+
T2 : tensor(bool)
+
Constrain output types to boolean tensors.
+
+ + +#### Examples + +
+float16 + +```python +node = onnx.helper.make_node( + "IsNaN", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.2, np.nan, np.inf, 2.8, -np.inf, np.inf], dtype=np.float16) +y = np.isnan(x) +expect(node, inputs=[x], outputs=[y], name="test_isnan_float16") +``` + +
+ + +
+isnan + +```python +node = onnx.helper.make_node( + "IsNaN", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.2, np.nan, np.inf, 2.8, -np.inf, np.inf], dtype=np.float32) +y = np.isnan(x) +expect(node, inputs=[x], outputs=[y], name="test_isnan") +``` + +
+ + +### **LRN** + + Local Response Normalization proposed in the [AlexNet paper](https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf). + It normalizes over local input regions. + The local region is defined across the channels. For an element `X[n, c, d1, ..., dk]` in a tensor + of shape `(N x C x D1 x D2, ..., Dk)`, its region is + `{X[n, i, d1, ..., dk] | max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2))}`. + + `square_sum[n, c, d1, ..., dk] = sum(X[n, i, d1, ..., dk] ^ 2)`, + where `max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2))`. + + `Y[n, c, d1, ..., dk] = X[n, c, d1, ..., dk] / (bias + alpha / size * square_sum[n, c, d1, ..., dk] ) ^ beta` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Attributes + +
+
alpha : float (default is 0.0001)
+
Scaling parameter.
+
beta : float (default is 0.75)
+
The exponent.
+
bias : float (default is 1.0)
+
+
size : int (required)
+
The number of channels to sum over
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor, which has the shape and type as input tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+default + +```python +alpha = 0.0001 +beta = 0.75 +bias = 1.0 +nsize = 3 +node = onnx.helper.make_node("LRN", inputs=["x"], outputs=["y"], size=3) +x = np.random.randn(5, 5, 5, 5).astype(np.float32) +square_sum = np.zeros((5, 5, 5, 5)).astype(np.float32) +for n, c, h, w in np.ndindex(x.shape): + square_sum[n, c, h, w] = sum( + x[ + n, + max(0, c - math.floor((nsize - 1) / 2)) : min( + 5, c + math.ceil((nsize - 1) / 2) + 1 + ), + h, + w, + ] + ** 2 + ) +y = x / ((bias + (alpha / nsize) * square_sum) ** beta) +expect(node, inputs=[x], outputs=[y], name="test_lrn_default") +``` + +
+ + +
+lrn + +```python +alpha = 0.0002 +beta = 0.5 +bias = 2.0 +nsize = 3 +node = onnx.helper.make_node( + "LRN", + inputs=["x"], + outputs=["y"], + alpha=alpha, + beta=beta, + bias=bias, + size=nsize, +) +x = np.random.randn(5, 5, 5, 5).astype(np.float32) +square_sum = np.zeros((5, 5, 5, 5)).astype(np.float32) +for n, c, h, w in np.ndindex(x.shape): + square_sum[n, c, h, w] = sum( + x[ + n, + max(0, c - math.floor((nsize - 1) / 2)) : min( + 5, c + math.ceil((nsize - 1) / 2) + 1 + ), + h, + w, + ] + ** 2 + ) +y = x / ((bias + (alpha / nsize) * square_sum) ** beta) +expect(node, inputs=[x], outputs=[y], name="test_lrn") +``` + +
+ + +### **LSTM** + + Computes an one-layer LSTM. This operator is usually supported via some + custom implementation such as CuDNN. + + Notations: + + * `X` - input tensor + * `i` - input gate + * `o` - output gate + * `f` - forget gate + * `c` - cell gate + * `t` - time step (t-1 means previous time step) + * `W[iofc]` - W parameter weight matrix for input, output, forget, and cell gates + * `R[iofc]` - R recurrence weight matrix for input, output, forget, and cell gates + * `Wb[iofc]` - W bias vectors for input, output, forget, and cell gates + * `Rb[iofc]` - R bias vectors for input, output, forget, and cell gates + * `P[iof]` - P peephole weight vector for input, output, and forget gates + * `WB[iofc]` - W parameter weight matrix for backward input, output, forget, and cell gates + * `RB[iofc]` - R recurrence weight matrix for backward input, output, forget, and cell gates + * `WBb[iofc]` - W bias vectors for backward input, output, forget, and cell gates + * `RBb[iofc]` - R bias vectors for backward input, output, forget, and cell gates + * `PB[iof]` - P peephole weight vector for backward input, output, and forget gates + * `H` - Hidden state + * `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + * Relu(x) - max(0, x) + * Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + * Sigmoid(x) - 1/(1 + e^{-x}) + + NOTE: Below are optional + + * Affine(x) - alpha*x + beta + * LeakyRelu(x) - x if x >= 0 else alpha * x + * ThresholdedRelu(x) - x if x >= alpha else 0 + * ScaledTanh(x) - alpha*Tanh(beta*x) + * HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + * Elu(x) - x if x >= 0 else alpha*(e^x - 1) + * Softsign(x) - x/(1 + |x|) + * Softplus(x) - log(1 + e^x) + + Equations (Default: f=Sigmoid, g=Tanh, h=Tanh): + + * it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi) + * ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf) + * ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc) + * Ct = ft (.) Ct-1 + it (.) ct + * ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Po (.) Ct + Wbo + Rbo) + * Ht = ot (.) h(Ct) + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 7, 14 + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings
+
A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
input_forget : int (default is 0)
+
Couple the input and forget gates if 1.
+
layout : int (default is 0)
+
The shape format of inputs X, initial_h, initial_c and outputs Y, Y_h, Y_c. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = initial_c.shape = Y_c.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = initial_c.shape = Y_c.shape = [batch_size, num_directions, hidden_size].
+
+ +#### Inputs (3 - 8) + +
+
X (differentiable) : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W (differentiable) : T
+
The weight tensor for the gates. Concatenation of `W[iofc]` and `WB[iofc]` (if bidirectional) along dimension 0. The tensor has shape `[num_directions, 4*hidden_size, input_size]`.
+
R (differentiable) : T
+
The recurrence weight tensor. Concatenation of `R[iofc]` and `RB[iofc]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 4*hidden_size, hidden_size]`.
+
B (optional, differentiable) : T
+
The bias tensor for input gate. Concatenation of `[Wb[iofc], Rb[iofc]]`, and `[WBb[iofc], RBb[iofc]]` (if bidirectional) along dimension 0. This tensor has shape `[num_directions, 8*hidden_size]`. Optional: If not specified - assumed to be 0.
+
sequence_lens (optional, non-differentiable) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional, non-differentiable) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
initial_c (optional, non-differentiable) : T
+
Optional initial value of the cell. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
P (optional, differentiable) : T
+
The weight tensor for peepholes. Concatenation of `P[iof]` and `PB[iof]` (if bidirectional) along dimension 0. It has shape `[num_directions, 3*hidde_size]`. Optional: If not specified - assumed to be 0.
+
+ +#### Outputs (0 - 3) + +
+
Y (optional, differentiable) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional, differentiable) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
Y_c (optional, differentiable) : T
+
The last output value of the cell. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ + +#### Examples + +
+batchwise + +```python +input = np.array([[[1.0, 2.0]], [[3.0, 4.0]], [[5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 7 +weight_scale = 0.3 +number_of_gates = 4 +layout = 1 + +node = onnx.helper.make_node( + "LSTM", + inputs=["X", "W", "R"], + outputs=["Y", "Y_h"], + hidden_size=hidden_size, + layout=layout, +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +lstm = LSTMHelper(X=input, W=W, R=R, layout=layout) +Y, Y_h = lstm.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y.astype(np.float32), Y_h.astype(np.float32)], + name="test_lstm_batchwise", +) +``` + +
+ + +
+defaults + +```python +input = np.array([[[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 3 +weight_scale = 0.1 +number_of_gates = 4 + +node = onnx.helper.make_node( + "LSTM", inputs=["X", "W", "R"], outputs=["", "Y_h"], hidden_size=hidden_size +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +lstm = LSTMHelper(X=input, W=W, R=R) +_, Y_h = lstm.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y_h.astype(np.float32)], + name="test_lstm_defaults", +) +``` + +
+ + +
+initial_bias + +```python +input = np.array([[[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]]]).astype( + np.float32 +) + +input_size = 3 +hidden_size = 4 +weight_scale = 0.1 +custom_bias = 0.1 +number_of_gates = 4 + +node = onnx.helper.make_node( + "LSTM", + inputs=["X", "W", "R", "B"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +# Adding custom bias +W_B = custom_bias * np.ones((1, number_of_gates * hidden_size)).astype( + np.float32 +) +R_B = np.zeros((1, number_of_gates * hidden_size)).astype(np.float32) +B = np.concatenate((W_B, R_B), 1) + +lstm = LSTMHelper(X=input, W=W, R=R, B=B) +_, Y_h = lstm.step() +expect( + node, + inputs=[input, W, R, B], + outputs=[Y_h.astype(np.float32)], + name="test_lstm_with_initial_bias", +) +``` + +
+ + +
+peepholes + +```python +input = np.array([[[1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0]]]).astype( + np.float32 +) + +input_size = 4 +hidden_size = 3 +weight_scale = 0.1 +number_of_gates = 4 +number_of_peepholes = 3 + +node = onnx.helper.make_node( + "LSTM", + inputs=["X", "W", "R", "B", "sequence_lens", "initial_h", "initial_c", "P"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +# Initializing Inputs +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) +B = np.zeros((1, 2 * number_of_gates * hidden_size)).astype(np.float32) +seq_lens = np.repeat(input.shape[0], input.shape[1]).astype(np.int32) +init_h = np.zeros((1, input.shape[1], hidden_size)).astype(np.float32) +init_c = np.zeros((1, input.shape[1], hidden_size)).astype(np.float32) +P = weight_scale * np.ones((1, number_of_peepholes * hidden_size)).astype( + np.float32 +) + +lstm = LSTMHelper( + X=input, W=W, R=R, B=B, P=P, initial_c=init_c, initial_h=init_h +) +_, Y_h = lstm.step() +expect( + node, + inputs=[input, W, R, B, seq_lens, init_h, init_c, P], + outputs=[Y_h.astype(np.float32)], + name="test_lstm_with_peepholes", +) +``` + +
+ + +### **LayerNormalization** + + This is layer normalization defined in ONNX as function. + The overall computation can be split into two stages. + The first stage is standardization, which makes the + normalized elements have zero mean and unit variances. + The computation required by standardization can be + described by the following equations. + ``` + Mean = ReduceMean(X) + D = Sub(X, Mean) + DD = Mul(D, D) + Var = ReduceMean(DD) + VarEps = Add(Var, epsilon) + StdDev = Sqrt(VarEps) + InvStdDev = Reciprocal(StdDev) + Normalized = Mul(D, InvStdDev) + ``` + where `normalized_axes` is `[axis, ..., rank of X - 1]`. + The variables `Var` and `StdDev` stand for variance and + standard deviation, respectively. The second output is + `Mean` and the last one is `InvStdDev`. + Depending on `stash_type` attribute, the actual computation + must happen in different floating-point precision. + For example, if `stash_type` is 1, this operator casts + all input variables to 32-bit float, perform the computation, and + finally cast `Normalized` back to the original type of `X`. + The second stage then scales and shifts the outcome of the + first stage using + ``` + NormalizedScaled = Mul(Normalized, Scale) + Y = Add(NormalizedScaled, B) + ``` + The second stage doesn't depends on `stash_type`. + All equations are in [this syntax](https://github.com/onnx/onnx/blob/main/docs/Syntax.md). + The same variable (i.e., input, output, and attribute) uses + the same name in the equations above and this operator's definition. + Let `d[i]` indicate the i-th dimension of `X`. + If `X`'s shape is `[d[0], ..., d[axis-1], d[axis], ..., d[rank-1]]`, + the shape of `Mean` and `InvStdDev` is `[d[0], ..., d[axis-1], 1, ..., 1]`. + `Y` and `X` have the same shape. This operator supports unidirectional broadcasting + (tensors `Scale` and `B` should be unidirectional broadcastable to tensor `X`); + for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
The first normalization dimension. If rank(X) is r, axis' allowed range is [-r, r). Negative value means counting dimensions from the back.
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
stash_type : int (default is 1)
+
Type of Mean and InvStdDev. This also specifies stage one's computation precision.
+
+ +#### Inputs (2 - 3) + +
+
X : T
+
Tensor to be normalized.
+
Scale : T
+
Scale tensor.
+
B (optional) : T
+
Bias tensor.
+
+ +#### Outputs (1 - 3) + +
+
Y : T
+
Normalized tensor.
+
Mean (optional) : U
+
Saved mean used during training to speed up gradient computation
+
InvStdDev (optional) : U
+
Saved inverse standard deviation used during training to speed up gradient computation.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types and output Y type to float tensors.
+
U : tensor(float), tensor(bfloat16)
+
Type of Mean and InvStdDev tensors.
+
+ + +#### Examples + +
+d + +```python +X = np.random.randn(3, 4).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + B = np.random.randn(*normalized_shape).astype(np.float32) + Y, mean, inv_std_dev = _layer_normalization(X, W, B, axis=axis) + + node = onnx.helper.make_node( + "LayerNormalization", + inputs=["X", "W", "B"], + outputs=["Y", "Mean", "InvStdDev"], + axis=axis, + ) + + if axis < 0: + name = f"test_layer_normalization_2d_axis_negative_{-axis}" + else: + name = f"test_layer_normalization_2d_axis{axis}" + + expect(node, inputs=[X, W, B], outputs=[Y, mean, inv_std_dev], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+ + +
+d_epsilon + +```python +epsilon = 1e-1 +X = np.random.randn(2, 3, 5).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + B = np.random.randn(*normalized_shape).astype(np.float32) + Y, mean, inv_std_dev = _layer_normalization(X, W, B, axis, epsilon) + node = onnx.helper.make_node( + "LayerNormalization", + inputs=["X", "W", "B"], + outputs=["Y", "Mean", "InvStdDev"], + axis=axis, + epsilon=epsilon, + ) + + if axis < 0: + name = f"test_layer_normalization_3d_axis_negative_{-axis}_epsilon" + else: + name = f"test_layer_normalization_3d_axis{axis}_epsilon" + + expect(node, inputs=[X, W, B], outputs=[Y, mean, inv_std_dev], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+ + +
+default_axis + +```python +X = np.random.randn(2, 3, 4, 5).astype(np.float32) + +# Default axis in LayerNormalization is -1. +normalized_shape = calculate_normalized_shape(X.shape, -1) +W = np.random.randn(*normalized_shape).astype(np.float32) +B = np.random.randn(*normalized_shape).astype(np.float32) +# Axis is default to -1 in the reference implementation. +Y, mean, inv_std_dev = _layer_normalization(X, W, B) + +# Not specifying axis attribute means -1. +node = onnx.helper.make_node( + "LayerNormalization", + inputs=["X", "W", "B"], + outputs=["Y", "Mean", "InvStdDev"], +) + +expect( + node, + inputs=[X, W, B], + outputs=[Y, mean, inv_std_dev], + name="test_layer_normalization_default_axis", +) +``` + +
+ + +
+layernormalization + +```python +X = np.random.randn(2, 3, 4, 5).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + B = np.random.randn(*normalized_shape).astype(np.float32) + Y, mean, inv_std_dev = _layer_normalization(X, W, B, axis) + + node = onnx.helper.make_node( + "LayerNormalization", + inputs=["X", "W", "B"], + outputs=["Y", "Mean", "InvStdDev"], + axis=axis, + ) + + if axis < 0: + name = f"test_layer_normalization_4d_axis_negative_{-axis}" + else: + name = f"test_layer_normalization_4d_axis{axis}" + + expect(node, inputs=[X, W, B], outputs=[Y, mean, inv_std_dev], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+ + +### **LeakyRelu** + + LeakyRelu takes input data (Tensor) and an argument alpha, and produces one + output data (Tensor) where the function `f(x) = alpha * x for x < 0`, + `f(x) = x for x >= 0`, is applied to the data tensor elementwise. + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Attributes + +
+
alpha : float (default is 0.01)
+
Coefficient of leakage.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+leakyrelu + +```python +node = onnx.helper.make_node( + "LeakyRelu", inputs=["x"], outputs=["y"], alpha=0.1 +) + +x = np.array([-1, 0, 1]).astype(np.float32) +# expected output [-0.1, 0., 1.] +y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * 0.1 +expect(node, inputs=[x], outputs=[y], name="test_leakyrelu_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * 0.1 +expect(node, inputs=[x], outputs=[y], name="test_leakyrelu") +``` + +
+ + +
+leakyrelu_default + +```python +default_alpha = 0.01 +node = onnx.helper.make_node( + "LeakyRelu", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * default_alpha +expect(node, inputs=[x], outputs=[y], name="test_leakyrelu_default") +``` + +
+ + +### **Less** + + Returns the tensor resulted from performing the `less` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 7, 9 + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ + +#### Examples + +
+less + +```python +node = onnx.helper.make_node( + "Less", + inputs=["x", "y"], + outputs=["less"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less") + +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.random.randn(3, 4, 5).astype(np.int8) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_int8") + +x = np.random.randn(3, 4, 5).astype(np.int16) +y = np.random.randn(3, 4, 5).astype(np.int16) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_uint64") +``` + +
+ + +
+less + +```python +node = onnx.helper.make_node( + "LessOrEqual", + inputs=["x", "y"], + outputs=["less_equal"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal") + +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.random.randn(3, 4, 5).astype(np.int8) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_int8") + +x = np.random.randn(3, 4, 5).astype(np.int16) +y = np.random.randn(3, 4, 5).astype(np.int16) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_uint64") +``` + +
+ + +
+less_broadcast + +```python +node = onnx.helper.make_node( + "Less", + inputs=["x", "y"], + outputs=["less"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_bcast") +``` + +
+ + +
+less_broadcast + +```python +node = onnx.helper.make_node( + "LessOrEqual", + inputs=["x", "y"], + outputs=["less_equal"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_bcast") +``` + +
+ + +### **LessOrEqual** + + Returns the tensor resulted from performing the `less_equal` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +Other versions of this operator: 12 + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input types to all numeric tensors.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ + +### **LinearAttention** + + Unified linear attention operator for autoregressive decoding (T=1) and prefill (T>1). + + The query, key, value, and (where applicable) decay/beta inputs use 3D packed format + [B, T, H*D], where heads are flattened into the last dimension; q_num_heads and + kv_num_heads are always required and are used to unpack to 4D internally for computation. + The optional past_state and present_state are 4D with shape (B, H_kv, d_k, d_v). + + Group-query attention (GQA) is supported: q_num_heads must be a positive multiple of + kv_num_heads. When q_num_heads == kv_num_heads this reduces to multi-headed linear + attention; when q_num_heads > kv_num_heads each KV head (and its recurrent state) is + shared by `q_num_heads / kv_num_heads` query heads (multi-query attention is the + special case kv_num_heads == 1). + + The update_rule attribute selects the recurrence type: + - "linear": S_t = S_{t-1} + k_t ⊗ v_t; o_t = scale * q_t^T S_t + - "gated": S_t = exp(g_t) * S_{t-1} + k_t ⊗ v_t; o_t = scale * q_t^T S_t + - "delta": S_t = S_{t-1} + β_t * k_t ⊗ (v_t - S_{t-1}^T k_t); o_t = scale * q_t^T S_t + - "gated_delta": S_t = exp(g_t) * S_{t-1} + β_t * k_t ⊗ (v_t - exp(g_t) * S_{t-1}^T k_t); o_t = scale * q_t^T S_t + + where g_t is the decay (in log-space), β_t is the update rate, and ⊗ denotes outer product. + + Semantics: Equivalent to running the recurrent update sequentially for each token, + but may be implemented using chunk-parallel algorithms for GPU efficiency. + + +#### Version + +This version of the operator has been available since version 27 of the default ONNX operator set. + +#### Attributes + +
+
chunk_size : int (default is 64)
+
Chunk size for the chunk-parallel WY decomposition during prefill (T>1). Tuning hint; does not affect output correctness.
+
kv_num_heads : int (required)
+
Number of key/value heads. Always required.
+
q_num_heads : int (required)
+
Number of query heads. Always required.
+
scale : float (default is 0.0)
+
Output scaling factor. When 0.0 (default), derives d_k = query.shape[-1] / q_num_heads and uses 1/sqrt(d_k). Set explicitly to override.
+
update_rule : string (default is gated_delta)
+
The update rule for the linear attention recurrence. One of: 'linear', 'gated', 'delta', 'gated_delta'. Default is 'gated_delta'.
+
+ +#### Inputs (3 - 6) + +
+
query (differentiable) : T
+
Query vectors with 3D packed shape (B, T, H_q * d_k). Heads are packed into the last dimension.
+
key (differentiable) : T
+
Key vectors with 3D packed shape (B, T, H_kv * d_k). Should be L2-normalized for delta/gated_delta modes.
+
value (differentiable) : T
+
Value vectors with 3D packed shape (B, T, H_kv * d_v).
+
past_state (optional, non-differentiable) : S
+
Recurrent state from previous step with shape (B, H_kv, d_k, d_v). Always 4D. If not provided, defaults to zeros.
+
decay (optional, differentiable) : T
+
Exponential decay gate in log-space. 3D packed shape: (B, T, H_kv * d_k) for per-key-dimension decay (GLA/RWKV-6), or (B, T, H_kv) for per-head scalar decay (DeltaNet/RetNet). Required for 'gated' and 'gated_delta' modes.
+
beta (optional, differentiable) : T
+
Update rate (sigmoid output). 3D packed shape: (B, T, H_kv) or (B, T, 1). Required for 'delta' and 'gated_delta' modes.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Attention output with 3D packed shape (B, T, H_q * d_v).
+
present_state (non-differentiable) : S
+
Updated recurrent state with shape (B, H_kv, d_k, d_v). Always 4D.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(bfloat16), tensor(float)
+
Constrain activation input and output types to float16, bfloat16, or float32 tensors.
+
S : tensor(float16), tensor(bfloat16), tensor(float)
+
Constrain state types to float16, bfloat16, or float32 tensors. Should be float32 or the same as T for numerical stability on long sequences.
+
+ + +#### Examples + +
+decode_step + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "past_state", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 1, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +past_state = np.random.randn(b, h_kv, d_k, d_v).astype(np.float32) * 0.1 +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + past_state=past_state, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, past_state, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_decode_step", + opset_imports=_OPSET, +) +``` + +
+ + +
+delta + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "", "beta"], + outputs=["output", "present_state"], + update_rule="delta", + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, + update_rule="delta", +) +expect( + node, + inputs=[query, key, value, beta], + outputs=[output, present_state], + name="test_linear_attention_delta", + opset_imports=_OPSET, +) +``` + +
+ + +
+explicit_scale + +```python +scale = 0.25 +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, + scale=scale, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, + scale=scale, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_explicit_scale", + opset_imports=_OPSET, +) +``` + +
+ + +
+fp16 + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=8, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 8, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float16) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float16), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float16) +decay = (-np.abs(np.random.randn(b, t, h_kv * d_k)) * 0.1).astype(np.float16) +beta = np.random.rand(b, t, h_kv).astype(np.float16) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_fp16", + opset_imports=_OPSET, +) +``` + +
+ + +
+gated + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay"], + outputs=["output", "present_state"], + update_rule="gated", + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = np.random.randn(b, t, h_kv * d_k).astype(np.float32) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +# Per-key-dim decay in log-space (negative -> decay). +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 + +output, present_state = _compute( + query, + key, + value, + decay=decay, + q_num_heads=h_q, + kv_num_heads=h_kv, + update_rule="gated", +) +expect( + node, + inputs=[query, key, value, decay], + outputs=[output, present_state], + name="test_linear_attention_gated", + opset_imports=_OPSET, +) +``` + +
+ + +
+gated_delta + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_gated_delta", + opset_imports=_OPSET, +) +``` + +
+ + +
+gated_delta_beta_scalar + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, 1).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_gated_delta_beta_scalar", + opset_imports=_OPSET, +) +``` + +
+ + +
+gated_delta_gqa + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=8, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 8, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_gated_delta_gqa", + opset_imports=_OPSET, +) +``` + +
+ + +
+gated_delta_mqa + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=8, + kv_num_heads=1, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 8, 1, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_gated_delta_mqa", + opset_imports=_OPSET, +) +``` + +
+ + +
+gated_per_head_decay + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay"], + outputs=["output", "present_state"], + update_rule="gated", + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = np.random.randn(b, t, h_kv * d_k).astype(np.float32) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +# Per-head scalar decay. +decay = -np.abs(np.random.randn(b, t, h_kv)).astype(np.float32) * 0.1 + +output, present_state = _compute( + query, + key, + value, + decay=decay, + q_num_heads=h_q, + kv_num_heads=h_kv, + update_rule="gated", +) +expect( + node, + inputs=[query, key, value, decay], + outputs=[output, present_state], + name="test_linear_attention_gated_per_head_decay", + opset_imports=_OPSET, +) +``` + +
+ + +
+linear + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value"], + outputs=["output", "present_state"], + update_rule="linear", + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = np.random.randn(b, t, h_kv * d_k).astype(np.float32) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + q_num_heads=h_q, + kv_num_heads=h_kv, + update_rule="linear", +) +expect( + node, + inputs=[query, key, value], + outputs=[output, present_state], + name="test_linear_attention_linear", + opset_imports=_OPSET, +) +``` + +
+ + +
+linear_t1_no_past + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value"], + outputs=["output", "present_state"], + update_rule="linear", + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 1, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = np.random.randn(b, t, h_kv * d_k).astype(np.float32) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + q_num_heads=h_q, + kv_num_heads=h_kv, + update_rule="linear", +) +expect( + node, + inputs=[query, key, value], + outputs=[output, present_state], + name="test_linear_attention_linear_t1_no_past", + opset_imports=_OPSET, +) +``` + +
+ + +
+no_past_explicit_zeros + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "past_state", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +past_state = np.zeros((b, h_kv, d_k, d_v), dtype=np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + past_state=past_state, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, past_state, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_no_past_explicit_zeros", + opset_imports=_OPSET, +) +``` + +
+ + +
+prefill_with_past + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "past_state", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +past_state = np.random.randn(b, h_kv, d_k, d_v).astype(np.float32) * 0.1 +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + past_state=past_state, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, past_state, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_prefill_with_past", + opset_imports=_OPSET, +) +``` + +
+ + +### **Log** + + Calculates the natural log of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The natural log of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+log + +```python +node = onnx.helper.make_node( + "Log", + inputs=["x"], + outputs=["y"], +) + +x = np.array([1, 10]).astype(np.float32) +y = np.log(x) # expected output [0., 2.30258512] +expect(node, inputs=[x], outputs=[y], name="test_log_example") + +x = np.exp(np.random.randn(3, 4, 5).astype(np.float32)) +y = np.log(x) +expect(node, inputs=[x], outputs=[y], name="test_log") +``` + +
+ + +### **LogSoftmax** + + The operator computes the log of softmax values for the given input: + + LogSoftmax(input, axis) = Log(Softmax(input, axis=axis)) + + The "axis" attribute indicates the dimension along which LogSoftmax + will be performed. The output tensor has the same shape + and contains the LogSoftmax values of the corresponding input. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 11 + +#### Attributes + +
+
axis : int (default is -1)
+
+Describes the dimension LogSoftmax will be performed on. +Negative value means counting dimensions +from the back. Accepted range is [-r, r-1] where r = rank(input). +
+
+ +#### Inputs + +
+
input (differentiable) : T
+
The input tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output values with the same shape as the input tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+logsoftmax + +```python +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], +) +x = np.array([[-1, 0, 1]]).astype(np.float32) +# expected output +# [[-2.4076061 -1.407606 -0.407606 ]] +y = logsoftmax(x) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_example_1") +``` + +
+ + +
+logsoftmax_axis + +```python +x = np.array([[0, 1, 2, 3], [10000, 10001, 10002, 10003]]).astype(np.float32) +# expected output +# [[-3.4401896 -2.4401896 -1.4401896 -0.44018966] +# [-3.4401896 -2.4401896 -1.4401896 -0.44018966]] +y = logsoftmax(x) + +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], +) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_large_number") + +x = np.abs(np.random.randn(3, 4, 5).astype(np.float32)) +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], + axis=0, +) +y = logsoftmax(x, axis=0) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_axis_0") + +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], + axis=1, +) +y = logsoftmax(x, axis=1) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_axis_1") + +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], + axis=2, +) +y = logsoftmax(x, axis=2) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_axis_2") + +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], + axis=-1, +) +y = logsoftmax(x, axis=-1) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_negative_axis") + +# default axis is -1 +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], +) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_default_axis") +``` + +
+ + +### **Loop** + + Generic Looping construct. This loop has multiple termination conditions: + + 1) Trip count. Iteration count specified at runtime. Set by + specifying the input M. Optional. Set to empty string to omit. + Note that a static trip count (specified at graph construction time) can be + specified by passing in a constant node for input M. + 2) Loop termination condition. This is an input to the op that determines + whether to run the first iteration and also a loop-carried dependency for + the body graph. The body graph must yield a value for the condition variable, + whether this input is provided or not. + + This table summarizes the operating modes of this operator with equivalent + C-style code: + + Operator inputs defined as (max_trip_count, condition_var). + + * input ("", ""): + for (int i=0; ; ++i) { + cond = ... // Note this value is ignored, but is required in the body + } + + * input ("", cond) // Note this is analogous to a while loop + bool cond = ...; + for (int i=0; cond; ++i) { + cond = ...; + } + + * input ("", 1) // Note this is analogous to a do-while loop + bool cond = true + for (int i=0; cond; ++i) { + cond = ...; + } + + * input (trip_count, "") // Note this is analogous to a for loop + int trip_count = ... + for (int i=0; i < trip_count; ++i) { + cond = ...; // ignored + } + + * input (trip_count, cond) + int trip_count = ...; + bool cond = ...; + for (int i=0; i < trip_count && cond; ++i) { + cond = ...; + } + + + *Sample usage - cond as well as trip count* + + graph predict-net { + %a = Constant[value = ]() + %b = Constant[value = ]() + %keepgoing = Constant[value = ]() + %max_trip_count = Constant[value = ]() + %keepgoing_out, %b_out, %user_defined_vals = Loop[body = ](%max_trip_count, %keepgoing, %b) + return + } + + graph body-net ( + %i[INT32, scalar] // iteration number + %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used + %b_in[INT32, scalar] // incoming value of loop-carried-dependency b + ) { + %my_local = Add(%a, %b_in) + %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b + %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition + %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated + return %keepgoing_out, %b_out, %user_defined_val + } + + *Sample equivalent C code* + + { + /* User-defined code (enclosing scope) */ + int a = 3, b = 6; + bool keepgoing = true; // Analogous to input cond + /* End user-defined code */ + + /* Implicitly-defined code */ + const int max_trip_count = 10; // Analogous to input M + int user_defined_vals[]; // Imagine this is resizable + /* End implicitly-defined code */ + /* initialize loop-carried variables and scan-output variables */ + bool keepgoing_out = keepgoing + int b_out = b + + for (int i=0; i < max_trip_count && keepgoing_out; ++i) { + /* Implicitly-defined code: bind actual parameter values + to formal parameter variables of loop-body */ + bool keepgoing_in = keepgoing_out; + bool b_in = b_out; + + /* User-defined code (loop body) */ + int my_local = a + b_in; // Reading value "a" from the enclosing scope is fine + b_out = a - b_in; + keepgoing_out = my_local > b_out; + user_defined_val = b_in + b_in; // b_in and b_out are different variables + /* End user-defined code */ + + /* Implicitly defined-code */ + user_defined_vals[i] = user_defined_val // accumulate scan-output values + } + // int t = my_local; // Can't do this. my_local is not accessible here. + + // The values below are bound to the output variables of the loop and therefore accessible + // b_out; user_defined_vals; keepgoing_out; + } + + There are several things of note in this code snippet: + + 1) Values from the enclosing scope (i.e. variable "a" here) are in scope and can + be referenced in the inputs of the loop. + 2) Any values computed in the loop body that needs to be used in a subsequent + iteration or after the loop are modeled using a pair of variables in the loop-body, + consisting of an input variable (eg., b_in) and an output variable (eg., b_out). + These are referred to as loop-carried dependences. The loop operation node + supplies the input value of the input variable for the first iteration, and + returns the output value of the output variable produced by the final + iteration. + 3) Scan_output variables are used to implicitly concatenate values computed across + all the iterations. In the above example, the value of user_defined_val computed + over all iterations are concatenated and returned as the value of user_defined_vals + after the loop. + 4) Values created in the body cannot be accessed in the enclosing scope, + except using the mechanism described above. + + Note that the semantics of this op support "diagonal" or "wavefront" execution. + (See Step 3 here for an example: + https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). + Frontends should emit multi-layer RNNs as a series of While operators (with + time being the inner looping dimension), with each successive layer consuming + the scan_outputs from the previous layer, possibly going through several + point-wise operators (e.g. dropout, residual connections, linear layer). + + The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13, 16, 19, 21, 23, 24 + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies...). It has 1+N+K outputs: (condition, loop carried dependencies..., scan_outputs...). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
+
+ +#### Inputs (2 - ∞) + +
+
M (optional) : I
+
A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
+
cond (optional) : B
+
A boolean termination condition. Optional. Pass empty string to skip.
+
v_initial (variadic, heterogeneous) : V
+
The initial values of any loop-carried dependencies (values that change across loop iterations)
+
+ +#### Outputs (1 - ∞) + +
+
v_final_and_scan_outputs (variadic, heterogeneous) : V
+
Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), seq(tensor(float4e2m1)), seq(tensor(float8e8m0)), seq(tensor(uint2)), seq(tensor(int2)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4)), optional(tensor(float4e2m1)), optional(tensor(float8e8m0)), optional(tensor(uint2)), optional(tensor(int2))
+
All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv13.
+
I : tensor(int64)
+
tensor of int64, which should be a scalar.
+
B : tensor(bool)
+
tensor of bool, which should be a scalar.
+
+ + +#### Examples + +
+loop_11 + +```python +# Given a tensor x of values [x1, ..., xN], and initial tensor y +# sum up its elements using a scan +# returning the final state (y+x1+x2+...+xN) as well the scan_output +# [y+x1, y+x1+x2, ..., y+x1+x2+...+xN] + +y_in = onnx.helper.make_tensor_value_info("y_in", onnx.TensorProto.FLOAT, [1]) +y_out = onnx.helper.make_tensor_value_info("y_out", onnx.TensorProto.FLOAT, [1]) +scan_out = onnx.helper.make_tensor_value_info( + "scan_out", onnx.TensorProto.FLOAT, [1] +) +cond_in = onnx.helper.make_tensor_value_info( + "cond_in", onnx.TensorProto.BOOL, [] +) +cond_out = onnx.helper.make_tensor_value_info( + "cond_out", onnx.TensorProto.BOOL, [] +) +iter_count = onnx.helper.make_tensor_value_info( + "iter_count", onnx.TensorProto.INT64, [] +) + +x = np.array([1, 2, 3, 4, 5]).astype(np.float32) +y = np.array([-2]).astype(np.float32) + +x_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["x"], + value=onnx.helper.make_tensor( + name="const_tensor_x", + data_type=onnx.TensorProto.FLOAT, + dims=x.shape, + vals=x.flatten().astype(float), + ), +) + +one_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["one"], + value=onnx.helper.make_tensor( + name="const_tensor_one", + data_type=onnx.TensorProto.INT64, + dims=(), + vals=[1], + ), +) + +i_add_node = onnx.helper.make_node( + "Add", inputs=["iter_count", "one"], outputs=["end"] +) + +start_unsqueeze_node = onnx.helper.make_node( + "Unsqueeze", inputs=["iter_count"], outputs=["slice_start"], axes=[0] +) + +end_unsqueeze_node = onnx.helper.make_node( + "Unsqueeze", inputs=["end"], outputs=["slice_end"], axes=[0] +) + +slice_node = onnx.helper.make_node( + "Slice", inputs=["x", "slice_start", "slice_end"], outputs=["slice_out"] +) + +y_add_node = onnx.helper.make_node( + "Add", inputs=["y_in", "slice_out"], outputs=["y_out"] +) + +identity_node = onnx.helper.make_node( + "Identity", inputs=["cond_in"], outputs=["cond_out"] +) + +scan_identity_node = onnx.helper.make_node( + "Identity", inputs=["y_out"], outputs=["scan_out"] +) + +loop_body = onnx.helper.make_graph( + [ + identity_node, + x_const_node, + one_const_node, + i_add_node, + start_unsqueeze_node, + end_unsqueeze_node, + slice_node, + y_add_node, + scan_identity_node, + ], + "loop_body", + [iter_count, cond_in, y_in], + [cond_out, y_out, scan_out], +) + +node = onnx.helper.make_node( + "Loop", + inputs=["trip_count", "cond", "y"], + outputs=["res_y", "res_scan"], + body=loop_body, +) + +trip_count = np.array(5).astype(np.int64) +res_y = np.array([13]).astype(np.float32) +cond = np.array(1).astype(bool) +res_scan = np.array([-1, 1, 4, 8, 13]).astype(np.float32).reshape((5, 1)) +expect( + node, + inputs=[trip_count, cond, y], + outputs=[res_y, res_scan], + name="test_loop11", + opset_imports=[onnx.helper.make_opsetid("", 11)], +) +``` + +
+ + +
+loop_13 + +```python +# Given a tensor x of values [x1, ..., xN], +# Return a sequence of tensors of +# [[x1], [x1, x2], ..., [x1, ..., xN]] + +seq_in = onnx.helper.make_tensor_sequence_value_info( + "seq_in", onnx.TensorProto.FLOAT, None +) +seq_out = onnx.helper.make_tensor_sequence_value_info( + "seq_out", onnx.TensorProto.FLOAT, None +) +cond_in = onnx.helper.make_tensor_value_info( + "cond_in", onnx.TensorProto.BOOL, [] +) +cond_out = onnx.helper.make_tensor_value_info( + "cond_out", onnx.TensorProto.BOOL, [] +) +iter_count = onnx.helper.make_tensor_value_info( + "iter_count", onnx.TensorProto.INT64, [] +) + +x = np.array([1, 2, 3, 4, 5]).astype(np.float32) + +x_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["x"], + value=onnx.helper.make_tensor( + name="const_tensor_x", + data_type=onnx.TensorProto.FLOAT, + dims=x.shape, + vals=x.flatten().astype(float), + ), +) + +one_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["one"], + value=onnx.helper.make_tensor( + name="const_tensor_one", + data_type=onnx.TensorProto.INT64, + dims=(), + vals=[1], + ), +) + +zero_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["slice_start"], + value=onnx.helper.make_tensor( + name="const_tensor_zero", + data_type=onnx.TensorProto.INT64, + dims=(1,), + vals=[0], + ), +) + +axes_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["axes"], + value=onnx.helper.make_tensor( + name="const_tensor_axes", + data_type=onnx.TensorProto.INT64, + dims=(), + vals=[0], + ), +) + +add_node = onnx.helper.make_node( + "Add", inputs=["iter_count", "one"], outputs=["end"] +) + +end_unsqueeze_node = onnx.helper.make_node( + "Unsqueeze", inputs=["end", "axes"], outputs=["slice_end"] +) + +slice_node = onnx.helper.make_node( + "Slice", inputs=["x", "slice_start", "slice_end"], outputs=["slice_out"] +) + +insert_node = onnx.helper.make_node( + "SequenceInsert", inputs=["seq_in", "slice_out"], outputs=["seq_out"] +) + +identity_node = onnx.helper.make_node( + "Identity", inputs=["cond_in"], outputs=["cond_out"] +) + +loop_body = onnx.helper.make_graph( + [ + identity_node, + x_const_node, + one_const_node, + zero_const_node, + add_node, + axes_node, + end_unsqueeze_node, + slice_node, + insert_node, + ], + "loop_body", + [iter_count, cond_in, seq_in], + [cond_out, seq_out], +) + +node = onnx.helper.make_node( + "Loop", + inputs=["trip_count", "cond", "seq_empty"], + outputs=["seq_res"], + body=loop_body, +) + +trip_count = np.array(5).astype(np.int64) +seq_empty: list[Any] = [] +seq_res = [x[: int(i)] for i in x] +cond = np.array(1).astype(bool) +expect( + node, + inputs=[trip_count, cond, seq_empty], + outputs=[seq_res], + name="test_loop13_seq", + opset_imports=[onnx.helper.make_opsetid("", 13)], + input_type_protos=[ + onnx.helper.make_tensor_type_proto( + onnx.TensorProto.INT64, trip_count.shape + ), + onnx.helper.make_tensor_type_proto(onnx.TensorProto.BOOL, cond.shape), + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, []) + ), + ], +) +``` + +
+ + +
+loop_16_none + +```python +# Given a tensor sequence of values [x1, ..., xN], and an initial optional sequence of tensors [x0], +# Return a concatenated sequence of tensors of +# [x0, [x1], [x1, x2], ..., [x1, ..., xN]] + +ten_in_tp = onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, []) +seq_in_tp = onnx.helper.make_sequence_type_proto(ten_in_tp) +opt_in_tp = onnx.helper.make_optional_type_proto(seq_in_tp) +opt_in = onnx.helper.make_value_info("opt_seq_in", opt_in_tp) +seq_out = onnx.helper.make_tensor_sequence_value_info( + "seq_out", onnx.TensorProto.FLOAT, [] +) +cond_in = onnx.helper.make_tensor_value_info( + "cond_in", onnx.TensorProto.BOOL, [] +) +cond_out = onnx.helper.make_tensor_value_info( + "cond_out", onnx.TensorProto.BOOL, [] +) +iter_count = onnx.helper.make_tensor_value_info( + "iter_count", onnx.TensorProto.INT64, [] +) + +x0 = np.array(0).astype(np.float32) +x = np.array([1, 2, 3, 4, 5]).astype(np.float32) + +optional_has_elem_node = onnx.helper.make_node( + "OptionalHasElement", inputs=["opt_seq_in"], outputs=["optional_has_elem"] +) + +optional_is_none = onnx.helper.make_node( + "Not", inputs=["optional_has_elem"], outputs=["optional_is_none"] +) + +optional_get_elem = onnx.helper.make_node( + "OptionalGetElement", inputs=["opt_seq_in"], outputs=["seq_in"] +) + +constant_in = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["constant_in"], + value=onnx.helper.make_tensor( + name="const_tensor", data_type=onnx.TensorProto.FLOAT, dims=(), vals=[0] + ), +) + +seq_const_in = onnx.helper.make_node( + "SequenceConstruct", inputs=["constant_in"], outputs=["init_seq_in"] +) + +then_seq_out = onnx.helper.make_tensor_sequence_value_info( + "init_seq_in", onnx.TensorProto.FLOAT, [] +) +then_body = onnx.helper.make_graph( + [constant_in, seq_const_in], "then_body", [], [then_seq_out] +) + +else_seq_out = onnx.helper.make_tensor_sequence_value_info( + "seq_in", onnx.TensorProto.FLOAT, [] +) +else_body = onnx.helper.make_graph( + [optional_get_elem], "else_body", [], [else_seq_out] +) + +if_node = onnx.helper.make_node( + "If", + inputs=["optional_is_none"], + outputs=["sequence"], + then_branch=then_body, + else_branch=else_body, +) + +x_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["x"], + value=onnx.helper.make_tensor( + name="const_tensor_x", + data_type=onnx.TensorProto.FLOAT, + dims=x.shape, + vals=x.flatten().astype(float), + ), +) + +one_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["one"], + value=onnx.helper.make_tensor( + name="const_tensor_one", + data_type=onnx.TensorProto.INT64, + dims=(), + vals=[1], + ), +) + +zero_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["slice_start"], + value=onnx.helper.make_tensor( + name="const_tensor_zero", + data_type=onnx.TensorProto.INT64, + dims=(1,), + vals=[0], + ), +) + +axes_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["axes"], + value=onnx.helper.make_tensor( + name="const_tensor_axes", + data_type=onnx.TensorProto.INT64, + dims=(), + vals=[0], + ), +) + +add_node = onnx.helper.make_node( + "Add", inputs=["iter_count", "one"], outputs=["end"] +) + +end_unsqueeze_node = onnx.helper.make_node( + "Unsqueeze", inputs=["end", "axes"], outputs=["slice_end"] +) + +slice_node = onnx.helper.make_node( + "Slice", inputs=["x", "slice_start", "slice_end"], outputs=["slice_out"] +) + +insert_node = onnx.helper.make_node( + "SequenceInsert", inputs=["sequence", "slice_out"], outputs=["seq_out"] +) + +identity_node = onnx.helper.make_node( + "Identity", inputs=["cond_in"], outputs=["cond_out"] +) + +loop_body = onnx.helper.make_graph( + [ + identity_node, + optional_has_elem_node, + optional_is_none, + if_node, + x_const_node, + one_const_node, + zero_const_node, + add_node, + axes_node, + end_unsqueeze_node, + slice_node, + insert_node, + ], + "loop_body", + [iter_count, cond_in, opt_in], + [cond_out, seq_out], +) + +node = onnx.helper.make_node( + "Loop", + inputs=["trip_count", "cond", "opt_seq"], + outputs=["seq_res"], + body=loop_body, +) + +trip_count = np.array(5).astype(np.int64) +cond = np.array(1).astype(bool) +seq_res = compute_loop_outputs(x, [x0], trip_count) +opt_seq_in: list[Any] = [x0] +expect( + node, + inputs=[trip_count, cond, opt_seq_in], + outputs=[seq_res], + name="test_loop16_seq_none", + opset_imports=[onnx.helper.make_opsetid("", 16)], + input_type_protos=[ + onnx.helper.make_tensor_type_proto( + onnx.TensorProto.INT64, trip_count.shape + ), + onnx.helper.make_tensor_type_proto(onnx.TensorProto.BOOL, cond.shape), + opt_in_tp, + ], +) +``` + +
+ + +### **LpNormalization** + + Given a matrix, apply Lp-normalization along the provided axis. + The output is computed as: `output = input / Lp_norm(input, axis)`. + When the Lp norm is zero (i.e., all elements along the axis are zero), + the output is defined to be zero to avoid division by zero. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Attributes + +
+
axis : int (default is -1)
+
The axis on which to apply normalization, -1 mean last axis.
+
p : int (default is 2)
+
The order of the normalization, only 1 or 2 are supported.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input matrix
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Matrix after normalization
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+default + +```python +node = onnx.helper.make_node("LpNormalization", inputs=["x"], outputs=["y"]) +x = np.array( + [[[1.0, 2.0, 2.0], [3.0, 4.0, 0.0]], [[0.0, 5.0, 5.0], [6.0, 8.0, 0.0]]], + dtype=np.float32, +) +lp_norm_default = np.sqrt(np.sum(x**2, axis=-1, keepdims=True)) +y = x / lp_norm_default +expect(node, inputs=[x], outputs=[y], name="test_lpnormalization_default") +``` + +
+ + +
+l1normalization_axis_0 + +```python +node = onnx.helper.make_node( + "LpNormalization", inputs=["x"], outputs=["y"], axis=0, p=1 +) +x = np.array([3.0, 4.0], dtype=np.float32) +l1_norm_axis_0 = np.sum(abs(x), axis=0, keepdims=True) +y = x / l1_norm_axis_0 +expect(node, inputs=[x], outputs=[y], name="test_l1normalization_axis_0") +``` + +
+ + +
+l1normalization_axis_1 + +```python +node = onnx.helper.make_node( + "LpNormalization", inputs=["x"], outputs=["y"], axis=1, p=1 +) +x = np.array([[3.0, 4.0], [6.0, 8.0]], dtype=np.float32) +l1_norm_axis_1 = np.sum(abs(x), axis=1, keepdims=True) +y = x / l1_norm_axis_1 +expect(node, inputs=[x], outputs=[y], name="test_l1normalization_axis_1") +``` + +
+ + +
+l1normalization_axis_last + +```python +node = onnx.helper.make_node( + "LpNormalization", inputs=["x"], outputs=["y"], axis=-1, p=1 +) +x = np.array( + [[[1.0, 2.0, 2.0], [3.0, 4.0, 0.0]], [[0.0, 5.0, 5.0], [6.0, 8.0, 0.0]]], + dtype=np.float32, +) +l1_norm_axis_last = np.sum(abs(x), axis=-1, keepdims=True) +y = x / l1_norm_axis_last +expect(node, inputs=[x], outputs=[y], name="test_l1normalization_axis_last") +``` + +
+ + +
+l2normalization_axis_0 + +```python +node = onnx.helper.make_node( + "LpNormalization", inputs=["x"], outputs=["y"], axis=0, p=2 +) +x = np.array( + [[[1.0, 2.0, 2.0], [3.0, 4.0, 0.0]], [[0.0, 5.0, 5.0], [6.0, 8.0, 0.0]]], + dtype=np.float32, +) +l2_norm_axis_0 = np.sqrt(np.sum(x**2, axis=0, keepdims=True)) +# When norm is 0, output is 0 (0/0 = 0) +y = np.where(l2_norm_axis_0 == 0, 0, x / l2_norm_axis_0) +expect(node, inputs=[x], outputs=[y], name="test_l2normalization_axis_0") +``` + +
+ + +
+l2normalization_axis_1 + +```python +node = onnx.helper.make_node( + "LpNormalization", inputs=["x"], outputs=["y"], axis=1, p=2 +) +x = np.array([[3.0, 4.0], [6.0, 8.0]], dtype=np.float32) +l2_norm_axis_1 = np.sqrt(np.sum(x**2, axis=1, keepdims=True)) +y = x / l2_norm_axis_1 +expect(node, inputs=[x], outputs=[y], name="test_l2normalization_axis_1") +``` + +
+ + +### **LpPool** + + LpPool consumes an input tensor X and applies Lp pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + Lp pooling consisting of computing the Lp norm on all values of a subset + of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape will be following: + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - {kernelSpatialShape}) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - {kernelSpatialShape}) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled `pad_shape[i]` is the sum of pads along axis `i`. + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - {kernelSpatialShape} + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + {kernelSpatialShape} - input_spatial_shape[i] + ``` + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 2, 11, 18 + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
p : int (default is 2)
+
p value of the Lp norm used to pool over the input data.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size.
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+lppool_1d_default + +```python +"""input_shape: [1, 3, 32] +output_shape: [1, 3, 31] +""" +p = 3 +kernel_shape = [2] +strides = [1] +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=kernel_shape, + strides=strides, + p=p, +) +x = np.random.randn(1, 3, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", p=p) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_1d_default") +``` + +
+ + +
+lppool_2d_default + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 31, 31] +""" +p = 4 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + p=p, +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = (2, 2) +strides = (1, 1) +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", p=p) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_default") +``` + +
+ + +
+lppool_2d_dilations + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +p = 2 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[1, 1], + dilations=[2, 2], + p=p, +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ] +).astype(np.float32) + +y = np.array( + [ + [ + [ + [14.560219778561036, 16.24807680927192], + [21.633307652783937, 23.49468024894146], + ] + ] + ] +).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_dilations") +``` + +
+ + +
+lppool_2d_pads + +```python +"""input_shape: [1, 3, 28, 28] +output_shape: [1, 3, 30, 30] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +p = 3 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[2, 2, 2, 2], + p=p, +) +x = np.random.randn(1, 3, 28, 28).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (3, 3) +strides = (1, 1) +pad_bottom = pad_top = pad_right = pad_left = 2 +pads = [pad_top, pad_left, pad_bottom, pad_right] +out_shape, extra_pads = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = np.pad( + x, + ( + (0, 0), + (0, 0), + (extra_pads[0], extra_pads[2]), + (extra_pads[1], extra_pads[3]), + ), + mode="constant", + constant_values=0, +) +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "LPPOOL", + pads_required=extra_pads, + pads=pads, + p=p, +) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_pads") +``` + +
+ + +
+lppool_2d_same_lower + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [1, 0, 1, 0] by axis +""" +p = 4 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_LOWER", + p=p, +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_LOWER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_LOWER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_bottom = pad_shape[0] // 2 +pad_top = pad_shape[0] - pad_bottom +pad_right = pad_shape[1] // 2 +pad_left = pad_shape[1] - pad_right +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=0, +) +pads = [pad_top, pad_left, pad_bottom, pad_right] +y = pool( + padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", pads, pads, p=p +) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_same_lower") +``` + +
+ + +
+lppool_2d_same_upper + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [0, 1, 0, 1] by axis +""" +p = 2 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_UPPER", + p=p, +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_UPPER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_UPPER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_top = pad_shape[0] // 2 +pad_bottom = pad_shape[0] - pad_top +pad_left = pad_shape[1] // 2 +pad_right = pad_shape[1] - pad_left +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=0, +) +pads = [pad_top, pad_left, pad_bottom, pad_right] +y = pool( + padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", pads, pads, p=p +) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_same_upper") +``` + +
+ + +
+lppool_2d_strides + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 10, 10] +""" +p = 2 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + strides=[3, 3], + p=p, +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = (5, 5) +strides = (3, 3) +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", p=p) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_strides") +``` + +
+ + +
+lppool_3d_default + +```python +"""input_shape: [1, 3, 32, 32, 32] +output_shape: [1, 3, 31, 31, 31] +""" +p = 3 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], + p=p, +) +x = np.random.randn(1, 3, 32, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = [2, 2, 2] +strides = [1, 1, 1] +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", p=p) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_3d_default") +``` + +
+ + +### **MatMul** + + Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 9 + +#### Inputs + +
+
A (differentiable) : T
+
N-dimensional matrix A
+
B (differentiable) : T
+
N-dimensional matrix B
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Matrix multiply results from A * B
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(bfloat16)
+
Constrain input and output types to float/int tensors.
+
+ + +#### Examples + +
+matmul + +```python +node = onnx.helper.make_node( + "MatMul", + inputs=["a", "b"], + outputs=["c"], +) + +# 2d +a = np.random.randn(3, 4).astype(np.float32) +b = np.random.randn(4, 3).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_2d") + +# 3d +a = np.random.randn(2, 3, 4).astype(np.float32) +b = np.random.randn(2, 4, 3).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_3d") + +# 4d +a = np.random.randn(1, 2, 3, 4).astype(np.float32) +b = np.random.randn(1, 2, 4, 3).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_4d") + +# broadcasting +a = np.random.randn(3, 1, 3, 4).astype(np.float32) +b = np.random.randn(1, 2, 4, 2).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_bcast") + +# 1d + 3d +a = np.random.randn(4).astype(np.float32) +b = np.random.randn(2, 4, 1).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_1d_3d") + +# 3d + 1d +a = np.random.randn(1, 2, 4, 3).astype(np.float32) +b = np.random.randn(3).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_4d_1d") + +# 1d + 1d +a = np.random.randn(3).astype(np.float32) +b = np.random.randn(3).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_1d_1d") +``` + +
+ + +### **MatMulInteger** + + Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html). + The production MUST never overflow. The accumulation may overflow if and only if in 32 bits. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Inputs (2 - 4) + +
+
A (non-differentiable) : T1
+
N-dimensional matrix A
+
B (non-differentiable) : T2
+
N-dimensional matrix B
+
a_zero_point (optional, non-differentiable) : T1
+
Zero point tensor for input 'A'. It's optional and default value is 0. It could be a scalar or N-D tensor. Scalar refers to per tensor quantization whereas N-D refers to per row quantization. If the input is 2D of shape [M, K] then zero point tensor may be an M element vector [zp_1, zp_2, ..., zp_M]. If the input is N-D tensor with shape [D1, D2, M, K] then zero point tensor may have shape [D1, D2, M, 1].
+
b_zero_point (optional, non-differentiable) : T2
+
Zero point tensor for input 'B'. It's optional and default value is 0. It could be a scalar or a N-D tensor, Scalar refers to per tensor quantization whereas N-D refers to per col quantization. If the input is 2D of shape [K, N] then zero point tensor may be an N element vector [zp_1, zp_2, ..., zp_N]. If the input is N-D tensor with shape [D1, D2, K, N] then zero point tensor may have shape [D1, D2, 1, N].
+
+ +#### Outputs + +
+
Y (non-differentiable) : T3
+
Matrix multiply results from A * B
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8)
+
Constrain input A data type to 8-bit integer tensor.
+
T2 : tensor(int8), tensor(uint8)
+
Constrain input B data type to 8-bit integer tensor.
+
T3 : tensor(int32)
+
Constrain output Y data type as 32-bit integer tensor.
+
+ + +#### Examples + +
+matmulinteger + +```python +node = onnx.helper.make_node( + "MatMulInteger", + inputs=["A", "B", "a_zero_point", "b_zero_point"], + outputs=["Y"], +) + +A = np.array( + [ + [11, 7, 3], + [10, 6, 2], + [9, 5, 1], + [8, 4, 0], + ], + dtype=np.uint8, +) + +a_zero_point = np.array([12], dtype=np.uint8) + +B = np.array( + [ + [1, 4], + [2, 5], + [3, 6], + ], + dtype=np.uint8, +) + +b_zero_point = np.array([0], dtype=np.uint8) + +output = np.array( + [ + [-38, -83], + [-44, -98], + [-50, -113], + [-56, -128], + ], + dtype=np.int32, +) + +expect( + node, + inputs=[A, B, a_zero_point, b_zero_point], + outputs=[output], + name="test_matmulinteger", +) +``` + +
+ + +### **Max** + + Element-wise max of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 8, 12 + +#### Inputs (1 - ∞) + +
+
data_0 (variadic, differentiable) : T
+
List of tensors for max.
+
+ +#### Outputs + +
+
max (differentiable) : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ + +#### Examples + +
+max + +```python +data_0 = np.array([3, 2, 1]).astype(np.float32) +data_1 = np.array([1, 4, 4]).astype(np.float32) +data_2 = np.array([2, 5, 3]).astype(np.float32) +result = np.array([3, 5, 4]).astype(np.float32) +node = onnx.helper.make_node( + "Max", + inputs=["data_0", "data_1", "data_2"], + outputs=["result"], +) +expect( + node, + inputs=[data_0, data_1, data_2], + outputs=[result], + name="test_max_example", +) + +node = onnx.helper.make_node( + "Max", + inputs=["data_0"], + outputs=["result"], +) +expect(node, inputs=[data_0], outputs=[data_0], name="test_max_one_input") + +result = np.maximum(data_0, data_1) +node = onnx.helper.make_node( + "Max", + inputs=["data_0", "data_1"], + outputs=["result"], +) +expect( + node, inputs=[data_0, data_1], outputs=[result], name="test_max_two_inputs" +) +``` + +
+ + +
+max_all_numeric_types + +```python +for op_dtype in all_numeric_dtypes: + data_0 = np.array([3, 2, 1]).astype(op_dtype) + data_1 = np.array([1, 4, 4]).astype(op_dtype) + result = np.array([3, 4, 4]).astype(op_dtype) + node = onnx.helper.make_node( + "Max", + inputs=["data_0", "data_1"], + outputs=["result"], + ) + expect( + node, + inputs=[data_0, data_1], + outputs=[result], + name=f"test_max_{np.dtype(op_dtype).name}", + ) +``` + +
+ + +### **MaxPool** + + MaxPool consumes an input tensor X and applies max pooling across + the tensor according to kernel sizes, stride sizes, and pad lengths. + max pooling consisting of computing the max on all values of a + subset of the input tensor according to the kernel size and downsampling the + data into the output tensor Y for further processing. The output spatial shape is calculated differently + depending on whether explicit padding is used, where pads is employed, or auto padding is used, where auto_pad is utilized. + With explicit padding (https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html?highlight=maxpool#torch.nn.MaxPool2d): + ``` + output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + or + ``` + output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1) + ``` + if ceil_mode is enabled. `pad_shape[i]` is the sum of pads along axis `i`. Sliding windows that would start in the right padded region are ignored. + + `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following when ceil_mode is enabled: + ``` + VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) + ``` + or when ceil_mode is disabled (https://www.tensorflow.org/api_docs/python/tf/keras/layers/AveragePooling2D): + ``` + VALID: output_spatial_shape[i] = floor((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i]) + 1 + SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = floor((input_spatial_shape[i] - 1) / strides_spatial_shape[i]) + 1 + ``` + And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`: + ``` + pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i] + ``` + The output of each pooling window is maximum number of elements exclude pad. + + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 8, 10, 11, 12 + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
ceil_mode : int (default is 0)
+
Whether to use ceil or floor (default) to compute the output shape.
+
dilations : list of ints
+
Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
storage_order : int (default is 0)
+
The storage order of the tensor. 0 is row major, and 1 is column major. This attribute is used only to convert an n-tuple index value into a single integer value for producing the second output.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
+ +#### Outputs (1 - 2) + +
+
Y (differentiable) : T
+
Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
+
Indices (optional, non-differentiable) : I
+
Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x ... x Dn).
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(uint8)
+
Constrain input and output types to float and 8 bit tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ + +#### Examples + +
+maxpool_1d_default + +```python +"""input_shape: [1, 3, 32] +output_shape: [1, 3, 31] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2], +) +x = np.random.randn(1, 3, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = [2] +strides = [1] +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX") + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_1d_default") +``` + +
+ + +
+maxpool_2d_ceil + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + strides=[2, 2], + ceil_mode=True, +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[11, 12], [15, 16]]]]).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_ceil") +``` + +
+ + +
+maxpool_2d_ceil_output_size_reduce_by_one + +```python +"""input_shape: [1, 1, 2, 2] +output_shape: [1, 1, 1, 1] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[1, 1], + strides=[2, 2], + ceil_mode=True, +) +x = np.array([[[[1, 2], [3, 4]]]]).astype(np.float32) +y = np.array([[[[1]]]]).astype(np.float32) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_maxpool_2d_ceil_output_size_reduce_by_one", +) +``` + +
+ + +
+maxpool_2d_default + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 31, 31] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = (2, 2) +strides = (1, 1) +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX") + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_default") +``` + +
+ + +
+maxpool_2d_dilations + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[1, 1], + dilations=[2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[11, 12], [15, 16]]]]).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_dilations") +``` + +
+ + +
+maxpool_2d_pads + +```python +"""input_shape: [1, 3, 28, 28] +output_shape: [1, 3, 30, 30] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[2, 2, 2, 2], +) +x = np.random.randn(1, 3, 28, 28).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (3, 3) +strides = (1, 1) +pad_bottom = pad_top = pad_right = pad_left = 2 +pads = [pad_top, pad_left, pad_bottom, pad_right] +out_shape, extra_pads = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=np.nan, +) + +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "MAX", + pads_required=extra_pads, + pads=pads, +) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_pads") +``` + +
+ + +
+maxpool_2d_precomputed_pads + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 5, 5] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + pads=[2, 2, 2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array( + [ + [ + [ + [13, 14, 15, 15, 15], + [18, 19, 20, 20, 20], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + ] + ] + ] +).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_precomputed_pads") +``` + +
+ + +
+maxpool_2d_precomputed_same_upper + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 3, 3] +pad_shape: [2, 2] -> [1, 1, 1, 1] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + strides=[2, 2], + auto_pad="SAME_UPPER", +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[7, 9, 10], [17, 19, 20], [22, 24, 25]]]]).astype(np.float32) + +expect( + node, inputs=[x], outputs=[y], name="test_maxpool_2d_precomputed_same_upper" +) +``` + +
+ + +
+maxpool_2d_precomputed_strides + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "MaxPool", inputs=["x"], outputs=["y"], kernel_shape=[2, 2], strides=[2, 2] +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[7, 9], [17, 19]]]]).astype(np.float32) + +expect( + node, inputs=[x], outputs=[y], name="test_maxpool_2d_precomputed_strides" +) +``` + +
+ + +
+maxpool_2d_same_lower + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [1, 0, 1, 0] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_LOWER", +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_LOWER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_LOWER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_bottom = pad_shape[0] // 2 +pad_top = pad_shape[0] - pad_bottom +pad_right = pad_shape[1] // 2 +pad_left = pad_shape[1] - pad_right +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=np.nan, +) +pads = [pad_top, pad_left, pad_bottom, pad_right] +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX", pads, pads) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_same_lower") +``` + +
+ + +
+maxpool_2d_same_upper + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [0, 1, 0, 1] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_UPPER", +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_UPPER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_UPPER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_top = pad_shape[0] // 2 +pad_bottom = pad_shape[0] - pad_top +pad_left = pad_shape[1] // 2 +pad_right = pad_shape[1] - pad_left +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=np.nan, +) +pads = [pad_top, pad_left, pad_bottom, pad_right] +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX", pads, pads) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_same_upper") +``` + +
+ + +
+maxpool_2d_strides + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 10, 10] +""" +node = onnx.helper.make_node( + "MaxPool", inputs=["x"], outputs=["y"], kernel_shape=[5, 5], strides=[3, 3] +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = (5, 5) +strides = (3, 3) +out_shape, pads = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX") + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_strides") +``` + +
+ + +
+maxpool_2d_uint8 + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 5, 5] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + pads=[2, 2, 2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.uint8) +y = np.array( + [ + [ + [ + [13, 14, 15, 15, 15], + [18, 19, 20, 20, 20], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + ] + ] + ] +).astype(np.uint8) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_uint8") +``` + +
+ + +
+maxpool_3d_default + +```python +"""input_shape: [1, 3, 32, 32, 32] +output_shape: [1, 3, 31, 31, 31] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], +) +x = np.random.randn(1, 3, 32, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = [2, 2, 2] +strides = [1, 1, 1] +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX") + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_3d_default") +``` + +
+ + +
+maxpool_3d_dilations + +```python +"""input_shape: [1, 1, 4, 4, 4] +output_shape: [1, 1, 2, 2, 2] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], + strides=[1, 1, 1], + dilations=[2, 2, 2], +) +x = np.array( + [ + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[[11, 12], [15, 16]], [[11, 12], [15, 16]]]]]).astype( + np.float32 +) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_3d_dilations") +``` + +
+ + +
+maxpool_3d_dilations_use_ref_impl + +```python +"""input_shape: [1, 1, 4, 4, 4] +output_shape: [1, 1, 2, 2, 2] +""" +dilations = [2, 2, 2] +kernel_shape = [2, 2, 2] +strides = [1, 1, 1] +ceil_mode = False +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], + strides=[1, 1, 1], + dilations=dilations, +) +x = np.array( + [ + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + ] + ] + ] +).astype(np.float32) + +x_shape = x.shape[2:] +out_shape, pads = get_output_shape_explicit_padding( + None, x_shape, kernel_shape, strides, dilations, ceil_mode=ceil_mode +) +padded = x +y = pool( + padded, + (1, 1, *x_shape), + kernel_shape, + strides, + out_shape, + "MAX", + pads_required=pads, + pads=None, + dilations=dilations, +) + +expect( + node, inputs=[x], outputs=[y], name="test_maxpool_3d_dilations_use_ref_impl" +) +``` + +
+ + +
+maxpool_3d_dilations_use_ref_impl_large + +```python +x_shape = (32, 32, 32) +dilations = (2, 2, 2) +kernel_shape = (5, 5, 5) +strides = (3, 3, 3) +ceil_mode = True + +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=kernel_shape, + strides=strides, + dilations=dilations, + ceil_mode=ceil_mode, +) + +x = np.random.randn(1, 1, *x_shape).astype(np.float32) +out_shape, pads = get_output_shape_explicit_padding( + None, x_shape, kernel_shape, strides, dilations, ceil_mode=ceil_mode +) +padded = np.pad( + x, + ( + (0, 0), + (0, 0), + (pads[0], pads[3]), + (pads[1], pads[4]), + (pads[2], pads[5]), + ), + mode="constant", + constant_values=0, +) +y = pool( + padded, + (1, 1, *x_shape), + kernel_shape, + strides, + out_shape, + "MAX", + pads_required=pads, + pads=None, + dilations=dilations, +) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_maxpool_3d_dilations_use_ref_impl_large", +) +``` + +
+ + +
+maxpool_with_argmax_2d_precomputed_pads + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 5, 5] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y", "z"], + kernel_shape=[5, 5], + pads=[2, 2, 2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array( + [ + [ + [ + [13, 14, 15, 15, 15], + [18, 19, 20, 20, 20], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + ] + ] + ] +).astype(np.float32) +z = np.array( + [ + [ + [ + [12, 13, 14, 14, 14], + [17, 18, 19, 19, 19], + [22, 23, 24, 24, 24], + [22, 23, 24, 24, 24], + [22, 23, 24, 24, 24], + ] + ] + ] +).astype(np.int64) + +expect( + node, + inputs=[x], + outputs=[y, z], + name="test_maxpool_with_argmax_2d_precomputed_pads", +) +``` + +
+ + +
+maxpool_with_argmax_2d_precomputed_strides + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y", "z"], + kernel_shape=[2, 2], + strides=[2, 2], + storage_order=1, +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[7, 9], [17, 19]]]]).astype(np.float32) +z = np.array([[[[6, 16], [8, 18]]]]).astype(np.int64) + +expect( + node, + inputs=[x], + outputs=[y, z], + name="test_maxpool_with_argmax_2d_precomputed_strides", +) +``` + +
+ + +### **MaxRoiPool** + + ROI max pool consumes an input tensor X and region of interests (RoIs) to + apply max pooling across each RoI, to produce output 4-D tensor of shape + (num_rois, channels, pooled_shape[0], pooled_shape[1]). + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Attributes + +
+
pooled_shape : list of ints (required)
+
ROI pool output shape (height, width).
+
spatial_scale : float (default is 1.0)
+
Multiplicative spatial scale factor to translate ROI coordinates from their input scale to the scale used when pooling.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
+
rois (non-differentiable) : T
+
RoIs (Regions of Interest) to pool over. Should be a 2-D tensor of shape (num_rois, 5) given as [[batch_id, x1, y1, x2, y2], ...].
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
RoI pooled output 4-D tensor of shape (num_rois, channels, pooled_shape[0], pooled_shape[1]).
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +### **MaxUnpool** + + MaxUnpool essentially computes the partial inverse of the MaxPool op. + The input information to this op is typically the output information from a MaxPool op. The first + input tensor X is the tensor that needs to be unpooled, which is typically the pooled tensor (first output) + from MaxPool. The second input tensor, I, contains the indices to the (locally maximal) elements corresponding + to the elements in the first input tensor X. Input tensor I is typically the second output of the MaxPool op. + The third (optional) input is a tensor that specifies the output size of the unpooling operation. + + MaxUnpool is intended to do 'partial' inverse of the MaxPool op. 'Partial' because all the non-maximal + values from the original input to MaxPool are set to zero in the output of the MaxUnpool op. Pooling + the result of an unpooling operation should give back the original input to the unpooling op. + + MaxUnpool can produce the same output size for several input sizes, which makes unpooling op ambiguous. + The third input argument, output_size, is meant to disambiguate the op and produce output tensor of + known/predictable size. + + In addition to the inputs, MaxUnpool takes three attributes, namely kernel_shape, strides, and pads, + which define the exact unpooling op. The attributes typically have the same values as the corresponding + pooling op that the unpooling op is trying to invert. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 9, 11 + +#### Attributes + +
+
kernel_shape : list of ints (required)
+
The size of the kernel along each axis.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. `pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs (2 - 3) + +
+
X (differentiable) : T1
+
Input data tensor that has to be unpooled. This tensor is typically the first output of the MaxPool op.Dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non-image case, the dimensions are in the form of (N x C x D1 x D2 ... Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
I (non-differentiable) : T2
+
Input data tensor containing the indices corresponding to elements in the first input tensor X.This tensor is typically the second output of the MaxPool op.Dimensions must be the same as input tensor X. The indices are linear, i.e. computed considering the tensor as flattened 1-D tensor, assuming row-major storage. Also, the linear indices should not consider padding. So the values in indices are in the range [0, N x C x D1 x ... x Dn).
+
output_shape (optional, non-differentiable) : T2
+
The shape of the output can be explicitly set which will cause pads values to be auto generated. If 'output_shape' is specified, 'pads' values are ignored.
+
+ +#### Outputs + +
+
output (differentiable) : T1
+
Output data tensor that contains the result of the unpooling.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T2 : tensor(int64)
+
Constrain index tensor to int64
+
+ + +#### Examples + +
+with_output_shape + +```python +node = onnx.helper.make_node( + "MaxUnpool", + inputs=["xT", "xI", "output_shape"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[2, 2], +) +xT = np.array([[[[5, 6], [7, 8]]]], dtype=np.float32) +xI = np.array([[[[5, 7], [13, 15]]]], dtype=np.int64) +output_shape = np.array((1, 1, 5, 5), dtype=np.int64) +y = np.array( + [ + [ + [ + [0, 0, 0, 0, 0], + [0, 5, 0, 6, 0], + [0, 0, 0, 0, 0], + [0, 7, 0, 8, 0], + [0, 0, 0, 0, 0], + ] + ] + ], + dtype=np.float32, +) +expect( + node, + inputs=[xT, xI, output_shape], + outputs=[y], + name="test_maxunpool_export_with_output_shape", +) +``` + +
+ + +
+without_output_shape + +```python +node = onnx.helper.make_node( + "MaxUnpool", + inputs=["xT", "xI"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[2, 2], +) +xT = np.array([[[[1, 2], [3, 4]]]], dtype=np.float32) +xI = np.array([[[[5, 7], [13, 15]]]], dtype=np.int64) +y = np.array( + [[[[0, 0, 0, 0], [0, 1, 0, 2], [0, 0, 0, 0], [0, 3, 0, 4]]]], + dtype=np.float32, +) +expect( + node, + inputs=[xT, xI], + outputs=[y], + name="test_maxunpool_export_without_output_shape", +) +``` + +
+ + +### **Mean** + + Element-wise mean of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 8 + +#### Inputs (1 - ∞) + +
+
data_0 (variadic, differentiable) : T
+
List of tensors for mean.
+
+ +#### Outputs + +
+
mean (differentiable) : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+mean + +```python +data_0 = np.array([3, 0, 2]).astype(np.float32) +data_1 = np.array([1, 3, 4]).astype(np.float32) +data_2 = np.array([2, 6, 6]).astype(np.float32) +result = np.array([2, 3, 4]).astype(np.float32) +node = onnx.helper.make_node( + "Mean", + inputs=["data_0", "data_1", "data_2"], + outputs=["result"], +) +expect( + node, + inputs=[data_0, data_1, data_2], + outputs=[result], + name="test_mean_example", +) + +node = onnx.helper.make_node( + "Mean", + inputs=["data_0"], + outputs=["result"], +) +expect(node, inputs=[data_0], outputs=[data_0], name="test_mean_one_input") + +result = np.divide(np.add(data_0, data_1), 2.0) +node = onnx.helper.make_node( + "Mean", + inputs=["data_0", "data_1"], + outputs=["result"], +) +expect( + node, inputs=[data_0, data_1], outputs=[result], name="test_mean_two_inputs" +) +``` + +
+ + +### **MeanVarianceNormalization** + + A MeanVarianceNormalization Function: Perform mean variance normalization + on the input tensor X using formula: `(X-EX)/sqrt(E(X-EX)^2)` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Attributes + +
+
axes : list of ints (default is ['0', '2', '3'])
+
A list of integers, along which to reduce. The default is to calculate along axes [0,2,3] for calculating mean and variance along each channel. Two variables with the same C-coordinate are associated with the same mean and variance.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ + +#### Examples + +
+meanvariancenormalization + +```python +node = onnx.helper.make_node( + "MeanVarianceNormalization", inputs=["X"], outputs=["Y"] +) + +input_data = np.array( + [ + [ + [[0.8439683], [0.5665144], [0.05836735]], + [[0.02916367], [0.12964272], [0.5060197]], + [[0.79538304], [0.9411346], [0.9546573]], + ], + [ + [[0.17730942], [0.46192095], [0.26480448]], + [[0.6746842], [0.01665257], [0.62473077]], + [[0.9240844], [0.9722341], [0.11965699]], + ], + [ + [[0.41356155], [0.9129373], [0.59330076]], + [[0.81929934], [0.7862604], [0.11799799]], + [[0.69248444], [0.54119414], [0.07513223]], + ], + ], + dtype=np.float32, +) + +# Calculate expected output data +data_mean = np.mean(input_data, axis=(0, 2, 3), keepdims=1) +data_mean_squared = np.power(data_mean, 2) +data_squared = np.power(input_data, 2) +data_squared_mean = np.mean(data_squared, axis=(0, 2, 3), keepdims=1) +std = np.sqrt(data_squared_mean - data_mean_squared) +expected_output = (input_data - data_mean) / (std + 1e-9) + +expect(node, inputs=[input_data], outputs=[expected_output], name="test_mvn") +``` + +
+ + +### **MelWeightMatrix** + + Generate a MelWeightMatrix that can be used to re-weight a Tensor containing a linearly sampled frequency spectra (from DFT or STFT) into num_mel_bins frequency information based on the [lower_edge_hertz, upper_edge_hertz] range on the mel scale. + This function defines the mel scale in terms of a frequency in hertz according to the following formula: + + mel(f) = 2595 * log10(1 + f/700) + + In the returned matrix, all the triangles (filterbanks) have a peak value of 1.0. + + The returned MelWeightMatrix can be used to right-multiply a spectrogram S of shape [frames, num_spectrogram_bins] of linear scale spectrum values (e.g. STFT magnitudes) to generate a "mel spectrogram" M of shape [frames, num_mel_bins]. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
output_datatype : int (default is 1)
+
The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T3. The default value is 1 = FLOAT.
+
+ +#### Inputs + +
+
num_mel_bins (non-differentiable) : T1
+
The number of bands in the mel spectrum.
+
dft_length (non-differentiable) : T1
+
The size of the original DFT. The size of the original DFT is used to infer the size of the onesided DFT, which is understood to be floor(dft_length/2) + 1, i.e. the spectrogram only contains the nonredundant DFT bins.
+
sample_rate (non-differentiable) : T1
+
Samples per second of the input signal used to create the spectrogram. Used to figure out the frequencies corresponding to each spectrogram bin, which dictates how they are mapped into the mel scale.
+
lower_edge_hertz (non-differentiable) : T2
+
Lower bound on the frequencies to be included in the mel spectrum. This corresponds to the lower edge of the lowest triangular band.
+
upper_edge_hertz (non-differentiable) : T2
+
The desired top edge of the highest frequency band.
+
+ +#### Outputs + +
+
output (non-differentiable) : T3
+
The Mel Weight Matrix. The output has the shape: [floor(dft_length/2) + 1][num_mel_bins].
+
+ +#### Type Constraints + +
+
T1 : tensor(int32), tensor(int64)
+
Constrain to integer tensors.
+
T2 : tensor(float), tensor(float16), tensor(double), tensor(bfloat16)
+
Constrain to float tensors
+
T3 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain to any numerical types.
+
+ + +#### Examples + +
+melweightmatrix + +```python +node = onnx.helper.make_node( + "MelWeightMatrix", + inputs=[ + "num_mel_bins", + "dft_length", + "sample_rate", + "lower_edge_hertz", + "upper_edge_hertz", + ], + outputs=["output"], +) + +num_mel_bins = np.int32(8) +dft_length = np.int32(16) +sample_rate = np.int32(8192) +lower_edge_hertz = np.float32(0) +upper_edge_hertz = np.float32(8192 / 2) + +num_spectrogram_bins = dft_length // 2 + 1 +frequency_bins = np.arange(0, num_mel_bins + 2) + +low_frequency_mel = 2595 * np.log10(1 + lower_edge_hertz / 700) +high_frequency_mel = 2595 * np.log10(1 + upper_edge_hertz / 700) +mel_step = (high_frequency_mel - low_frequency_mel) / frequency_bins.shape[0] + +frequency_bins = frequency_bins * mel_step + low_frequency_mel +frequency_bins = 700 * (np.power(10, (frequency_bins / 2595)) - 1) +frequency_bins = ((dft_length + 1) * frequency_bins) // sample_rate +frequency_bins = frequency_bins.astype(int) + +output = np.zeros((num_spectrogram_bins, num_mel_bins)) +output.flags.writeable = True + +for i in range(num_mel_bins): + lower_frequency_value = frequency_bins[i] # left + center_frequency_point = frequency_bins[i + 1] # center + higher_frequency_point = frequency_bins[i + 2] # right + low_to_center = center_frequency_point - lower_frequency_value + if low_to_center == 0: + output[center_frequency_point, i] = 1 + else: + for j in range(lower_frequency_value, center_frequency_point + 1): + output[j, i] = float(j - lower_frequency_value) / float( + low_to_center + ) + center_to_high = higher_frequency_point - center_frequency_point + if center_to_high > 0: + for j in range(center_frequency_point, higher_frequency_point): + output[j, i] = float(higher_frequency_point - j) / float( + center_to_high + ) + +# Expected output +# 1.000000, 1.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 1.000000, 1.000000, 0.000000, 0.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 1.000000, 0.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 1.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 1.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 1.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, +output = output.astype(np.float32) +expect( + node, + inputs=[ + num_mel_bins, + dft_length, + sample_rate, + lower_edge_hertz, + upper_edge_hertz, + ], + outputs=[output], + name="test_melweightmatrix", +) +``` + +
+ + +### **Min** + + Element-wise min of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 8, 12 + +#### Inputs (1 - ∞) + +
+
data_0 (variadic, differentiable) : T
+
List of tensors for min.
+
+ +#### Outputs + +
+
min (differentiable) : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ + +#### Examples + +
+min + +```python +data_0 = np.array([3, 2, 1]).astype(np.float32) +data_1 = np.array([1, 4, 4]).astype(np.float32) +data_2 = np.array([2, 5, 0]).astype(np.float32) +result = np.array([1, 2, 0]).astype(np.float32) +node = onnx.helper.make_node( + "Min", + inputs=["data_0", "data_1", "data_2"], + outputs=["result"], +) +expect( + node, + inputs=[data_0, data_1, data_2], + outputs=[result], + name="test_min_example", +) + +node = onnx.helper.make_node( + "Min", + inputs=["data_0"], + outputs=["result"], +) +expect(node, inputs=[data_0], outputs=[data_0], name="test_min_one_input") + +result = np.minimum(data_0, data_1) +node = onnx.helper.make_node( + "Min", + inputs=["data_0", "data_1"], + outputs=["result"], +) +expect( + node, inputs=[data_0, data_1], outputs=[result], name="test_min_two_inputs" +) +``` + +
+ + +
+min_all_numeric_types + +```python +for op_dtype in all_numeric_dtypes: + data_0 = np.array([3, 2, 1]).astype(op_dtype) + data_1 = np.array([1, 4, 4]).astype(op_dtype) + result = np.array([1, 2, 1]).astype(op_dtype) + node = onnx.helper.make_node( + "Min", + inputs=["data_0", "data_1"], + outputs=["result"], + ) + expect( + node, + inputs=[data_0, data_1], + outputs=[result], + name=f"test_min_{np.dtype(op_dtype).name}", + ) +``` + +
+ + +### **Mish** + + Mish: A Self Regularized Non-Monotonic Neural Activation Function. + + Perform the linear unit element-wise on the input tensor X using formula: + + ``` + mish(x) = x * tanh(softplus(x)) = x * tanh(ln(1 + e^{x})) + ``` + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 18 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input X and output types to float tensors.
+
+ + +#### Examples + +
+mish + +```python +node = onnx.helper.make_node("Mish", inputs=["X"], outputs=["Y"]) + +input_data = np.linspace(-10, 10, 10000, dtype=np.float32) + +# Calculate expected output data +expected_output = input_data * np.tanh(np.log1p(np.exp(input_data))) + +expect(node, inputs=[input_data], outputs=[expected_output], name="test_mish") +``` + +
+ + +### **Mod** + + Performs an element-wise binary modulo operation. + The semantics and supported data types depend on the value of the `fmod` attribute which must be `0` (default), or `1`. + + If the `fmod` attribute is set to `0`, `T` is constrained to integer data types and the semantics follow that of the Python `%`-operator. + The sign of the result is that of the divisor. + + If `fmod` is set to `1`, the behavior of this operator follows that of the `fmod` function in C and `T` is constrained to floating point data types. + The result of this operator is the remainder of the division operation `x / y` where `x` and `y` are respective elements of `A` and `B`. The result is exactly the value `x - n * y`, where `n` is `x / y` with its fractional part truncated. + The returned value has the same sign as `x` (except if `x` is `-0`) and is less or equal to `|y|` in magnitude. + The following special cases apply when `fmod` is set to `1`: + - If `x` is `-0` and `y` is greater than zero, either `+0` or `-0` may be returned. + - If `x` is `±∞` and `y` is not `NaN`, `NaN` is returned. + - If `y` is `±0` and `x` is not `NaN`, `NaN` should be returned. + - If `y` is `±∞` and `x` is finite, `x` is returned. + - If either argument is `NaN`, `NaN` is returned. + + This operator supports **multidirectional (i.e., NumPy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 10 + +#### Attributes + +
+
fmod : int (default is 0)
+
Whether the operator should behave like fmod (default=0 meaning it will do integer mods); Set this to 1 to force fmod treatment
+
+ +#### Inputs + +
+
A (differentiable) : T
+
Dividend tensor
+
B (non-differentiable) : T
+
Divisor tensor
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Remainder tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to high-precision numeric tensors.
+
+ + +#### Examples + +
+mod_broadcast + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.arange(0, 30).reshape([3, 2, 5]).astype(np.int32) +y = np.array([7]).astype(np.int32) +z = np.mod(x, y) +# array([[[0, 1, 2, 3, 4], +# [5, 6, 0, 1, 2]], + +# [[3, 4, 5, 6, 0], +# [1, 2, 3, 4, 5]], + +# [[6, 0, 1, 2, 3], +# [4, 5, 6, 0, 1]]], dtype=int32) +expect(node, inputs=[x, y], outputs=[z], name="test_mod_broadcast") +``` + +
+ + +
+mod_int64_fmod + +```python +node = onnx.helper.make_node("Mod", inputs=["x", "y"], outputs=["z"], fmod=1) + +x = np.array([-4, 7, 5, 4, -7, 8]).astype(np.int64) +y = np.array([2, -3, 8, -2, 3, 5]).astype(np.int64) +z = np.fmod(x, y) # expected output [ 0, 1, 5, 0, -1, 3] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_int64_fmod") +``` + +
+ + +
+mod_mixed_sign_float16 + +```python +node = onnx.helper.make_node("Mod", inputs=["x", "y"], outputs=["z"], fmod=1) + +x = np.array([-4.3, 7.2, 5.0, 4.3, -7.2, 8.0]).astype(np.float16) +y = np.array([2.1, -3.4, 8.0, -2.1, 3.4, 5.0]).astype(np.float16) +z = np.fmod( + x, y +) # expected output [-0.10156, 0.3984 , 5. , 0.10156, -0.3984 , 3.] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_float16") +``` + +
+ + +
+mod_mixed_sign_float32 + +```python +node = onnx.helper.make_node("Mod", inputs=["x", "y"], outputs=["z"], fmod=1) + +x = np.array([-4.3, 7.2, 5.0, 4.3, -7.2, 8.0]).astype(np.float32) +y = np.array([2.1, -3.4, 8.0, -2.1, 3.4, 5.0]).astype(np.float32) +z = np.fmod( + x, y +) # expected output [-0.10000038, 0.39999962, 5. , 0.10000038, -0.39999962, 3.] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_float32") +``` + +
+ + +
+mod_mixed_sign_float64 + +```python +node = onnx.helper.make_node("Mod", inputs=["x", "y"], outputs=["z"], fmod=1) + +x = np.array([-4.3, 7.2, 5.0, 4.3, -7.2, 8.0]).astype(np.float64) +y = np.array([2.1, -3.4, 8.0, -2.1, 3.4, 5.0]).astype(np.float64) +z = np.fmod(x, y) # expected output [-0.1, 0.4, 5. , 0.1, -0.4, 3.] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_float64") +``` + +
+ + +
+mod_mixed_sign_int16 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([-4, 7, 5, 4, -7, 8]).astype(np.int16) +y = np.array([2, -3, 8, -2, 3, 5]).astype(np.int16) +z = np.mod(x, y) # expected output [ 0, -2, 5, 0, 2, 3] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_int16") +``` + +
+ + +
+mod_mixed_sign_int32 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([-4, 7, 5, 4, -7, 8]).astype(np.int32) +y = np.array([2, -3, 8, -2, 3, 5]).astype(np.int32) +z = np.mod(x, y) # expected output [ 0, -2, 5, 0, 2, 3] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_int32") +``` + +
+ + +
+mod_mixed_sign_int64 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([-4, 7, 5, 4, -7, 8]).astype(np.int64) +y = np.array([2, -3, 8, -2, 3, 5]).astype(np.int64) +z = np.mod(x, y) # expected output [ 0, -2, 5, 0, 2, 3] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_int64") +``` + +
+ + +
+mod_mixed_sign_int8 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([-4, 7, 5, 4, -7, 8]).astype(np.int8) +y = np.array([2, -3, 8, -2, 3, 5]).astype(np.int8) +z = np.mod(x, y) # expected output [ 0, -2, 5, 0, 2, 3] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_int8") +``` + +
+ + +
+mod_uint16 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([4, 7, 5]).astype(np.uint16) +y = np.array([2, 3, 8]).astype(np.uint16) +z = np.mod(x, y) # expected output [0, 1, 5] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_uint16") +``` + +
+ + +
+mod_uint32 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([4, 7, 5]).astype(np.uint32) +y = np.array([2, 3, 8]).astype(np.uint32) +z = np.mod(x, y) # expected output [0, 1, 5] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_uint32") +``` + +
+ + +
+mod_uint64 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([4, 7, 5]).astype(np.uint64) +y = np.array([2, 3, 8]).astype(np.uint64) +z = np.mod(x, y) # expected output [0, 1, 5] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_uint64") +``` + +
+ + +
+mod_uint8 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([4, 7, 5]).astype(np.uint8) +y = np.array([2, 3, 8]).astype(np.uint8) +z = np.mod(x, y) # expected output [0, 1, 5] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_uint8") +``` + +
+ + +### **Mul** + + Performs element-wise binary multiplication (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + + (Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 7, 13 + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ + +#### Examples + +
+mul + +```python +node = onnx.helper.make_node( + "Mul", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.float32) +z = x * y # expected output [4., 10., 18.] +expect(node, inputs=[x, y], outputs=[z], name="test_mul_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.int8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int8) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_int8") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.int16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int16) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_int16") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_uint8") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_uint16") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_uint32") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_uint64") +``` + +
+ + +
+mul_broadcast + +```python +node = onnx.helper.make_node( + "Mul", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_bcast") +``` + +
+ + +### **Multinomial** + + Generate a tensor of samples from a multinomial distribution according to the probabilities + of each of the possible outcomes. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 7 + +#### Attributes + +
+
dtype : int (default is 6)
+
(Optional) The data type for the elements of the output tensor, if not specified, we will use int32.
+
sample_size : int (default is 1)
+
Number of times to sample.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
Input tensor with shape [batch_size, class_size], where class_size is the number of all possible outcomes. Each value along the axis zero represents the unnormalized log-probability of each corresponding outcome in a batch.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor with shape [batch_size, sample_size], where sample_size is the number of times to sample. Each value along the axis zero represents the outcome of the corresponding sample in a batch.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
T2 : tensor(int32), tensor(int64)
+
Constrain output types to integral tensors.
+
+ + +### **Neg** + + Neg takes one input data (Tensor) and produces one output data + (Tensor) where each element flipped sign, y = -x, is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(int32), tensor(int8), tensor(int16), tensor(int64), tensor(float16), tensor(double), tensor(bfloat16)
+
Constrain input and output types to signed numeric tensors.
+
+ + +#### Examples + +
+neg + +```python +node = onnx.helper.make_node( + "Neg", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-4, 2]).astype(np.float32) +y = np.negative(x) # expected output [4., -2.], +expect(node, inputs=[x], outputs=[y], name="test_neg_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.negative(x) +expect(node, inputs=[x], outputs=[y], name="test_neg") +``` + +
+ + +### **NegativeLogLikelihoodLoss** + + A NegativeLogLikelihoodLoss operator computes (weighted) negative log likelihood loss. + Its "input" tensor has the shape of (N, C, d1, d2, ..., dk) where k >= 0. + The "input" tensor contains log-probabilities for input[n, :, d_1, d_2,..., d_k] being in a class of [0, C). + The operator's "target" input tensor has the shape of (N, d1, d2, ..., dk). It encodes class labels (one of C classes) + or it may contain a special value (indicated by an attribute ignore_index) for N x d1 x d2 x ... x dk samples. + The loss value for input[n, :, d_1, d_2,...d_k] being classified as class c = target[n][d_1][d_2]...[d_k] is computed as: + + ``` + loss[n][d_1][d_2]...[d_k] = -input[n][c][d_1][d_2]...[d_k]. + ``` + + When an optional "weight" is provided, the sample loss is calculated as: + + ``` + loss[n][d_1][d_2]...[d_k] = -input[n][c][d_1][d_2]...[d_k] * weight[c]. + ``` + + loss is zero for the case when target-value equals ignore_index. + + ``` + loss[n][d_1][d_2]...[d_k] = 0, when target[n][d_1][d_2]...[d_k] = ignore_index + ``` + + If "reduction" attribute is set to "none", the operator's output will be the above loss with shape (N, d1, d2, ..., dk). + If "reduction" attribute is set to "mean" (the default attribute value), the output loss is (weight) averaged: + + ``` + mean(loss), if "weight" is not provided, + ``` + + or if weight is provided, + + ``` + sum(loss) / sum(weight[target[n][d_1][d_2]...[d_k]]]), for all samples. + ``` + + If "reduction" attribute is set to "sum", the output is a scalar: `sum(loss)`. + + See also https://pytorch.org/docs/stable/nn.html#torch.nn.NLLLoss. + + Example 1: + + ``` + // negative log likelihood loss, "none" reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + + loss = np.zeros((N, d1)) + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] + + // print(loss) + // [[-3. -2.] + // [-0. -2.]] + ``` + + Example 2: + + ``` + // weighted negative log likelihood loss, sum reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + weight = [0.2, 0.3, 0.1] + loss = np.zeros((N, d1)) + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] * weight[c] + + loss = np.sum(loss) + // print(loss) + // -1.1 + ``` + + Example 3: + + ``` + // weighted negative log likelihood loss, mean reduction + N, C, d1 = 2, 3, 2 + input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]], + [[0.0, 1.0], [2.0, 2.0], [1.0, 2]]] + target = [[2, 1], [0, 2]] + weight = [0.2, 0.3, 0.1] + loss = np.zeros((N, d1)) + weight_total = 0 + for n in range(N): + for d_1 in range(d1): + c = target[n][d_1] + loss[n][d_1] = -input[n][c][d_1] * weight[c] + weight_total = weight_total + weight[c] + + loss = np.sum(loss) / weight_total + // print(loss) + // -1.57 + ``` + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 12, 13 + +#### Attributes + +
+
ignore_index : int
+
Specifies a target value that is ignored and does not contribute to the input gradient. It's an optional value.
+
reduction : string (default is mean)
+
Type of reduction to apply to loss: none, sum, mean (default). 'none': the output is the loss for each sample. 'sum': the output will be summed. 'mean': the sum of the output will be divided by the sum of applied weights.
+
+ +#### Inputs (2 - 3) + +
+
input (differentiable) : T
+
Input tensor of shape (N, C) or (N, C, d1, d2, ..., dk).
+
target (non-differentiable) : Tind
+
Target tensor of shape (N) or (N, d1, d2, ..., dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
+
weight (optional, non-differentiable) : T
+
Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
+
+ +#### Outputs + +
+
loss (differentiable) : T
+
The negative log likelihood loss
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input, weight, and output types to floating-point tensors.
+
Tind : tensor(int32), tensor(int64)
+
Constrain target to integer types
+
+ + +#### Examples + +
+input_shape_is_NC + +```python +reduction = "none" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C = 3, 5 +np.random.seed(0) +input = np.random.rand(N, C).astype(np.float32) +target = np.random.randint(0, high=C, size=(N,)).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NC", +) +``` + +
+ + +
+input_shape_is_NCd1 + +```python +reduction = "mean" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C, d1 = 3, 5, 2 +np.random.seed(0) +input = np.random.rand(N, C, d1).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, d1)).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1", +) +``` + +
+ + +
+input_shape_is_NCd1_ii + +```python +reduction = "mean" +ignore_index = np.int64(1) +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, d1 = 3, 5, 2 +np.random.seed(0) +input = np.random.rand(N, C, d1).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, d1)).astype(np.int64) +target[0][0] = np.int64(1) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1_ii", +) +``` + +
+ + +
+input_shape_is_NCd1_mean_weight_negative_ii + +```python +reduction = "mean" +ignore_index = np.int64(-1) + +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1 = 3, 5, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1)).astype(np.int64) +target[0][0] = -1 +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1_mean_weight_negative_ii", +) +``` + +
+ + +
+input_shape_is_NCd1_weight + +```python +reduction = "mean" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, +) + +N, C, d1 = 3, 5, 2 +np.random.seed(0) +input = np.random.rand(N, C, d1).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, d1)).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1_weight", +) +``` + +
+ + +
+input_shape_is_NCd1_weight_ii + +```python +reduction = "mean" +ignore_index = np.int64(1) +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, d1 = 3, 5, 2 +np.random.seed(0) +input = np.random.rand(N, C, d1).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, d1)).astype(np.int64) +target[0][0] = np.int64(1) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1_weight_ii", +) +``` + +
+ + +
+input_shape_is_NCd1d2 + +```python +reduction = "none" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2", +) +``` + +
+ + +
+input_shape_is_NCd1d2_no_weight_reduction_mean_ii + +```python +reduction = "mean" +ignore_index = np.int64(1) +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) +target[0][0][0] = np.int64(1) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_no_weight_reduction_mean_ii", +) +``` + +
+ + +
+input_shape_is_NCd1d2_reduction_mean + +```python +reduction = "mean" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_reduction_mean", +) +``` + +
+ + +
+input_shape_is_NCd1d2_reduction_sum + +```python +reduction = "sum" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_reduction_sum", +) +``` + +
+ + +
+input_shape_is_NCd1d2_with_weight + +```python +reduction = "none" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_with_weight", +) +``` + +
+ + +
+input_shape_is_NCd1d2_with_weight_reduction_mean + +```python +reduction = "mean" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_with_weight_reduction_mean", +) +``` + +
+ + +
+input_shape_is_NCd1d2_with_weight_reduction_sum + +```python +reduction = "sum" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_with_weight_reduction_sum", +) +``` + +
+ + +
+input_shape_is_NCd1d2_with_weight_reduction_sum_ii + +```python +reduction = "sum" +ignore_index = np.int64(0) +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) +target[0][0][0] = np.int64(0) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_with_weight_reduction_sum_ii", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3_none_no_weight_negative_ii + +```python +reduction = "none" +ignore_index = np.int64(-5) + +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1, dim2, dim3 = 3, 5, 6, 6, 5 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2, dim3).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2, dim3)).astype( + np.int64 +) +target[0][0][0][0] = -5 + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2d3_none_no_weight_negative_ii", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3_sum_weight_high_ii + +```python +reduction = "sum" +ignore_index = np.int64(10) + +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C = 3, 5 +np.random.seed(0) +input = np.random.rand(N, C).astype(np.float32) +target = np.random.randint(0, high=C, size=(N)).astype(np.int64) +target[0] = 10 +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2d3_sum_weight_high_ii", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3d4d5_mean_weight + +```python +reduction = "mean" + +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +target = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2d3d4d5_mean_weight", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3d4d5_none_no_weight + +```python +reduction = "none" + +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +target = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2d3d4d5_none_no_weight", +) +``` + +
+ + +### **NonMaxSuppression** + + Filter out boxes that have high intersection-over-union (IOU) overlap with previously selected boxes. + Bounding boxes with score less than score_threshold are removed. Bounding box format is indicated by attribute center_point_box. + Boxes are suppressed if their IOU with a previously selected box is strictly greater than iou_threshold (i.e., boxes with IOU exactly equal to the threshold are kept). + Note that this algorithm is agnostic to where the origin is in the coordinate system and more generally is invariant to + orthogonal transformations and translations of the coordinate system; thus translating or reflections of the coordinate system + result in the same boxes being selected by the algorithm. + The selected_indices output is a set of integers indexing into the input collection of bounding boxes representing the selected boxes. + The bounding box coordinates corresponding to the selected indices can then be obtained using the Gather or GatherND operation. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +Other versions of this operator: 10 + +#### Attributes + +
+
center_point_box : int (default is 0)
+
Integer indicate the format of the box data. The default is 0. 0 - the box data is supplied as [y1, x1, y2, x2] where (y1, x1) and (y2, x2) are the coordinates of any diagonal pair of box corners and the coordinates can be provided as normalized (i.e., lying in the interval [0, 1]) or absolute. Mostly used for TF models. 1 - the box data is supplied as [x_center, y_center, width, height]. Mostly used for Pytorch models.
+
+ +#### Inputs (2 - 5) + +
+
boxes : tensor(float)
+
An input tensor with shape [num_batches, spatial_dimension, 4]. The single box data format is indicated by center_point_box.
+
scores : tensor(float)
+
An input tensor with shape [num_batches, num_classes, spatial_dimension]
+
max_output_boxes_per_class (optional) : tensor(int64)
+
Integer representing the maximum number of boxes to be selected per batch per class. It is a scalar. Default to 0, which means no output.
+
iou_threshold (optional) : tensor(float)
+
Float representing the threshold for deciding whether boxes overlap too much with respect to IOU. Boxes with IoU strictly greater than this threshold are suppressed. It is scalar. Value range [0, 1]. Default to 0.
+
score_threshold (optional) : tensor(float)
+
Float representing the threshold for deciding when to remove boxes based on score. It is a scalar.
+
+ +#### Outputs + +
+
selected_indices : tensor(int64)
+
selected indices from the boxes tensor. [num_selected_indices, 3], the selected index format is [batch_index, class_index, box_index].
+
+ +#### Type Constraints + + + +#### Examples + +
+nonmaxsuppression_center_point_box_format + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], + center_point_box=1, +) +boxes = np.array( + [ + [ + [0.5, 0.5, 1.0, 1.0], + [0.5, 0.6, 1.0, 1.0], + [0.5, 0.4, 1.0, 1.0], + [0.5, 10.5, 1.0, 1.0], + [0.5, 10.6, 1.0, 1.0], + [0.5, 100.5, 1.0, 1.0], + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 3], [0, 0, 0], [0, 0, 5]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_center_point_box_format", +) +``` + +
+ + +
+nonmaxsuppression_flipped_coordinates + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [1.0, 1.0, 0.0, 0.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, 0.9, 1.0, -0.1], + [0.0, 10.0, 1.0, 11.0], + [1.0, 10.1, 0.0, 11.1], + [1.0, 101.0, 0.0, 100.0], + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 3], [0, 0, 0], [0, 0, 5]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_flipped_coordinates", +) +``` + +
+ + +
+nonmaxsuppression_identical_boxes + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + ] + ] +).astype(np.float32) +scores = np.array( + [[[0.9, 0.9, 0.9, 0.9, 0.9, 0.9, 0.9, 0.9, 0.9, 0.9]]] +).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 0]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_identical_boxes", +) +``` + +
+ + +
+nonmaxsuppression_iou_threshold_boundary + +```python +"""Test boundary condition where IoU exactly equals threshold. + +This test verifies that the comparison is strict (>), not inclusive (>=). +When IoU exactly equals the threshold, boxes should be KEPT, not suppressed. +This follows PyTorch's NMS implementation. +""" +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +# Two boxes with 50% overlap in each dimension +# box1=[0,0,1,1], box2=[0.5,0.5,1.5,1.5] +# Intersection area = 0.5 * 0.5 = 0.25 +# Union area = 1.0 + 1.0 - 0.25 = 1.75 +# IoU = 0.25 / 1.75 (exact value computed below as float32) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], # box 0 + [0.5, 0.5, 1.5, 1.5], # box 1 - overlaps box 0 + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.8]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +# Compute the exact IoU value and use it as threshold +# This ensures the threshold exactly equals the IoU +exact_iou = np.float32(0.25 / 1.75) +iou_threshold = np.array([exact_iou]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +# Both boxes should be selected because IoU == threshold (not > threshold) +selected_indices = np.array([[0, 0, 0], [0, 0, 1]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_iou_threshold_boundary", +) +``` + +
+ + +
+nonmaxsuppression_limit_output_size + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]]).astype(np.float32) +max_output_boxes_per_class = np.array([2]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 3], [0, 0, 0]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_limit_output_size", +) +``` + +
+ + +
+nonmaxsuppression_single_box + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array([[[0.0, 0.0, 1.0, 1.0]]]).astype(np.float32) +scores = np.array([[[0.9]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 0]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_single_box", +) +``` + +
+ + +
+nonmaxsuppression_suppress_by_IOU + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 3], [0, 0, 0], [0, 0, 5]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_suppress_by_IOU", +) +``` + +
+ + +
+nonmaxsuppression_suppress_by_IOU_and_scores + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.4]).astype(np.float32) +selected_indices = np.array([[0, 0, 3], [0, 0, 0]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_suppress_by_IOU_and_scores", +) +``` + +
+ + +
+nonmaxsuppression_two_batches + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ], + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ], + ] +).astype(np.float32) +scores = np.array( + [[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]], [[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]] +).astype(np.float32) +max_output_boxes_per_class = np.array([2]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array( + [[0, 0, 3], [0, 0, 0], [1, 0, 3], [1, 0, 0]] +).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_two_batches", +) +``` + +
+ + +
+nonmaxsuppression_two_classes + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ] + ] +).astype(np.float32) +scores = np.array( + [[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3], [0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]] +).astype(np.float32) +max_output_boxes_per_class = np.array([2]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array( + [[0, 0, 3], [0, 0, 0], [0, 1, 3], [0, 1, 0]] +).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_two_classes", +) +``` + +
+ + +### **NonZero** + + Returns the indices of the elements that are non-zero + (in row-major order - by dimension). + NonZero behaves similar to numpy.nonzero: + https://docs.scipy.org/doc/numpy/reference/generated/numpy.nonzero.html, + but for scalar input, NonZero produces output shape (0, N) instead of (1, N), which is different from Numpy's behavior. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Inputs + +
+
X (non-differentiable) : T
+
input
+
+ +#### Outputs + +
+
Y (non-differentiable) : tensor(int64)
+
output
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to all tensor types.
+
+ + +#### Examples + +
+nonzero + +```python +node = onnx.helper.make_node( + "NonZero", + inputs=["condition"], + outputs=["result"], +) + +condition = np.array([[1, 0], [1, 1]], dtype=bool) +result = np.array( + np.nonzero(condition), dtype=np.int64 +) # expected output [[0, 1, 1], [0, 0, 1]] +expect(node, inputs=[condition], outputs=[result], name="test_nonzero_example") +``` + +
+ + +### **Not** + + Returns the negation of the input tensor element-wise. + +#### Version + +This version of the operator has been available since version 1 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input/output to boolean tensors.
+
+ + +#### Examples + +
+not + +```python +node = onnx.helper.make_node( + "Not", + inputs=["x"], + outputs=["not"], +) + +# 2d +x = (np.random.randn(3, 4) > 0).astype(bool) +expect(node, inputs=[x], outputs=[np.logical_not(x)], name="test_not_2d") + +# 3d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +expect(node, inputs=[x], outputs=[np.logical_not(x)], name="test_not_3d") + +# 4d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +expect(node, inputs=[x], outputs=[np.logical_not(x)], name="test_not_4d") +``` + +
+ + +### **OneHot** + + Produces a one-hot tensor based on inputs. + The locations represented by the index values in the 'indices' input tensor will have 'on_value' + and the other locations will have 'off_value' in the output tensor, where 'on_value' and 'off_value' + are specified as part of required input argument 'values', which is a two-element tensor of format + [off_value, on_value]. The rank of the output tensor will be one greater than the rank of the + input tensor. The additional dimension is for one-hot representation. The additional dimension will + be inserted at the position specified by 'axis'. If 'axis' is not specified then then additional + dimension will be inserted as the innermost dimension, i.e. axis=-1. The size of the additional + dimension is specified by required scalar input 'depth'. The type of the output tensor is the same + as the type of the 'values' input. Any entries in the 'indices' input tensor with values outside + the range [-depth, depth-1] will result in one-hot representation with all 'off_value' values in the + output tensor. + + when axis = 0: + output[input[i, j, k], i, j, k] = 1 for all i, j, k and 0 otherwise. + + when axis = -1: + output[i, j, k, input[i, j, k]] = 1 for all i, j, k and 0 otherwise. + + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Attributes + +
+
axis : int (default is -1)
+
(Optional) Axis along which one-hot representation in added. Default: axis=-1. axis=-1 means that the additional dimension will be inserted as the innermost/last dimension in the output tensor. Negative value means counting dimensions from the back. Accepted range is [-r-1, r] where r = rank(indices).
+
+ +#### Inputs + +
+
indices (non-differentiable) : T1
+
Input tensor containing indices. Any entries in the 'indices' input tensor with values outside the range [-depth, depth-1] will result in one-hot representation with all 'off_value' values in the output tensor.In case 'indices' is of non-integer type, the values will be casted to int64 before use.
+
depth (non-differentiable) : T2
+
Scalar or Rank 1 tensor containing exactly one element, specifying the number of classes in one-hot tensor. This is also the size of the one-hot dimension (specified by 'axis' attribute) added on in the output tensor. The values in the 'indices' input tensor are expected to be in the range [-depth, depth-1]. In case 'depth' is of non-integer type, it will be casted to int64 before use.
+
values (non-differentiable) : T3
+
Rank 1 tensor containing exactly two elements, in the format [off_value, on_value], where 'on_value' is the value used for filling locations specified in 'indices' input tensor, and 'off_value' is the value used for filling locations other than those specified in 'indices' input tensor.
+
+ +#### Outputs + +
+
output (non-differentiable) : T3
+
Tensor of rank one greater than input tensor 'indices', i.e. rank(output) = rank(indices) + 1. The data type for the elements of the output tensor is the same as the type of input 'values' is used.
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input to only numeric types.
+
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input to only numeric types.
+
T3 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type.
+
+ + +#### Examples + +
+with_axis + +```python +axisValue = 1 +on_value = 3 +off_value = 1 +output_type = np.float32 +node = onnx.helper.make_node( + "OneHot", + inputs=["indices", "depth", "values"], + outputs=["y"], + axis=axisValue, +) +indices = np.array([[1, 9], [2, 4]], dtype=np.float32) +depth = np.float32(10) +values = np.array([off_value, on_value], dtype=output_type) +y = one_hot(indices, depth, axis=axisValue, dtype=output_type) +y = y * (on_value - off_value) + off_value +expect( + node, + inputs=[indices, depth, values], + outputs=[y], + name="test_onehot_with_axis", +) +``` + +
+ + +
+with_negative_axis + +```python +axisValue = -2 +on_value = 3 +off_value = 1 +output_type = np.float32 +node = onnx.helper.make_node( + "OneHot", + inputs=["indices", "depth", "values"], + outputs=["y"], + axis=axisValue, +) +indices = np.array([[1, 9], [2, 4]], dtype=np.float32) +depth = np.float32(10) +values = np.array([off_value, on_value], dtype=output_type) +y = one_hot(indices, depth, axis=axisValue, dtype=output_type) +y = y * (on_value - off_value) + off_value +expect( + node, + inputs=[indices, depth, values], + outputs=[y], + name="test_onehot_with_negative_axis", +) +``` + +
+ + +
+with_negative_indices + +```python +axisValue = 1 +on_value = 3 +off_value = 1 +output_type = np.float32 +node = onnx.helper.make_node( + "OneHot", + inputs=["indices", "depth", "values"], + outputs=["y"], + axis=axisValue, +) +indices = np.array([0, -7, -8], dtype=np.int64) + +# print(y) +# [[3. 1. 1. 1. 1. 1. 1. 1. 1. 1.] +# [1. 1. 1. 3. 1. 1. 1. 1. 1. 1.] +# [1. 1. 3. 1. 1. 1. 1. 1. 1. 1.]] + +depth = np.float32(10) +values = np.array([off_value, on_value], dtype=output_type) +y = one_hot(indices, depth, axis=axisValue, dtype=output_type) +y = y * (on_value - off_value) + off_value +expect( + node, + inputs=[indices, depth, values], + outputs=[y], + name="test_onehot_negative_indices", +) +``` + +
+ + +
+with_out_of_range_indices + +```python +axisValue = 1 +on_value = 3 +off_value = 1 +output_type = np.float32 +node = onnx.helper.make_node( + "OneHot", + inputs=["indices", "depth", "values"], + outputs=["y"], + axis=axisValue, +) +# Indices outside [-depth, depth-1] map to an all-off_value row. +indices = np.array([5, -6, -1], dtype=np.int64) + +# print(y) +# [[1. 1. 1. 1. 1.] +# [1. 1. 1. 1. 1.] +# [1. 1. 1. 1. 3.]] + +depth = np.float32(5) +values = np.array([off_value, on_value], dtype=output_type) +y = one_hot(indices, depth, axis=axisValue, dtype=output_type) +y = y * (on_value - off_value) + off_value +expect( + node, + inputs=[indices, depth, values], + outputs=[y], + name="test_onehot_out_of_range_indices", +) +``` + +
+ + +
+without_axis + +```python +on_value = 5 +off_value = 2 +output_type = np.int32 +node = onnx.helper.make_node( + "OneHot", inputs=["indices", "depth", "values"], outputs=["y"] +) +indices = np.array([0, 7, 8], dtype=np.int64) +depth = np.float32(12) +values = np.array([off_value, on_value], dtype=output_type) +y = one_hot(indices, depth, dtype=output_type) +y = y * (on_value - off_value) + off_value +expect( + node, + inputs=[indices, depth, values], + outputs=[y], + name="test_onehot_without_axis", +) +``` + +
+ + +### **Optional** + + Constructs an optional-type value containing either an empty optional of a certain type specified by the attribute, + or a non-empty value containing the input element. + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +#### Attributes + +
+
type : type_proto
+
Type of the element in the optional output
+
+ +#### Inputs (0 - 1) + +
+
input (optional) : V
+
The input element.
+
+ +#### Outputs + +
+
output : O
+
The optional output enclosing the input element.
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input type to all tensor and sequence types.
+
O : optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128))
+
Constrain output type to all optional tensor or optional sequence types.
+
+ + +### **OptionalGetElement** + + If the input is a tensor or sequence type, it returns the input. + If the input is an optional type, it outputs the element in the input. + It is an error if the input is an empty optional-type (i.e. does not have an element) and the behavior is undefined in this case. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 15 + +#### Inputs + +
+
input : O
+
The optional input.
+
+ +#### Outputs + +
+
output : V
+
Output element in the optional input.
+
+ +#### Type Constraints + +
+
O : optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input type to optional tensor and optional sequence types.
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain output type to all tensor or sequence types.
+
+ + +### **OptionalHasElement** + + Returns true if (1) the input is an optional-type and contains an element, + or, (2) the input is a tensor or sequence type. + If the input is not provided or is an empty optional-type, this op returns false. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 15 + +#### Inputs (0 - 1) + +
+
input (optional) : O
+
The optional input.
+
+ +#### Outputs + +
+
output : B
+
A scalar boolean tensor. If true, it indicates that optional-type input contains an element. Otherwise, it is empty.
+
+ +#### Type Constraints + +
+
O : optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input type to optional tensor and optional sequence types.
+
B : tensor(bool)
+
Constrain output to a boolean tensor.
+
+ + +#### Examples + +
+empty + +```python +optional = None + +tensor_type_proto = onnx.helper.make_tensor_type_proto( + elem_type=onnx.TensorProto.INT32, shape=[] +) +optional_type_proto = onnx.helper.make_optional_type_proto(tensor_type_proto) + +# OptionalHasElement takes a tensor or optional as input +for input_type_proto in [tensor_type_proto, optional_type_proto]: + input_name_options = { + "empty": "optional_input", + "empty_no_input_name": "", + "empty_no_input": None, + } + for test_name_surfix, input_name in input_name_options.items(): + if input_type_proto == tensor_type_proto and input_name: + # the input tensor cannot be empty if input name is provided. + continue + node = onnx.helper.make_node( + "OptionalHasElement", + inputs=[] if input_name is None else [input_name], + outputs=["output"], + ) + output = optional_has_element_reference_implementation(optional) + test_name = ( + "test_optional_has_element_" + + test_name_surfix + + ( + "_optional_input" + if input_type_proto == optional_type_proto + else "_tensor_input" + ) + ) + expect( + node, + inputs=[optional] if input_name else [], + outputs=[output], + input_type_protos=[input_type_proto] if input_name else [], + name=test_name, + ) +``` + +
+ + +
+get_element_sequence + +```python +optional = [np.array([1, 2, 3, 4]).astype(np.int32)] +tensor_type_proto = onnx.helper.make_tensor_type_proto( + elem_type=onnx.TensorProto.INT32, + shape=[ + 4, + ], +) +seq_type_proto = onnx.helper.make_sequence_type_proto(tensor_type_proto) +optional_type_proto = onnx.helper.make_optional_type_proto(seq_type_proto) + +node = onnx.helper.make_node( + "OptionalGetElement", inputs=["optional_input"], outputs=["output"] +) +output = optional_get_element_reference_implementation(optional) +expect( + node, + inputs=[optional], + outputs=[output], + input_type_protos=[optional_type_proto], + name="test_optional_get_element_optional_sequence", +) +expect( + node, + inputs=[optional], + outputs=[output], + input_type_protos=[seq_type_proto], + name="test_optional_get_element_sequence", +) +``` + +
+ + +
+get_element_tensor + +```python +optional = np.array([1, 2, 3, 4]).astype(np.float32) +tensor_type_proto = onnx.helper.make_tensor_type_proto( + elem_type=onnx.TensorProto.FLOAT, + shape=[ + 4, + ], +) +optional_type_proto = onnx.helper.make_optional_type_proto(tensor_type_proto) + +node = onnx.helper.make_node( + "OptionalGetElement", inputs=["optional_input"], outputs=["output"] +) +output = optional_get_element_reference_implementation(optional) +expect( + node, + inputs=[optional], + outputs=[output], + input_type_protos=[optional_type_proto], + name="test_optional_get_element_optional_tensor", +) +expect( + node, + inputs=[optional], + outputs=[output], + input_type_protos=[tensor_type_proto], + name="test_optional_get_element_tensor", +) +``` + +
+ + +
+optionalhaselement + +```python +optional = np.array([1, 2, 3, 4]).astype(np.float32) +tensor_type_proto = onnx.helper.make_tensor_type_proto( + elem_type=onnx.TensorProto.FLOAT, + shape=[ + 4, + ], +) +optional_type_proto = onnx.helper.make_optional_type_proto(tensor_type_proto) + +# OptionalHasElement takes a tensor or optional as input +for input_type_protos in [tensor_type_proto, optional_type_proto]: + node = onnx.helper.make_node( + "OptionalHasElement", inputs=["optional_input"], outputs=["output"] + ) + output = optional_has_element_reference_implementation(optional) + test_name = "test_optional_has_element_" + ( + "optional_input" + if input_type_protos == optional_type_proto + else "tensor_input" + ) + expect( + node, + inputs=[optional], + outputs=[output], + input_type_protos=[optional_type_proto], + name=test_name, + ) +``` + +
+ + +### **Or** + + Returns the tensor resulted from performing the `or` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input to boolean tensor.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ + +#### Examples + +
+or + +```python +node = onnx.helper.make_node( + "Or", + inputs=["x", "y"], + outputs=["or"], +) + +# 2d +x = (np.random.randn(3, 4) > 0).astype(bool) +y = (np.random.randn(3, 4) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or2d") + +# 3d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(3, 4, 5) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or3d") + +# 4d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or4d") +``` + +
+ + +
+or_broadcast + +```python +node = onnx.helper.make_node( + "Or", + inputs=["x", "y"], + outputs=["or"], +) + +# 3d vs 1d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(5) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or_bcast3v1d") + +# 3d vs 2d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(4, 5) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or_bcast3v2d") + +# 4d vs 2d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(5, 6) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or_bcast4v2d") + +# 4d vs 3d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(4, 5, 6) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or_bcast4v3d") + +# 4d vs 4d +x = (np.random.randn(1, 4, 1, 6) > 0).astype(bool) +y = (np.random.randn(3, 1, 5, 6) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or_bcast4v4d") +``` + +
+ + +### **PRelu** + + PRelu takes input data (Tensor) and slope tensor as input, and produces one + output data (Tensor) where the function `f(x) = slope * x for x < 0`, + `f(x) = x for x >= 0`., is applied to the data tensor elementwise. + This operator supports **unidirectional broadcasting** (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 7, 9 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
slope (differentiable) : T
+
Slope tensor. The shape of slope can be smaller than first input X; if so, its shape must be unidirectional broadcastable to X
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor (same size as X)
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64)
+
Constrain input and output types to float/int tensors.
+
+ + +#### Examples + +
+prelu + +```python +node = onnx.helper.make_node( + "PRelu", + inputs=["x", "slope"], + outputs=["y"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +slope = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * slope + +expect(node, inputs=[x, slope], outputs=[y], name="test_prelu_example") +``` + +
+ + +
+prelu_broadcast + +```python +node = onnx.helper.make_node( + "PRelu", + inputs=["x", "slope"], + outputs=["y"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +slope = np.random.randn(5).astype(np.float32) +y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * slope + +expect(node, inputs=[x, slope], outputs=[y], name="test_prelu_broadcast") +``` + +
+ + +### **Pad** + + Given a tensor containing the data to be padded (`data`), a tensor containing the number of start and end pad values for axis (`pads`), (optionally) a `mode`, and (optionally) `constant_value`, + a padded tensor (`output`) is generated. + + The four supported `modes` are (similar to corresponding modes supported by `numpy.pad`): + + 1) `constant`(default) - pads with a given constant value as specified by `constant_value` (which defaults to 0, empty string, or False) + + 2) `reflect` - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis + + 3) `edge` - pads with the edge values of array + + 4) `wrap` - wrap-around padding as if the data tensor forms a torus + + + Example 1 (`constant` mode): + + Insert 0 pads to the beginning of the second dimension. + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'constant' + + constant_value = 0.0 + + output = [ + [0.0, 0.0, 1.0, 1.2], + [0.0, 0.0, 2.3, 3.4], + [0.0, 0.0, 4.5, 5.7], + ] + ``` + + Example 2 (`reflect` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'reflect' + + output = [ + [1.0, 1.2, 1.0, 1.2], + [2.3, 3.4, 2.3, 3.4], + [4.5, 5.7, 4.5, 5.7], + ] + ``` + + Example 3 (`edge` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [0, 2, 0, 0] + + mode = 'edge' + + output = [ + [1.0, 1.0, 1.0, 1.2], + [2.3, 2.3, 2.3, 3.4], + [4.5, 4.5, 4.5, 5.7], + ] + ``` + + Example 4 (`wrap` mode): + + ``` + data = [ + [1.0, 1.2], + [2.3, 3.4], + [4.5, 5.7], + ] + + pads = [2, 1, 1, 1] + + mode = 'wrap' + + output = [ + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + [3.4, 2.3, 3.4, 2.3], + [5.7, 4.5, 5.7, 4.5], + [1.2, 1.0, 1.2, 1.0], + ] + ``` + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 2, 11, 13, 18, 19, 21, 23, 24 + +#### Attributes + +
+
mode : string (default is constant)
+
Supported modes: `constant`(default), `reflect`, `edge`, `wrap`
+
+ +#### Inputs (2 - 4) + +
+
data (differentiable) : T
+
Input tensor.
+
pads (non-differentiable) : tensor(int64)
+
Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. `pads` should be a 1D tensor of shape [2 * num_axes] where `num_axes` refers to the number of elements in the `axes` input or the input rank if `axes` are not provided explicitly. `pads` format should be: [x1_begin, x2_begin, ..., x1_end, x2_end,...], where xi_begin is the number of pad values added at the beginning of axis `axes[i]` and xi_end, the number of pad values added at the end of axis `axes[i]`.
+
constant_value (optional, non-differentiable) : T
+
(Optional) A scalar value to be used if the mode chosen is `constant` (by default it is 0, empty string or False).
+
axes (optional, non-differentiable) : Tind
+
1-D tensor of axes that `pads` apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated. If not provided, all axes are assumed (`[0, 1, ..., input_rank-1]`).
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor after padding.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types up to IRv13.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ + +#### Examples + +
+constant_pad + +```python +node = onnx.helper.make_node( + "Pad", inputs=["x", "pads", "value"], outputs=["y"], mode="constant" +) +x = np.random.randn(1, 3, 4, 5).astype(np.float32) +pads = np.array([0, 0, 1, 3, 0, 0, 2, 4]).astype( + np.int64 +) # pad order [x1_begin, x2_begin, ..., x1_end, x2_end, ...] +value = np.float32(1.2) +y = pad_impl(x, pads, "constant", 1.2) + +expect(node, inputs=[x, pads, value], outputs=[y], name="test_constant_pad") +``` + +
+ + +
+constant_pad_axes + +```python +node = onnx.helper.make_node( + "Pad", inputs=["x", "pads", "value", "axes"], outputs=["y"], mode="constant" +) +x = np.random.randn(1, 3, 4, 5).astype(np.float32) +pads = np.array([0, 3, 0, 4]).astype( + np.int64 +) # pad order [x1_begin, x2_begin, ..., x1_end, x2_end, ...] +value = np.float32(1.2) +axes = np.array([1, 3], dtype=np.int64) +y = pad_impl( + x, + pads, + "constant", + 1.2, + [1, 3], +) + +expect( + node, + inputs=[x, pads, value, axes], + outputs=[y], + name="test_constant_pad_axes", +) +``` + +
+ + +
+constant_pad_negative_axes + +```python +node = onnx.helper.make_node( + "Pad", inputs=["x", "pads", "value", "axes"], outputs=["y"], mode="constant" +) +x = np.random.randn(1, 3, 4, 5).astype(np.float32) +pads = np.array([0, 3, 0, 4]).astype( + np.int64 +) # pad order [x1_begin, x2_begin, ..., x1_end, x2_end, ...] +value = np.float32(1.2) +axes = np.array([-3, -1], dtype=np.int64) +y = pad_impl( + x, + pads, + "constant", + 1.2, + [-3, -1], +) + +expect( + node, + inputs=[x, pads, value, axes], + outputs=[y], + name="test_constant_pad_negative_axes", +) +``` + +
+ + +
+reflection_edge_and_wrap_pad + +```python +for mode in ("edge", "reflect", "wrap"): + node = onnx.helper.make_node( + "Pad", inputs=["x", "pads"], outputs=["y"], mode=mode + ) + x = np.random.randn(1, 3, 4, 5).astype(np.int32) + pads = np.array([0, 0, 1, 1, 0, 0, 1, 1]).astype( + np.int64 + ) # pad order [x1_begin, x2_begin, ..., x1_end, x2_end, ...] + y = pad_impl(x, pads, mode) + + expect(node, inputs=[x, pads], outputs=[y], name=f"test_{mode}_pad") +``` + +
+ + +### **Pow** + + Pow takes input data (Tensor) and exponent Tensor, and + produces one output data (Tensor) where the function `f(x) = x^exponent`, + is applied to the data tensor elementwise. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 15 of the default ONNX operator set. + +Other versions of this operator: 1, 7, 12, 13 + +#### Inputs + +
+
X (differentiable) : T
+
First operand, base of the exponent.
+
Y (differentiable) : T1
+
Second operand, power of the exponent.
+
+ +#### Outputs + +
+
Z (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input X and output types to float/int tensors.
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input Y types to float/int tensors.
+
+ + +#### Examples + +
+pow + +```python +node = onnx.helper.make_node( + "Pow", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.float32) +z = pow(x, y) # expected output [1., 32., 729.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_example") + +x = np.arange(60).reshape(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = pow(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_pow") +``` + +
+ + +
+pow_broadcast + +```python +node = onnx.helper.make_node( + "Pow", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array(2).astype(np.float32) +z = pow(x, y) # expected output [1., 4., 9.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_bcast_scalar") + +node = onnx.helper.make_node( + "Pow", + inputs=["x", "y"], + outputs=["z"], +) +x = np.array([[1, 2, 3], [4, 5, 6]]).astype(np.float32) +y = np.array([1, 2, 3]).astype(np.float32) +# expected output [[1, 4, 27], [4, 25, 216]] +z = pow(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_pow_bcast_array") +``` + +
+ + +
+types + +```python +node = onnx.helper.make_node( + "Pow", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.int64) +z = pow(x, y) # expected output [1., 32., 729.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_float32_int64") + +x = np.array([1, 2, 3]).astype(np.int64) +y = np.array([4, 5, 6]).astype(np.float32) +z = pow(x, y) # expected output [1, 32, 729] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_int64_float32") + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.int32) +z = pow(x, y) # expected output [1., 32., 729.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_float32_int32") + +x = np.array([1, 2, 3]).astype(np.int32) +y = np.array([4, 5, 6]).astype(np.float32) +z = pow(x, y) # expected output [1, 32, 729] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_int32_float32") + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.uint64) +z = pow(x, y) # expected output [1., 32., 729.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_float32_uint64") + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.uint32) +z = pow(x, y) # expected output [1., 32., 729.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_float32_uint32") + +x = np.array([1, 2, 3]).astype(np.int64) +y = np.array([4, 5, 6]).astype(np.int64) +z = pow(x, y) # expected output [1, 32, 729] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_int64_int64") + +x = np.array([1, 2, 3]).astype(np.int32) +y = np.array([4, 5, 6]).astype(np.int32) +z = pow(x, y) # expected output [1, 32, 729] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_int32_int32") +``` + +
+ + +### **QLinearConv** + + The convolution operator consumes a quantized input tensor, its scale and zero point, + a quantized filter, its scale and zero point, and output's scale and zero point, + and computes the quantized output. Each scale and zero-point pair must have same shape. + It means they must be either scalars (per tensor) or 1-D tensors (per output channel). + Each input or output and its related zero point must have same type. + When bias is present it must be quantized using scale = input scale * weight scale and + zero point as 0. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
auto_pad : string (default is NOTSET)
+
auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that `output_shape[i] = ceil(input_shape[i] / strides[i])` for each axis `i`. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
+
dilations : list of ints
+
dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each spatial axis.
+
group : int (default is 1)
+
number of groups input channels and output channels are divided into. default is 1.
+
kernel_shape : list of ints
+
The shape of the convolution kernel. If not present, should be inferred from input 'w'.
+
pads : list of ints
+
Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0.The value represent the number of pixels added to the beginning and end part of the corresponding axis.`pads` format should be as follow [x1_begin, x2_begin...x1_end, x2_end,...], where xi_begin the number ofpixels added at the beginning of axis `i` and xi_end, the number of pixels added at the end of axis `i`.This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaultsto 0 along start and end of each spatial axis.
+
strides : list of ints
+
Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
+
+ +#### Inputs (8 - 9) + +
+
x : T1
+
Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
+
x_scale : tensor(float)
+
Scale tensor for input 'x'. It's a scalar, which means a per-tensor/layer quantization.
+
x_zero_point : T1
+
Zero point tensor for input 'x'. It's a scalar, which means a per-tensor/layer quantization.
+
w : T2
+
The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
+
w_scale : tensor(float)
+
Scale tensor for input 'w'. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it's a 1-D tensor, its number of elements should be equal to the number of output channels (M).
+
w_zero_point : T2
+
Zero point tensor for input 'w'. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it's a 1-D tensor, its number of elements should be equal to the number of output channels (M).
+
y_scale : tensor(float)
+
Scale tensor for output 'y'. It's a scalar, which means a per-tensor/layer quantization.
+
y_zero_point : T3
+
Zero point tensor for output 'y'. It's a scalar, which means a per-tensor/layer quantization.
+
B (optional) : T4
+
Optional 1D bias to be added to the convolution, has size of M. Bias must be quantized using scale = x_scale * w_scale and zero_point = 0
+
+ +#### Outputs + +
+
y : T3
+
Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
+
+ +#### Type Constraints + +
+
T1 : tensor(int8), tensor(uint8)
+
Constrain input type to 8-bit integer tensor.
+
T2 : tensor(int8), tensor(uint8)
+
Constrain filter type to 8-bit integer tensor.
+
T3 : tensor(int8), tensor(uint8)
+
Constrain output type to 8-bit integer tensor.
+
T4 : tensor(int32)
+
Constrain bias type to 32-bit integer tensor.
+
+ + +#### Examples + +
+qlinearconv + +```python +node = onnx.helper.make_node( + "QLinearConv", + inputs=[ + "x", + "x_scale", + "x_zero_point", + "w", + "w_scale", + "w_zero_point", + "y_scale", + "y_zero_point", + ], + outputs=["y"], +) + +x = np.array( + [ + [255, 174, 162, 25, 203, 168, 58], + [15, 59, 237, 95, 129, 0, 64], + [56, 242, 153, 221, 168, 12, 166], + [232, 178, 186, 195, 237, 162, 237], + [188, 39, 124, 77, 80, 102, 43], + [127, 230, 21, 83, 41, 40, 134], + [255, 154, 92, 141, 42, 148, 247], + ], + dtype=np.uint8, +).reshape((1, 1, 7, 7)) + +x_scale = np.float32(0.00369204697) +x_zero_point = np.uint8(132) + +w = np.array([0], dtype=np.uint8).reshape((1, 1, 1, 1)) + +w_scale = np.array([0.00172794575], dtype=np.float32) +w_zero_point = np.array([255], dtype=np.uint8) + +y_scale = np.float32(0.00162681262) +y_zero_point = np.uint8(123) + +output = np.array( + [ + [0, 81, 93, 230, 52, 87, 197], + [240, 196, 18, 160, 126, 255, 191], + [199, 13, 102, 34, 87, 243, 89], + [23, 77, 69, 60, 18, 93, 18], + [67, 216, 131, 178, 175, 153, 212], + [128, 25, 234, 172, 214, 215, 121], + [0, 101, 163, 114, 213, 107, 8], + ], + dtype=np.uint8, +).reshape((1, 1, 7, 7)) + +expect( + node, + inputs=[ + x, + x_scale, + x_zero_point, + w, + w_scale, + w_zero_point, + y_scale, + y_zero_point, + ], + outputs=[output], + name="test_qlinearconv", +) +``` + +
+ + +### **QLinearMatMul** + + Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html). + It consumes two quantized input tensors, their scales and zero points, scale and zero point of output, + and computes the quantized output. The quantization formula is y = saturate((x / y_scale) + y_zero_point). + For (x / y_scale), it is rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details. + Scale and zero point must have same shape. They must be either scalar (per tensor) or N-D tensor + (per row for 'a' and per column for 'b'). Scalar refers to per tensor quantization whereas N-D refers to per row + or per column quantization. If the input is 2D of shape [M, K] then zero point and scale tensor may be + an M element vector [v_1, v_2, ..., v_M] for per row quantization and K element vector of shape [v_1, v_2, ..., v_K] + for per column quantization. If the input is N-D tensor with shape [D1, D2, M, K] then zero point and scale tensor may + have shape [D1, D2, M, 1] for per row quantization and shape [D1, D2, 1, K] for per column quantization. + Production must never overflow, and accumulation may overflow if and only if in 32 bits. + +#### Version + +This version of the operator has been available since version 21 of the default ONNX operator set. + +Other versions of this operator: 10 + +#### Inputs + +
+
a (non-differentiable) : T1
+
N-dimensional quantized matrix a
+
a_scale (non-differentiable) : TS
+
scale of quantized input a
+
a_zero_point (non-differentiable) : T1
+
zero point of quantized input a
+
b (non-differentiable) : T2
+
N-dimensional quantized matrix b
+
b_scale (non-differentiable) : TS
+
scale of quantized input b
+
b_zero_point (non-differentiable) : T2
+
zero point of quantized input b
+
y_scale (non-differentiable) : TS
+
scale of quantized output y
+
y_zero_point (non-differentiable) : T3
+
zero point of quantized output y
+
+ +#### Outputs + +
+
y (non-differentiable) : T3
+
Quantized matrix multiply results from a * b
+
+ +#### Type Constraints + +
+
TS : tensor(float), tensor(float16), tensor(bfloat16)
+
Constrain scales.
+
T1 : tensor(int8), tensor(uint8), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
The type of input a and its zeropoint.
+
T2 : tensor(int8), tensor(uint8), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
The type of input b and its zeropoint.
+
T3 : tensor(int8), tensor(uint8), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz)
+
The type of the output and its zeropoint.
+
+ + +#### Examples + +
+int + +```python +for quant_type_name in ["uint8", "int8"]: + quant_type = getattr(np, quant_type_name) + for dtype_name in ["float32", "float16"]: + dtype = getattr(np, dtype_name) + node = onnx.helper.make_node( + "QLinearMatMul", + inputs=[ + "a", + "a_scale", + "a_zero_point", + "b", + "b_scale", + "b_zero_point", + "y_scale", + "y_zero_point", + ], + outputs=["y"], + ) + + # 2D + a = np.array([[208, 236, 0, 238], [3, 214, 255, 29]]) + if quant_type == np.int8: + a -= 127 + a = a.astype(quant_type) + + a_scale = np.array([0.0066], dtype=dtype) + a_zero_point = np.array( + [113 - 127] if quant_type == np.int8 else [113], dtype=quant_type + ) + + b = np.array( + [[152, 51, 244], [60, 26, 255], [0, 127, 246], [127, 254, 247]] + ) + if quant_type == np.int8: + b -= 127 + b = b.astype(quant_type) + + b_scale = np.array([0.00705], dtype=dtype) + b_zero_point = np.array( + [114 - 127] if quant_type == np.int8 else [114], dtype=quant_type + ) + + y_scale = np.array([0.0107], dtype=dtype) + y_zero_point = np.array( + [118 - 127] if quant_type == np.int8 else [118], dtype=quant_type + ) + + if quant_type == np.int8: + output = np.array([[41, -12, -9], [1, -75, -128]]) + else: + output = np.array([[168, 115, 255], [1, 66, 151]]) + output = output.astype(quant_type) + + expect( + node, + inputs=[ + a, + a_scale, + a_zero_point, + b, + b_scale, + b_zero_point, + y_scale, + y_zero_point, + ], + outputs=[output], + name=f"test_qlinearmatmul_2D_{quant_type_name}_{dtype_name}", + ) + + # 3D + a = np.array( + [ + [[208, 236, 0, 238], [3, 214, 255, 29]], + [[208, 236, 0, 238], [3, 214, 255, 29]], + ], + ) + if quant_type == np.int8: + a -= 127 + a = a.astype(quant_type) + + a_scale = np.array([0.0066], dtype=dtype) + a_zero_point = np.array( + [113 - 127] if quant_type == np.int8 else [113], dtype=quant_type + ) + + b = np.array( + [ + [[152, 51, 244], [60, 26, 255], [0, 127, 246], [127, 254, 247]], + [[152, 51, 244], [60, 26, 255], [0, 127, 246], [127, 254, 247]], + ], + ) + if quant_type == np.int8: + b -= 127 + b = b.astype(quant_type) + + b_scale = np.array([0.00705], dtype=dtype) + b_zero_point = np.array([114], dtype=quant_type) + + y_scale = np.array([0.0107], dtype=dtype) + y_zero_point = np.array( + [118 - 127] if quant_type == np.int8 else [118], dtype=quant_type + ) + + if quant_type == np.int8: + output = np.array( + [ + [[-86, -128, -128], [115, 39, -121]], + [[-86, -128, -128], [115, 39, -121]], + ] + ) + else: + output = np.array( + [ + [[168, 115, 255], [1, 66, 151]], + [[168, 115, 255], [1, 66, 151]], + ] + ) + output = output.astype(quant_type) + + expect( + node, + inputs=[ + a, + a_scale, + a_zero_point, + b, + b_scale, + b_zero_point, + y_scale, + y_zero_point, + ], + outputs=[output], + name=f"test_qlinearmatmul_3D_{quant_type_name}_{dtype_name}", + ) +``` + +
+ + +### **QuantizeLinear** + + The linear quantization operator consumes a high-precision tensor, a scale, and a zero point to compute the + low-precision/quantized tensor. The scale factor and zero point must have the same shape, determining the quantization + granularity. The quantization formula is `y = saturate((x / y_scale) + y_zero_point)`. + + Saturation is done according to: + - uint16: [0, 65535] + - int16: [-32768, 32767] + - uint8: [0, 255] + - int8: [-128, 127] + - uint4: [0, 15] + - int4: [-8, 7] + - uint2: [0, 3] + - int2: [-2, 1] + + For `(x / y_scale)`, it rounds to the nearest even. Refer to https://en.wikipedia.org/wiki/Rounding for details. + + `y_zero_point` and `y` must have the same type. `y_zero_point` is usually not used for quantization to float8 and 4bit types, but the quantization + formula remains the same for consistency, and the type of the attribute `y_zero_point` still determines the quantization type. + `x` and `y_scale` are allowed to have different types. The type of `y_scale` determines the precision of the division operation between `x` and + `y_scale`, unless the `precision` attribute is specified. + + There are three supported quantization granularities, determined by the shape of `y_scale`. + In all cases, `y_zero_point` must have the same shape as `y_scale`. + - Per-tensor (per-layer) quantization: `y_scale` is a scalar. + - Per-axis quantization: The scale must be a 1-D tensor, with the length of the quantization axis. For an input shape + `(D0, ..., Di, ..., Dn)` and `axis=i`, `y_scale` is a 1-D tensor of length `Di`. + - Blocked quantization: The scale's shape is identical to the input's shape, except for one dimension, in which + blocking is performed. Given `x` shape `(D0, ..., Di, ..., Dn)`, `axis=i`, and block size `B`: `y_scale` shape is + `(D0, ..., ceil(Di/B), ..., Dn)`. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 10, 13, 19, 21, 23, 24 + +#### Attributes + +
+
axis : int (default is 1)
+
(Optional) The axis of the dequantizing dimension of the input tensor. Used only for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is `[-r, r-1]` where `r = rank(input)`. When the rank of the input is 1, per-tensor quantization is applied, rendering the axis unnecessary in this scenario.
+
block_size : int (default is 0)
+
(Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given `x` shape `(D0, ..., Di, ..., Dn)`, `y_scale` shape `(S0, ... Si, ...Sn)` and `axis=i`, the accepted range is `[ceil(Di/Si), ceil(Di/(Si-1))-1]`
+
output_dtype : int (default is 0)
+
(Optional) The output data type. If not supplied, the output data type is inferred from `y_zero_point` data type (`T3`). If neither `output_dtype` nor `y_zero_point` are supplied, output data type is uint8. If both `output_dtype` and `y_zero_point` are specified, `output_dtype` must be `T3`.
+
precision : int (default is 0)
+
(Optional) The precision of the division operation between `x` and `y_scale`. If not provided, it will be the same as the type of `y_scale`.
+
saturate : int (default is 1)
+
The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 quantization (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz). It is true by default. All cases are fully described in two tables inserted in the operator description.
+
+ +#### Inputs (2 - 3) + +
+
x : T1
+
N-D full precision Input tensor to be quantized.
+
y_scale : T2
+
Scale for doing quantization to get `y`. For per-tensor/layer quantization the scale is a scalar, for per-axis quantization it is a 1-D Tensor and for blocked quantization it has the same shape as the input, except for one dimension in which blocking is performed.
+
y_zero_point (optional) : T3
+
Zero point for doing quantization to get `y`. Shape must match `y_scale`. Default is uint8 with zero point of 0 if it's not specified.
+
+ +#### Outputs + +
+
y : T3
+
N-D quantized output tensor. It has same shape as input `x`.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(float16), tensor(bfloat16), tensor(int32)
+
The type of the input 'x'.
+
T2 : tensor(float), tensor(float16), tensor(bfloat16), tensor(int32), tensor(float8e8m0)
+
The type of the input 'y_scale'.
+
T3 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(uint2), tensor(int2)
+
The type of the input `y_zero_point` and the output `y`.
+
+ + +#### Examples + +
+axis + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array( + [ + [ + [[-162, 10], [-100, 232], [-20, -50]], + [[-76, 0], [0, 252], [32, -44]], + [[245, -485], [-960, -270], [-375, -470]], + ], + ], + dtype=np.float32, +) +y_scale = np.array([2, 4, 5], dtype=np.float32) +y_zero_point = np.array([84, 24, 196], dtype=np.uint8) +y = (x / y_scale.reshape(1, 3, 1, 1) + y_zero_point.reshape(1, 3, 1, 1)).astype( + np.uint8 +) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_axis", +) +``` + +
+ + +
+blocked_asymmetric + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=1, + block_size=2, +) + +x = np.array( + [ + [6.0, 12.0, 50.0, 5.0], + [1.0, 8.0, 4.0, 5.0], + [0.0, 20.0, 10.0, 4.0], + ], + dtype=np.float32, +) +y_scale = np.array( + [ + [1.5, 2.5], + [3.0, 4.9], + [5.1, 6.9], + ], + dtype=np.float32, +) +y_zero_point = np.array( + [ + [0, 1], + [1, 0], + [2, 3], + ], + dtype=np.uint8, +) +# x.shape = (3, 4) +# y_scale.shape = (3, 2) +assert y_scale.shape == y_zero_point.shape +block_axis = 1 +# The block shape is [x.shape[i] // y_scale.shape[i] for i in range(len(x.shape))] = (1, 2) +assert all( + x.shape[i] == y_scale.shape[i] + for i in range(len(x.shape)) + if i != block_axis +) +assert x.shape[block_axis] % y_scale.shape[block_axis] == 0 +repeats = x.shape[block_axis] // y_scale.shape[block_axis] + +# Create element-wise scale and zero point +y_scale_elementwise = np.repeat(y_scale, repeats=repeats, axis=block_axis) +y_zero_point_elementwise = np.repeat( + y_zero_point, repeats=repeats, axis=block_axis +) + +y = np.rint(x / y_scale_elementwise + y_zero_point_elementwise).astype(np.uint8) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_blocked_asymmetric", +) +``` + +
+ + +
+blocked_symmetric + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale"], + outputs=["y"], + axis=1, + block_size=2, + output_dtype=TensorProto.INT16, +) + +x = np.array( + [ + [6.0, -8, -10, 5.0], + [1.0, 8.0, 4.0, 5.0], + [0.0, 20.0, 10.0, 4.0], + ], + dtype=np.float32, +) + +y_scale = np.array( + [ + [1.5, 2.5], + [3.0, 4.9], + [5.1, 6.9], + ], + dtype=np.float32, +) + +# x.shape = (3, 4) +# y_scale.shape = (3, 2) + +block_axis = 1 +# The block shape is [x.shape[i] // y_scale.shape[i] for i in range(len(x.shape))] = (1, 2) +assert all( + x.shape[i] == y_scale.shape[i] + for i in range(len(x.shape)) + if i != block_axis +) +assert x.shape[block_axis] % y_scale.shape[block_axis] == 0 +repeats = x.shape[block_axis] // y_scale.shape[block_axis] + +# Create element-wise scale and zero point +y_scale_elementwise = np.repeat(y_scale, repeats=repeats, axis=block_axis) + +y_val = np.clip( + np.rint(x / y_scale_elementwise), a_min=-32768, a_max=32767 +).astype(np.int16) +y = make_tensor( + "y", + TensorProto.INT16, + x.shape, + y_val, +) +expect( + node, + inputs=[x, y_scale], + outputs=[y], + name="test_quantizelinear_blocked_symmetric", +) +``` + +
+ + +
+e4m3fn + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array([0.0, 1.0, 2.0, 100000.0, 200.0]).astype(np.float32) +y_scale = np.float32(2) +y_zero_point = make_tensor("y_zero_point", TensorProto.FLOAT8E4M3FN, [1], [0]) +y = make_tensor("y", TensorProto.FLOAT8E4M3FN, [5], [0, 0.5, 1, 448, 96]) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_e4m3fn", +) +``` + +
+ + +
+e5m2 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array([0.0, 1.0, 2.0, 100000.0, 200.0]).astype(np.float32) +y_scale = np.float32(2) +y_zero_point = make_tensor("y_zero_point", TensorProto.FLOAT8E5M2, [1], [0.0]) +y = make_tensor("y", TensorProto.FLOAT8E5M2, [5], [0, 0.5, 1, 49152, 96]) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_e5m2", +) +``` + +
+ + +
+float4e2m1 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=0, +) + +x = np.array( + [ + [0.0, 2.5, 4.8, 8.6], + [-30, -20, 6, 9], + [-0.0, -2.5, -4.8, -8.6], + ] +).astype(np.float32) + +y_scale = np.asarray([2.0, 3.0, 4.0], dtype=np.float32) +y_zero_point = make_tensor( + "y_zero_point", + TensorProto.FLOAT4E2M1, + y_scale.shape, + np.zeros_like(y_scale), +) +y = make_tensor( + "y", + TensorProto.FLOAT4E2M1, + x.shape, + [0, 1, 2, 4, -6, -6, 2, 3, 0, -0.5, -1, -2], +) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_float4e2m1", +) +``` + +
+ + +
+int16 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array( + [ + 0.0, + -514.0, + 3.0, + -3.0, + 2.9, + -2.9, + 3.1, + -3.1, + 65022.0, + -66046.0, + 65023.0, + -66047.0, + 65024.0, + -66048.0, + 70000.0, + -70000.0, + ] +).astype(np.float32) +y_scale = np.float32(2.0) +y_zero_point = np.int16(256) +y = np.array( + [ + 256, + -1, + 258, + 254, + 257, + 255, + 258, + 254, + 32767, + -32767, + 32767, + -32768, + 32767, + -32768, + 32767, + -32768, + ] +).astype(np.int16) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_int16", +) +``` + +
+ + +
+int2 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=0, +) +x = np.array( + [ + [0.0, 2.5, 4.8, 8.6], + [-4.0, -3.0, 1.0, 2.0], + [-0.0, -2.5, -4.8, -8.6], + ], + dtype=np.float32, +) +y_scale = np.asarray([2.0, 3.0, 4.0], dtype=np.float32) +y_zero_point = make_tensor( + "y_zero_point", TensorProto.INT2, y_scale.shape, np.zeros_like(y_scale) +) +y = make_tensor( + "y", TensorProto.INT2, x.shape, [0, 1, 1, 1, -1, -1, 0, 1, 0, -1, -1, -2] +) +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_int2", +) +``` + +
+ + +
+int4 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=0, +) + +x = np.array( + [ + [0.0, 2.5, 4.8, 8.6], + [-30, -20, 6, 9], + [12, 15, 16, 40], + ] +).astype(np.float32) + +y_scale = np.asarray([2.0, 3.0, 4.0], dtype=np.float32) +y_zero_point = make_tensor( + "y_zero_point", TensorProto.INT4, y_scale.shape, np.ones_like(y_scale) +) +y = make_tensor( + "y", TensorProto.INT4, x.shape, [1, 2, 3, 5, -8, -6, 3, 4, 4, 5, 5, 7] +) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_int4", +) +``` + +
+ + +
+quantizelinear + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array([0, 2, 3, 1000, -254, -1000]).astype(np.float32) +y_scale = np.float32(2) +y_zero_point = np.uint8(128) +y = np.array([128, 129, 130, 255, 1, 0]).astype(np.uint8) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear", +) +``` + +
+ + +
+uint16 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array( + [ + 0.0, + -128.0, + 3.0, + -3.0, + 2.9, + -2.9, + 3.1, + -3.1, + 65536.0, + -65534.0, + 70000.0, + -70000.0, + ] +).astype(np.float32) +y_scale = np.float32(2.0) +y_zero_point = np.uint16(32767) +y = np.array( + [ + 32767, + 32703, + 32769, + 32765, + 32768, + 32766, + 32769, + 32765, + 65535, + 0, + 65535, + 0, + ] +).astype(np.uint16) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_uint16", +) +``` + +
+ + +
+uint2 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=0, +) + +x = np.array( + [ + [0.0, 2.5, 4.8, 8.6], + [-2.0, -1.0, 1.0, 3.0], + [4.0, 5.0, 6.0, 7.0], + ], + dtype=np.float32, +) +y_scale = np.asarray([2.0, 3.0, 4.0], dtype=np.float32) +y_zero_point = make_tensor( + "y_zero_point", TensorProto.UINT2, y_scale.shape, np.zeros_like(y_scale) +) +y = make_tensor( + "y", TensorProto.UINT2, x.shape, [0, 1, 2, 3, 0, 0, 0, 1, 1, 1, 2, 2] +) +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_uint2", +) +``` + +
+ + +
+uint4 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=0, +) + +x = np.array( + [ + [0.0, 2.5, 4.8, 8.6], + [-30, -20, 6, 9], + [12, 15, 16, 40], + ] +).astype(np.float32) + +y_scale = np.asarray([2.0, 3.0, 4.0], dtype=np.float32) +y_zero_point = make_tensor( + "y_zero_point", TensorProto.UINT4, y_scale.shape, np.ones_like(y_scale) +) +y = make_tensor( + "y", TensorProto.UINT4, x.shape, [1, 2, 3, 5, 0, 0, 3, 4, 4, 5, 5, 11] +) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_uint4", +) +``` + +
+ + +### **RMSNormalization** + + This is RMS normalization defined in ONNX as function as described in the paper https://arxiv.org/pdf/1910.07467. + The overall computation can be split into two stages. The root mean squared norm is taken over the last D dimensions, + where D is the dimension of normalized_shape. For example, if normalized_shape is (3, 5) (a 2-dimensional shape), + the rms norm is computed over the last 2 dimensions of the input. The computation required by standardization can be + described by the following equations. + ``` + XSquared = Mul(X, X) + XSquaredMean = ReduceMean(XSquared) + MeanSquareEpsilon = Add(XSquaredMean, epsilon) + RMS = Sqrt(MeanSquareEpsilon) + Normalized = Div(X, RMS) + ``` + where `normalized_axes` is `[axis, ..., rank of X - 1]`. The variables `RMS` stand for root mean square, + Depending on `stash_type` attribute, the actual computation + must happen in different floating-point precision. + For example, if `stash_type` is 1, this operator casts + all input variables to 32-bit float, perform the computation, and + finally cast `Normalized` back to the original type of `X`. + The second stage then scales the outcome of the first stage using: + ``` + Y= Mul(Normalized, Scale) + ``` + Let `d[i]` indicate the i-th dimension of `X`. + If `X`'s shape is `[d[0], ..., d[axis-1], d[axis], ..., d[rank-1]]`, + the shape of `RMS` is `[d[0], ..., d[axis-1], 1, ..., 1]`. + `Y` and `X` have the same shape. This operator supports unidirectional broadcasting + (`Scale` should be unidirectional broadcastable to tensor `X`); + for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -1)
+
The first normalization dimension. If rank(X) is r, axis' allowed range is [-r, r). Negative value means counting dimensions from the back.
+
epsilon : float (default is 1e-05)
+
The epsilon value to use to avoid division by zero.
+
stash_type : int (default is 1)
+
The floating-point precision used in stage one of the computation.
+
+ +#### Inputs + +
+
X : T
+
The input tensor to be normalized. In general, the shape is (D1, D2, ... , Dn) for n-dimensional data, where the root mean squared norm is taken over the last D dimensions, D is determined by the axis attribute.
+
scale : V
+
Scale tensor. Scale tensor shape should be broadcastable to the normalized shape.
+
+ +#### Outputs + +
+
Y : V
+
Output data tensor. Same shape as X
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input X type to float tensors.
+
V : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain output Y and scale type to float tensors.
+
+ + +#### Examples + +
+d + +```python +X = np.random.randn(3, 4).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + Y = _rms_normalization(X, W, axis=axis) + + node = onnx.helper.make_node( + "RMSNormalization", + inputs=["X", "W"], + outputs=["Y"], + axis=axis, + ) + + if axis < 0: + name = f"test_rms_normalization_2d_axis_negative_{-axis}" + else: + name = f"test_rms_normalization_2d_axis{axis}" + + expect(node, inputs=[X, W], outputs=[Y], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+ + +
+d_epsilon + +```python +epsilon = 1e-1 +X = np.random.randn(2, 3, 5).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + Y = _rms_normalization(X, W, axis=axis, epsilon=epsilon) + node = onnx.helper.make_node( + "RMSNormalization", + inputs=["X", "W"], + outputs=["Y"], + axis=axis, + epsilon=epsilon, + ) + + if axis < 0: + name = f"test_rms_normalization_3d_axis_negative_{-axis}_epsilon" + else: + name = f"test_rms_normalization_3d_axis{axis}_epsilon" + + expect(node, inputs=[X, W], outputs=[Y], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+ + +
+default_axis + +```python +X = np.random.randn(2, 3, 4, 5).astype(np.float32) + +# Default axis in RMSNormalization is -1. +normalized_shape = calculate_normalized_shape(X.shape, -1) +W = np.random.randn(*normalized_shape).astype(np.float32) +# Axis is default to -1 in the reference implementation. +Y = _rms_normalization(X, W) + +# Not specifying axis attribute means -1. +node = onnx.helper.make_node( + "RMSNormalization", + inputs=["X", "W"], + outputs=["Y"], +) + +expect( + node, + inputs=[X, W], + outputs=[Y], + name="test_rms_normalization_default_axis", +) +``` + +
+ + +
+rmsnormalization + +```python +X = np.random.randn(2, 3, 4, 5).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + Y = _rms_normalization(X, W, axis=axis) + + node = onnx.helper.make_node( + "RMSNormalization", + inputs=["X", "W"], + outputs=["Y"], + axis=axis, + ) + + if axis < 0: + name = f"test_rms_normalization_4d_axis_negative_{-axis}" + else: + name = f"test_rms_normalization_4d_axis{axis}" + + expect(node, inputs=[X, W], outputs=[Y], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+ + +### **RNN** + + Computes an one-layer simple RNN. This operator is usually supported + via some custom implementation such as CuDNN. + + Notations: + + * `X` - input tensor + * `i` - input gate + * `t` - time step (t-1 means previous time step) + * `Wi` - W parameter weight matrix for input gate + * `Ri` - R recurrence weight matrix for input gate + * `Wbi` - W parameter bias vector for input gate + * `Rbi` - R parameter bias vector for input gate + * `WBi` - W parameter weight matrix for backward input gate + * `RBi` - R recurrence weight matrix for backward input gate + * `WBbi` - WR bias vectors for backward input gate + * `RBbi` - RR bias vectors for backward input gate + * `H` - Hidden state + * `num_directions` - 2 if direction == bidirectional else 1 + + Activation functions: + + * Relu(x) - max(0, x) + * Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x}) + * Sigmoid(x) - 1/(1 + e^{-x}) + + NOTE: Below are optional + + * Affine(x) - alpha*x + beta + * LeakyRelu(x) - x if x >= 0 else alpha * x + * ThresholdedRelu(x) - x if x >= alpha else 0 + * ScaledTanh(x) - alpha*Tanh(beta*x) + * HardSigmoid(x) - min(max(alpha*x + beta, 0), 1) + * Elu(x) - x if x >= 0 else alpha*(e^x - 1) + * Softsign(x) - x/(1 + |x|) + * Softplus(x) - log(1 + e^x) + + Equations (Default: f=Tanh): + + * Ht = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Wbi + Rbi) + This operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 7, 14 + +#### Attributes + +
+
activation_alpha : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
+
activation_beta : list of floats
+
Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
+
activations : list of strings (default is ['Tanh', 'Tanh'])
+
One (or two if bidirectional) activation function for input gate. The activation function must be one of the activation functions specified above. Optional: Default `Tanh` if not specified.
+
clip : float
+
Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
+
direction : string (default is forward)
+
Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
+
hidden_size : int
+
Number of neurons in the hidden layer
+
layout : int (default is 0)
+
The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size].
+
+ +#### Inputs (3 - 6) + +
+
X (differentiable) : T
+
The input sequences packed (and potentially padded) into one 3-D tensor with the shape of `[seq_length, batch_size, input_size]`.
+
W (differentiable) : T
+
The weight tensor for input gate. Concatenation of `Wi` and `WBi` (if bidirectional). The tensor has shape `[num_directions, hidden_size, input_size]`.
+
R (differentiable) : T
+
The recurrence weight tensor. Concatenation of `Ri` and `RBi` (if bidirectional). The tensor has shape `[num_directions, hidden_size, hidden_size]`.
+
B (optional, differentiable) : T
+
The bias tensor for input gate. Concatenation of `[Wbi, Rbi]` and `[WBbi, RBbi]` (if bidirectional). The tensor has shape `[num_directions, 2*hidden_size]`. Optional: If not specified - assumed to be 0.
+
sequence_lens (optional, non-differentiable) : T1
+
Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length `seq_length`. It has shape `[batch_size]`.
+
initial_h (optional, non-differentiable) : T
+
Optional initial value of the hidden. If not specified - assumed to be 0. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Outputs (0 - 2) + +
+
Y (optional, differentiable) : T
+
A tensor that concats all the intermediate output values of the hidden. It has shape `[seq_length, num_directions, batch_size, hidden_size]`.
+
Y_h (optional, differentiable) : T
+
The last output value of the hidden. It has shape `[num_directions, batch_size, hidden_size]`.
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
T1 : tensor(int32)
+
Constrain seq_lens to integer tensor.
+
+ + +#### Examples + +
+batchwise + +```python +input = np.array([[[1.0, 2.0]], [[3.0, 4.0]], [[5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 4 +weight_scale = 0.5 +layout = 1 + +node = onnx.helper.make_node( + "RNN", + inputs=["X", "W", "R"], + outputs=["Y", "Y_h"], + hidden_size=hidden_size, + layout=layout, +) + +W = weight_scale * np.ones((1, hidden_size, input_size)).astype(np.float32) +R = weight_scale * np.ones((1, hidden_size, hidden_size)).astype(np.float32) + +rnn = RNNHelper(X=input, W=W, R=R, layout=layout) +Y, Y_h = rnn.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y.astype(np.float32), Y_h.astype(np.float32)], + name="test_simple_rnn_batchwise", +) +``` + +
+ + +
+defaults + +```python +input = np.array([[[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 4 +weight_scale = 0.1 + +node = onnx.helper.make_node( + "RNN", inputs=["X", "W", "R"], outputs=["", "Y_h"], hidden_size=hidden_size +) + +W = weight_scale * np.ones((1, hidden_size, input_size)).astype(np.float32) +R = weight_scale * np.ones((1, hidden_size, hidden_size)).astype(np.float32) + +rnn = RNNHelper(X=input, W=W, R=R) +_, Y_h = rnn.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y_h.astype(np.float32)], + name="test_simple_rnn_defaults", +) +``` + +
+ + +
+initial_bias + +```python +input = np.array([[[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]]]).astype( + np.float32 +) + +input_size = 3 +hidden_size = 5 +custom_bias = 0.1 +weight_scale = 0.1 + +node = onnx.helper.make_node( + "RNN", + inputs=["X", "W", "R", "B"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +W = weight_scale * np.ones((1, hidden_size, input_size)).astype(np.float32) +R = weight_scale * np.ones((1, hidden_size, hidden_size)).astype(np.float32) + +# Adding custom bias +W_B = custom_bias * np.ones((1, hidden_size)).astype(np.float32) +R_B = np.zeros((1, hidden_size)).astype(np.float32) +B = np.concatenate((W_B, R_B), axis=1) + +rnn = RNNHelper(X=input, W=W, R=R, B=B) +_, Y_h = rnn.step() +expect( + node, + inputs=[input, W, R, B], + outputs=[Y_h.astype(np.float32)], + name="test_simple_rnn_with_initial_bias", +) +``` + +
+ + +
+seq_length + +```python +input = np.array( + [ + [[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]], + [[10.0, 11.0, 12.0], [13.0, 14.0, 15.0], [16.0, 17.0, 18.0]], + ] +).astype(np.float32) + +input_size = 3 +hidden_size = 5 + +node = onnx.helper.make_node( + "RNN", + inputs=["X", "W", "R", "B"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +W = np.random.randn(1, hidden_size, input_size).astype(np.float32) +R = np.random.randn(1, hidden_size, hidden_size).astype(np.float32) + +# Adding custom bias +W_B = np.random.randn(1, hidden_size).astype(np.float32) +R_B = np.random.randn(1, hidden_size).astype(np.float32) +B = np.concatenate((W_B, R_B), axis=1) + +rnn = RNNHelper(X=input, W=W, R=R, B=B) +_, Y_h = rnn.step() +expect( + node, + inputs=[input, W, R, B], + outputs=[Y_h.astype(np.float32)], + name="test_rnn_seq_length", +) +``` + +
+ + +### **RandomNormal** + + Generate a tensor with random values drawn from a normal distribution. The shape + of the tensor is specified by the `shape` argument and the parameter of the normal distribution + specified by `mean` and `scale`. + + The data type is specified by the 'dtype' argument. The 'dtype' argument must + be one of the data types specified in the 'DataType' enum field in the + TensorProto message. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Attributes + +
+
dtype : int (default is 1)
+
The data type for the elements of the output tensor. Default is TensorProto::FLOAT.
+
mean : float (default is 0.0)
+
The mean of the normal distribution.
+
scale : float (default is 1.0)
+
The standard deviation of the normal distribution.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
shape : list of ints (required)
+
The shape of the output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor of random values drawn from normal distribution
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ + +### **RandomNormalLike** + + Generate a tensor with random values drawn from a normal distribution. + The shape of the output tensor is copied from the shape of the input tensor, + and the parameters of the normal distribution are specified by `mean` and `scale`. + + The data type is specified by the 'dtype' argument, or copied from the input tensor if not provided. + The 'dtype' argument must be one of the data types specified in the 'DataType' enum field in the + TensorProto message, and be valid as an output type. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Attributes + +
+
dtype : int
+
(Optional) The data type for the elements of the output tensor, if not specified, we will use the data type of the input tensor.
+
mean : float (default is 0.0)
+
The mean of the normal distribution.
+
scale : float (default is 1.0)
+
The standard deviation of the normal distribution.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
Input tensor to copy shape and optionally type information from.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of random values drawn from normal distribution
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type. If the dtype attribute is not provided this must be a valid output type.
+
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ + +### **RandomUniform** + + Generate a tensor with random values drawn from a uniform distribution. The shape + of the tensor is specified by the `shape` argument and the range by `low` and `high`. + + The data type is specified by the 'dtype' argument. The 'dtype' argument must + be one of the data types specified in the 'DataType' enum field in the + TensorProto message. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Attributes + +
+
dtype : int (default is 1)
+
The data type for the elements of the output tensor. If not specified, default is TensorProto::FLOAT.
+
high : float (default is 1.0)
+
Upper boundary of the output values.
+
low : float (default is 0.0)
+
Lower boundary of the output values.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
shape : list of ints (required)
+
The shape of the output tensor.
+
+ +#### Inputs + + +#### Outputs + +
+
output : T
+
Output tensor of random values drawn from uniform distribution
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ + +### **RandomUniformLike** + + Generate a tensor with random values drawn from a uniform distribution. + The shape of the output tensor is copied from the shape of the input tensor, + and the parameters of the uniform distribution are specified by `low` and `high`. + + The data type is specified by the 'dtype' argument, or copied from the input tensor if not provided. + The 'dtype' argument must be one of the data types specified in the 'DataType' enum field in the + TensorProto message and be valid as an output type. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Attributes + +
+
dtype : int
+
(Optional) The data type for the elements of the output tensor, if not specified, we will use the data type of the input tensor.
+
high : float (default is 1.0)
+
Upper boundary of the output values.
+
low : float (default is 0.0)
+
Lower boundary of the output values.
+
seed : float
+
(Optional) Seed to the random generator, if not specified we will auto generate one.
+
+ +#### Inputs + +
+
input : T1
+
Input tensor to copy shape and optionally type information from.
+
+ +#### Outputs + +
+
output : T2
+
Output tensor of random values drawn from uniform distribution
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type. If the dtype attribute is not provided this must be a valid output type.
+
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain output types to float tensors.
+
+ + +### **Range** + + Generate a tensor containing a sequence of numbers that begin at `start` and extends by increments of `delta` + up to `limit` (exclusive). + + The number of elements in the output of range is computed as below: + ``` + number_of_elements = max( ceil( (limit - start) / delta ) , 0 ) + ``` + The pseudocode determining the contents of the output is shown below: + ``` + for(int i=0; i11 + +#### Attributes + +
+
stash_type : int (default is 1)
+
The data type used for intermediate computation when T is float16 or bfloat16. Defaults to 1 (float). Has no effect for other types.
+
+ +#### Inputs + +
+
start : T
+
Scalar. First entry for the range of output values.
+
limit : T
+
Scalar. Exclusive upper limit for the range of output values.
+
delta : T
+
Scalar. Value to step by.
+
+ +#### Outputs + +
+
output : T
+
A 1-D tensor with same type as the inputs containing generated range of values.
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(double), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(bfloat16)
+
Constrain input types to common numeric type tensors.
+
+ + +#### Examples + +
+range_bfloat16_type_positive_delta + +```python +node = onnx.helper.make_node( + "Range", + inputs=["start", "limit", "delta"], + outputs=["output"], +) + +start = np.array(1.0, dtype=ml_dtypes.bfloat16) +limit = np.array(5.0, dtype=ml_dtypes.bfloat16) +delta = np.array(2.0, dtype=ml_dtypes.bfloat16) + +output = np.arange(1.0, 5.0, 2.0, dtype=np.float32).astype( + ml_dtypes.bfloat16 +) # expected output [1.0, 3.0] as bfloat16 + +expect( + node, + inputs=[start, limit, delta], + outputs=[output], + name="test_range_bfloat16_type_positive_delta", +) +``` + +
+ + +
+range_float16_type_positive_delta + +```python +node = onnx.helper.make_node( + "Range", + inputs=["start", "limit", "delta"], + outputs=["output"], +) + +start = np.float16(1) +limit = np.float16(5) +delta = np.float16(2) + +output = np.arange( + start, limit, delta, dtype=np.float16 +) # expected output [1.0, 3.0] +expect( + node, + inputs=[start, limit, delta], + outputs=[output], + name="test_range_float16_type_positive_delta", +) +``` + +
+ + +
+range_float_type_positive_delta + +```python +node = onnx.helper.make_node( + "Range", + inputs=["start", "limit", "delta"], + outputs=["output"], +) + +start = np.float32(1) +limit = np.float32(5) +delta = np.float32(2) + +output = np.arange( + start, limit, delta, dtype=np.float32 +) # expected output [1.0, 3.0] +expect( + node, + inputs=[start, limit, delta], + outputs=[output], + name="test_range_float_type_positive_delta", +) +``` + +
+ + +
+range_int32_type_negative_delta + +```python +node = onnx.helper.make_node( + "Range", + inputs=["start", "limit", "delta"], + outputs=["output"], +) + +start = np.int32(10) +limit = np.int32(6) +delta = np.int32(-3) + +output = np.arange( + start, limit, delta, dtype=np.int32 +) # expected output [10, 7] +expect( + node, + inputs=[start, limit, delta], + outputs=[output], + name="test_range_int32_type_negative_delta", +) +``` + +
+ + +### **Reciprocal** + + Reciprocal takes one input data (Tensor) and produces one output data + (Tensor) where the reciprocal is, y = 1/x, is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+reciprocal + +```python +node = onnx.helper.make_node( + "Reciprocal", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-4, 2]).astype(np.float32) +y = np.reciprocal(x) # expected output [-0.25, 0.5], +expect(node, inputs=[x], outputs=[y], name="test_reciprocal_example") + +x = np.random.rand(3, 4, 5).astype(np.float32) + 0.5 +y = np.reciprocal(x) +expect(node, inputs=[x], outputs=[y], name="test_reciprocal") +``` + +
+ + +### **ReduceL1** + + Computes the L1 norm of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13 + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ + +#### Examples + +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL1", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sum(a=np.abs(data), axis=None, keepdims=keepdims == 1) +# print(reduced) +# [[[78.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(a=np.abs(data), axis=None, keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_default_axes_keepdims_random", +) +``` + +
+ + +
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([2], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceL1", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[3., 7.], [11., 15.], [19., 23.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_do_not_keepdims_random", +) +``` + +
+ + +
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceL1", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +reduced = np.array(np.zeros(reduced_shape, dtype=np.float32)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_empty_set", +) +``` + +
+ + +
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL1", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[3.], [7.]], [[11.], [15.]], [[19.], [23.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_keep_dims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_keep_dims_random", +) +``` + +
+ + +
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL1", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[3.], [7.]], [[11.], [15.]], [[19.], [23.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_negative_axes_keep_dims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_negative_axes_keep_dims_random", +) +``` + +
+ + +### **ReduceL2** + + Computes the L2 norm of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13 + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ + +#### Examples + +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL2", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sqrt(np.sum(a=np.square(data), axis=None, keepdims=keepdims == 1)) +# print(reduced) +# [[[25.49509757]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sqrt(np.sum(a=np.square(data), axis=None, keepdims=keepdims == 1)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_default_axes_keepdims_random", +) +``` + +
+ + +
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([2], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceL2", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) +# print(reduced) +# [[2.23606798, 5.], +# [7.81024968, 10.63014581], +# [13.45362405, 16.2788206]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_do_not_keepdims_random", +) +``` + +
+ + +
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceL2", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +reduced = np.array(np.zeros(reduced_shape, dtype=np.float32)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_empty_set", +) +``` + +
+ + +
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL2", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) +# print(reduced) +# [[[2.23606798], [5.]] +# [[7.81024968], [10.63014581]] +# [[13.45362405], [16.2788206 ]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_keep_dims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_keep_dims_random", +) +``` + +
+ + +
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL2", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) +# print(reduced) +# [[[2.23606798], [5.]] +# [[7.81024968], [10.63014581]] +# [[13.45362405], [16.2788206 ]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_negative_axes_keep_dims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_negative_axes_keep_dims_random", +) +``` + +
+ + +### **ReduceLogSum** + + Computes the log sum of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or undefined otherwise. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13 + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ + +#### Examples + +
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceLogSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +zero = np.array(np.zeros(reduced_shape, dtype=np.float32)) +reduced = np.log(zero) # -inf + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_empty_set", +) +``` + +
+ + +
+keepdims + +```python +node = onnx.helper.make_node( + "ReduceLogSum", inputs=["data", "axes"], outputs=["reduced"] +) +data = np.random.ranf([3, 4, 5]).astype(np.float32) +reduced = np.log(np.sum(data, keepdims=True)) +axes = np.array([], dtype=np.int64) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_default", +) +``` + +
+ + +
+negative_axes_keepdims + +```python +axes = np.array([-2], dtype=np.int64) +node = onnx.helper.make_node( + "ReduceLogSum", inputs=["data", "axes"], outputs=["reduced"] +) +data = np.random.ranf([3, 4, 5]).astype(np.float32) +reduced = np.log(np.sum(data, axis=tuple(axes), keepdims=True)) +# print(reduced) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_negative_axes", +) +``` + +
+ + +
+nokeepdims + +```python +shape = [3, 4, 5] +axes = np.array([2, 1], dtype=np.int64) + +node = onnx.helper.make_node( + "ReduceLogSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=0, +) +data = np.random.ranf(shape).astype(np.float32) +reduced = np.log(np.sum(data, axis=tuple(axes), keepdims=False)) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_desc_axes", +) + +axes = np.array([0, 1], dtype=np.int64) +node = onnx.helper.make_node( + "ReduceLogSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=0, +) +data = np.random.ranf(shape).astype(np.float32) +reduced = np.log(np.sum(data, axis=tuple(axes), keepdims=False)) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_asc_axes", +) +``` + +
+ + +### **ReduceLogSumExp** + + Computes the log sum exponent of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or undefined otherwise. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13 + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ + +#### Examples + +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceLogSumExp", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], dtype=np.double +) +reduced = np.log(np.sum(np.exp(data), axis=None, keepdims=keepdims == 1)) +# print(reduced) +# [[[60.00671387]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.double) +reduced = np.log(np.sum(np.exp(data), axis=None, keepdims=keepdims == 1)) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_default_axes_keepdims_random", +) +``` + +
+ + +
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 +node = onnx.helper.make_node( + "ReduceLogSumExp", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], dtype=np.double +) +reduced = np.log(np.sum(np.exp(data), axis=tuple(axes), keepdims=keepdims == 1)) +# print(reduced) +# [[20., 2.31326175] +# [40.00004578, 2.31326175] +# [60.00671387, 2.31326175]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.double) +reduced = np.log(np.sum(np.exp(data), axis=tuple(axes), keepdims=keepdims == 1)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_do_not_keepdims_random", +) +``` + +
+ + +
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceLogSumExp", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +zero = np.array(np.zeros(reduced_shape, dtype=np.float32)) +reduced = np.log(zero) # -inf + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_empty_set", +) +``` + +
+ + +
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 +node = onnx.helper.make_node( + "ReduceLogSumExp", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], dtype=np.double +) +reduced = np.log(np.sum(np.exp(data), axis=tuple(axes), keepdims=keepdims == 1)) +# print(reduced) +# [[[20., 2.31326175]] +# [[40.00004578, 2.31326175]] +# [[60.00671387, 2.31326175]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.double) +reduced = np.log(np.sum(np.exp(data), axis=tuple(axes), keepdims=keepdims == 1)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_keepdims_random", +) +``` + +
+ + +
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 +node = onnx.helper.make_node( + "ReduceLogSumExp", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], dtype=np.double +) +reduced = np.log(np.sum(np.exp(data), axis=tuple(axes), keepdims=keepdims == 1)) +# print(reduced) +# [[[20., 2.31326175]] +# [[40.00004578, 2.31326175]] +# [[60.00671387, 2.31326175]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_negative_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.double) +reduced = np.log( + np.sum(np.exp(data), axis=tuple(axes.tolist()), keepdims=keepdims == 1) +) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_negative_axes_keepdims_random", +) +``` + +
+ + +### **ReduceMax** + + Computes the max of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields minus infinity (if supported by the datatype) or the minimum value of the data type otherwise. + + + If the input data type is Boolean, the comparison should consider `False < True`. + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 12, 13, 18 + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(uint8), tensor(int8), tensor(bool)
+
Constrain input and output types to numeric and Boolean tensors.
+
+ + +#### Examples + +
+bool_inputs + +```python +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMax", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[True, True], [True, False], [False, True], [False, False]], +) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=bool(keepdims)) +# print(reduced) +# [[True], +# [True], +# [True], +# [False]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_bool_inputs", +) +``` + +
+ + +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = None +keepdims = 1 +node = onnx.helper.make_node( + "ReduceMax", inputs=["data"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.maximum.reduce(data, axis=axes, keepdims=keepdims == 1) + +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_max_default_axes_keepdim_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.maximum.reduce(data, axis=axes, keepdims=keepdims == 1) + +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_max_default_axes_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+ + +
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceMax", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[20., 2.] +# [40., 2.] +# [60., 2.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_do_not_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_do_not_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+ + +
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceMax", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +one = np.array(np.ones(reduced_shape, dtype=np.float32)) +zero = np.array(np.zeros(reduced_shape, dtype=np.float32)) +reduced = -(one / zero) # -inf + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_empty_set", +) +``` + +
+ + +
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMax", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[20., 2.]] +# [[40., 2.]] +# [[60., 2.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+ + +
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMax", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[20., 2.]] +# [[40., 2.]] +# [[60., 2.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_negative_axes_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_negative_axes_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+ + +### **ReduceMean** + + Computes the mean of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields undefined. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13 + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ + +#### Examples + +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMean", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.mean(data, axis=None, keepdims=keepdims == 1) +# print(reduced) +# [[[18.25]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.mean(data, axis=None, keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_default_axes_keepdims_random", +) +``` + +
+ + +
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceMean", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[12.5, 1.5] +# [35., 1.5] +# [57.5, 1.5]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_do_not_keepdims_random", +) +``` + +
+ + +
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMean", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[12.5, 1.5]] +# [[35., 1.5]] +# [[57.5, 1.5]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_keepdims_random", +) +``` + +
+ + +
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMean", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[12.5, 1.5]] +# [[35., 1.5]] +# [[57.5, 1.5]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_negative_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_negative_axes_keepdims_random", +) +``` + +
+ + +### **ReduceMin** + + Computes the min of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields plus infinity (if supported by the datatype) or the maximum value of the data type otherwise. + + + If the input data type is Boolean, the comparison should consider `False < True`. + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 12, 13, 18 + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(uint8), tensor(int8), tensor(bool)
+
Constrain input and output types to numeric and Boolean tensors.
+
+ + +#### Examples + +
+bool_inputs + +```python +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMin", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[True, True], [True, False], [False, True], [False, False]], +) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=bool(keepdims)) +# print(reduced) +# [[ True], +# [False], +# [False], +# [False]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_bool_inputs", +) +``` + +
+ + +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = None +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMin", inputs=["data"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.minimum.reduce(data, axis=axes, keepdims=keepdims == 1) +# print(reduced) +# [[[1.]]] + +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_min_default_axes_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.minimum.reduce(data, axis=axes, keepdims=keepdims == 1) + +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_min_default_axes_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+ + +
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceMin", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[5., 1.] +# [30., 1.] +# [55., 1.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_do_not_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_do_not_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+ + +
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceMin", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +one = np.array(np.ones(reduced_shape, dtype=np.float32)) +zero = np.array(np.zeros(reduced_shape, dtype=np.float32)) +reduced = one / zero # inf + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_empty_set", +) +``` + +
+ + +
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMin", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[5., 1.]] +# [[30., 1.]] +# [[55., 1.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+ + +
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMin", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[5., 1.]] +# [[30., 1.]] +# [[55., 1.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_negative_axes_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_negative_axes_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+ + +### **ReduceProd** + + Computes the product of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 1. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13 + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ + +#### Examples + +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = None +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceProd", inputs=["data"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.prod(data, axis=axes, keepdims=keepdims == 1) +# print(reduced) +# [[[4.790016e+08]]] + +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_prod_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.prod(data, axis=axes, keepdims=keepdims == 1) +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_prod_default_axes_keepdims_random", +) +``` + +
+ + +
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceProd", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[3., 8.] +# [35., 48.] +# [99., 120.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_do_not_keepdims_random", +) +``` + +
+ + +
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceProd", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +reduced = np.array(np.ones(reduced_shape, dtype=np.float32)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_empty_set", +) +``` + +
+ + +
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceProd", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[3., 8.]] +# [[35., 48.]] +# [[99., 120.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_keepdims_random", +) +``` + +
+ + +
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceProd", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[3., 8.]] +# [[35., 48.]] +# [[99., 120.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_negative_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_negative_axes_keepdims_random", +) +``` + +
+ + +### **ReduceSum** + + Computes the sum of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 11 + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ + +#### Examples + +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSum", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(data, axis=None, keepdims=keepdims == 1) +# print(reduced) +# [[[78.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(data, axis=None, keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_default_axes_keepdims_random", +) +``` + +
+ + +
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceSum", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) +# print(reduced) +# [[4., 6.] +# [12., 14.] +# [20., 22.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_do_not_keepdims_random", +) +``` + +
+ + +
+empty_axes_input_noop + +```python +shape = [3, 2, 2] +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, + noop_with_empty_axes=True, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +axes = np.array([], dtype=np.int64) +reduced = np.array(data) +# print(reduced) +# [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_empty_axes_input_noop_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.array(data) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_empty_axes_input_noop", +) +``` + +
+ + +
+empty_set + +```python +"""Test case with the reduced-axis of size zero.""" +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +reduced = np.array(np.zeros(reduced_shape, dtype=np.float32)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_empty_set", +) +``` + +
+ + +
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSum", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) +# print(reduced) +# [[[4., 6.]] +# [[12., 14.]] +# [[20., 22.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_keepdims_random", +) +``` + +
+ + +
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSum", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) +# print(reduced) +# [[[4., 6.]] +# [[12., 14.]] +# [[20., 22.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_negative_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_negative_axes_keepdims_random", +) +``` + +
+ + +
+non_reduced_axis_zero + +```python +"""Test case with the non-reduced-axis of size zero.""" +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 0, 1] + +node = onnx.helper.make_node( + "ReduceSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([2], dtype=np.int64) +reduced = np.array([], dtype=np.float32).reshape(reduced_shape) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_empty_set_non_reduced_axis_zero", +) +``` + +
+ + +### **ReduceSumSquare** + + Computes the sum square of the input tensor's elements along the provided axes. The resulting + tensor has the same rank as the input if `keepdims` equals 1. If `keepdims` equals 0, then + the resulting tensor has the reduced dimension pruned. Input tensors of rank zero are + valid. Reduction over an empty set of values yields 0. + + + The above behavior is similar to numpy, with the exception that numpy defaults `keepdims` + to `False` instead of `True`. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13 + +#### Attributes + +
+
keepdims : int (default is 1)
+
Keep the reduced dimension or not, default 1 means keep reduced dimension.
+
noop_with_empty_axes : int (default is 0)
+
Defines behavior when axes is not provided or is empty. If false (default), reduction happens over all axes (similar to the case when `axis=None` in numpy). If true, reduction happens over an empty set of axes (similar to the case when `axis=()` in numpy). Note that reduction over an empty set of axes means that the reduction step behaves like a no-op (identity function), but composite-reduction operators will still perform the non-reduction steps as needed. Thus, ReduceLogSum returns the Log of input tensor, and ReduceSumSquare returns the Square of the input tensor, in this case.
+
+ +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
An input tensor.
+
axes (optional, non-differentiable) : tensor(int64)
+
Optional input list of integers, along which to reduce. The default is to reduce over empty axes. When axes is empty (either not provided or explicitly empty), behavior depends on 'noop_with_empty_axes': reduction over all axes if 'noop_with_empty_axes' is false, and reduction over the empty set of axes when 'noop_with_empty_axes' is true. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
reduced (differentiable) : T
+
Reduced output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
+ + +#### Examples + +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSumSquare", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(np.square(data), axis=None, keepdims=keepdims == 1) +# print(reduced) +# [[[650.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(np.square(data), axis=None, keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_default_axes_keepdims_random", +) +``` + +
+ + +
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceSumSquare", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[10., 20.] +# [74., 100.] +# [202., 244.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_do_not_keepdims_random", +) +``` + +
+ + +
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceSumSquare", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +reduced = np.array(np.zeros(reduced_shape, dtype=np.float32)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_empty_set", +) +``` + +
+ + +
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSumSquare", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[10., 20.]] +# [[74., 100.]] +# [[202., 244.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_keepdims_random", +) +``` + +
+ + +
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSumSquare", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[10., 20.s]] +# [[74., 100.]] +# [[202., 244.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_negative_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_negative_axes_keepdims_random", +) +``` + +
+ + +### **RegexFullMatch** + + RegexFullMatch performs a full regex match on each element of the input tensor. If an element fully matches the regex pattern specified as an attribute, the corresponding element in the output is True and it is False otherwise. [RE2](https://github.com/google/re2/wiki/Syntax) regex syntax is used. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
pattern : string
+
Regex pattern to match on. This must be valid RE2 syntax.
+
+ +#### Inputs + +
+
X (non-differentiable) : T1
+
Tensor with strings to match on.
+
+ +#### Outputs + +
+
Y (non-differentiable) : T2
+
Tensor of bools indicating if each input string fully matches the regex pattern specified.
+
+ +#### Type Constraints + +
+
T1 : tensor(string)
+
Inputs must be UTF-8 strings
+
T2 : tensor(bool)
+
Outputs are bools and are True where there is a full regex match and False otherwise.
+
+ + +#### Examples + +
+basic + +```python +node = onnx.helper.make_node( + "RegexFullMatch", + inputs=["X"], + outputs=["Y"], + pattern=r"www\.[\w.-]+\.\bcom\b", +) + +x = np.array(["www.google.com", "www.facebook.com", "www.bbc.co.uk"]).astype( + object +) +result = np.array([True, True, False]) +expect(node, inputs=[x], outputs=[result], name="test_regex_full_match_basic") +``` + +
+ + +
+match_email_domain + +```python +node = onnx.helper.make_node( + "RegexFullMatch", + inputs=["X"], + outputs=["Y"], + pattern=r"(\W|^)[\w.\-]{0,25}@(yahoo|gmail)\.com(\W|$)", +) + +x = np.array( + [ + ["account@gmail.com", "account@hotmail.com"], + ["not email", "account2@yahoo.com"], + ] +).astype(object) +result = np.array([[True, False], [False, True]]) +expect( + node, + inputs=[x], + outputs=[result], + name="test_regex_full_match_email_domain", +) +``` + +
+ + +
+match_empty + +```python +node = onnx.helper.make_node( + "RegexFullMatch", + inputs=["X"], + outputs=["Y"], + pattern=r"(\W|^)[\w.\-]{0,25}@(yahoo|gmail)\.com(\W|$)", +) + +x = np.array([[], []]).astype(object) +result = np.array([[], []]).astype(bool) +expect( + node, + inputs=[x], + outputs=[result], + name="test_regex_full_match_empty", +) +``` + +
+ + +### **Relu** + + Relu takes one input data (Tensor) and produces one output data + (Tensor) where the rectified linear function, y = max(0, x), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 13 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(int32), tensor(int8), tensor(int16), tensor(int64), tensor(float16), tensor(double), tensor(bfloat16)
+
Constrain input and output types to signed numeric tensors.
+
+ + +#### Examples + +
+relu + +```python +node = onnx.helper.make_node( + "Relu", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + +expect(node, inputs=[x], outputs=[y], name="test_relu") +``` + +
+ + +### **Reshape** + + Reshape the input tensor similar to numpy.reshape. + First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. + At most one dimension of the new shape can be -1. In this case, the value is + inferred from the size of the tensor and the remaining dimensions. A dimension + could also be 0, in which case the actual dimension value is unchanged (i.e. taken + from the input tensor). If 'allowzero' is set, and the new shape includes 0, the + dimension will be set explicitly to zero (i.e. not taken from input tensor). + Shape (second input) could be an empty shape, which means converting to a scalar. + The input tensor's shape and the output tensor's shape are required to have the same number of elements. + + If the attribute 'allowzero' is set, it is invalid for the specified shape to + contain both a zero value and -1, as the value of the dimension corresponding + to -1 cannot be determined uniquely. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 5, 13, 14, 19, 21, 23, 24 + +#### Attributes + +
+
allowzero : int (default is 0)
+
(Optional) By default, when any value in the 'shape' input is equal to zero the corresponding dimension value is copied from the input tensor dynamically. allowzero=1 indicates that if any value in the 'shape' input is set to zero, the zero value is honored, similar to NumPy.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
shape (non-differentiable) : tensor(int64)
+
Specified shape for output.
+
+ +#### Outputs + +
+
reshaped (differentiable) : T
+
Reshaped data.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types.
+
+ + +#### Examples + +
+allowzero + +```python +original_shape = [0, 3, 4] +test_cases = { + "allowzero_reordered": np.array([3, 4, 0], dtype=np.int64), +} +data = np.random.random_sample(original_shape).astype(np.float32) + +for test_name, shape in test_cases.items(): + node = onnx.helper.make_node( + "Reshape", + inputs=["data", "shape"], + outputs=["reshaped"], + allowzero=1, # if allowzero=1, final shape = (3, 4, 0) + # if allowzero=0, final shape = (3, 4, 4) + ) + + reshaped = reshape_reference_implementation(data, shape, allowzero=1) + + expect( + node, + inputs=[data, shape], + outputs=[reshaped], + name="test_reshape_" + test_name, + ) +``` + +
+ + +
+reshape + +```python +original_shape = [2, 3, 4] +test_cases = { + "reordered_all_dims": np.array([4, 2, 3], dtype=np.int64), + "reordered_last_dims": np.array([2, 4, 3], dtype=np.int64), + "reduced_dims": np.array([2, 12], dtype=np.int64), + "extended_dims": np.array([2, 3, 2, 2], dtype=np.int64), + "one_dim": np.array([24], dtype=np.int64), + "negative_dim": np.array([2, -1, 2], dtype=np.int64), + "negative_extended_dims": np.array([-1, 2, 3, 4], dtype=np.int64), + "zero_dim": np.array([2, 0, 4, 1], dtype=np.int64), + "zero_and_negative_dim": np.array([2, 0, 1, -1], dtype=np.int64), +} +data = np.random.random_sample(original_shape).astype(np.float32) + +for test_name, shape in test_cases.items(): + node = onnx.helper.make_node( + "Reshape", + inputs=["data", "shape"], + outputs=["reshaped"], + ) + + reshaped = reshape_reference_implementation(data, shape) + + expect( + node, + inputs=[data, shape], + outputs=[reshaped], + name="test_reshape_" + test_name, + ) +``` + +
+ + +### **Resize** + + Resize the input tensor. In general, it calculates every value in the output tensor as a weighted average of neighborhood (a.k.a. sampling locations) in the input tensor. + Each dimension value of the output tensor is: + ``` + output_dimension = floor(input_dimension * (roi_end - roi_start) * scale) + ``` + if input \"sizes\" is not specified. + +#### Version + +This version of the operator has been available since version 19 of the default ONNX operator set. + +Other versions of this operator: 10, 11, 13, 18 + +#### Attributes + +
+
antialias : int (default is 0)
+
If set to 1, "linear" and "cubic" interpolation modes will use an antialiasing filter when downscaling. Antialiasing is achieved by stretching the resampling filter by a factor max(1, 1 / scale), which means that when downsampling, more input pixels contribute to an output pixel.
+
axes : list of ints
+
If provided, it specifies a subset of axes that 'roi', 'scales' and 'sizes' refer to. If not provided, all axes are assumed [0, 1, ..., r-1], where r = rank(data). Non-specified dimensions are interpreted as non-resizable. Negative value means counting dimensions from the back. Accepted range is [-r, r-1], where r = rank(data). Behavior is undefined if an axis is repeated.
+
coordinate_transformation_mode : string (default is half_pixel)
+
+This attribute describes how to transform the coordinate in the resized tensor to the coordinate in the original tensor. + +The coordinate of each dimension is transformed individually. Let's describe a case using axis x as an example. +Denote `x_resized` as the coordinate of axis x in the resized tensor, + `x_original` as the coordinate of axis x in the original tensor, + `length_original` as the length of the original tensor in axis x, + `length_resized` as the length of the resized tensor in axis x, + `scale = length_resized / length_original`, + `output_width` the target length on the axis x which can be a fractional number when it is calculated out of a scale factor, + and `output_width_int` the effective output width as an integer. + +if coordinate_transformation_mode is `"half_pixel"`, +``` +x_original = (x_resized + 0.5) / scale - 0.5 +``` + +if coordinate_transformation_mode is `"half_pixel_symmetric"`, +``` +adjustment = output_width_int / output_width +center = input_width / 2 +offset = center * (1 - adjustment) +x_ori = offset + (x + 0.5) / scale - 0.5 +``` + +if coordinate_transformation_mode is `"pytorch_half_pixel"`, +``` +x_original = length_resized > 1 ? (x_resized + 0.5) / scale - 0.5 : 0 +``` + +if coordinate_transformation_mode is `"align_corners"`, +``` +x_original = x_resized * (length_original - 1) / (length_resized - 1) +``` + +if coordinate_transformation_mode is `"asymmetric"`, +``` +x_original = x_resized / scale +``` + +if coordinate_transformation_mode is `"tf_crop_and_resize"`, +``` +x_original = length_resized > 1 ? start_x * (length_original - 1) + x_resized * (end_x - start_x) * (length_original - 1) / (length_resized - 1) : 0.5 * (start_x + end_x) * (length_original - 1) +``` +.
+
cubic_coeff_a : float (default is -0.75)
+
The coefficient 'a' used in cubic interpolation. Two common choice are -0.5 (in some cases of TensorFlow) and -0.75 (in PyTorch). Check out Equation (4) in https://ieeexplore.ieee.org/document/1163711 for the details. This attribute is valid only if mode is "cubic".
+
exclude_outside : int (default is 0)
+
If set to 1, the weight of sampling locations outside the tensor will be set to 0 and the weight will be renormalized so that their sum is 1.0. The default value is 0.
+
extrapolation_value : float (default is 0.0)
+
When coordinate_transformation_mode is "tf_crop_and_resize" and x_original is outside the range [0, length_original - 1], this value is used as the corresponding output value. Default is 0.0f.
+
keep_aspect_ratio_policy : string (default is stretch)
+
+This attribute describes how to interpret the `sizes` input with regard to keeping the original aspect ratio of the input, and it is not applicable when +the `scales` input is used. + +Given a set of `sizes`, associated with a subset of `axes` (explicitly provided or default), and assuming `d = axes[i]`, with `i` being the index of the provided `sizes`. + +If `keep_aspect_ratio_policy` is `"stretch"`, the original aspect ratio is disregarded, and the input is resized to the specified size: +`out_size[d] = sizes[i]` + +If `keep_aspect_ratio_policy` is `"not_larger"`, the sizes are adjusted so that no extent of the output is larger than the specified size, while keeping the original aspect ratio: +``` +scale = Min(sizes[i] / in_size[d]) +out_size[d] = round_int(scale * in_size[d]) +``` + +If `keep_aspect_ratio_policy` is `"not_smaller"`, the sizes are adjusted so that no extent of the output is smaller than the specified size, while keeping the original aspect ratio: +``` +scale = Max(sizes[i] / in_size[d]) +out_size[d] = round_int(scale * in_size[d]) +``` + +For non-resizable axes (those not specified in `axes`), the output size will be equal to the input size. + +Note: `round_int` stands for computing the nearest integer value, rounding halfway cases up.
+
mode : string (default is nearest)
+
Three interpolation modes: "nearest" (default), "linear" and "cubic". The "linear" mode includes linear interpolation for 1D tensor and N-linear interpolation for N-D tensor (for example, bilinear interpolation for 2D tensor). The "cubic" mode includes cubic interpolation for 1D tensor and N-cubic interpolation for N-D tensor (for example, bicubic interpolation for 2D tensor).
+
nearest_mode : string (default is round_prefer_floor)
+
Four modes: "round_prefer_floor" (default, as known as round half down), "round_prefer_ceil" (as known as round half up), "floor", "ceil". Only used by nearest interpolation. It indicates how to get "nearest" pixel in input tensor from x_original, so this attribute is valid only if "mode" is "nearest".
+
+ +#### Inputs (1 - 4) + +
+
X (differentiable) : T1
+
N-D tensor
+
roi (optional, non-differentiable) : T2
+
1-D tensor given as [start1, ..., startN, end1, ..., endN], where N is the rank of X or the length of axes, if provided. The RoIs' coordinates are normalized in the coordinate system of the input image. It only takes effect when coordinate_transformation_mode is "tf_crop_and_resize"
+
scales (optional, non-differentiable) : tensor(float)
+
The scale array along each dimension. It takes value greater than 0. If it's less than 1, it's sampling down, otherwise, it's upsampling. The number of elements of 'scales' should be the same as the rank of input 'X' or the length of 'axes', if provided. One of 'scales' and 'sizes' MUST be specified and it is an error if both are specified. If 'sizes' is needed, the user can use an empty string as the name of 'scales' in this operator's input list.
+
sizes (optional, non-differentiable) : tensor(int64)
+
Target size of the output tensor. Its interpretation depends on the 'keep_aspect_ratio_policy' value.The number of elements of 'sizes' should be the same as the rank of input 'X', or the length of 'axes', if provided. Only one of 'scales' and 'sizes' can be specified.
+
+ +#### Outputs + +
+
Y (differentiable) : T1
+
N-D tensor after resizing
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input 'X' and output 'Y' to all tensor types.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Constrain roi type to float or double.
+
+ + +#### Examples + +
+resize_downsample_scales_cubic + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.8, 0.8], dtype=np.float32) + +# [[[[ 1.47119141 2.78125 4.08251953] +# [ 6.71142578 8.02148438 9.32275391] +# [11.91650391 13.2265625 14.52783203]]]] +output = interpolate_nd( + data, lambda x, _: cubic_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_cubic", +) +``` + +
+ + +
+resize_downsample_scales_cubic_A_n0p5_exclude_outside + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + cubic_coeff_a=-0.5, + exclude_outside=True, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.8, 0.8], dtype=np.float32) + +# [[[[ 1.36812675 2.6695014 4.0133367 ] +# [ 6.57362535 7.875 9.2188353 ] +# [11.94896657 13.25034122 14.59417652]]]] +output = interpolate_nd( + data, + lambda x, _: cubic_coeffs(x, A=-0.5), + scale_factors=scales, + exclude_outside=True, +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_cubic_A_n0p5_exclude_outside", +) +``` + +
+ + +
+resize_downsample_scales_cubic_align_corners + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + coordinate_transformation_mode="align_corners", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.8, 0.8], dtype=np.float32) + +# [[[[ 1. 2.39519159 3.79038317] +# [ 6.58076634 7.97595793 9.37114951] +# [12.16153268 13.55672427 14.95191585]]]] +output = interpolate_nd( + data, + lambda x, _: cubic_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="align_corners", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_cubic_align_corners", +) +``` + +
+ + +
+resize_downsample_scales_cubic_antialias + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + antialias=1, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.6, 0.6], dtype=np.float32) + +# [[[[ 2.5180721 4.2858863] +# [ 9.589329 11.357142 ]]]] +output = interpolate_nd( + data, cubic_coeffs_antialias, scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_cubic_antialias", +) +``` + +
+ + +
+resize_downsample_scales_linear + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.6, 0.6], dtype=np.float32) + +# [[[[2.6666665 4.3333331]]]] +output = interpolate_nd( + data, lambda x, _: linear_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_linear", +) +``` + +
+ + +
+resize_downsample_scales_linear_align_corners + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="align_corners", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.6, 0.6], dtype=np.float32) + +# [[[[1. 3.142857]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="align_corners", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_linear_align_corners", +) +``` + +
+ + +
+resize_downsample_scales_linear_antialias + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", + antialias=1, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.6, 0.6], dtype=np.float32) + +# [[[[ 2.875 4.5 ] +# [ 9.375 11. ]]]] +output = interpolate_nd( + data, linear_coeffs_antialias, scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_linear_antialias", +) +``` + +
+ + +
+resize_downsample_scales_linear_half_pixel_symmetric + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="half_pixel_symmetric", +) + +data = np.array([[[[1, 2, 3, 4]]]], dtype=np.float32) +scales = np.array([1.0, 1.0, 1.0, 0.6], dtype=np.float32) + +# [[[[1.6666667, 3.3333333]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="half_pixel_symmetric", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_linear_half_pixel_symmetric", +) +``` + +
+ + +
+resize_downsample_scales_nearest + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="nearest", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.6, 0.6], dtype=np.float32) + +# [[[[1. 3.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_nearest", +) +``` + +
+ + +
+resize_downsample_sizes_cubic + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="cubic", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 3, 3], dtype=np.int64) + +# [[[[ 1.63078704 3.00462963 4.37847222] +# [ 7.12615741 8.5 9.87384259] +# [12.62152778 13.99537037 15.36921296]]]] +output = interpolate_nd( + data, lambda x, _: cubic_coeffs(x), output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_cubic", +) +``` + +
+ + +
+resize_downsample_sizes_cubic_antialias + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="cubic", + antialias=1, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 3, 3], dtype=np.int64) + +# [[[[ 1.7750092 3.1200073 4.4650054] +# [ 7.1550016 8.5 9.844998 ] +# [12.534994 13.8799925 15.224991 ]]]] +output = interpolate_nd(data, cubic_coeffs_antialias, output_size=sizes).astype( + np.float32 +) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_cubic_antialias", +) +``` + +
+ + +
+resize_downsample_sizes_linear_antialias + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="linear", + antialias=1, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 3, 3], dtype=np.int64) + +# [[[[ 2.3636363 3.590909 4.818182 ] +# [ 7.2727275 8.5 9.727273 ] +# [12.181818 13.409091 14.636364 ]]]] +output = interpolate_nd( + data, linear_coeffs_antialias, output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_linear_antialias", +) +``` + +
+ + +
+resize_downsample_sizes_linear_pytorch_half_pixel + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="pytorch_half_pixel", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 3, 1], dtype=np.int64) + +# [[[[ 1.6666666] +# [ 7. ] +# [12.333333 ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + output_size=sizes, + coordinate_transformation_mode="pytorch_half_pixel", +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_linear_pytorch_half_pixel", +) +``` + +
+ + +
+resize_downsample_sizes_nearest + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 1, 3], dtype=np.int64) + +# [[[[1. 2. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_nearest", +) +``` + +
+ + +
+resize_downsample_sizes_nearest_not_larger + +```python +keep_aspect_ratio_policy = "not_larger" +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 3], dtype=np.int64) # Results in 1x2 + +# [[[[1. 3.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x), + output_size=sizes, + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_nearest_not_larger", +) +``` + +
+ + +
+resize_downsample_sizes_nearest_not_smaller + +```python +keep_aspect_ratio_policy = "not_smaller" +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 3], dtype=np.int64) # Results in 2x3 + +# [[[[1. 2. 4.] +# [5. 6. 8.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x), + output_size=sizes, + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_nearest_not_smaller", +) +``` + +
+ + +
+resize_tf_crop_and_resize + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "roi", "", "sizes"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="tf_crop_and_resize", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +# Note: for some rois, the result may be different with that of TF for inaccurate floating point +roi = np.array([0, 0, 0.4, 0.6, 1, 1, 0.6, 0.8], dtype=np.float32) +sizes = np.array([1, 1, 3, 3], dtype=np.int64) + +# [[[[ 7.6000004 7.9 8.2 ] +# [ 8.8 9.1 9.400001 ] +# [10. 10.3 10.6 ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + output_size=sizes, + roi=roi, + coordinate_transformation_mode="tf_crop_and_resize", +).astype(np.float32) + +expect( + node, + inputs=[data, roi, sizes], + outputs=[output], + name="test_resize_tf_crop_and_resize", +) +``` + +
+ + +
+resize_tf_crop_and_resize_axes_2_3 + +```python +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "roi", "", "sizes"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="tf_crop_and_resize", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +# Note: for some rois, the result may be different with that of TF for inaccurate floating point +roi = np.array([0.4, 0.6, 0.6, 0.8], dtype=np.float32) +sizes = np.array([3, 3], dtype=np.int64) + +# [[[[ 7.6000004 7.9 8.2 ] +# [ 8.8 9.1 9.400001 ] +# [10. 10.3 10.6 ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + output_size=sizes, + roi=roi, + axes=axes, + coordinate_transformation_mode="tf_crop_and_resize", +).astype(np.float32) + +expect( + node, + inputs=[data, roi, sizes], + outputs=[output], + name="test_resize_tf_crop_and_resize_axes_2_3", +) +``` + +
+ + +
+resize_tf_crop_and_resize_axes_3_2 + +```python +axes = [3, 2] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "roi", "", "sizes"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="tf_crop_and_resize", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +# Note: for some rois, the result may be different with that of TF for inaccurate floating point +roi = np.array([0.6, 0.4, 0.8, 0.6], dtype=np.float32) +sizes = np.array([3, 3], dtype=np.int64) + +# [[[[ 7.6000004 7.9 8.2 ] +# [ 8.8 9.1 9.400001 ] +# [10. 10.3 10.6 ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + output_size=sizes, + roi=roi, + axes=axes, + coordinate_transformation_mode="tf_crop_and_resize", +).astype(np.float32) + +expect( + node, + inputs=[data, roi, sizes], + outputs=[output], + name="test_resize_tf_crop_and_resize_axes_3_2", +) +``` + +
+ + +
+resize_tf_crop_and_resize_extrapolation_value + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "roi", "", "sizes"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="tf_crop_and_resize", + extrapolation_value=10.0, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +# Note: for some rois, the result may be different with that of TF for inaccurate floating point +roi = np.array([0, 0, 0.4, 0.6, 1, 1, 1.2, 1.7], dtype=np.float32) +sizes = np.array([1, 1, 3, 3], dtype=np.int64) + +# [[[[ 7.6000004 10. 10. ] +# [12.400001 10. 10. ] +# [10. 10. 10. ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + output_size=sizes, + roi=roi, + coordinate_transformation_mode="tf_crop_and_resize", + extrapolation_value=10.0, +).astype(np.float32) + +expect( + node, + inputs=[data, roi, sizes], + outputs=[output], + name="test_resize_tf_crop_and_resize_extrapolation_value", +) +``` + +
+ + +
+resize_upsample_scales_cubic + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[ 0.47265625 0.76953125 1.24609375 1.875 2.28125 +# 2.91015625 3.38671875 3.68359375] +# [ 1.66015625 1.95703125 2.43359375 3.0625 3.46875 +# 4.09765625 4.57421875 4.87109375] +# [ 3.56640625 3.86328125 4.33984375 4.96875 5.375 +# 6.00390625 6.48046875 6.77734375] +# [ 6.08203125 6.37890625 6.85546875 7.484375 7.890625 +# 8.51953125 8.99609375 9.29296875] +# [ 7.70703125 8.00390625 8.48046875 9.109375 9.515625 +# 10.14453125 10.62109375 10.91796875] +# [10.22265625 10.51953125 10.99609375 11.625 12.03125 +# 12.66015625 13.13671875 13.43359375] +# [12.12890625 12.42578125 12.90234375 13.53125 13.9375 +# 14.56640625 15.04296875 15.33984375] +# [13.31640625 13.61328125 14.08984375 14.71875 15.125 +# 15.75390625 16.23046875 16.52734375]]]] +output = interpolate_nd( + data, lambda x, _: cubic_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_cubic", +) +``` + +
+ + +
+resize_upsample_scales_cubic_A_n0p5_exclude_outside + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + cubic_coeff_a=-0.5, + exclude_outside=True, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[ 0.55882353 0.81494204 1.35698249 1.89705882 2.39705882 +# 2.93713516 3.47917561 3.73529412] +# [ 1.58329755 1.83941606 2.38145651 2.92153285 3.42153285 +# 3.96160918 4.50364964 4.75976814] +# [ 3.75145936 4.00757787 4.54961832 5.08969466 5.58969466 +# 6.12977099 6.67181144 6.92792995] +# [ 5.91176471 6.16788321 6.70992366 7.25 7.75 +# 8.29007634 8.83211679 9.08823529] +# [ 7.91176471 8.16788321 8.70992366 9.25 9.75 +# 10.29007634 10.83211679 11.08823529] +# [10.07207005 10.32818856 10.87022901 11.41030534 11.91030534 +# 12.45038168 12.99242213 13.24854064] +# [12.24023186 12.49635036 13.03839082 13.57846715 14.07846715 +# 14.61854349 15.16058394 15.41670245] +# [13.26470588 13.52082439 14.06286484 14.60294118 15.10294118 +# 15.64301751 16.18505796 16.44117647]]]] +output = interpolate_nd( + data, + lambda x, _: cubic_coeffs(x, A=-0.5), + scale_factors=scales, + exclude_outside=True, +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_cubic_A_n0p5_exclude_outside", +) +``` + +
+ + +
+resize_upsample_scales_cubic_align_corners + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + coordinate_transformation_mode="align_corners", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[ 1. 1.34110787 1.80029155 2.32944606 2.67055394 +# 3.19970845 3.65889213 4. ] +# [ 2.36443149 2.70553936 3.16472303 3.69387755 4.03498542 +# 4.56413994 5.02332362 5.36443149] +# [ 4.20116618 4.54227405 5.00145773 5.53061224 5.87172012 +# 6.40087464 6.86005831 7.20116618] +# [ 6.31778426 6.65889213 7.1180758 7.64723032 7.98833819 +# 8.51749271 8.97667638 9.31778426] +# [ 7.68221574 8.02332362 8.48250729 9.01166181 9.35276968 +# 9.8819242 10.34110787 10.68221574] +# [ 9.79883382 10.13994169 10.59912536 11.12827988 11.46938776 +# 11.99854227 12.45772595 12.79883382] +# [11.63556851 11.97667638 12.43586006 12.96501458 13.30612245 +# 13.83527697 14.29446064 14.63556851] +# [13. 13.34110787 13.80029155 14.32944606 14.67055394 +# 15.19970845 15.65889213 16. ]]]] +output = interpolate_nd( + data, + lambda x, _: cubic_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="align_corners", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_cubic_align_corners", +) +``` + +
+ + +
+resize_upsample_scales_cubic_asymmetric + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + coordinate_transformation_mode="asymmetric", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[ 1. 1.40625 2. 2.5 3. 3.59375 4. +# 4.09375] +# [ 2.625 3.03125 3.625 4.125 4.625 5.21875 5.625 +# 5.71875] +# [ 5. 5.40625 6. 6.5 7. 7.59375 8. +# 8.09375] +# [ 7. 7.40625 8. 8.5 9. 9.59375 10. +# 10.09375] +# [ 9. 9.40625 10. 10.5 11. 11.59375 12. +# 12.09375] +# [11.375 11.78125 12.375 12.875 13.375 13.96875 14.375 +# 14.46875] +# [13. 13.40625 14. 14.5 15. 15.59375 16. +# 16.09375] +# [13.375 13.78125 14.375 14.875 15.375 15.96875 16.375 +# 16.46875]]]] +output = interpolate_nd( + data, + lambda x, _: cubic_coeffs(x, A=-0.75), + scale_factors=scales, + coordinate_transformation_mode="asymmetric", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_cubic_asymmetric", +) +``` + +
+ + +
+resize_upsample_scales_linear + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[1. 1.25 1.75 2. ] +# [1.5 1.75 2.25 2.5 ] +# [2.5 2.75 3.25 3.5 ] +# [3. 3.25 3.75 4. ]]]] +output = interpolate_nd( + data, lambda x, _: linear_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_linear", +) +``` + +
+ + +
+resize_upsample_scales_linear_align_corners + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="align_corners", +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[1. 1.33333333 1.66666667 2. ] +# [1.66666667 2. 2.33333333 2.66666667] +# [2.33333333 2.66666667 3. 3.33333333] +# [3. 3.33333333 3.66666667 4. ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="align_corners", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_linear_align_corners", +) +``` + +
+ + +
+resize_upsample_scales_linear_half_pixel_symmetric + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="half_pixel_symmetric", +) + +data = np.array([[[[1, 2], [3, 4]]]], dtype=np.float32) +scales = np.array([1.0, 1.0, 2.3, 2.94], dtype=np.float32) + +# [[[[1. , 1.15986395, 1.5 , 1.84013605, 2. ], +# [1.56521738, 1.72508133, 2.06521738, 2.40535343, 2.56521738], +# [2.43478262, 2.59464657, 2.93478262, 3.27491867, 3.43478262], +# [3. , 3.15986395, 3.5 , 3.84013605, 4. ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="half_pixel_symmetric", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_linear_half_pixel_symmetric", +) +``` + +
+ + +
+resize_upsample_scales_nearest + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="nearest", +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 3.0], dtype=np.float32) + +# [[[[1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 2. 2. 2.] +# [3. 3. 3. 4. 4. 4.] +# [3. 3. 3. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_nearest", +) +``` + +
+ + +
+resize_upsample_scales_nearest_axes_2_3 + +```python +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="nearest", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([2.0, 3.0], dtype=np.float32) + +# [[[[1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 2. 2. 2.] +# [3. 3. 3. 4. 4. 4.] +# [3. 3. 3. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), scale_factors=scales, axes=axes +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_nearest_axes_2_3", +) +``` + +
+ + +
+resize_upsample_scales_nearest_axes_3_2 + +```python +axes = [3, 2] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="nearest", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([3.0, 2.0], dtype=np.float32) + +# [[[[1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 2. 2. 2.] +# [3. 3. 3. 4. 4. 4.] +# [3. 3. 3. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), scale_factors=scales, axes=axes +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_nearest_axes_3_2", +) +``` + +
+ + +
+resize_upsample_sizes_cubic + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="cubic", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 9, 10], dtype=np.int64) + +# [[[[ 0.45507922 0.64057922 0.97157922 1.42257922 1.90732922 +# 2.22332922 2.70807922 3.15907922 3.49007922 3.67557922] +# [ 1.39437963 1.57987963 1.91087963 2.36187963 2.84662963 +# 3.16262963 3.64737963 4.09837963 4.42937963 4.61487963] +# [ 2.95130693 3.13680693 3.46780693 3.91880693 4.40355693 +# 4.71955693 5.20430693 5.65530693 5.98630693 6.17180693] +# [ 5.20525069 5.39075069 5.72175069 6.17275069 6.65750069 +# 6.97350069 7.45825069 7.90925069 8.24025069 8.42575069] +# [ 6.88975 7.07525 7.40625 7.85725 8.342 +# 8.658 9.14275 9.59375 9.92475 10.11025 ] +# [ 8.57424931 8.75974931 9.09074931 9.54174931 10.02649931 +# 10.34249931 10.82724931 11.27824931 11.60924931 11.79474931] +# [10.82819307 11.01369307 11.34469307 11.79569307 12.28044307 +# 12.59644307 13.08119307 13.53219307 13.86319307 14.04869307] +# [12.38512037 12.57062037 12.90162037 13.35262037 13.83737037 +# 14.15337037 14.63812037 15.08912037 15.42012037 15.60562037] +# [13.32442078 13.50992078 13.84092078 14.29192078 14.77667078 +# 15.09267078 15.57742078 16.02842078 16.35942078 16.54492078]]]] +output = interpolate_nd( + data, lambda x, _: cubic_coeffs(x), output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_cubic", +) +``` + +
+ + +
+resize_upsample_sizes_nearest + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 7, 8], dtype=np.int64) + +# [[[[1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest", +) +``` + +
+ + +
+resize_upsample_sizes_nearest_axes_2_3 + +```python +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([7, 8], dtype=np.int64) + +# [[[[1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), output_size=sizes, axes=axes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_axes_2_3", +) +``` + +
+ + +
+resize_upsample_sizes_nearest_axes_3_2 + +```python +axes = [3, 2] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([8, 7], dtype=np.int64) + +# [[[[1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), output_size=sizes, axes=axes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_axes_3_2", +) +``` + +
+ + +
+resize_upsample_sizes_nearest_ceil_half_pixel + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + coordinate_transformation_mode="half_pixel", + nearest_mode="ceil", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 8, 8], dtype=np.int64) + +# [[[[ 1. 2. 2. 3. 3. 4. 4. 4.] +# [ 5. 6. 6. 7. 7. 8. 8. 8.] +# [ 5. 6. 6. 7. 7. 8. 8. 8.] +# [ 9. 10. 10. 11. 11. 12. 12. 12.] +# [ 9. 10. 10. 11. 11. 12. 12. 12.] +# [13. 14. 14. 15. 15. 16. 16. 16.] +# [13. 14. 14. 15. 15. 16. 16. 16.] +# [13. 14. 14. 15. 15. 16. 16. 16.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x, mode="ceil"), output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_ceil_half_pixel", +) +``` + +
+ + +
+resize_upsample_sizes_nearest_floor_align_corners + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + coordinate_transformation_mode="align_corners", + nearest_mode="floor", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 8, 8], dtype=np.int64) + +# [[[[ 1. 1. 1. 2. 2. 3. 3. 4.] +# [ 1. 1. 1. 2. 2. 3. 3. 4.] +# [ 1. 1. 1. 2. 2. 3. 3. 4.] +# [ 5. 5. 5. 6. 6. 7. 7. 8.] +# [ 5. 5. 5. 6. 6. 7. 7. 8.] +# [ 9. 9. 9. 10. 10. 11. 11. 12.] +# [ 9. 9. 9. 10. 10. 11. 11. 12.] +# [13. 13. 13. 14. 14. 15. 15. 16.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x, mode="floor"), + output_size=sizes, + coordinate_transformation_mode="align_corners", +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_floor_align_corners", +) +``` + +
+ + +
+resize_upsample_sizes_nearest_not_larger + +```python +keep_aspect_ratio_policy = "not_larger" +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([7, 8], dtype=np.int64) # Results in 7x7 + +# [[[[1. 1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2.] +# [3. 3. 3. 3. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x), + output_size=sizes, + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_not_larger", +) +``` + +
+ + +
+resize_upsample_sizes_nearest_not_smaller + +```python +keep_aspect_ratio_policy = "not_smaller" +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([7, 8], dtype=np.int64) # Results in 8x8 + +# [[[[1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x), + output_size=sizes, + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_not_smaller", +) +``` + +
+ + +
+resize_upsample_sizes_nearest_round_prefer_ceil_asymmetric + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + coordinate_transformation_mode="asymmetric", + nearest_mode="round_prefer_ceil", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 8, 8], dtype=np.int64) + +# [[[[ 1. 2. 2. 3. 3. 4. 4. 4.] +# [ 5. 6. 6. 7. 7. 8. 8. 8.] +# [ 5. 6. 6. 7. 7. 8. 8. 8.] +# [ 9. 10. 10. 11. 11. 12. 12. 12.] +# [ 9. 10. 10. 11. 11. 12. 12. 12.] +# [13. 14. 14. 15. 15. 16. 16. 16.] +# [13. 14. 14. 15. 15. 16. 16. 16.] +# [13. 14. 14. 15. 15. 16. 16. 16.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x, mode="round_prefer_ceil"), + output_size=sizes, + coordinate_transformation_mode="asymmetric", +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_round_prefer_ceil_asymmetric", +) +``` + +
+ + +### **ReverseSequence** + + Reverse batch of sequences having different lengths specified by `sequence_lens`. + + For each slice i iterating on batch axis, the operator reverses the first sequence_lens[i] elements on time axis, + and copies elements whose index's beyond sequence_lens[i] to the output. So the output slice i contains reversed + sequences on the first sequence_lens[i] elements, then have original values copied for the other elements. + + Example 1: + input = [[0.0, 4.0, 8.0, 12.0], + [1.0, 5.0, 9.0, 13.0], + [2.0, 6.0, 10.0, 14.0], + [3.0, 7.0, 11.0, 15.0]] + sequence_lens = [4, 3, 2, 1] + time_axis = 0 + batch_axis = 1 + + output = [[3.0, 6.0, 9.0, 12.0], + [2.0, 5.0, 8.0, 13.0], + [1.0, 4.0, 10.0, 14.0], + [0.0, 7.0, 11.0, 15.0]] + + Example 2: + input = [[0.0, 1.0, 2.0, 3.0 ], + [4.0, 5.0, 6.0, 7.0 ], + [8.0, 9.0, 10.0, 11.0], + [12.0, 13.0, 14.0, 15.0]] + sequence_lens = [1, 2, 3, 4] + time_axis = 1 + batch_axis = 0 + + output = [[0.0, 1.0, 2.0, 3.0 ], + [5.0, 4.0, 6.0, 7.0 ], + [10.0, 9.0, 8.0, 11.0], + [15.0, 14.0, 13.0, 12.0]] + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
batch_axis : int (default is 1)
+
(Optional) Specify which axis is batch axis. Must be one of 1 (default), or 0.
+
time_axis : int (default is 0)
+
(Optional) Specify which axis is time axis. Must be one of 0 (default), or 1.
+
+ +#### Inputs + +
+
input : T
+
Tensor of rank r >= 2.
+
sequence_lens : tensor(int64)
+
Tensor specifying lengths of the sequences in a batch. It has shape `[batch_size]`.
+
+ +#### Outputs + +
+
Y : T
+
Tensor with same shape of input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input and output types can be of any tensor type.
+
+ + +#### Examples + +
+reversesequence_batch + +```python +node = onnx.helper.make_node( + "ReverseSequence", + inputs=["x", "sequence_lens"], + outputs=["y"], + time_axis=1, + batch_axis=0, +) +x = np.array( + [ + [0.0, 1.0, 2.0, 3.0], + [4.0, 5.0, 6.0, 7.0], + [8.0, 9.0, 10.0, 11.0], + [12.0, 13.0, 14.0, 15.0], + ], + dtype=np.float32, +) +sequence_lens = np.array([1, 2, 3, 4], dtype=np.int64) + +y = np.array( + [ + [0.0, 1.0, 2.0, 3.0], + [5.0, 4.0, 6.0, 7.0], + [10.0, 9.0, 8.0, 11.0], + [15.0, 14.0, 13.0, 12.0], + ], + dtype=np.float32, +) + +expect( + node, + inputs=[x, sequence_lens], + outputs=[y], + name="test_reversesequence_batch", +) +``` + +
+ + +
+reversesequence_time + +```python +node = onnx.helper.make_node( + "ReverseSequence", + inputs=["x", "sequence_lens"], + outputs=["y"], + time_axis=0, + batch_axis=1, +) +x = np.array( + [ + [0.0, 4.0, 8.0, 12.0], + [1.0, 5.0, 9.0, 13.0], + [2.0, 6.0, 10.0, 14.0], + [3.0, 7.0, 11.0, 15.0], + ], + dtype=np.float32, +) +sequence_lens = np.array([4, 3, 2, 1], dtype=np.int64) + +y = np.array( + [ + [3.0, 6.0, 9.0, 12.0], + [2.0, 5.0, 8.0, 13.0], + [1.0, 4.0, 10.0, 14.0], + [0.0, 7.0, 11.0, 15.0], + ], + dtype=np.float32, +) + +expect( + node, + inputs=[x, sequence_lens], + outputs=[y], + name="test_reversesequence_time", +) +``` + +
+ + +### **RoiAlign** + + Region of Interest (RoI) align operation described in the + [Mask R-CNN paper](https://arxiv.org/abs/1703.06870). + RoiAlign consumes an input tensor X and region of interests (rois) + to apply pooling across each RoI; it produces a 4-D tensor of shape + (num_rois, C, output_height, output_width). + + RoiAlign is proposed to avoid the misalignment by removing + quantizations while converting from original image into feature + map and from feature map into RoI feature; in each ROI bin, + the value of the sampled locations are computed directly + through bilinear interpolation. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 10, 16 + +#### Attributes + +
+
coordinate_transformation_mode : string (default is half_pixel)
+
Allowed values are 'half_pixel' and 'output_half_pixel'. Use the value 'half_pixel' to pixel shift the input coordinates by -0.5 (the recommended behavior). Use the value 'output_half_pixel' to omit the pixel shift for the input (use this for a backward-compatible behavior).
+
mode : string (default is avg)
+
The pooling method. Two modes are supported: 'avg' and 'max'. Default is 'avg'.
+
output_height : int (default is 1)
+
default 1; Pooled output Y's height.
+
output_width : int (default is 1)
+
default 1; Pooled output Y's width.
+
sampling_ratio : int (default is 0)
+
Number of sampling points in the interpolation grid used to compute the output value of each pooled output bin. If > 0, then exactly sampling_ratio x sampling_ratio grid points are used. If == 0, then an adaptive number of grid points are used (computed as ceil(roi_width / output_width), and likewise for height). Default is 0.
+
spatial_scale : float (default is 1.0)
+
Multiplicative spatial scale factor to translate ROI coordinates from their input spatial scale to the scale used when pooling, i.e., spatial scale of the input feature map X relative to the input image. E.g.; default is 1.0f.
+
+ +#### Inputs + +
+
X : T1
+
Input data tensor from the previous operator; 4-D feature map of shape (N, C, H, W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
+
rois : T1
+
RoIs (Regions of Interest) to pool over; rois is 2-D input of shape (num_rois, 4) given as [[x1, y1, x2, y2], ...]. The RoIs' coordinates are in the coordinate system of the input image. Each coordinate set has a 1:1 correspondence with the 'batch_indices' input.
+
batch_indices : T2
+
1-D tensor of shape (num_rois,) with each element denoting the index of the corresponding image in the batch.
+
+ +#### Outputs + +
+
Y : T1
+
RoI pooled output, 4-D tensor of shape (num_rois, C, output_height, output_width). The r-th batch element Y[r-1] is a pooled feature map corresponding to the r-th RoI X[r-1].
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain types to float tensors.
+
T2 : tensor(int64)
+
Constrain types to int tensors.
+
+ + +#### Examples + +
+roialign_aligned_false + +```python +node = onnx.helper.make_node( + "RoiAlign", + inputs=["X", "rois", "batch_indices"], + outputs=["Y"], + spatial_scale=1.0, + output_height=5, + output_width=5, + sampling_ratio=2, + coordinate_transformation_mode="output_half_pixel", +) + +X, batch_indices, rois = get_roi_align_input_values() +# (num_rois, C, output_height, output_width) +Y = np.array( + [ + [ + [ + [0.4664, 0.4466, 0.3405, 0.5688, 0.6068], + [0.3714, 0.4296, 0.3835, 0.5562, 0.3510], + [0.2768, 0.4883, 0.5222, 0.5528, 0.4171], + [0.4713, 0.4844, 0.6904, 0.4920, 0.8774], + [0.6239, 0.7125, 0.6289, 0.3355, 0.3495], + ] + ], + [ + [ + [0.3022, 0.4305, 0.4696, 0.3978, 0.5423], + [0.3656, 0.7050, 0.5165, 0.3172, 0.7015], + [0.2912, 0.5059, 0.6476, 0.6235, 0.8299], + [0.5916, 0.7389, 0.7048, 0.8372, 0.8893], + [0.6227, 0.6153, 0.7097, 0.6154, 0.4585], + ] + ], + [ + [ + [0.2384, 0.3379, 0.3717, 0.6100, 0.7601], + [0.3767, 0.3785, 0.7147, 0.9243, 0.9727], + [0.5749, 0.5826, 0.5709, 0.7619, 0.8770], + [0.5355, 0.2566, 0.2141, 0.2796, 0.3600], + [0.4365, 0.3504, 0.2887, 0.3661, 0.2349], + ] + ], + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, rois, batch_indices], + outputs=[Y], + name="test_roialign_aligned_false", +) +``` + +
+ + +
+roialign_aligned_true + +```python +node = onnx.helper.make_node( + "RoiAlign", + inputs=["X", "rois", "batch_indices"], + outputs=["Y"], + spatial_scale=1.0, + output_height=5, + output_width=5, + sampling_ratio=2, + coordinate_transformation_mode="half_pixel", +) + +X, batch_indices, rois = get_roi_align_input_values() +# (num_rois, C, output_height, output_width) +Y = np.array( + [ + [ + [ + [0.5178, 0.3434, 0.3229, 0.4474, 0.6344], + [0.4031, 0.5366, 0.4428, 0.4861, 0.4023], + [0.2512, 0.4002, 0.5155, 0.6954, 0.3465], + [0.3350, 0.4601, 0.5881, 0.3439, 0.6849], + [0.4932, 0.7141, 0.8217, 0.4719, 0.4039], + ] + ], + [ + [ + [0.3070, 0.2187, 0.3337, 0.4880, 0.4870], + [0.1871, 0.4914, 0.5561, 0.4192, 0.3686], + [0.1433, 0.4608, 0.5971, 0.5310, 0.4982], + [0.2788, 0.4386, 0.6022, 0.7000, 0.7524], + [0.5774, 0.7024, 0.7251, 0.7338, 0.8163], + ] + ], + [ + [ + [0.2393, 0.4075, 0.3379, 0.2525, 0.4743], + [0.3671, 0.2702, 0.4105, 0.6419, 0.8308], + [0.5556, 0.4543, 0.5564, 0.7502, 0.9300], + [0.6626, 0.5617, 0.4813, 0.4954, 0.6663], + [0.6636, 0.3721, 0.2056, 0.1928, 0.2478], + ] + ], + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, rois, batch_indices], + outputs=[Y], + name="test_roialign_aligned_true", +) +``` + +
+ + +
+roialign_mode_max + +```python +X = np.array( + [ + [ + [ + [ + 0.2764, + 0.715, + 0.1958, + 0.3416, + 0.4638, + 0.0259, + 0.2963, + 0.6518, + 0.4856, + 0.725, + ], + [ + 0.9637, + 0.0895, + 0.2919, + 0.6753, + 0.0234, + 0.6132, + 0.8085, + 0.5324, + 0.8992, + 0.4467, + ], + [ + 0.3265, + 0.8479, + 0.9698, + 0.2471, + 0.9336, + 0.1878, + 0.4766, + 0.4308, + 0.34, + 0.2162, + ], + [ + 0.0206, + 0.172, + 0.2155, + 0.4394, + 0.0653, + 0.3406, + 0.7724, + 0.3921, + 0.2541, + 0.5799, + ], + [ + 0.4062, + 0.2194, + 0.4473, + 0.4687, + 0.7109, + 0.9327, + 0.9815, + 0.632, + 0.1728, + 0.6119, + ], + [ + 0.3097, + 0.1283, + 0.4984, + 0.5068, + 0.4279, + 0.0173, + 0.4388, + 0.043, + 0.4671, + 0.7119, + ], + [ + 0.1011, + 0.8477, + 0.4726, + 0.1777, + 0.9923, + 0.4042, + 0.1869, + 0.7795, + 0.9946, + 0.9689, + ], + [ + 0.1366, + 0.3671, + 0.7011, + 0.6234, + 0.9867, + 0.5585, + 0.6985, + 0.5609, + 0.8788, + 0.9928, + ], + [ + 0.5697, + 0.8511, + 0.6711, + 0.9406, + 0.8751, + 0.7496, + 0.165, + 0.1049, + 0.1559, + 0.2514, + ], + [ + 0.7012, + 0.4056, + 0.7879, + 0.3461, + 0.0415, + 0.2998, + 0.5094, + 0.3727, + 0.5482, + 0.0502, + ], + ] + ] + ], + dtype=np.float32, +) +rois = np.array( + [[0.0, 0.0, 9.0, 9.0], [0.0, 5.0, 4.0, 9.0], [5.0, 5.0, 9.0, 9.0]], + dtype=np.float32, +) +batch_indices = np.array([0, 0, 0], dtype=np.int64) + +Y = np.array( + [ + [ + [ + [0.3445228, 0.37310338, 0.37865096, 0.446696, 0.37991184], + [0.4133513, 0.5455125, 0.6651902, 0.55805874, 0.27110294], + [0.21223956, 0.40924096, 0.8417618, 0.792561, 0.37196714], + [0.46835402, 0.39741728, 0.8012819, 0.4969306, 0.5495158], + [0.3595896, 0.5196813, 0.5403741, 0.23814403, 0.19992709], + ] + ], + [ + [ + [0.30517197, 0.5086199, 0.3189761, 0.4054401, 0.47630402], + [0.50862, 0.8477, 0.37808004, 0.24936005, 0.79384017], + [0.17620805, 0.29368007, 0.44870415, 0.4987201, 0.63148826], + [0.51066005, 0.8511, 0.5368801, 0.9406, 0.70008016], + [0.4487681, 0.51066035, 0.5042561, 0.5643603, 0.42004836], + ] + ], + [ + [ + [0.21062402, 0.3510401, 0.37416005, 0.5967599, 0.46507207], + [0.32336006, 0.31180006, 0.6236001, 0.9946, 0.7751202], + [0.35744014, 0.5588001, 0.35897616, 0.7030401, 0.6353923], + [0.5996801, 0.27940005, 0.17948808, 0.35152006, 0.31769615], + [0.3598083, 0.40752012, 0.2385281, 0.43856013, 0.26313624], + ] + ], + ], + dtype=np.float32, +) + +node = onnx.helper.make_node( + "RoiAlign", + inputs=["X", "rois", "batch_indices"], + mode="max", + outputs=["Y"], + spatial_scale=1.0, + output_height=5, + output_width=5, + sampling_ratio=2, + coordinate_transformation_mode="output_half_pixel", +) + +expect( + node, + inputs=[X, rois, batch_indices], + outputs=[Y], + name="test_roialign_mode_max", +) +``` + +
+ + +### **RotaryEmbedding** + + RotaryEmbedding is the implementation of rotary positional embeddings (RoPE) based on the paper https://arxiv.org/pdf/2104.09864. + The key advantage of RoPE is that it allows the model to understand both the absolute position of a token and the relative distances + between tokens. This is achieved through a rotational mechanism where the extent of rotation is computed based on the token's absolute position (position_ids). + + The rotational mechanism is defined by sine and cosine functions that are used to represent the rotation angles. + For each token in the sequence, its positional embedding is computed by rotating its embedding vector. This is done by splitting the + embedding vector either into two halves or interleaving every alternate token and applying the rotation matrix to each half of the embedding vector. + The rotation matrix is parameterized by the token's position in the sequence. The rotated halves of the embedding vector are concatenated + to form the final positional embedding for each token. The rotated positional embeddings are used in the self-attention mechanism. + The rotation ensures that the model captures both absolute and relative positional information. + + Rotary embeddings are defined using the following algorithm: + + ```python + def rotary_embedding( + input: np.ndarray, + cos_cache: np.ndarray, + sin_cache: np.ndarray, + position_ids: np.ndarray | None = None, + interleaved=None, + rotary_embedding_dim=None, + num_heads=None, + ) -> np.ndarray: + original_input_shape = input.shape + # First ensure input to be processed has shape [batch_size, seq_len, num_heads, head_size] + if len(input.shape) == 4: + input = np.transpose(input, (0, 2, 1, 3)) + batch_size = input.shape[0] + sequence_length = input.shape[1] + if len(input.shape) == 3: + hidden_size = input.shape[2] + assert num_heads != 0 + head_size = int(hidden_size / num_heads) + new_shape = [batch_size, sequence_length, num_heads, head_size] + input = np.reshape(input, new_shape) + assert len(input.shape) == 4 + head_size = input.shape[3] + + # Fully or partially perform rotation on input based on rotary_embedding_dim attribute + if rotary_embedding_dim is None or rotary_embedding_dim == 0: + # If rotary_embedding_dim not provided, perform full rotation by using head_size + rotary_embedding_dim = head_size + x_rotate = input[:, :, :, :rotary_embedding_dim] + x_not_rotate = input[:, :, :, rotary_embedding_dim:] + rotary_embedding_dim_half = int(rotary_embedding_dim / 2) + + # Retrieve sin and cos caches using position ids + if position_ids is not None: + cos_cache = cos_cache[ + position_ids + ] # Shape: [batch_size, sequence_length, rotary_embedding_dim/2] + sin_cache = sin_cache[ + position_ids + ] # Shape: [batch_size, sequence_length, rotary_embedding_dim/2] + + # Shape: [batch_size, sequence_length, rotary_embedding_dim/2] + if cos_cache.shape[-1] != rotary_embedding_dim_half: + raise ValueError( + f"Last dimension of cos cache ({cos_cache.shape[-1]}) does not match rotary_embedding_dim/2 ({rotary_embedding_dim_half})." + ) + if sin_cache.shape[-1] != rotary_embedding_dim_half: + raise ValueError( + f"Last dimension of sin cache ({sin_cache.shape[-1]}) does not match rotary_embedding_dim/2 ({rotary_embedding_dim_half})." + ) + + cos_cache = np.expand_dims( + cos_cache, axis=2 + ) # Shape: [batch_size, sequence_length, 1, rotary_embedding_dim/2] + sin_cache = np.expand_dims( + sin_cache, axis=2 + ) # Shape: [batch_size, sequence_length, 1, rotary_embedding_dim/2] + + # Either divide the input in halves or interleave (based on interleaved attribute) + if interleaved: + x1 = x_rotate[:, :, :, 0::2] + x2 = x_rotate[:, :, :, 1::2] + else: + x1, x2 = np.split(x_rotate, 2, axis=-1) + + # Calculate real and imaginary values + real = (cos_cache * x1) - (sin_cache * x2) + imag = (sin_cache * x1) + (cos_cache * x2) + + # Inserted rotated embeddings back to the original input + if interleaved: + # x_rotate[:, :, :, 0::2] = real + # x_rotate[:, :, :, 1::2] = imag + real = np.expand_dims(real, axis=-1) + imag = np.expand_dims(imag, axis=-1) + x_rotate_concat = np.concatenate((real, imag), axis=-1) + x_rotate = np.reshape(x_rotate_concat, x_rotate.shape) + else: + x_rotate = np.concatenate((real, imag), axis=-1) + output = np.concatenate((x_rotate, x_not_rotate), axis=-1) + if len(original_input_shape) == 3: + output = np.reshape(output, original_input_shape) + else: + output = np.transpose(output, (0, 2, 1, 3)) + return output + ``` + +#### Version + +This version of the operator has been available since version 23 of the default ONNX operator set. + +#### Attributes + +
+
interleaved : int (default is 0)
+
Rotate using interleaved pattern. Default value is 0 (False).
+
num_heads : int
+
Number of attention heads. Must be provided when input is a 3D tensor.
+
rotary_embedding_dim : int (default is 0)
+
Rotary embedding dimension used to apply partial rotary embeddings.
+
+ +#### Inputs (3 - 4) + +
+
X : T
+
The input tensor representing the token embeddings. 4D tensor with shape `(batch_size, num_heads, sequence_length, head_size)` or 3D tensor with shape `(batch_size, sequence_length, hidden_size)`. For cases with a 4D input tensor, `head_size` has to be even. For cases with a 3D input tensor, `num_heads` attribute must be provided and `hidden_size` must be an even multiple of `num_heads` where `hidden_size = num_heads * head_size`
+
cos_cache : T
+
The cosine values for the rotation. 2D tensor with shape `(max_position_id_plus_1, head_size / 2)` for full rotation or `(max_position_id_plus_1, rotary_embedding_dim / 2)` for partial rotation when `position_ids` are provided. 3D tensor with shape `(batch_size, sequence_length, head_size / 2)` for full rotation or `(batch_size, sequence_length, rotary_embedding_dim / 2)` for partial rotation when `position_ids` are not provided. `max_position_id_plus_1` is a parameter to the model.
+
sin_cache : T
+
The sine values for the rotation. 2D tensor with shape `(max_position_id_plus_1, head_size / 2)` for full rotation or `(max_position_id_plus_1, rotary_embedding_dim / 2)` for partial rotation when `position_ids` are provided. 3D tensor with shape `(batch_size, sequence_length, head_size / 2)` for full rotation or `(batch_size, sequence_length, rotary_embedding_dim / 2)` for partial rotation when `position_ids` are not provided. `max_position_id_plus_1` is a parameter to the model.
+
position_ids (optional) : M
+
The position indices for the tokens. 2D tensor with shape `(batch_size, sequence_length)`
+
+ +#### Outputs + +
+
Y : T
+
Tensor with same shape as input.
+
+ +#### Type Constraints + +
+
T : tensor(float), tensor(float16), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
M : tensor(int64)
+
Constrain input and output types to integer tensors.
+
+ + +#### Examples + +
+rotary_embedding + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache", "position_ids"], + outputs=["output"], +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +position_ids_data = np.random.uniform(0, 50, (2, 3)).astype(np.int64) +sin_cache_data = np.random.rand(50, 4).astype(np.float32) +cos_cache_data = np.random.rand(50, 4).astype(np.float32) + +expected_output = rotary_embedding( + input_data, cos_cache_data, sin_cache_data, position_ids=position_ids_data +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data, position_ids_data], + outputs=[expected_output], + name="test_rotary_embedding", +) +``` + +
+ + +
+rotary_embedding_3d_input + +```python +num_heads = 4 +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache", "position_ids"], + outputs=["output"], + num_heads=num_heads, +) + +input_data = np.random.rand(2, 3, 32).astype(np.float32) +position_ids_data = np.random.uniform(0, 50, (2, 3)).astype(np.int64) +sin_cache_data = np.random.rand(50, 4).astype(np.float32) +cos_cache_data = np.random.rand(50, 4).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + position_ids=position_ids_data, + num_heads=num_heads, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data, position_ids_data], + outputs=[expected_output], + name="test_rotary_embedding_3d_input", +) +``` + +
+ + +
+rotary_embedding_interleaved + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache", "position_ids"], + outputs=["output"], + interleaved=1, +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +position_ids_data = np.random.uniform(0, 50, (2, 3)).astype(np.int64) +sin_cache_data = np.random.rand(50, 4).astype(np.float32) +cos_cache_data = np.random.rand(50, 4).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + position_ids=position_ids_data, + interleaved=1, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data, position_ids_data], + outputs=[expected_output], + name="test_rotary_embedding_interleaved", +) +``` + +
+ + +
+rotary_embedding_no_position_ids + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache"], + outputs=["output"], +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +sin_cache_data = np.random.rand(2, 3, 4).astype(np.float32) +cos_cache_data = np.random.rand(2, 3, 4).astype(np.float32) + +expected_output = rotary_embedding(input_data, cos_cache_data, sin_cache_data) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data], + outputs=[expected_output], + name="test_rotary_embedding_no_position_ids", +) +``` + +
+ + +
+rotary_embedding_no_position_ids_interleaved + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache"], + outputs=["output"], + interleaved=1, +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +sin_cache_data = np.random.rand(2, 3, 4).astype(np.float32) +cos_cache_data = np.random.rand(2, 3, 4).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + interleaved=1, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data], + outputs=[expected_output], + name="test_rotary_embedding_no_position_ids_interleaved", +) +``` + +
+ + +
+rotary_embedding_no_position_ids_rotary_dim + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache"], + outputs=["output"], + rotary_embedding_dim=4, +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +sin_cache_data = np.random.rand(2, 3, 2).astype(np.float32) +cos_cache_data = np.random.rand(2, 3, 2).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + rotary_embedding_dim=4, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data], + outputs=[expected_output], + name="test_rotary_embedding_no_position_ids_rotary_dim", +) +``` + +
+ + +
+rotary_embedding_with_interleaved_rotary_dim + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache", "position_ids"], + outputs=["output"], + rotary_embedding_dim=4, + interleaved=1, +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +position_ids_data = np.random.uniform(0, 50, (2, 3)).astype(np.int64) +sin_cache_data = np.random.rand(50, 2).astype(np.float32) +cos_cache_data = np.random.rand(50, 2).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + position_ids=position_ids_data, + interleaved=1, + rotary_embedding_dim=4, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data, position_ids_data], + outputs=[expected_output], + name="test_rotary_embedding_with_interleaved_rotary_dim", +) +``` + +
+ + +
+rotary_embedding_with_rotary_dim + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache", "position_ids"], + outputs=["output"], + rotary_embedding_dim=4, +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +position_ids_data = np.random.uniform(0, 50, (2, 3)).astype(np.int64) +sin_cache_data = np.random.rand(50, 2).astype(np.float32) +cos_cache_data = np.random.rand(50, 2).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + position_ids=position_ids_data, + rotary_embedding_dim=4, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data, position_ids_data], + outputs=[expected_output], + name="test_rotary_embedding_with_rotary_dim", +) +``` + +
+ + +### **Round** + + Round takes one input Tensor and rounds the values, element-wise, meaning + it finds the nearest integer for each value. + In case of halves, the rule is to round them to the nearest even integer. + If input x is integral, +0, -0, NaN, or infinite, x itself is returned. + The output tensor has the same shape and type as the input. + + Examples: + ``` + round([0.9]) = [1.0] + round([2.5]) = [2.0] + round([2.3]) = [2.0] + round([1.5]) = [2.0] + round([-4.5]) = [-4.0] + ``` + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 11 + +#### Inputs + +
+
X (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (non-differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+round + +```python +node = onnx.helper.make_node( + "Round", + inputs=["x"], + outputs=["y"], +) + +x = np.array( + [ + 0.1, + 0.5, + 0.9, + 1.2, + 1.5, + 1.8, + 2.3, + 2.5, + 2.7, + -1.1, + -1.5, + -1.9, + -2.2, + -2.5, + -2.8, + ] +).astype(np.float32) + +# expected output +y = np.array( + [ + 0.0, + 0.0, + 1.0, + 1.0, + 2.0, + 2.0, + 2.0, + 2.0, + 3.0, + -1.0, + -2.0, + -2.0, + -2.0, + -2.0, + -3.0, + ] +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_round") +``` + +
+ + +### **STFT** + + Computes the Short-time Fourier Transform of the signal. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
onesided : int (default is 1)
+
If onesided is 1, only values for w in [0, 1, 2, ..., floor(n_fft/2) + 1] are returned because the real-to-complex Fourier transform satisfies the conjugate symmetry, i.e., X[m, w] = X[m,w]=X[m,n_fft-w]*. Note if the input or window tensors are complex, then onesided output is not possible. Enabling onesided with real inputs performs a Real-valued fast Fourier transform (RFFT).When invoked with real or complex valued input, the default value is 1. Values can be 0 or 1.
+
+ +#### Inputs (2 - 4) + +
+
signal (non-differentiable) : T1
+
Input tensor representing a real or complex valued signal. For real input, the following shape is expected: [batch_size][signal_length][1]. For complex input, the following shape is expected: [batch_size][signal_length][2], where [batch_size][signal_length][0] represents the real component and [batch_size][signal_length][1] represents the imaginary component of the signal.
+
frame_step (non-differentiable) : T2
+
The number of samples to step between successive DFTs.
+
window (optional, non-differentiable) : T1
+
A tensor representing the window that will be slid over the signal.The window must have rank 1 with shape: [window_shape]. It's an optional value.
+
frame_length (optional, non-differentiable) : T2
+
A scalar representing the size of the DFT. It's an optional value.
+
+ +#### Outputs + +
+
output (non-differentiable) : T1
+
The Short-time Fourier Transform of the signals.If onesided is 1, the output has the shape: [batch_size][frames][dft_unique_bins][2], where dft_unique_bins is frame_length // 2 + 1 (the unique components of the DFT) If onesided is 0, the output has the shape: [batch_size][frames][frame_length][2], where frame_length is the length of the DFT.
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(float16), tensor(double), tensor(bfloat16)
+
Constrain signal and output to float tensors.
+
T2 : tensor(int32), tensor(int64)
+
Constrain scalar length types to int64_t.
+
+ + +#### Examples + +
+stft + +```python +signal = np.arange(0, 128, dtype=np.float32).reshape(1, 128, 1) +length = np.array(16).astype(np.int64) +onesided_length = (length >> 1) + 1 +step = np.array(8).astype(np.int64) + +no_window = "" # optional input, not supplied +node = onnx.helper.make_node( + "STFT", + inputs=["signal", "frame_step", no_window, "frame_length"], + outputs=["output"], +) + +nstfts = ((signal.shape[1] - length) // step) + 1 +# [batch_size][frames][frame_length][2] +output = np.empty([1, nstfts, onesided_length, 2], dtype=np.float32) +for i in range(nstfts): + start = i * step + stop = i * step + length + complex_out = np.fft.fft(signal[0, start:stop, 0])[0:onesided_length] + output[0, i] = np.stack((complex_out.real, complex_out.imag), axis=1) + +output = output.astype(signal.dtype) +expect(node, inputs=[signal, step, length], outputs=[output], name="test_stft") + +node = onnx.helper.make_node( + "STFT", + inputs=["signal", "frame_step", "window"], + outputs=["output"], +) + +# Test with window +a0 = 0.5 +a1 = 0.5 +window = a0 + a1 * np.cos( + 2 * np.pi * np.arange(0, length, 1, dtype=np.float32) / length +) +nstfts = 1 + (signal.shape[1] - window.shape[0]) // step + +# [batch_size][frames][frame_length][2] +output = np.empty([1, nstfts, onesided_length, 2], dtype=np.float32) +for i in range(nstfts): + start = i * step + stop = i * step + length + complex_out = np.fft.fft(signal[0, start:stop, 0] * window)[ + 0:onesided_length + ] + output[0, i] = np.stack((complex_out.real, complex_out.imag), axis=1) +window = window.astype(signal.dtype) +output = output.astype(signal.dtype) +expect( + node, + inputs=[signal, step, window], + outputs=[output], + name="test_stft_with_window", +) +``` + +
+ + +### **Scan** + + Scan can be used to iterate over one or more scan_input tensors, + constructing zero or more scan_output tensors. It combines ideas from general recurrences, + functional programming constructs such as scan, fold, map, and zip, and is intended to enable + generalizations of RNN-like constructs for sequence-to-sequence processing. + Other tensors (referred to as state_variables here) can be used to carry a state + when iterating from one element to another (similar to hidden-state in RNNs, also referred + to as loop-carried dependences in the context of loops). + Many common usages involve a single scan_input tensor (where functionality + similar to scan, fold and map can be obtained). When more than one scan_input is used, + a behavior similar to zip is obtained. + + The attribute body must be a graph, specifying the computation to be performed in + every iteration. It takes as input the current values of the state_variables and + the current iterated element of the scan_inputs. It must return the (updated) values + of the state_variables and zero or more scan_output_element tensors. The values of the + scan_output_element tensors are concatenated over all the iterations to produce the + scan_output values of the scan construct (similar to the concatenated intermediate + hidden-state values of RNN-like constructs). All the output tensors (state_variables as + well as scan_output_element tensors) are required to have the same shape in each iteration + of the loop (a restriction imposed to enable efficient memory allocation). + + Note that the iterated element passed to the body subgraph does not have a sequence + axis. It will have a rank one less than the rank of the corresponding scan_input. + + The scan operation returns the final values of the state_variables as well as the + scan_outputs. + + The optional attribute scan_input_directions specifies the direction (forward or backward) + for each scan input. If this attribute is omitted, all sequences are scanned in the forward + direction. A bidirectional scan may be performed by specifying the same tensor input twice + in the scan_inputs, once with a forward direction, and once with a backward direction. + + The scan_output of the operation is produced by concatenating the scan_output_element + values produced by the body in each iteration. The optional attribute scan_output_directions + specifies the direction in which scan_output is constructed (by appending or prepending the + scan_output_element to scan_output in each iteration) for each scan_output. If this attribute + is omitted, the scan_output_element is appended to the scan_output in each iteration. + + The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. + If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the + batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. + Note that scanning a non-zero axis may be less efficient than scanning axis zero. + + The optional attribute scan_output_axes specifies the axis along which the scan_outputs + are accumulated for each scan_output. For example, if axis 1 is the time axis (to be + scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis + value of 1. + + Note that because of the ONNX restriction that only the last parameter of an operator can + be variadic, the initial-states and scan-inputs are listed together as one input parameter. + Similarly, the final-states and scan-outputs are listed together as one output parameter. + The attribute num_scan_inputs indicates the number M of scan-inputs. + + The behavior of + + Scan < + num_scan_inputs = m, + body = loop-body, + scan_input_axes = [axis_1, ..., axis_m] + > (init_1, ..., init_n, scan_1, ..., scan_m) + + is equivalent to the following pseudo-code: + + // scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i + // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. + sequence_length = scan_1.shape[axis_1]; + + // initialize state-variables + st_1 = init_1; ... st_n = init_n; + // initialize scan-output variables: [] denotes an empty tensor + scan_out_1 = []; ...; scan_out_k = []; + // identify number of iterations: + + // execute loop + for (int t = 0; t < sequence_length; ++t) { + // generate the scan-input elements: the notation T[t] indicates the sub-tensor + // of rank one less than T obtained by indexing T at position t along axis k. + si_1 = scan_1[t]; + ... ; + si_m = scan_m[t]; + // execute loop-body + st_1, ..., st_n, so_1, ..., so_k = loop-body(st_1, ..., st_n, si_1, ..., si_m) + // accumulate the scan-output elements + scan_out_1 = Concat(scan_out_1, so_1); ... ; scan_out_k = Concat(scan_out_k, so_k); + } + + return st_1, ..., st_n, scan_out_1, ..., scan_out_k; + + *Sample usage: Encoding RNN using a Scan* + + The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, + recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can + be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes + %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these + values are computed in the outer graph, they need to be passed in as extra state_variables. + + graph rnn-encoding { + %H_0 = ... + %X = ... + %Y_h, %Y = Scan[body = , num_scan_inputs=1](%H_0, %X) + return %Y, %Y_h + } + + graph rnn-cell-1 ( + %H_tminus1[FLOAT, tensor] + %X_t[FLOAT, tensor] + ) { + %Wi = ... + %Ri = ... + %Wbi = ... + %Rbi = ... + %t1 = X_t * (Wi^T) + %t2 = H_tminus1*(Ri^T) + %t3 = Add(%t1, %t2) + %t4 = Add(%t3, %Wbi) + %t5 = Add(%t4, %Rbi) + %Ht = Tanh(%t5) + %Accumulate = Identity(%Ht) + return %Ht, %Accumulate + } + + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 8, 9, 11, 16, 19, 21, 23, 24 + +#### Attributes + +
+
body : graph (required)
+
The graph run each iteration. It has N+M inputs: (loop state variables..., scan_input_elts...). It has N+K outputs: (loop state variables..., scan_output_elts...). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
+
num_scan_inputs : int (required)
+
An attribute specifying the number of scan_inputs M.
+
scan_input_axes : list of ints
+
An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
scan_input_directions : list of ints
+
An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
+
scan_output_axes : list of ints
+
An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
+
scan_output_directions : list of ints
+
An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
+
+ +#### Inputs (1 - ∞) + +
+
initial_state_and_scan_inputs (variadic, heterogeneous) : V
+
Initial values of the loop's N state variables followed by M scan_inputs
+
+ +#### Outputs (1 - ∞) + +
+
final_state_and_scan_outputs (variadic, heterogeneous) : V
+
Final values of the loop's N state variables followed by K scan_outputs
+
+ +#### Type Constraints + +
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
All Tensor types up to IRv13.
+
+ + +#### Examples + +
+scan_8 + +```python +# Given an input sequence [x1, ..., xN], sum up its elements using a scan +# returning the final state (x1+x2+...+xN) as well the scan_output +# [x1, x1+x2, ..., x1+x2+...+xN] +# Note: the first input (sequence_lens) is optional and omitted via "". +node = onnx.parser.parse_node( + """ + y, z = Scan ("", initial, x) < + num_scan_inputs = 1, + body = scan_body (float[2] sum_in, float[2] next) + => (float[2] sum_out, float[2] scan_out) + { + sum_out = Add(sum_in, next) + scan_out = Identity(sum_out) + } + > + """ +) +# create inputs for batch-size 1, sequence-length 3, inner dimension 2 +initial = np.array([0, 0]).astype(np.float32).reshape((1, 2)) +x = np.array([1, 2, 3, 4, 5, 6]).astype(np.float32).reshape((1, 3, 2)) +# final state computed = [1 + 3 + 5, 2 + 4 + 6] +y = np.array([9, 12]).astype(np.float32).reshape((1, 2)) +# scan-output computed +z = np.array([1, 2, 4, 6, 9, 12]).astype(np.float32).reshape((1, 3, 2)) + +expect( + node, + inputs=[initial, x], + outputs=[y, z], + name="test_scan_sum", + opset_imports=[onnx.helper.make_opsetid("", 8)], +) +``` + +
+ + +
+scan_9 + +```python +# Given an input sequence [x1, ..., xN], sum up its elements using a scan +# returning the final state (x1+x2+...+xN) as well the scan_output +# [x1, x1+x2, ..., x1+x2+...+xN] +node = onnx.parser.parse_node( + """ + y, z = Scan (initial, x) < + num_scan_inputs = 1, + body = scan_body (float[2] sum_in, float[2] next) + => (float[2] sum_out, float[2] scan_out) + { + sum_out = Add(sum_in, next) + scan_out = Identity(sum_out) + } + > + """ +) +# create inputs for sequence-length 3, inner dimension 2 +initial = np.array([0, 0]).astype(np.float32).reshape((2,)) +x = np.array([1, 2, 3, 4, 5, 6]).astype(np.float32).reshape((3, 2)) +# final state computed = [1 + 3 + 5, 2 + 4 + 6] +y = np.array([9, 12]).astype(np.float32).reshape((2,)) +# scan-output computed +z = np.array([1, 2, 4, 6, 9, 12]).astype(np.float32).reshape((3, 2)) + +expect( + node, + inputs=[initial, x], + outputs=[y, z], + name="test_scan9_sum", + opset_imports=[onnx.helper.make_opsetid("", 9)], +) +``` + +
+ + +
+scan_9_multi_state + +```python +# Scan with two state variables: running sum and running product. +# This exercises the case where num_loop_state_vars (2) differs from +# num_scan_inputs (1). +# +# Body inputs: sum_in (state), prod_in (state), next (scan) +# Body outputs: sum_out (state), prod_out (state), scan_out (scan) +node = onnx.parser.parse_node( + """ + y_sum, y_prod, z = Scan (initial_sum, initial_prod, x) < + num_scan_inputs = 1, + body = scan_body (float[2] sum_in, float[2] prod_in, float[2] next) + => (float[2] sum_out, float[2] prod_out, float[2] scan_out) + { + sum_out = Add(sum_in, next) + prod_out = Mul(prod_in, next) + scan_out = Identity(sum_out) + } + > + """ +) +# x = [[1, 2], [3, 4], [5, 6]] +initial_sum = np.array([0, 0]).astype(np.float32) +initial_prod = np.array([1, 1]).astype(np.float32) +x = np.array([1, 2, 3, 4, 5, 6]).astype(np.float32).reshape((3, 2)) +# final sum = [1+3+5, 2+4+6] = [9, 12] +y_sum = np.array([9, 12]).astype(np.float32) +# final product = [1*3*5, 2*4*6] = [15, 48] +y_prod = np.array([15, 48]).astype(np.float32) +# scan output (running sum) = [[1,2], [4,6], [9,12]] +z = np.array([1, 2, 4, 6, 9, 12]).astype(np.float32).reshape((3, 2)) + +expect( + node, + inputs=[initial_sum, initial_prod, x], + outputs=[y_sum, y_prod, z], + name="test_scan9_multi_state", + opset_imports=[onnx.helper.make_opsetid("", 9)], +) +``` + +
+ + +
+scan_9_scalar + +```python +# Scan with scalar state and scan output to verify that output +# shapes are not distorted (e.g. (T,) not (T, 1)). +node = onnx.parser.parse_node( + """ + y, z = Scan (initial, x) < + num_scan_inputs = 1, + body = scan_body (float sum_in, float next) + => (float sum_out, float scan_out) + { + sum_out = Add(sum_in, next) + scan_out = Identity(sum_out) + } + > + """ +) +initial = np.float32(0.0) +x = np.array([1, 2, 3, 4, 5]).astype(np.float32) +# final state = 1+2+3+4+5 = 15 +y = np.float32(15.0) +# scan output = [1, 3, 6, 10, 15], shape (5,) +z = np.array([1, 3, 6, 10, 15]).astype(np.float32) + +expect( + node, + inputs=[initial, x], + outputs=[y, z], + name="test_scan9_scalar", + opset_imports=[onnx.helper.make_opsetid("", 9)], +) +``` + +
+ + +### **Scatter** (deprecated) + + This operator is deprecated. Please use ScatterElements, which provides the same functionality. + + Scatter takes three inputs `data`, `updates`, and `indices` of the same + rank r >= 1 and an optional attribute axis that identifies an axis of `data` + (by default, the outer-most axis, that is axis 0). The output of the operation + is produced by creating a copy of the input `data`, and then updating its value + to values specified by `updates` at specific index positions specified by + `indices`. Its output shape is the same as the shape of `data`. + + For each entry in `updates`, the target index in `data` is obtained by combining + the corresponding entry in `indices` with the index of the entry itself: the + index-value for dimension = axis is obtained from the value of the corresponding + entry in `indices` and the index-value for dimension != axis is obtained from the + index of the entry itself. + + For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry + is performed as below: + ``` + output[indices[i][j]][j] = updates[i][j] if axis = 0, + output[i][indices[i][j]] = updates[i][j] if axis = 1, + ``` + + This operator is the inverse of GatherElements. It is similar to Torch's Scatter operation. + + Example 1: + ``` + data = [ + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + ] + indices = [ + [1, 0, 2], + [0, 2, 1], + ] + updates = [ + [1.0, 1.1, 1.2], + [2.0, 2.1, 2.2], + ] + output = [ + [2.0, 1.1, 0.0] + [1.0, 0.0, 2.2] + [0.0, 2.1, 1.2] + ] + ``` + Example 2: + ``` + data = [[1.0, 2.0, 3.0, 4.0, 5.0]] + indices = [[1, 3]] + updates = [[1.1, 2.1]] + axis = 1 + output = [[1.0, 1.1, 3.0, 2.1, 5.0]] + ``` + +#### Version + +This version of the operator has been deprecated since version 11 of the default ONNX operator set. + +Other versions of this operator: 9 + + +#### Examples + +
+scatter_with_axis + +```python +axis = 1 +node = onnx.helper.make_node( + "Scatter", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 3]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter(data, indices, updates, axis=axis) +# print(y) produces +# [[1.0, 1.1, 3.0, 2.1, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_with_axis", + opset_imports=[helper.make_opsetid("", 10)], +) +``` + +
+ + +
+scatter_without_axis + +```python +node = onnx.helper.make_node( + "Scatter", + inputs=["data", "indices", "updates"], + outputs=["y"], +) +data = np.zeros((3, 3), dtype=np.float32) +indices = np.array([[1, 0, 2], [0, 2, 1]], dtype=np.int64) +updates = np.array([[1.0, 1.1, 1.2], [2.0, 2.1, 2.2]], dtype=np.float32) + +y = scatter(data, indices, updates) +# print(y) produces +# [[2.0, 1.1, 0.0], +# [1.0, 0.0, 2.2], +# [0.0, 2.1, 1.2]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_without_axis", + opset_imports=[helper.make_opsetid("", 10)], +) +``` + +
+ + +### **ScatterElements** + + ScatterElements takes three inputs `data`, `updates`, and `indices` of the same + rank r >= 1 and an optional attribute axis that identifies an axis of `data` + (by default, the outer-most axis, that is axis 0). The output of the operation + is produced by creating a copy of the input `data`, and then updating its value + to values specified by `updates` at specific index positions specified by + `indices`. Its output shape is the same as the shape of `data`. + + For each entry in `updates`, the target index in `data` is obtained by combining + the corresponding entry in `indices` with the index of the entry itself: the + index-value for dimension = axis is obtained from the value of the corresponding + entry in `indices` and the index-value for dimension != axis is obtained from the + index of the entry itself. + + `reduction` allows specification of an optional reduction operation, which is applied to all values in `updates` + tensor into `output` at the specified `indices`. + In cases where `reduction` is set to "none", indices should not have duplicate entries: that is, if idx1 != idx2, + then indices[idx1] != indices[idx2]. For instance, in a 2-D tensor case, the update + corresponding to the [i][j] entry is performed as below: + ``` + output[indices[i][j]][j] = updates[i][j] if axis = 0, + output[i][indices[i][j]] = updates[i][j] if axis = 1, + ``` + When `reduction` is set to some reduction function `f`, the update corresponding to the [i][j] entry is performed as below: + ``` + output[indices[i][j]][j] = f(output[indices[i][j]][j], updates[i][j]) if axis = 0, + output[i][indices[i][j]] = f(output[i][indices[i][j]], updates[i][j]) if axis = 1, + ``` + where the `f` is `+`, `*`, `max` or `min` as specified. + + This operator is the inverse of GatherElements. It is similar to Torch's Scatter operation. + + (Opset 18 change): Adds max/min to the set of allowed reduction ops. + + Example 1: + ``` + data = [ + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + [0.0, 0.0, 0.0], + ] + indices = [ + [1, 0, 2], + [0, 2, 1], + ] + updates = [ + [1.0, 1.1, 1.2], + [2.0, 2.1, 2.2], + ] + output = [ + [2.0, 1.1, 0.0] + [1.0, 0.0, 2.2] + [0.0, 2.1, 1.2] + ] + ``` + Example 2: + ``` + data = [[1.0, 2.0, 3.0, 4.0, 5.0]] + indices = [[1, 3]] + updates = [[1.1, 2.1]] + axis = 1 + output = [[1.0, 1.1, 3.0, 2.1, 5.0]] + ``` + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 11, 13, 16 + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
reduction : string (default is none)
+
Type of reduction to apply: none (default), add, mul, max, min. 'none': no reduction applied. 'add': reduction using the addition operation. 'mul': reduction using the multiplication operation.'max': reduction using the maximum operation.'min': reduction using the minimum operation.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : Tind
+
Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
+
updates (differentiable) : T
+
Tensor of rank r >=1 (same rank and shape as indices)
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank r >= 1 (same rank as input).
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input and output types can be of any tensor type.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ + +#### Examples + +
+scatter_elements_with_axis + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 3]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis) +# print(y) produces +# [[1.0, 1.1, 3.0, 2.1, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_axis", +) +``` + +
+ + +
+scatter_elements_with_duplicate_indices + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, + reduction="add", +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 1]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis, reduction="add") +# print(y) produces +# [[1.0, 5.2, 3.0, 4.0, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_duplicate_indices", +) +``` + +
+ + +
+scatter_elements_with_negative_indices + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, -3]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis) +# print(y) produces +# [[1.0, 1.1, 2.1, 4.0, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_negative_indices", +) +``` + +
+ + +
+scatter_elements_with_reduction_max + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, + reduction="max", +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 1]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis, reduction="max") +# print(y) produces +# [[1.0, 2.1, 3.0, 4.0, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_reduction_max", +) +``` + +
+ + +
+scatter_elements_with_reduction_min + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, + reduction="min", +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 1]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis, reduction="min") +# print(y) produces +# [[1.0, 1.1, 3.0, 4.0, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_reduction_min", +) +``` + +
+ + +
+scatter_elements_with_reduction_mul + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, + reduction="mul", +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 1]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis, reduction="mul") +# print(y) produces +# [[1.0, 4.62, 3.0, 4.0, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_reduction_mul", +) +``` + +
+ + +
+scatter_elements_without_axis + +```python +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], +) +data = np.zeros((3, 3), dtype=np.float32) +indices = np.array([[1, 0, 2], [0, 2, 1]], dtype=np.int64) +updates = np.array([[1.0, 1.1, 1.2], [2.0, 2.1, 2.2]], dtype=np.float32) + +y = scatter_elements(data, indices, updates) +# print(y) produces +# [[2.0, 1.1, 0.0], +# [1.0, 0.0, 2.2], +# [0.0, 2.1, 1.2]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_without_axis", +) +``` + +
+ + +### **ScatterND** + + ScatterND takes three inputs `data` tensor of rank r >= 1, `indices` tensor of rank q >= 1, + and `updates` tensor of rank q + r - indices.shape[-1] - 1. The output of the operation + is produced by creating a copy of the input `data`, and then updating its value to values + specified by `updates` at specific index positions specified by `indices`. Its output shape + is the same as the shape of `data`. + + `indices` is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of `indices`. + `indices` is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into `data`. + Hence, k can be a value at most the rank of `data`. When k equals rank(data), each update entry specifies an + update to a single element of the tensor. When k is less than rank(data) each update entry specifies an + update to a slice of the tensor. Index values are allowed to be negative, as per the usual + convention for counting backwards from the end, but are expected in the valid range. + + `updates` is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the + first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. + The remaining dimensions of `updates` correspond to the dimensions of the + replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, + corresponding to the trailing (r-k) dimensions of `data`. Thus, the shape of `updates` + must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation + of shapes. + + The `output` is calculated via the following equation: + + ``` + output = np.copy(data) + update_indices = indices.shape[:-1] + for idx in np.ndindex(update_indices): + output[tuple(indices[idx])] = updates[idx] + ``` + + The order of iteration in the above loop is not specified. + In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. + This ensures that the output value does not depend on the iteration order. + + `reduction` allows specification of an optional reduction operation, which is applied to all values in `updates` + tensor into `output` at the specified `indices`. + In cases where `reduction` is set to "none", indices should not have duplicate entries: that is, if idx1 != idx2, + then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order. + When `reduction` is set to some reduction function `f`, `output` is calculated as follows: + + ``` + output = np.copy(data) + update_indices = indices.shape[:-1] + for idx in np.ndindex(update_indices): + output[tuple(indices[idx])] = f(output[tuple(indices[idx])], updates[idx]) + ``` + + where the `f` is `+`, `*`, `max` or `min` as specified. + + This operator is the inverse of GatherND. + + (Opset 18 change): Adds max/min to the set of allowed reduction ops. + + Example 1: + ``` + data = [1, 2, 3, 4, 5, 6, 7, 8] + indices = [[4], [3], [1], [7]] + updates = [9, 10, 11, 12] + output = [1, 11, 3, 10, 9, 6, 7, 12] + ``` + + Example 2: + ``` + data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] + indices = [[0], [2]] + updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]] + output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] + ``` + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 11, 13, 16 + +#### Attributes + +
+
reduction : string (default is none)
+
Type of reduction to apply: none (default), add, mul, max, min. 'none': no reduction applied. 'add': reduction using the addition operation. 'mul': reduction using the addition operation. 'max': reduction using the maximum operation.'min': reduction using the minimum operation.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
Tensor of rank r >= 1.
+
indices (non-differentiable) : tensor(int64)
+
Tensor of rank q >= 1.
+
updates (differentiable) : T
+
Tensor of rank q + r - indices_shape[-1] - 1.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of rank r >= 1.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to any tensor type.
+
+ + +#### Examples + +
+scatternd + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], +) +data = np.array( + [ + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + ], + dtype=np.float32, +) +indices = np.array([[0], [2]], dtype=np.int64) +updates = np.array( + [ + [[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + ], + dtype=np.float32, +) +# Expecting output as np.array( +# [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], +# [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], +# [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], +# [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates) +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd", +) +``` + +
+ + +
+scatternd_add + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="add", +) +data = np.array( + [ + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + ], + dtype=np.float32, +) +indices = np.array([[0], [0]], dtype=np.int64) +updates = np.array( + [ + [[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + ], + dtype=np.float32, +) +# Expecting output as np.array( +# [[[7, 8, 9, 10], [13, 14, 15, 16], [18, 17, 16, 15], [16, 15, 14, 13]], +# [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], +# [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], +# [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="add") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_add", +) +``` + +
+ + +
+scatternd_max + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="max", +) +data = np.array( + [ + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + ], + dtype=np.float32, +) +indices = np.array([[0], [0]], dtype=np.int64) +updates = np.array( + [ + [[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + ], + dtype=np.float32, +) +# Expecting output as np.array( +# [[[5, 5, 5, 5], [6, 6, 7, 8], [8, 7, 7, 7], [8, 8 ,8, 8]], +# [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], +# [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], +# [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="max") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_max", +) +``` + +
+ + +
+scatternd_max_with_element_indices + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="max", +) +data = np.array([[1, 2], [3, 4]], dtype=np.float32) +# Indices address individual elements (index rank == data rank), +# which exercises the reduction at the element level. +indices = np.array([[0, 0], [1, 1]], dtype=np.int64) +updates = np.array([5, 1], dtype=np.float32) +# Expecting output as np.array([[5, 2], [3, 4]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="max") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_max_with_element_indices", +) +``` + +
+ + +
+scatternd_min + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="min", +) +data = np.array( + [ + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + ], + dtype=np.float32, +) +indices = np.array([[0], [0]], dtype=np.int64) +updates = np.array( + [ + [[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + ], + dtype=np.float32, +) +# Expecting output as np.array( +# [[[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 3, 2, 1]], +# [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], +# [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], +# [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="min") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_min", +) +``` + +
+ + +
+scatternd_min_with_element_indices + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="min", +) +data = np.array([[1, 2], [3, 4]], dtype=np.float32) +indices = np.array([[0, 0], [1, 1]], dtype=np.int64) +updates = np.array([5, 1], dtype=np.float32) +# Expecting output as np.array([[1, 2], [3, 1]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="min") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_min_with_element_indices", +) +``` + +
+ + +
+scatternd_multiply + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="mul", +) +data = np.array( + [ + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + ], + dtype=np.float32, +) +indices = np.array([[0], [0]], dtype=np.int64) +updates = np.array( + [ + [[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + ], + dtype=np.float32, +) +# Expecting output as np.array( +# [[[5, 10, 15, 20], [60, 72, 84, 96], [168, 147, 126, 105], [128, 96, 64, 32]], +# [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], +# [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], +# [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="mul") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_multiply", +) +``` + +
+ + +### **Selu** + + Selu takes one input data (Tensor) and produces one output data + (Tensor) where the scaled exponential linear unit function, + `y = gamma * (alpha * e^x - alpha) for x <= 0`, `y = gamma * x for x > 0`, + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Attributes + +
+
alpha : float (default is 1.67326)
+
Coefficient of SELU default to 1.67326319217681884765625 (i.e., float32 approximation of 1.6732632423543772848170429916717).
+
gamma : float (default is 1.0507)
+
Coefficient of SELU default to 1.05070102214813232421875 (i.e., float32 approximation of 1.0507009873554804934193349852946).
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+selu + +```python +node = onnx.helper.make_node( + "Selu", inputs=["x"], outputs=["y"], alpha=2.0, gamma=3.0 +) + +x = np.array([-1, 0, 1]).astype(np.float32) +# expected output [-3.79272318, 0., 3.] +y = ( + np.clip(x, 0, np.inf) * 3.0 + + (np.exp(np.clip(x, -np.inf, 0)) - 1) * 2.0 * 3.0 +) +expect(node, inputs=[x], outputs=[y], name="test_selu_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = ( + np.clip(x, 0, np.inf) * 3.0 + + (np.exp(np.clip(x, -np.inf, 0)) - 1) * 2.0 * 3.0 +) +expect(node, inputs=[x], outputs=[y], name="test_selu") +``` + +
+ + +
+selu_default + +```python +default_alpha = 1.67326319217681884765625 +default_gamma = 1.05070102214813232421875 +node = onnx.helper.make_node( + "Selu", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = ( + np.clip(x, 0, np.inf) * default_gamma + + (np.exp(np.clip(x, -np.inf, 0)) - 1) * default_alpha * default_gamma +) +expect(node, inputs=[x], outputs=[y], name="test_selu_default") +``` + +
+ + +### **SequenceAt** + + Outputs a tensor copy from the tensor at 'position' in 'input_sequence'. + Accepted range for 'position' is in `[-n, n - 1]`, where `n` is the number of tensors in 'input_sequence'. + Negative value means counting positions from the back. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
input_sequence : S
+
Input sequence.
+
position : I
+
Position of the tensor in the sequence. Negative value means counting positions from the back. Accepted range in `[-n, n - 1]`, where `n` is the number of tensors in 'input_sequence'. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
+
+ +#### Outputs + +
+
tensor : T
+
Output tensor at the specified position in the input sequence.
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain to any tensor type.
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type.
+
I : tensor(int32), tensor(int64)
+
Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
+
+ + +### **SequenceConstruct** + + Construct a tensor sequence containing 'inputs' tensors. + All tensors in 'inputs' must have the same data type. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs (1 - ∞) + +
+
inputs (variadic) : T
+
Tensors.
+
+ +#### Outputs + +
+
output_sequence : S
+
Sequence enclosing the input tensors.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input types to any tensor type.
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain output types to any tensor type.
+
+ + +### **SequenceEmpty** + + Construct an empty tensor sequence, with given data type. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
dtype : int
+
(Optional) The data type of the tensors in the output sequence. The default type is 'float'.
+
+ +#### Inputs + + +#### Outputs + +
+
output : S
+
Empty sequence.
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain output types to any tensor type.
+
+ + +### **SequenceErase** + + Outputs a tensor sequence that removes the tensor at 'position' from 'input_sequence'. + Accepted range for 'position' is in `[-n, n - 1]`, where `n` is the number of tensors in 'input_sequence'. + Negative value means counting positions from the back. + 'position' is optional, by default it erases the last tensor from 'input_sequence'. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs (1 - 2) + +
+
input_sequence : S
+
Input sequence.
+
position (optional) : I
+
Position of the tensor in the sequence. Negative value means counting positions from the back. Accepted range in `[-n, n - 1]`, where `n` is the number of tensors in 'input_sequence'. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
+
+ +#### Outputs + +
+
output_sequence : S
+
Output sequence that has the tensor at the specified position removed.
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain to any tensor type.
+
I : tensor(int32), tensor(int64)
+
Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
+
+ + +### **SequenceInsert** + + Outputs a tensor sequence that inserts 'tensor' into 'input_sequence' at 'position'. + 'tensor' must have the same data type as 'input_sequence'. + Accepted range for 'position' is in `[-n, n]`, where `n` is the number of tensors in 'input_sequence'. + Negative value means counting positions from the back. + 'position' is optional, by default it inserts 'tensor' to the back of 'input_sequence'. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs (2 - 3) + +
+
input_sequence : S
+
Input sequence.
+
tensor : T
+
Input tensor to be inserted into the input sequence.
+
position (optional) : I
+
Position in the sequence where the new tensor is inserted. It is optional and default is to insert to the back of the sequence. Negative value means counting positions from the back. Accepted range in `[-n, n]`, where `n` is the number of tensors in 'input_sequence'. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
+
+ +#### Outputs + +
+
output_sequence : S
+
Output sequence that contains the inserted tensor at given position.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain to any tensor type.
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain to any tensor type.
+
I : tensor(int32), tensor(int64)
+
Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
+
+ + +#### Examples + +
+sequenceinsert + +```python +test_cases = { + "at_back": [np.array([10, 11, 12]).astype(np.int64)], + "at_front": [np.array([-2, -1, 0]), np.array([0]).astype(np.int64)], +} +sequence = [ + np.array([1, 2, 3, 4]).astype(np.int64), + np.array([5, 6, 7]).astype(np.int64), + np.array([8, 9]).astype(np.int64), +] + +for test_name, test_inputs in test_cases.items(): + tensor = test_inputs[0].astype(np.int64) + + if len(test_inputs) > 1: + node = onnx.helper.make_node( + "SequenceInsert", + inputs=["sequence", "tensor", "position"], + outputs=["output_sequence"], + ) + position = test_inputs[1] + inserted = sequence_insert_reference_implementation( + sequence, tensor, position + ) + expect( + node, + inputs=[sequence, tensor, position], + outputs=[inserted], + name="test_sequence_insert_" + test_name, + ) + else: + node = onnx.helper.make_node( + "SequenceInsert", + inputs=["sequence", "tensor"], + outputs=["output_sequence"], + ) + inserted = sequence_insert_reference_implementation(sequence, tensor) + expect( + node, + inputs=[sequence, tensor], + outputs=[inserted], + name="test_sequence_insert_" + test_name, + ) +``` + +
+ + +### **SequenceLength** + + Produces a scalar(tensor of empty shape) containing the number of tensors in 'input_sequence'. + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Inputs + +
+
input_sequence : S
+
Input sequence.
+
+ +#### Outputs + +
+
length : I
+
Length of input sequence. It must be a scalar(tensor of empty shape).
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain to any tensor type.
+
I : tensor(int64)
+
Constrain output to integral tensor. It must be a scalar(tensor of empty shape).
+
+ + +### **SequenceMap** + + Applies a sub-graph to each sample in the input sequence(s). + + Inputs can be either tensors or sequences, with the exception of the first input which must + be a sequence. The length of the first input sequence will determine the number of samples in the + outputs. Any other sequence inputs should have the same number of samples. The number of inputs + and outputs, should match the one of the subgraph. + + For each i-th element in the output, a sample will be extracted from the input sequence(s) at + the i-th position and the sub-graph will be applied to it. + The outputs will contain the outputs of the sub-graph for each sample, in the same order as in + the input. + + This operator assumes that processing each sample is independent and could executed in parallel + or in any order. Users cannot expect any specific ordering in which each subgraph is computed. + +#### Version + +This version of the operator has been available since version 17 of the default ONNX operator set. + +#### Attributes + +
+
body : graph (required)
+
The graph to be run for each sample in the sequence(s). It should have as many inputs and outputs as inputs and outputs to the SequenceMap function.
+
+ +#### Inputs (1 - ∞) + +
+
input_sequence : S
+
Input sequence.
+
additional_inputs (variadic, heterogeneous) : V
+
Additional inputs to the graph
+
+ +#### Outputs (1 - ∞) + +
+
out_sequence (variadic, heterogeneous) : S
+
Output sequence(s)
+
+ +#### Type Constraints + +
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain input types to any sequence type.
+
V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain to any tensor or sequence type.
+
+ + +#### Examples + +
+sequence_map_add_1_sequence_1_tensor + +```python +body = onnx.helper.make_graph( + [onnx.helper.make_node("Add", ["in0", "in1"], ["out0"])], + "seq_map_body", + [ + onnx.helper.make_tensor_value_info( + "in0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "in1", onnx.TensorProto.FLOAT, ["N"] + ), + ], + [onnx.helper.make_tensor_value_info("out0", onnx.TensorProto.FLOAT, ["N"])], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["x0", "x1"], outputs=["y0"], body=body +) + +x0 = [np.random.uniform(0.0, 1.0, 10).astype(np.float32) for k in range(3)] +x1 = np.random.uniform(0.0, 1.0, 10).astype(np.float32) +y0 = [x0[i] + x1 for i in range(3)] +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), +] +expect( + node, + inputs=[x0, x1], + outputs=[y0], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_add_1_sequence_1_tensor", +) +``` + +
+ + +
+sequence_map_add_2_sequences + +```python +body = onnx.helper.make_graph( + [onnx.helper.make_node("Add", ["in0", "in1"], ["out0"])], + "seq_map_body", + [ + onnx.helper.make_tensor_value_info( + "in0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "in1", onnx.TensorProto.FLOAT, ["N"] + ), + ], + [onnx.helper.make_tensor_value_info("out0", onnx.TensorProto.FLOAT, ["N"])], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["x0", "x1"], outputs=["y0"], body=body +) + +N = [np.random.randint(1, 10) for _ in range(3)] +x0 = [np.random.uniform(0.0, 1.0, N[k]).astype(np.float32) for k in range(3)] +x1 = [np.random.uniform(0.0, 1.0, N[k]).astype(np.float32) for k in range(3)] +y0 = [x0[k] + x1[k] for k in range(3)] +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), +] +expect( + node, + inputs=[x0, x1], + outputs=[y0], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_add_2_sequences", +) +``` + +
+ + +
+sequence_map_extract_shapes + +```python +body = onnx.helper.make_graph( + [onnx.helper.make_node("Shape", ["x"], ["shape"])], + "seq_map_body", + [ + onnx.helper.make_tensor_value_info( + "x", onnx.TensorProto.FLOAT, ["H", "W", "C"] + ) + ], + [onnx.helper.make_tensor_value_info("shape", onnx.TensorProto.INT64, [3])], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["in_seq"], outputs=["shapes"], body=body +) + +shapes = [ + np.array([40, 30, 3], dtype=np.int64), + np.array([20, 10, 3], dtype=np.int64), + np.array([10, 5, 3], dtype=np.int64), +] +x0 = [np.zeros(shape, dtype=np.float32) for shape in shapes] +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto( + onnx.TensorProto.FLOAT, ["H", "W", "C"] + ) + ), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.INT64, [3]) + ), +] +expect( + node, + inputs=[x0], + outputs=[shapes], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_extract_shapes", +) +``` + +
+ + +
+sequence_map_identity_1_sequence + +```python +body = onnx.helper.make_graph( + [onnx.helper.make_node("Identity", ["in0"], ["out0"])], + "seq_map_body", + [onnx.helper.make_tensor_value_info("in0", onnx.TensorProto.FLOAT, ["N"])], + [onnx.helper.make_tensor_value_info("out0", onnx.TensorProto.FLOAT, ["M"])], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["x"], outputs=["y"], body=body +) + +x = [np.random.uniform(0.0, 1.0, 10).astype(np.float32) for _ in range(3)] +y = x +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), +] +expect( + node, + inputs=[x], + outputs=[y], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_identity_1_sequence", +) +``` + +
+ + +
+sequence_map_identity_1_sequence_1_tensor + +```python +body = onnx.helper.make_graph( + [ + onnx.helper.make_node("Identity", ["in0"], ["out0"]), + onnx.helper.make_node("Identity", ["in1"], ["out1"]), + ], + "seq_map_body", + [ + onnx.helper.make_tensor_value_info( + "in0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "in1", onnx.TensorProto.FLOAT, ["M"] + ), + ], + [ + onnx.helper.make_tensor_value_info( + "out0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "out1", onnx.TensorProto.FLOAT, ["M"] + ), + ], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["x0", "x1"], outputs=["y0", "y1"], body=body +) + +x0 = [ + np.random.uniform(0.0, 1.0, np.random.randint(1, 10)).astype(np.float32) + for _ in range(3) +] +x1 = np.random.uniform(0.0, 1.0, np.random.randint(1, 10)).astype(np.float32) +y0 = x0 +y1 = [x1 for _ in range(3)] +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["M"]), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["M"]) + ), +] +expect( + node, + inputs=[x0, x1], + outputs=[y0, y1], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_identity_1_sequence_1_tensor", +) +``` + +
+ + +
+sequence_map_identity_2_sequences + +```python +body = onnx.helper.make_graph( + [ + onnx.helper.make_node("Identity", ["in0"], ["out0"]), + onnx.helper.make_node("Identity", ["in1"], ["out1"]), + ], + "seq_map_body", + [ + onnx.helper.make_tensor_value_info( + "in0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "in1", onnx.TensorProto.FLOAT, ["M"] + ), + ], + [ + onnx.helper.make_tensor_value_info( + "out0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "out1", onnx.TensorProto.FLOAT, ["M"] + ), + ], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["x0", "x1"], outputs=["y0", "y1"], body=body +) + +x0 = [ + np.random.uniform(0.0, 1.0, np.random.randint(1, 10)).astype(np.float32) + for _ in range(3) +] +x1 = [ + np.random.uniform(0.0, 1.0, np.random.randint(1, 10)).astype(np.float32) + for _ in range(3) +] +y0 = x0 +y1 = x1 +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["M"]) + ), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["M"]) + ), +] +expect( + node, + inputs=[x0, x1], + outputs=[y0, y1], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_identity_2_sequences", +) +``` + +
+ + +### **Shape** + + Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. + Optional attributes start and end can be used to compute a slice of the input tensor's shape. + If start axis is omitted, the slice starts from axis 0. + The end axis, if specified, is exclusive (and the returned value will not include the size of that axis). + If the end axis is omitted, the axes upto the last one will be included. + Negative axes indicate counting back from the last axis. + Note that axes will be clamped to the range [0, r], where r is the + rank of the input tensor if they are out-of-range (after adding r in the case of + negative axis). Thus, specifying any end value > r is equivalent to specifying an end + value of r, and specifying any start value < -r is equivalent to specifying a start + value of 0. If start > end, the result will be an empty shape. + + Examples: + + ``` + Input tensor with shape: [2, 3, 4] + No attributes specified. + Output: [2, 3, 4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: -1 + Output: [4] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + end: -1 + Output: [2, 3] + ``` + + ``` + Input tensor with shape: [2, 3, 4] + start: 1 + end: 2 + Output: [3] + ``` + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 13, 15, 19, 21, 23, 24 + +#### Attributes + +
+
end : int
+
(Optional) Ending axis for slicing the shape. Negative value means counting dimensions from the back. If omitted, sizes of all axes upto (including) the last one will be included.
+
start : int (default is 0)
+
(Optional) Starting axis for slicing the shape. Default value is 0.Negative value means counting dimensions from the back.
+
+ +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
shape (non-differentiable) : T1
+
Shape of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor.
+
+ + +#### Examples + +
+shape + +```python +x = np.array( + [ + [1, 2, 3], + [4, 5, 6], + ] +).astype(np.float32) +test_shape("_example", x) # preserve names of original test cases + +x = np.random.randn(3, 4, 5).astype(np.float32) + +test_shape("", x) # preserve names of original test cases + +test_shape("_start_1", x, start=1) + +test_shape("_end_1", x, end=1) + +test_shape("_start_negative_1", x, start=-1) + +test_shape("_end_negative_1", x, end=-1) + +test_shape("_start_1_end_negative_1", x, start=1, end=-1) + +test_shape("_start_1_end_2", x, start=1, end=2) + +test_shape("_clip_start", x, start=-10) + +test_shape("_clip_end", x, end=10) + +test_shape("_start_greater_than_end", x, start=2, end=1) +``` + +
+ + +### **Shrink** + + Shrink takes one input data (Tensor) and produces one Tensor output, + having same datatype and shape with input. It has two attributes, lambd and + bias. The formula of this operator is: If x < -lambd, y = x + bias; + If x > lambd, y = x - bias; Otherwise, y = 0. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
bias : float (default is 0.0)
+
The bias value added to output. Default is 0.
+
lambd : float (default is 0.5)
+
The lambd value for the Shrink formulation. Default is 0.5.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
The input data as Tensor.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double)
+
Constrain input to only numeric types.
+
+ + +#### Examples + +
+hard_shrink + +```python +node = onnx.helper.make_node( + "Shrink", + inputs=["x"], + outputs=["y"], + lambd=1.5, +) +X = np.arange(-2.0, 2.1, dtype=np.float32) +Y = np.array([-2, 0, 0, 0, 2], dtype=np.float32) +expect(node, inputs=[X], outputs=[Y], name="test_shrink_hard") +``` + +
+ + +
+soft_shrink + +```python +node = onnx.helper.make_node( + "Shrink", + inputs=["x"], + outputs=["y"], + lambd=1.5, + bias=1.5, +) +X = np.arange(-2.0, 2.1, dtype=np.float32) +Y = np.array([-0.5, 0, 0, 0, 0.5], dtype=np.float32) +expect(node, inputs=[X], outputs=[Y], name="test_shrink_soft") +``` + +
+ + +### **Sigmoid** + + Sigmoid takes one input data (Tensor) and produces one output data + (Tensor) where the sigmoid function, y = 1 / (1 + exp(-x)), is applied to the + tensor elementwise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+sigmoid + +```python +node = onnx.helper.make_node( + "Sigmoid", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = 1.0 / ( + 1.0 + np.exp(np.negative(x)) +) # expected output [0.26894143, 0.5, 0.7310586] +expect(node, inputs=[x], outputs=[y], name="test_sigmoid_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = 1.0 / (1.0 + np.exp(np.negative(x))) +expect(node, inputs=[x], outputs=[y], name="test_sigmoid") +``` + +
+ + +### **Sign** + + Calculate the sign of the given input tensor element-wise. + If input > 0, output 1. if input < 0, output -1. if input == 0, output 0. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Inputs + +
+
input (non-differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (non-differentiable) : T
+
The sign of the input tensor computed element-wise. It has the same shape and type of the input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ + +#### Examples + +
+sign + +```python +node = onnx.helper.make_node( + "Sign", + inputs=["x"], + outputs=["y"], +) + +x = np.array(range(-5, 6)).astype(np.float32) +y = np.sign(x) +expect(node, inputs=[x], outputs=[y], name="test_sign") +``` + +
+ + +### **Sin** + + Calculates the sine of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 7 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The sine of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+sin + +```python +node = onnx.helper.make_node( + "Sin", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.sin(x) +expect(node, inputs=[x], outputs=[y], name="test_sin_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.sin(x) +expect(node, inputs=[x], outputs=[y], name="test_sin") +``` + +
+ + +### **Sinh** + + Calculates the hyperbolic sine of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic sine values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+sinh + +```python +node = onnx.helper.make_node( + "Sinh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.sinh(x) # expected output [-1.17520118, 0., 1.17520118] +expect(node, inputs=[x], outputs=[y], name="test_sinh_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.sinh(x) +expect(node, inputs=[x], outputs=[y], name="test_sinh") +``` + +
+ + +### **Size** + + Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 13, 19, 21, 23, 24 + +#### Inputs + +
+
data (non-differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
size (non-differentiable) : T1
+
Total number of elements of the input tensor
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Input tensor can be of arbitrary type.
+
T1 : tensor(int64)
+
Constrain output to int64 tensor, which should be a scalar though.
+
+ + +#### Examples + +
+size + +```python +node = onnx.helper.make_node( + "Size", + inputs=["x"], + outputs=["y"], +) + +x = np.array( + [ + [1, 2, 3], + [4, 5, 6], + ] +).astype(np.float32) +y = np.array(6).astype(np.int64) + +expect(node, inputs=[x], outputs=[y], name="test_size_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.array(x.size).astype(np.int64) + +expect(node, inputs=[x], outputs=[y], name="test_size") +``` + +
+ + +### **Slice** + + Produces a slice of the input tensor along multiple axes. Similar to numpy: + https://numpy.org/doc/stable/user/basics.indexing.html?highlight=slice#slicing-and-striding + + Slice uses the `starts`, `ends`, `axes` and `steps` inputs to select a sub-tensor + of its input `data` tensor. + + An effective `starts[i]`, `ends[i]`, and `steps[i]` must be computed for each `i` + in `[0, ... r-1]` where `r = rank(input)` as follows: + + If `axes` are omitted, they are set to `[0, ..., r-1]`. + If `steps` are omitted, they are set to `[1, ..., 1]` of length `len(starts)` + + The effective values are initialized as `start[i] = 0`, `ends[i] = dims[i]` where + `dims` are the dimensions of `input` and `steps[i] = 1`. + + All negative elements of `axes` are made non-negative by adding `r` to them, where + `r =rank(input)`. + + All negative values in `starts[i]` and `ends[i]` have `dims[axes[i]]` added to them, + where `dims` are the dimensions of `input`. Then `start[axes[i]]` is the adjusted + `starts[i]` is clamped into the range `[0, dims[axes[i]]]` for positive stepping + and `[0, dims[axes[i]]-1]` for negative stepping. + + The clamping for the adjusted `ends[i]` depends on the sign of `steps[i]` and must + accommodate copying 0 through `dims[axes[i]]` elements, so for positive stepping + `ends[axes[i]]` is clamped to `[0, dims[axes[i]]]`, while for negative stepping it + is clamped to `[-1, dims[axes[i]]-1]`. + + Finally, `steps[axes[i]] = steps[i]`. + + For slicing to the end of a dimension with unknown size, it is recommended to pass + in `INT_MAX` when slicing forward and 'INT_MIN' when slicing backward. + + Example 1: + + ``` + data = [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + axes = [0, 1] + starts = [1, 0] + ends = [2, 3] + steps = [1, 2] + result = [ + [5, 7], + ] + ``` + + Example 2: + + ``` + data = [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + starts = [0, 1] + ends = [-1, 1000] + result = [ + [2, 3, 4], + ] + ``` + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 10, 11 + +#### Inputs (3 - 5) + +
+
data (differentiable) : T
+
Tensor of data to extract slices from.
+
starts (non-differentiable) : Tind
+
1-D tensor of starting indices of corresponding axis in `axes`
+
ends (non-differentiable) : Tind
+
1-D tensor of ending indices (exclusive) of corresponding axis in `axes`
+
axes (optional, non-differentiable) : Tind
+
1-D tensor of axes that `starts` and `ends` apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated.
+
steps (optional, non-differentiable) : Tind
+
1-D tensor of slice step of corresponding axis in `axes`. Negative value means slicing backward. 'steps' cannot be 0. Defaults to 1s.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Sliced data tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
Tind : tensor(int32), tensor(int64)
+
Constrain indices to integer types
+
+ + +#### Examples + +
+slice + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes", "steps"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +y = x[0:3, 0:10] +starts = np.array([0, 0], dtype=np.int64) +ends = np.array([3, 10], dtype=np.int64) +axes = np.array([0, 1], dtype=np.int64) +steps = np.array([1, 1], dtype=np.int64) + +expect( + node, inputs=[x, starts, ends, axes, steps], outputs=[y], name="test_slice" +) +``` + +
+ + +
+slice_default_axes + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([0, 0, 3], dtype=np.int64) +ends = np.array([20, 10, 4], dtype=np.int64) +y = x[:, :, 3:4] + +expect( + node, inputs=[x, starts, ends], outputs=[y], name="test_slice_default_axes" +) +``` + +
+ + +
+slice_default_steps + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([0, 0, 3], dtype=np.int64) +ends = np.array([20, 10, 4], dtype=np.int64) +axes = np.array([0, 1, 2], dtype=np.int64) +y = x[:, :, 3:4] + +expect( + node, + inputs=[x, starts, ends, axes], + outputs=[y], + name="test_slice_default_steps", +) +``` + +
+ + +
+slice_end_out_of_bounds + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes", "steps"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([1], dtype=np.int64) +ends = np.array([1000], dtype=np.int64) +axes = np.array([1], dtype=np.int64) +steps = np.array([1], dtype=np.int64) +y = x[:, 1:1000] + +expect( + node, + inputs=[x, starts, ends, axes, steps], + outputs=[y], + name="test_slice_end_out_of_bounds", +) +``` + +
+ + +
+slice_neg + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes", "steps"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([0], dtype=np.int64) +ends = np.array([-1], dtype=np.int64) +axes = np.array([1], dtype=np.int64) +steps = np.array([1], dtype=np.int64) +y = x[:, 0:-1] + +expect( + node, + inputs=[x, starts, ends, axes, steps], + outputs=[y], + name="test_slice_neg", +) +``` + +
+ + +
+slice_neg_steps + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes", "steps"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([20, 10, 4], dtype=np.int64) +ends = np.array([0, 0, 1], dtype=np.int64) +axes = np.array([0, 1, 2], dtype=np.int64) +steps = np.array([-1, -3, -2]).astype(np.int64) +y = x[20:0:-1, 10:0:-3, 4:1:-2] + +expect( + node, + inputs=[x, starts, ends, axes, steps], + outputs=[y], + name="test_slice_neg_steps", +) +``` + +
+ + +
+slice_negative_axes + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([0, 0, 3], dtype=np.int64) +ends = np.array([20, 10, 4], dtype=np.int64) +axes = np.array([0, -2, -1], dtype=np.int64) +y = x[:, :, 3:4] + +expect( + node, + inputs=[x, starts, ends, axes], + outputs=[y], + name="test_slice_negative_axes", +) +``` + +
+ + +
+slice_start_out_of_bounds + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes", "steps"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([1000], dtype=np.int64) +ends = np.array([1000], dtype=np.int64) +axes = np.array([1], dtype=np.int64) +steps = np.array([1], dtype=np.int64) +y = x[:, 1000:1000] + +expect( + node, + inputs=[x, starts, ends, axes, steps], + outputs=[y], + name="test_slice_start_out_of_bounds", +) +``` + +
+ + +### **Softmax** + + The operator computes the normalized exponential values for the given input: + + Softmax(input, axis) = Exp(input) / ReduceSum(Exp(input), axis=axis, keepdims=1) + + The "axis" attribute indicates the dimension along which Softmax + will be performed. The output tensor has the same shape + and contains the Softmax values of the corresponding input. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 11 + +#### Attributes + +
+
axis : int (default is -1)
+
+Describes the dimension Softmax will be performed on. +Negative value means counting dimensions +from the back. Accepted range is [-r, r-1] where r = rank(input). +
+
+ +#### Inputs + +
+
input (differentiable) : T
+
The input tensor of rank >= axis.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The output values with the same shape as the input tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+softmax + +```python +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], +) +x = np.array([[-1, 0, 1]]).astype(np.float32) +# expected output [[0.09003058, 0.24472848, 0.66524094]] +y = softmax(x, axis=1) +expect(node, inputs=[x], outputs=[y], name="test_softmax_example") +``` + +
+ + +
+softmax_axis + +```python +x = np.array([[0, 1, 2, 3], [10000, 10001, 10002, 10003]]).astype(np.float32) +# expected output +# [[0.032058604 0.08714432 0.23688284 0.6439143 ] +# [0.032058604 0.08714432 0.23688284 0.6439143 ]] +y = softmax(x) + +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], +) +expect(node, inputs=[x], outputs=[y], name="test_softmax_large_number") + +x = np.abs(np.random.randn(3, 4, 5).astype(np.float32)) +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], + axis=0, +) +y = softmax(x, axis=0) +expect(node, inputs=[x], outputs=[y], name="test_softmax_axis_0") + +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], + axis=1, +) +y = softmax(x, axis=1) +expect(node, inputs=[x], outputs=[y], name="test_softmax_axis_1") + +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], + axis=2, +) +y = softmax(x, axis=2) +expect(node, inputs=[x], outputs=[y], name="test_softmax_axis_2") + +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], + axis=-1, +) +y = softmax(x, axis=-1) +expect(node, inputs=[x], outputs=[y], name="test_softmax_negative_axis") + +# default axis is -1 +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], +) +expect(node, inputs=[x], outputs=[y], name="test_softmax_default_axis") +``` + +
+ + +### **SoftmaxCrossEntropyLoss** + + Loss function that measures the softmax cross entropy + between 'scores' and 'labels'. + This operator first computes a loss tensor whose shape is identical to the labels input. + If the input is 2-D with shape (N, C), the loss tensor may be a N-element vector L = (l_1, l_2, ..., l_N). + If the input is N-D tensor with shape (N, C, D1, D2, ..., Dk), + the loss tensor L may have (N, D1, D2, ..., Dk) as its shape and L[i,][j_1][j_2]...[j_k] denotes a scalar element in L. + After L is available, this operator can optionally do a reduction operator. + + * shape(scores): (N, C) where C is the number of classes, or (N, C, D1, D2,..., Dk), + with K >= 1 in case of K-dimensional loss. + * shape(labels): (N) where each value is 0 <= labels[i] <= C-1, or (N, D1, D2,..., Dk), + with K >= 1 in case of K-dimensional loss. + + The loss for one sample, l_i, can calculated as follows: + ``` + l[i][d1][d2]...[dk] = -y[i][c][d1][d2]..[dk], where i is the index of classes. + ``` + or + ``` + l[i][d1][d2]...[dk] = -y[i][c][d1][d2]..[dk] * weights[c], if 'weights' is provided. + ``` + + loss is zero for the case when label-value equals ignore_index. + ``` + l[i][d1][d2]...[dk] = 0, when labels[n][d1][d2]...[dk] = ignore_index + ``` + + where: + ``` + p = Softmax(scores) + y = Log(p) + c = labels[i][d1][d2]...[dk] + ``` + + Finally, L is optionally reduced: + + * If reduction = 'none', the output is L with shape (N, D1, D2, ..., Dk). + * If reduction = 'sum', the output is scalar: Sum(L). + * If reduction = 'mean', the output is scalar: ReduceMean(L), or if weight is provided: `ReduceSum(L) / ReduceSum(W)`, + where tensor W is of shape `(N, D1, D2, ..., Dk)` and `W[n][d1][d2]...[dk] = weights[labels[i][d1][d2]...[dk]]`. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 12 + +#### Attributes + +
+
ignore_index : int
+
Specifies a target value that is ignored and does not contribute to the input gradient. It's an optional value.
+
reduction : string (default is mean)
+
Type of reduction to apply to loss: none, sum, mean(default). 'none': no reduction will be applied, 'sum': the output will be summed. 'mean': the sum of the output will be divided by the number of elements in the output.
+
+ +#### Inputs (2 - 3) + +
+
scores (differentiable) : T
+
The predicted outputs with shape [batch_size, class_size], or [batch_size, class_size, D1, D2 , ..., Dk], where K is the number of dimensions.
+
labels (non-differentiable) : Tind
+
The ground truth output tensor, with shape [batch_size], or [batch_size, D1, D2, ..., Dk], where K is the number of dimensions. Labels element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the label values should either be in the range [0, C) or have the value ignore_index.
+
weights (optional, non-differentiable) : T
+
A manual rescaling weight given to each class. If given, it has to be a 1D Tensor assigning weight to each of the classes. Otherwise, it is treated as if having all ones.
+
+ +#### Outputs (1 - 2) + +
+
output (differentiable) : T
+
Weighted loss float Tensor. If reduction is 'none', this has the shape of [batch_size], or [batch_size, D1, D2, ..., Dk] in case of K-dimensional loss. Otherwise, it is a scalar.
+
log_prob (optional, differentiable) : T
+
Log probability tensor. If the output of softmax is prob, its value is log(prob).
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
Tind : tensor(int32), tensor(int64)
+
Constrain target to integer types
+
+ + +#### Examples + +
+input_shape_is_NCd1_mean_weight_negative_ii + +```python +reduction = "mean" +ignore_index = np.int64(-1) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1 = 3, 5, 6 +np.random.seed(0) +x = np.random.rand(N, C, dim1).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N, dim1)).astype(np.int64) +labels[0][0] = -1 +weight = np.random.rand(C).astype(np.float32) + +sce = softmaxcrossentropy( + x, labels, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[x, labels, weight], + outputs=[sce], + name="test_sce_NCd1_mean_weight_negative_ii", +) +``` + +
+ + +
+input_shape_is_NCd1_mean_weight_negative_ii_log_prob + +```python +reduction = "mean" +ignore_index = np.int64(-1) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1 = 3, 5, 6 +np.random.seed(0) +x = np.random.rand(N, C, dim1).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N, dim1)).astype(np.int64) +labels[0][0] = -1 +weight = np.random.rand(C).astype(np.float32) + +loss, log_prob = softmaxcrossentropy( + x, + labels, + weight=weight, + reduction=reduction, + ignore_index=ignore_index, + get_log_prob=True, +) + +expect( + node, + inputs=[x, labels, weight], + outputs=[loss, log_prob], + name="test_sce_NCd1_mean_weight_negative_ii_log_prob", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3_none_no_weight_negative_ii + +```python +reduction = "none" +ignore_index = np.int64(-5) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1, dim2, dim3 = 3, 5, 6, 6, 5 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N, dim1, dim2, dim3)).astype( + np.int64 +) +labels[0][0][0][0] = -5 + +sce = softmaxcrossentropy( + x, labels, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[x, labels], + outputs=[sce], + name="test_sce_NCd1d2d3_none_no_weight_negative_ii", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3_none_no_weight_negative_ii_log_prob + +```python +reduction = "none" +ignore_index = np.int64(-5) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1, dim2, dim3 = 3, 5, 6, 6, 5 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N, dim1, dim2, dim3)).astype( + np.int64 +) +labels[0][0][0][0] = -5 + +loss, log_prob = softmaxcrossentropy( + x, labels, reduction=reduction, ignore_index=ignore_index, get_log_prob=True +) + +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3_sum_weight_high_ii + +```python +reduction = "sum" +ignore_index = np.int64(10) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C = 3, 5 +np.random.seed(0) +x = np.random.rand(N, C).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N)).astype(np.int64) +labels[0] = 10 +weight = np.random.rand(C).astype(np.float32) + +sce = softmaxcrossentropy( + x, labels, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[x, labels, weight], + outputs=[sce], + name="test_sce_NCd1d2d3_sum_weight_high_ii", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3_sum_weight_high_ii_log_prob + +```python +reduction = "sum" +ignore_index = np.int64(10) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C = 3, 5 +np.random.seed(0) +x = np.random.rand(N, C).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N)).astype(np.int64) +labels[0] = 10 +weight = np.random.rand(C).astype(np.float32) + +loss, log_prob = softmaxcrossentropy( + x, + labels, + weight=weight, + reduction=reduction, + ignore_index=ignore_index, + get_log_prob=True, +) + +expect( + node, + inputs=[x, labels, weight], + outputs=[loss, log_prob], + name="test_sce_NCd1d2d3_sum_weight_high_ii_log_prob", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3d4d5_mean_weight + +```python +reduction = "mean" + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +labels = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +sce = softmaxcrossentropy(x, labels, weight=weight, reduction=reduction) + +expect( + node, + inputs=[x, labels, weight], + outputs=[sce], + name="test_sce_NCd1d2d3d4d5_mean_weight", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3d4d5_mean_weight_log_prob + +```python +reduction = "mean" + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +labels = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +loss, log_prob = softmaxcrossentropy( + x, labels, weight=weight, reduction=reduction, get_log_prob=True +) + +expect( + node, + inputs=[x, labels, weight], + outputs=[loss, log_prob], + name="test_sce_NCd1d2d3d4d5_mean_weight_log_prob", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3d4d5_none_no_weight + +```python +reduction = "none" + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +labels = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) + +sce = softmaxcrossentropy(x, labels, reduction=reduction) + +expect( + node, + inputs=[x, labels], + outputs=[sce], + name="test_sce_NCd1d2d3d4d5_none_no_weight", +) +``` + +
+ + +
+input_shape_is_NCd1d2d3d4d5_none_no_weight_log_prob + +```python +reduction = "none" + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +labels = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) + +loss, log_prob = softmaxcrossentropy( + x, labels, reduction=reduction, get_log_prob=True +) + +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_NCd1d2d3d4d5_none_no_weight_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_mean + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels) + +# Check results +expect(node, inputs=[x, labels], outputs=[sce], name="test_sce_mean") +``` + +
+ + +
+softmaxcrossentropy_mean_3d + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +y = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, y) + +# Check results +expect(node, inputs=[x, y], outputs=[sce], name="test_sce_mean_3d") +``` + +
+ + +
+softmaxcrossentropy_mean_3d_log_prob + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +y = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy(x, y, get_log_prob=True) + +# Check results +expect( + node, + inputs=[x, y], + outputs=[loss, log_prob], + name="test_sce_mean_3d_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_mean_log_prob + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy(x, labels, get_log_prob=True) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_mean_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_mean_no_weights_ii + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +labels[0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, ignore_index=ignore_index) + +# Check results +expect( + node, inputs=[x, labels], outputs=[sce], name="test_sce_mean_no_weight_ii" +) +``` + +
+ + +
+softmaxcrossentropy_mean_no_weights_ii_3d + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) +labels[0][0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, ignore_index=ignore_index) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[sce], + name="test_sce_mean_no_weight_ii_3d", +) +``` + +
+ + +
+softmaxcrossentropy_mean_no_weights_ii_3d_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) +labels[0][0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, ignore_index=ignore_index, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_mean_no_weight_ii_3d_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_mean_no_weights_ii_4d + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2, 7).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2, 7)).astype(np.int64) +labels[0][0][0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy( + x, labels, reduction=reduction, ignore_index=ignore_index +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[sce], + name="test_sce_mean_no_weight_ii_4d", +) +``` + +
+ + +
+softmaxcrossentropy_mean_no_weights_ii_4d_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2, 7).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2, 7)).astype(np.int64) +labels[0][0][0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, reduction=reduction, ignore_index=ignore_index, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_mean_no_weight_ii_4d_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_mean_no_weights_ii_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +labels[0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, ignore_index=ignore_index, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_mean_no_weight_ii_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_mean_weights + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, weight=weights) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[sce], + name="test_sce_mean_weight", +) +``` + +
+ + +
+softmaxcrossentropy_mean_weights_ii + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(0) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +labels[0] = np.int64(0) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, weight=weights, ignore_index=ignore_index) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[sce], + name="test_sce_mean_weight_ii", +) +``` + +
+ + +
+softmaxcrossentropy_mean_weights_ii_3d + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(1) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) +labels[0][0] = np.int64(1) +weights = np.array([0.2, 0.3, 0.6, 0.1, 0.5], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, weight=weights, ignore_index=ignore_index) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[sce], + name="test_sce_mean_weight_ii_3d", +) +``` + +
+ + +
+softmaxcrossentropy_mean_weights_ii_3d_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(1) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) +labels[0][0] = np.int64(1) +weights = np.array([0.2, 0.3, 0.6, 0.1, 0.5], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, weight=weights, ignore_index=ignore_index, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[loss, log_prob], + name="test_sce_mean_weight_ii_3d_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_mean_weights_ii_4d + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2, 7).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2, 7)).astype(np.int64) +labels[0][0][0] = np.int64(2) +weights = np.array([0.2, 0.3, 0.6, 0.1, 0.5], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy( + x, labels, reduction=reduction, weight=weights, ignore_index=ignore_index +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[sce], + name="test_sce_mean_weight_ii_4d", +) +``` + +
+ + +
+softmaxcrossentropy_mean_weights_ii_4d_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2, 7).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2, 7)).astype(np.int64) +labels[0][0][0] = np.int64(2) +weights = np.array([0.2, 0.3, 0.6, 0.1, 0.5], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, + labels, + reduction=reduction, + weight=weights, + ignore_index=ignore_index, + get_log_prob=True, +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[loss, log_prob], + name="test_sce_mean_weight_ii_4d_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_mean_weights_ii_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(0) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +labels[0] = np.int64(0) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, weight=weights, ignore_index=ignore_index, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[loss, log_prob], + name="test_sce_mean_weight_ii_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_mean_weights_log_prob + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, weight=weights, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[loss, log_prob], + name="test_sce_mean_weight_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_none + +```python +# Define operator attributes. +reduction = "none" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, reduction="none") + +# Check results +expect(node, inputs=[x, labels], outputs=[sce], name="test_sce_none") +``` + +
+ + +
+softmaxcrossentropy_none_log_prob + +```python +# Define operator attributes. +reduction = "none" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, reduction="none", get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_none_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_none_weights + +```python +# Define operator attributes. +reduction = "none" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, weight=weights, reduction="none") + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[sce], + name="test_sce_none_weights", +) +``` + +
+ + +
+softmaxcrossentropy_none_weights_log_prob + +```python +# Define operator attributes. +reduction = "none" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, weight=weights, reduction="none", get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[loss, log_prob], + name="test_sce_none_weights_log_prob", +) +``` + +
+ + +
+softmaxcrossentropy_sum + +```python +# Define operator attributes. +reduction = "sum" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, reduction="sum") + +# Check results +expect(node, inputs=[x, labels], outputs=[sce], name="test_sce_sum") +``` + +
+ + +
+softmaxcrossentropy_sum_log_prob + +```python +# Define operator attributes. +reduction = "sum" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, reduction="sum", get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_sum_log_prob", +) +``` + +
+ + +### **Softplus** + + Softplus takes one input data (Tensor) and produces one output data + (Tensor) where the softplus function, y = ln(exp(x) + 1), is applied to + the tensor elementwise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+softplus + +```python +node = onnx.helper.make_node( + "Softplus", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.log( + np.exp(x) + 1 +) # expected output [0.31326166, 0.69314718, 1.31326163] +expect(node, inputs=[x], outputs=[y], name="test_softplus_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.log(np.exp(x) + 1) +expect(node, inputs=[x], outputs=[y], name="test_softplus") +``` + +
+ + +### **Softsign** + + Calculates the softsign (x/(1+|x|)) of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The softsign (x/(1+|x|)) values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+softsign + +```python +node = onnx.helper.make_node( + "Softsign", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.array([-0.5, 0, 0.5]).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_softsign_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = x / (1 + np.abs(x)) +expect(node, inputs=[x], outputs=[y], name="test_softsign") +``` + +
+ + +### **SpaceToDepth** + + SpaceToDepth rearranges blocks of spatial data into depth. More specifically, + this op outputs a copy of the input tensor where values from the height and width dimensions + are moved to the depth dimension. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Attributes + +
+
blocksize : int (required)
+
Blocks of [blocksize, blocksize] are moved.
+
+ +#### Inputs + +
+
input (differentiable) : T
+
Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor of [N, C * blocksize * blocksize, H/blocksize, W/blocksize].
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ + +#### Examples + +
+example + +```python +node = onnx.helper.make_node( + "SpaceToDepth", + inputs=["x"], + outputs=["y"], + blocksize=2, +) + +# (1, 1, 4, 6) input tensor +x = np.array( + [ + [ + [ + [0, 6, 1, 7, 2, 8], + [12, 18, 13, 19, 14, 20], + [3, 9, 4, 10, 5, 11], + [15, 21, 16, 22, 17, 23], + ] + ] + ] +).astype(np.float32) + +# (1, 4, 2, 3) output tensor +y = np.array( + [ + [ + [[0, 1, 2], [3, 4, 5]], + [[6, 7, 8], [9, 10, 11]], + [[12, 13, 14], [15, 16, 17]], + [[18, 19, 20], [21, 22, 23]], + ] + ] +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_spacetodepth_example") +``` + +
+ + +
+spacetodepth + +```python +b, c, h, w = shape = (2, 2, 6, 6) +blocksize = 2 +node = onnx.helper.make_node( + "SpaceToDepth", + inputs=["x"], + outputs=["y"], + blocksize=blocksize, +) +x = np.random.random_sample(shape).astype(np.float32) +tmp = np.reshape( + x, [b, c, h // blocksize, blocksize, w // blocksize, blocksize] +) +tmp = np.transpose(tmp, [0, 3, 5, 1, 2, 4]) +y = np.reshape(tmp, [b, c * (blocksize**2), h // blocksize, w // blocksize]) +expect(node, inputs=[x], outputs=[y], name="test_spacetodepth") +``` + +
+ + +### **Split** + + Split a tensor into a list of tensors, along the specified 'axis'. + Either input 'split' or the attribute 'num_outputs' should be specified, but not both. + If the attribute 'num_outputs' is specified, then the tensor is split into equal sized parts. + If the tensor is not evenly splittable into `num_outputs`, the last chunk will be smaller. + If the input 'split' is specified, it indicates the sizes of each output in the split. + +#### Version + +This version of the operator has been available since version 18 of the default ONNX operator set. + +Other versions of this operator: 1, 2, 11, 13 + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1] where r = rank(input).
+
num_outputs : int
+
Number of outputs to split parts of the tensor into. If the tensor is not evenly splittable the last chunk will be smaller.
+
+ +#### Inputs (1 - 2) + +
+
input (differentiable) : T
+
The tensor to split
+
split (optional, non-differentiable) : tensor(int64)
+
Optional length of each output. Values should be >= 0.Sum of the values must be equal to the dim value at 'axis' specified.
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, differentiable) : T
+
One or more outputs forming list of tensors after splitting
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ + +#### Examples + +
+1d_opset13 + +```python +node_input = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float32) + +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2", "output_3"], + axis=0, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0]).astype(np.float32), + np.array([5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_1d_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2"], + axis=0, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0, 5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_1d_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) +``` + +
+ + +
+1d_opset18 + +```python +node_input = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float32) + +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2", "output_3"], + axis=0, + num_outputs=3, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0]).astype(np.float32), + np.array([5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_1d_opset18", +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2"], + axis=0, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0, 5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_1d_opset18", +) +``` + +
+ + +
+1d_uneven_split_opset18 + +```python +node_input = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0]).astype(np.float32) + +# If axis is not specified, split is applied on default axis 0 +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2", "output_3", "output_4"], + num_outputs=4, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0]).astype(np.float32), + np.array([5.0, 6.0]).astype(np.float32), + np.array([7.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_1d_uneven_split_opset18", +) +``` + +
+ + +
+2d_opset13 + +```python +node_input = np.array( + [[1.0, 2.0, 3.0, 4.0, 5.0, 6.0], [7.0, 8.0, 9.0, 10.0, 11.0, 12.0]] +).astype(np.float32) + +node = onnx.helper.make_node( + "Split", inputs=["input"], outputs=["output_1", "output_2"], axis=1 +) + +expected_outputs = [ + np.array([[1.0, 2.0, 3.0], [7.0, 8.0, 9.0]]).astype(np.float32), + np.array([[4.0, 5.0, 6.0], [10.0, 11.0, 12.0]]).astype(np.float32), +] + +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_2d_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2"], + axis=1, +) + +expected_outputs = [ + np.array([[1.0, 2.0], [7.0, 8.0]]).astype(np.float32), + np.array([[3.0, 4.0, 5.0, 6.0], [9.0, 10.0, 11.0, 12.0]]).astype( + np.float32 + ), +] + +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_2d_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) +``` + +
+ + +
+2d_opset18 + +```python +node_input = np.array( + [[1.0, 2.0, 3.0, 4.0, 5.0, 6.0], [7.0, 8.0, 9.0, 10.0, 11.0, 12.0]] +).astype(np.float32) + +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2"], + axis=1, + num_outputs=2, +) + +expected_outputs = [ + np.array([[1.0, 2.0, 3.0], [7.0, 8.0, 9.0]]).astype(np.float32), + np.array([[4.0, 5.0, 6.0], [10.0, 11.0, 12.0]]).astype(np.float32), +] + +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_2d", +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2"], + axis=1, +) + +expected_outputs = [ + np.array([[1.0, 2.0], [7.0, 8.0]]).astype(np.float32), + np.array([[3.0, 4.0, 5.0, 6.0], [9.0, 10.0, 11.0, 12.0]]).astype( + np.float32 + ), +] + +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_2d_opset18", +) +``` + +
+ + +
+2d_uneven_split_opset18 + +```python +node_input = np.array( + [ + [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0], + [9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0], + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2", "output_3"], + axis=1, + num_outputs=3, +) + +expected_outputs = [ + np.array([[1.0, 2.0, 3.0], [9.0, 10.0, 11.0]]).astype(np.float32), + np.array([[4.0, 5.0, 6.0], [12.0, 13.0, 14.0]]).astype(np.float32), + np.array([[7.0, 8.0], [15.0, 16.0]]).astype(np.float32), +] + +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_2d_uneven_split_opset18", +) +``` + +
+ + +
+default_values_opset13 + +```python +node_input = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float32) + +# If axis is not specified, split is applied on default axis 0 +node = onnx.helper.make_node( + "Split", inputs=["input"], outputs=["output_1", "output_2", "output_3"] +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0]).astype(np.float32), + np.array([5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_default_axis_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", inputs=["input", "split"], outputs=["output_1", "output_2"] +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0, 5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_default_axis_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) +``` + +
+ + +
+default_values_opset18 + +```python +node_input = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float32) + +# If axis is not specified, split is applied on default axis 0 +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2", "output_3"], + num_outputs=3, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0]).astype(np.float32), + np.array([5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_default_axis_opset18", +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", inputs=["input", "split"], outputs=["output_1", "output_2"] +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0, 5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_default_axis_opset18", +) +``` + +
+ + +
+zero_size_splits_opset13 + +```python +# 1-dimensional tensor with dimension_size=0 +node_input = np.array([]).astype(np.float32) + +# Split empty tensor to tensors of size zero +split = np.array([0, 0, 0]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2", "output_3"], +) + +expected_outputs = [ + np.array([]).astype(np.float32), + np.array([]).astype(np.float32), + np.array([]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_zero_size_splits_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) +``` + +
+ + +
+zero_size_splits_opset18 + +```python +# 1-dimensional tensor with dimension_size=0 +node_input = np.array([]).astype(np.float32) + +# Split empty tensor to tensors of size zero +split = np.array([0, 0, 0]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2", "output_3"], +) + +expected_outputs = [ + np.array([]).astype(np.float32), + np.array([]).astype(np.float32), + np.array([]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_zero_size_splits_opset18", +) +``` + +
+ + +### **SplitToSequence** + + Split a tensor into a sequence of tensors, along the specified 'axis'. + Lengths of the parts can be specified using the optional argument 'split'. + If the argument `split' is not specified, a default scalar value of 1 + is used as the value of `split'. + 'split' must contain only positive numbers. + 'split' is either a scalar (tensor of empty shape), or a 1-D tensor. + If 'split' is a scalar, then 'input' will be split into chunks all of size 'split' + if possible. The last chunk alone may be smaller than 'split' if the 'input' size + along the given axis 'axis' is not divisible by 'split'. + If 'split' is a 1-dimensional tensor, the input tensor is split into 'size(split)' chunks, + with lengths of the parts on 'axis' specified in 'split'. In this scenario, the sum of entries + in 'split' must be equal to the dimension size of input tensor on 'axis'. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +Other versions of this operator: 11 + +#### Attributes + +
+
axis : int (default is 0)
+
Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1].
+
keepdims : int (default is 1)
+
Keep the split dimension or not. Default 1, which means we keep split dimension. If input 'split' is specified, this attribute is ignored.
+
+ +#### Inputs (1 - 2) + +
+
input : T
+
The tensor to split
+
split (optional) : I
+
Length of each output. It can be either a scalar(tensor of empty shape), or a 1-D tensor. All values must be >= 0.
+
+ +#### Outputs + +
+
output_sequence : S
+
One or more outputs forming a sequence of tensors after splitting
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input types to all tensor types.
+
I : tensor(int32), tensor(int64)
+
Constrain split size to integral tensor.
+
S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128))
+
Constrain output types to all tensor types.
+
+ + +#### Examples + +
+nokeepdims + +```python +data = np.arange(18).reshape((3, 6)).astype(np.float32) + +node = onnx.helper.make_node( + "SplitToSequence", + ["data"], + ["seq"], + axis=1, + keepdims=0, +) + +expected_outputs = [[data[:, i] for i in range(data.shape[1])]] + +expect( + node, + inputs=[data], + outputs=expected_outputs, + name="test_split_to_sequence_nokeepdims", +) +``` + +
+ + +
+with_split_1 + +```python +data = np.arange(18).reshape((3, 6)).astype(np.float32) +split = np.array(2, dtype=np.int64) + +node = onnx.helper.make_node( + "SplitToSequence", ["data", "split"], ["seq"], axis=1 +) + +expected_outputs = [ + [ + np.array([[0.0, 1.0], [6.0, 7.0], [12.0, 13.0]], dtype=np.float32), + np.array([[2.0, 3.0], [8.0, 9.0], [14.0, 15.0]], dtype=np.float32), + np.array([[4.0, 5.0], [10.0, 11.0], [16.0, 17.0]], dtype=np.float32), + ] +] + +expect( + node, + inputs=[data, split], + outputs=expected_outputs, + name="test_split_to_sequence_1", +) +``` + +
+ + +
+with_split_2 + +```python +data = np.arange(18).reshape((3, 6)).astype(np.float32) +split = np.array([1, 2], dtype=np.int64) + +node = onnx.helper.make_node( + "SplitToSequence", ["data", "split"], ["seq"], axis=0 +) + +expected_outputs = [ + [ + data[:1], + data[1:], + ] +] + +expect( + node, + inputs=[data, split], + outputs=expected_outputs, + name="test_split_to_sequence_2", +) +``` + +
+ + +### **Sqrt** + + Square root takes one input data (Tensor) and produces one output data + (Tensor) where the square root is, y = x^0.5, is applied to + the tensor elementwise. If x is negative, then it will return NaN. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+sqrt + +```python +node = onnx.helper.make_node( + "Sqrt", + inputs=["x"], + outputs=["y"], +) + +x = np.array([1, 4, 9]).astype(np.float32) +y = np.sqrt(x) # expected output [1., 2., 3.] +expect(node, inputs=[x], outputs=[y], name="test_sqrt_example") + +x = np.abs(np.random.randn(3, 4, 5).astype(np.float32)) +y = np.sqrt(x) +expect(node, inputs=[x], outputs=[y], name="test_sqrt") +``` + +
+ + +### **Squeeze** + + Remove single-dimensional entries from the shape of a tensor. + Takes an input `axes` with a list of axes to squeeze. + If `axes` is not provided, all the single dimensions will be removed from + the shape. If an axis is selected with shape entry not equal to one, an error is raised. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13, 21, 23, 24 + +#### Inputs (1 - 2) + +
+
data (differentiable) : T
+
Tensors with at least max(dims) dimensions.
+
axes (optional, non-differentiable) : tensor(int64)
+
1D tensor of integers indicating the dimensions to squeeze. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
+
+ +#### Outputs + +
+
squeezed (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types up to IRv13.
+
+ + +#### Examples + +
+squeeze + +```python +node = onnx.helper.make_node( + "Squeeze", + inputs=["x", "axes"], + outputs=["y"], +) +x = np.random.randn(1, 3, 4, 5).astype(np.float32) +axes = np.array([0], dtype=np.int64) +y = np.squeeze(x, axis=0) + +expect(node, inputs=[x, axes], outputs=[y], name="test_squeeze") +``` + +
+ + +
+squeeze_negative_axes + +```python +node = onnx.helper.make_node( + "Squeeze", + inputs=["x", "axes"], + outputs=["y"], +) +x = np.random.randn(1, 3, 1, 5).astype(np.float32) +axes = np.array([-2], dtype=np.int64) +y = np.squeeze(x, axis=-2) +expect(node, inputs=[x, axes], outputs=[y], name="test_squeeze_negative_axes") +``` + +
+ + +### **StringConcat** + + StringConcat concatenates string tensors elementwise (with NumPy-style broadcasting support) + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Inputs + +
+
X (non-differentiable) : T
+
Tensor to prepend in concatenation
+
Y (non-differentiable) : T
+
Tensor to append in concatenation
+
+ +#### Outputs + +
+
Z (non-differentiable) : T
+
Concatenated string tensor
+
+ +#### Type Constraints + +
+
T : tensor(string)
+
Inputs and outputs must be UTF-8 strings
+
+ + +#### Examples + +
+stringconcat + +```python +node = onnx.helper.make_node( + "StringConcat", + inputs=["x", "y"], + outputs=["result"], +) +x = np.array(["abc", "def"]).astype("object") +y = np.array([".com", ".net"]).astype("object") +result = np.array(["abc.com", "def.net"]).astype("object") + +expect(node, inputs=[x, y], outputs=[result], name="test_string_concat") + +x = np.array(["cat", "dog", "snake"]).astype("object") +y = np.array(["s"]).astype("object") +result = np.array(["cats", "dogs", "snakes"]).astype("object") + +expect( + node, + inputs=[x, y], + outputs=[result], + name="test_string_concat_broadcasting", +) + +x = np.array("cat").astype("object") +y = np.array("s").astype("object") +result = np.array("cats").astype("object") + +expect( + node, + inputs=[x, y], + outputs=[result], + name="test_string_concat_zero_dimensional", +) + +x = np.array(["abc", ""]).astype("object") +y = np.array(["", "abc"]).astype("object") +result = np.array(["abc", "abc"]).astype("object") + +expect( + node, + inputs=[x, y], + outputs=[result], + name="test_string_concat_empty_string", +) + +x = np.array(["的", "中"]).astype("object") +y = np.array(["的", "中"]).astype("object") +result = np.array(["的的", "中中"]).astype("object") + +expect( + node, + inputs=[x, y], + outputs=[result], + name="test_string_concat_utf8", +) +``` + +
+ + +### **StringNormalizer** + + StringNormalization performs string operations for basic cleaning. + This operator has only one input (denoted by X) and only one output + (denoted by Y). This operator first examines the elements in the X, + and removes elements specified in "stopwords" attribute. + After removing stop words, the intermediate result can be further lowercased, + uppercased, or just returned depending the "case_change_action" attribute. + This operator only accepts [C]- and [1, C]-tensor. + If all elements in X are dropped, the output will be the empty value of string tensor with shape [1] + if input shape is [C] and shape [1, 1] if input shape is [1, C]. + +#### Version + +This version of the operator has been available since version 10 of the default ONNX operator set. + +#### Attributes + +
+
case_change_action : string (default is NONE)
+
string enum that cases output to be lowercased/uppercases/unchanged. Valid values are "LOWER", "UPPER", "NONE". Default is "NONE"
+
is_case_sensitive : int (default is 0)
+
Boolean. Whether the identification of stop words in X is case-sensitive. Default is false
+
locale : string
+
Environment dependent string that denotes the locale according to which output strings needs to be upper/lowercased.Default en_US or platform specific equivalent as decided by the implementation.
+
stopwords : list of strings
+
List of stop words. If not set, no word would be removed from X.
+
+ +#### Inputs + +
+
X : tensor(string)
+
UTF-8 strings to normalize
+
+ +#### Outputs + +
+
Y : tensor(string)
+
UTF-8 Normalized strings
+
+ +#### Type Constraints + + + +#### Examples + +
+monday_casesensintive_lower + +```python +input = np.array(["monday", "tuesday", "wednesday", "thursday"]).astype(object) +output = np.array(["tuesday", "wednesday", "thursday"]).astype(object) +stopwords = ["monday"] + +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + case_change_action="LOWER", + is_case_sensitive=1, + stopwords=stopwords, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_export_monday_casesensintive_lower", +) +``` + +
+ + +
+monday_casesensintive_nochangecase + +```python +input = np.array(["monday", "tuesday", "wednesday", "thursday"]).astype(object) +output = np.array(["tuesday", "wednesday", "thursday"]).astype(object) +stopwords = ["monday"] + +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + is_case_sensitive=1, + stopwords=stopwords, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_export_monday_casesensintive_nochangecase", +) +``` + +
+ + +
+monday_casesensintive_upper + +```python +input = np.array(["monday", "tuesday", "wednesday", "thursday"]).astype(object) +output = np.array(["TUESDAY", "WEDNESDAY", "THURSDAY"]).astype(object) +stopwords = ["monday"] + +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + case_change_action="UPPER", + is_case_sensitive=1, + stopwords=stopwords, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_export_monday_casesensintive_upper", +) +``` + +
+ + +
+monday_empty_output + +```python +input = np.array(["monday", "monday"]).astype(object) +output = np.array([""]).astype(object) +stopwords = ["monday"] + +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + case_change_action="UPPER", + is_case_sensitive=1, + stopwords=stopwords, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_export_monday_empty_output", +) +``` + +
+ + +
+monday_insensintive_upper_twodim + +```python +input = ( + np.array( + ["Monday", "tuesday", "wednesday", "Monday", "tuesday", "wednesday"] + ) + .astype(object) + .reshape([1, 6]) +) + +# It does upper case cecedille, accented E +# and german umlaut but fails +# with german eszett +output = ( + np.array(["TUESDAY", "WEDNESDAY", "TUESDAY", "WEDNESDAY"]) + .astype(object) + .reshape([1, 4]) +) +stopwords = ["monday"] + +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + case_change_action="UPPER", + stopwords=stopwords, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_export_monday_insensintive_upper_twodim", +) +``` + +
+ + +
+nostopwords_nochangecase + +```python +input = np.array(["monday", "tuesday"]).astype(object) +output = input + +# No stopwords. This is a NOOP +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + is_case_sensitive=1, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_nostopwords_nochangecase", +) +``` + +
+ + +### **StringSplit** + + StringSplit splits a string tensor's elements into substrings based on a delimiter attribute and a maxsplit attribute. + + The first output of this operator is a tensor of strings representing the substrings from splitting each input string on the `delimiter` substring. This tensor has one additional rank compared to the input tensor in order to store the substrings for each input element (where the input tensor is not empty). Note that, in order to ensure the same number of elements are present in the final dimension, this tensor will pad empty strings as illustrated in the examples below. Consecutive delimiters are not grouped together and are deemed to delimit empty strings, except if the `delimiter` is unspecified or is the empty string (""). In the case where the `delimiter` is unspecified or the empty string, consecutive whitespace characters are regarded as a single separator and leading or trailing whitespace is removed in the output. + + The second output tensor represents the number of substrings generated. `maxsplit` can be used to limit the number of splits performed - after the `maxsplit`th split if the string is not fully split, the trailing suffix of input string after the final split point is also added. For elements where fewer splits are possible than specified in `maxsplit`, it has no effect. + +#### Version + +This version of the operator has been available since version 20 of the default ONNX operator set. + +#### Attributes + +
+
delimiter : string
+
Delimiter to split on. If left unset or set to the empty string (""), the input is split on consecutive whitespace.
+
maxsplit : int
+
Maximum number of splits (from left to right). If left unset (or if the number of possible splits are less than maxsplit), it will make as many splits as possible. Note that the maximum possible number of substrings returned with `maxsplit` specified is `maxsplit+1` since the remaining suffix after the `maxsplit`th split is included in the output.
+
+ +#### Inputs + +
+
X (non-differentiable) : T1
+
Tensor of strings to split.
+
+ +#### Outputs + +
+
Y (non-differentiable) : T2
+
Tensor of substrings representing the outcome of splitting the strings in the input on the delimiter. Note that to ensure the same number of elements are present in the final rank, this tensor will pad any necessary empty strings.
+
Z (non-differentiable) : T3
+
The number of substrings generated for each input element.
+
+ +#### Type Constraints + +
+
T1 : tensor(string)
+
The input must be a UTF-8 string tensor
+
T2 : tensor(string)
+
Tensor of substrings.
+
T3 : tensor(int64)
+
The number of substrings generated.
+
+ + +#### Examples + +
+basic + +```python +node = onnx.helper.make_node( + "StringSplit", + inputs=["x"], + outputs=["substrings", "length"], + delimiter=".", + maxsplit=None, +) + +x = np.array(["abc.com", "def.net"]).astype(object) + +substrings = np.array([["abc", "com"], ["def", "net"]]).astype(object) + +length = np.array([2, 2], dtype=np.int64) + +expect( + node, + inputs=[x], + outputs=[substrings, length], + name="test_string_split_basic", +) +``` + +
+ + +
+consecutive_delimiters + +```python +node = onnx.helper.make_node( + "StringSplit", + inputs=["x"], + outputs=["substrings", "length"], + delimiter="-", + maxsplit=None, +) + +x = np.array(["o-n-n--x-", "o-n----nx"]).astype(object) + +substrings = np.array( + [["o", "n", "n", "", "x", ""], ["o", "n", "", "", "", "nx"]] +).astype(object) + +length = np.array([6, 6], dtype=np.int64) + +expect( + node, + inputs=[x], + outputs=[substrings, length], + name="test_string_split_consecutive_delimiters", +) +``` + +
+ + +
+empty_string_delimiter + +```python +for delimiter, test_name in ( + ("", "test_string_split_empty_string_delimiter"), + (None, "test_string_split_no_delimiter"), +): + node = onnx.helper.make_node( + "StringSplit", + inputs=["x"], + outputs=["substrings", "length"], + delimiter=delimiter, + maxsplit=None, + ) + + x = np.array( + ["hello world !", " hello world !", " hello world ! "] + ).astype(object) + + substrings = np.array( + [ + ["hello", "world", "!"], + ["hello", "world", "!"], + ["hello", "world", "!"], + ] + ).astype(object) + + length = np.array([3, 3, 3], dtype=np.int64) + + expect( + node, + inputs=[x], + outputs=[substrings, length], + name=test_name, + ) +``` + +
+ + +
+empty_string_split + +```python +node = onnx.helper.make_node( + "StringSplit", + inputs=["x"], + outputs=["substrings", "length"], + delimiter=None, + maxsplit=None, +) + +x = np.array([]).astype(object) + +substrings = np.array([]).astype(object).reshape(0, 0) + +length = np.array([], dtype=np.int64) + +expect( + node, + inputs=[x], + outputs=[substrings, length], + name="test_string_split_empty_tensor", + output_type_protos=[ + onnx.helper.make_tensor_type_proto(onnx.TensorProto.STRING, (0, None)), + None, + ], +) +``` + +
+ + +
+maxsplit + +```python +node = onnx.helper.make_node( + "StringSplit", + inputs=["x"], + outputs=["substrings", "length"], + maxsplit=2, +) + +x = np.array( + [["hello world", "def.net"], ["o n n x", "the quick brown fox"]] +).astype(object) + +substrings = np.array( + [ + [["hello", "world", ""], ["def.net", "", ""]], + [["o", "n", "n x"], ["the", "quick", "brown fox"]], + ] +).astype(object) + +length = np.array([[2, 1], [3, 3]], np.int64) + +expect( + node, + inputs=[x], + outputs=[substrings, length], + name="test_string_split_maxsplit", +) +``` + +
+ + +### **Sub** + + Performs element-wise binary subtraction (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + + (Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 7, 13 + +#### Inputs + +
+
A (differentiable) : T
+
First operand.
+
B (differentiable) : T
+
Second operand.
+
+ +#### Outputs + +
+
C (differentiable) : T
+
Result, has same element type as two inputs
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to all numeric tensors.
+
+ + +#### Examples + +
+sub + +```python +node = onnx.helper.make_node( + "Sub", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([3, 2, 1]).astype(np.float32) +z = x - y # expected output [-2., 0., 2.] +expect(node, inputs=[x, y], outputs=[z], name="test_sub_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.int8) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.int8) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_int8") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.int16) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.int16) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_int16") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.uint8) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_uint8") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.uint16) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_uint16") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.uint32) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_uint32") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.uint64) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_uint64") +``` + +
+ + +
+sub_broadcast + +```python +node = onnx.helper.make_node( + "Sub", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_bcast") +``` + +
+ + +### **Sum** + + Element-wise sum of each of the input tensors (with Numpy-style broadcasting support). + All inputs and outputs must have the same data type. + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6, 8 + +#### Inputs (1 - ∞) + +
+
data_0 (variadic, differentiable) : T
+
List of tensors for sum.
+
+ +#### Outputs + +
+
sum (differentiable) : T
+
Output tensor.
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+sum + +```python +data_0 = np.array([3, 0, 2]).astype(np.float32) +data_1 = np.array([1, 3, 4]).astype(np.float32) +data_2 = np.array([2, 6, 6]).astype(np.float32) +result = np.array([6, 9, 12]).astype(np.float32) +node = onnx.helper.make_node( + "Sum", + inputs=["data_0", "data_1", "data_2"], + outputs=["result"], +) +expect( + node, + inputs=[data_0, data_1, data_2], + outputs=[result], + name="test_sum_example", +) + +node = onnx.helper.make_node( + "Sum", + inputs=["data_0"], + outputs=["result"], +) +expect(node, inputs=[data_0], outputs=[data_0], name="test_sum_one_input") + +result = np.add(data_0, data_1) +node = onnx.helper.make_node( + "Sum", + inputs=["data_0", "data_1"], + outputs=["result"], +) +expect( + node, inputs=[data_0, data_1], outputs=[result], name="test_sum_two_inputs" +) +``` + +
+ + +### **Swish** + + Swish function takes one input data (Tensor) and produces one output data (Tensor) of the same shape, + where $Swish(x) = x * sigmoid(alpha * x)$. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Coefficient to multiply with input before sigmoid.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(float16), tensor(float), tensor(bfloat16), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+swish + +```python +node = onnx.helper.make_node( + "Swish", + inputs=["x"], + outputs=["y"], + alpha=1.0, # pass alpha as attribute +) + +x = np.array([3, 4, 5], dtype=np.float32) +y = swish(x, alpha=1.0) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_swish", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +### **Tan** + + Calculates the tangent of the given input tensor, element-wise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 7 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The tangent of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+tan + +```python +node = onnx.helper.make_node( + "Tan", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.tan(x) +expect(node, inputs=[x], outputs=[y], name="test_tan_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.tan(x) +expect(node, inputs=[x], outputs=[y], name="test_tan") +``` + +
+ + +### **Tanh** + + Calculates the hyperbolic tangent of the given input tensor element-wise. + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
output (differentiable) : T
+
The hyperbolic tangent values of the input tensor computed element-wise
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+tanh + +```python +node = onnx.helper.make_node( + "Tanh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.tanh(x) # expected output [-0.76159418, 0., 0.76159418] +expect(node, inputs=[x], outputs=[y], name="test_tanh_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.tanh(x) +expect(node, inputs=[x], outputs=[y], name="test_tanh") +``` + +
+ + +### **TensorScatter** + + TensorScatter is a generic tensor update operation, motivated by the requirements for KV cache updates for Attention + ops commonly found in LLMs. It is a functional operation that models an in-place update to a KV cache buffer. + + The past and present cache tensors have the same shape (batch_size, D1, D2, ..., max_sequence_length, ..., Dn), with + the sequence dimension (indicated by the `axis` attribute) being max_sequence_length, so the sizes of these tensors do + not need to grow between iterations. The `update` tensor's shape only differs from the cache tensors in the sequence + dimension: (batch_size, D1, D2, ..., sequence_length, ..., Dn), where sequence_length <= max_sequence_length. + + The optional `write_indices` input indicates the write index for each sample in the batch, assumed to be zero + if not provided. When the `mode` attribute is set to "circular", the write index is modulo max_sequence_length. + The operation can be described using the following pseudocode: + + ``` + for prefix_idx in np.ndindex(past_cache.shape[:axis]): + batch_idx = prefix_idx[0] + for sequence_idx in range(sequence_length): + cache_idx = (*prefix_idx, write_indices[batch_idx] + sequence_idx) + if mode == "circular": + cache_idx = tuple(np.mod(np.asarray(cache_idx), max_sequence_length)) + update_idx = (*prefix_idx, sequence_idx) + present_cache[cache_idx] = update[update_idx] + ``` + + During the prefill phase of attention, only the first two inputs are needed. During the decode phase, `write_indices` + is also needed so that the incoming key or value update can be appended after the last valid token for each sample + in the batch. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +#### Attributes + +
+
axis : int (default is -2)
+
Sequence dimension of the `past_cache` and `update` tensors. It cannot be 0 (the batch dimension). Default is -2.
+
mode : string (default is linear)
+
Write mode of cache update. Supported modes include `linear` and `circular`. `linear` mode requires write_indices+sequence_length<=max_sequence_length. For `circular` mode, the updates happen in wrap-around fashion, ie, the update index is modulo `max_sequence_length`
+
+ +#### Inputs (2 - 3) + +
+
past_cache (differentiable) : T
+
Past state cache for key or value with shape `(batch_size, D1, D2, ..., max_sequence_length, ..., Dn)`.
+
update (differentiable) : T
+
New update tensor with shape `(batch_size, D1, D2, ..., sequence_length, ..., Dn)`.
+
write_indices (optional, non-differentiable) : tensor(int64)
+
Write indices for the incoming update tensor in the cache. Shape is `(batch_size,)`. Assumed to be all zeros if not provided.
+
+ +#### Outputs + +
+
present_cache (differentiable) : T
+
Updated cache. Same shape as `past_cache`.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0)
+
Constrain input and output types to any tensor type.
+
+ + +#### Examples + +
+tensorscatter + +```python +node = onnx.helper.make_node( + "TensorScatter", + inputs=["past_cache", "update", "write_indices"], + outputs=["present_cache"], + mode="linear", +) +past_cache = np.array( + [ + [[[1, 2, 3, 4, 5], [5, 6, 7, 8, 9], [8, 7, 6, 5, 4], [4, 3, 2, 1, 0]]], + [[[1, 2, 3, 4, 5], [5, 6, 7, 8, 9], [8, 7, 6, 5, 4], [4, 3, 2, 1, 0]]], + ], + dtype=np.float32, +) +update = np.array( + [ + [[[5, 5, 5, 5, 5]]], + [[[1, 1, 1, 1, 1]]], + ], + dtype=np.float32, +) +write_indices = np.array([1, 2], dtype=np.int64) +present_cache = np.array( + [ + [[[1, 2, 3, 4, 5], [5, 5, 5, 5, 5], [8, 7, 6, 5, 4], [4, 3, 2, 1, 0]]], + [[[1, 2, 3, 4, 5], [5, 6, 7, 8, 9], [1, 1, 1, 1, 1], [4, 3, 2, 1, 0]]], + ], + dtype=np.float32, +) +expect( + node, + inputs=[past_cache, update, write_indices], + outputs=[present_cache], + name="test_tensorscatter", +) +``` + +
+ + +
+tensorscatter_3d + +```python +node = onnx.helper.make_node( + "TensorScatter", + inputs=["past_cache", "update", "write_indices"], + outputs=["present_cache"], +) +past_cache = np.array( + [ + [ + [1, 2, 3, 4, 5], + [5, 6, 7, 8, 9], + [8, 7, 6, 5, 4], + [5, 4, 3, 2, 1], + ], + [ + [1, 2, 3, 4, 5], + [5, 6, 7, 8, 9], + [8, 7, 6, 5, 4], + [5, 4, 3, 2, 1], + ], + [ + [1, 2, 3, 4, 5], + [5, 6, 7, 8, 9], + [8, 7, 6, 5, 4], + [5, 4, 3, 2, 1], + ], + ], + dtype=np.float32, +) +update = np.array( + [ + [ + [4, 4, 4, 4, 4], + [5, 5, 5, 5, 5], + ], + [ + [6, 6, 6, 6, 6], + [7, 7, 7, 7, 7], + ], + [ + [2, 2, 2, 2, 2], + [3, 3, 3, 3, 3], + ], + ], + dtype=np.float32, +) +write_indices = np.array([1, 2, 0], dtype=np.int64) +present_cache = np.array( + [ + [ + [1, 2, 3, 4, 5], + [4, 4, 4, 4, 4], + [5, 5, 5, 5, 5], + [5, 4, 3, 2, 1], + ], + [ + [1, 2, 3, 4, 5], + [5, 6, 7, 8, 9], + [6, 6, 6, 6, 6], + [7, 7, 7, 7, 7], + ], + [ + [2, 2, 2, 2, 2], + [3, 3, 3, 3, 3], + [8, 7, 6, 5, 4], + [5, 4, 3, 2, 1], + ], + ], + dtype=np.float32, +) +expect( + node, + inputs=[past_cache, update, write_indices], + outputs=[present_cache], + name="test_tensorscatter_3d", +) +``` + +
+ + +
+tensorscatter_circular + +```python +node = onnx.helper.make_node( + "TensorScatter", + inputs=["past_cache", "update", "write_indices"], + outputs=["present_cache"], + mode="circular", +) +past_cache = np.array( + [ + [[[1, 2, 3, 4, 5], [5, 6, 7, 8, 9], [8, 7, 6, 5, 4], [4, 3, 2, 1, 0]]], + [[[1, 2, 3, 4, 5], [5, 6, 7, 8, 9], [8, 7, 6, 5, 4], [4, 3, 2, 1, 0]]], + ], + dtype=np.float32, +) +update = np.array( + [ + [ + [ + [5, 5, 5, 5, 5], + [6, 6, 6, 6, 6], + ] + ], + [ + [ + [1, 1, 1, 1, 1], + [2, 2, 2, 2, 2], + ] + ], + ], + dtype=np.float32, +) +write_indices = np.array([1, 3], dtype=np.int64) +present_cache = np.array( + [ + [[[1, 2, 3, 4, 5], [5, 5, 5, 5, 5], [6, 6, 6, 6, 6], [4, 3, 2, 1, 0]]], + [[[2, 2, 2, 2, 2], [5, 6, 7, 8, 9], [8, 7, 6, 5, 4], [1, 1, 1, 1, 1]]], + ], + dtype=np.float32, +) +expect( + node, + inputs=[past_cache, update, write_indices], + outputs=[present_cache], + name="test_tensorscatter_circular", +) +``` + +
+ + +### **TfIdfVectorizer** + + This transform extracts n-grams from the input sequence and save them as a vector. Input can + be either a 1-D or 2-D tensor. For 1-D input, output is the n-gram representation of that input. + For 2-D input, the output is also a 2-D tensor whose i-th row is the n-gram representation of the i-th input row. + More specifically, if input shape is [C], the corresponding output shape would be [max(ngram_indexes) + 1]. + If input shape is [N, C], this operator produces a [N, max(ngram_indexes) + 1]-tensor. + + In contrast to standard n-gram extraction, here, the indexes of extracting an n-gram from the original + sequence are not necessarily consecutive numbers. The discontinuity between indexes are controlled by the number of skips. + If the number of skips is 2, we should skip two tokens when scanning through the original sequence. + Let's consider an example. Assume that input sequence is [94, 17, 36, 12, 28] and the number of skips is 2. + The associated 2-grams are [94, 12] and [17, 28] respectively indexed by [0, 3] and [1, 4]. + If the number of skips becomes 0, the 2-grams generated are [94, 17], [17, 36], [36, 12], [12, 28] + indexed by [0, 1], [1, 2], [2, 3], [3, 4], respectively. + + The output vector (denoted by Y) stores the count of each n-gram; + Y[ngram_indexes[i]] indicates the times that the i-th n-gram is found. The attribute ngram_indexes is used to determine the mapping + between index i and the corresponding n-gram's output coordinate. If pool_int64s is [94, 17, 17, 36], ngram_indexes is [1, 0], + ngram_counts=[0, 0], then the Y[0] (first element in Y) and Y[1] (second element in Y) are the counts of [17, 36] and [94, 17], + respectively. An n-gram which cannot be found in pool_strings/pool_int64s should be ignored and has no effect on the output. + Note that we may consider all skips up to S when generating the n-grams. + + The examples used above are true if mode is "TF". If mode is "IDF", all the counts larger than 1 would be truncated to 1 and + the i-th element in weights would be used to scale (by multiplication) the count of the i-th n-gram in pool. If mode is "TFIDF", + this operator first computes the counts of all n-grams and then scale them by the associated values in the weights attribute. + + Only one of pool_strings and pool_int64s can be set. If pool_int64s is set, the input should be an integer tensor. + If pool_strings is set, the input must be a string tensor. + +#### Version + +This version of the operator has been available since version 9 of the default ONNX operator set. + +#### Attributes + +
+
max_gram_length : int (required)
+
Maximum n-gram length. If this value is 3, 3-grams will be used to generate the output.
+
max_skip_count : int (required)
+
Maximum number of items (integers/strings) to be skipped when constructing an n-gram from X. If max_skip_count=1, min_gram_length=2, max_gram_length=3, this operator may generate 2-grams with skip_count=0 and skip_count=1, and 3-grams with skip_count=0 and skip_count=1
+
min_gram_length : int (required)
+
Minimum n-gram length. If this value is 2 and max_gram_length is 3, output may contain counts of 2-grams and 3-grams.
+
mode : string (required)
+
The weighting criteria. It can be one of "TF" (term frequency), "IDF" (inverse document frequency), and "TFIDF" (the combination of TF and IDF)
+
ngram_counts : list of ints (required)
+
The starting indexes of 1-grams, 2-grams, and so on in pool. It is useful when determining the boundary between two consecutive collections of n-grams. For example, if ngram_counts is [0, 17, 36], the first index (zero-based) of 1-gram/2-gram/3-gram in pool are 0/17/36. This format is essentially identical to CSR (or CSC) sparse matrix format, and we choose to use this due to its popularity.
+
ngram_indexes : list of ints (required)
+
list of int64s (type: AttributeProto::INTS). This list is parallel to the specified 'pool_*' attribute. The i-th element in ngram_indexes indicate the coordinate of the i-th n-gram in the output tensor.
+
pool_int64s : list of ints
+
List of int64 n-grams learned from the training set. Either this or pool_strings attributes must be present but not both. It's an 1-D tensor starting with the collections of all 1-grams and ending with the collections of n-grams. The i-th element in pool stores the n-gram that should be mapped to coordinate ngram_indexes[i] in the output vector.
+
pool_strings : list of strings
+
List of strings n-grams learned from the training set. Either this or pool_int64s attributes must be present but not both. It's an 1-D tensor starting with the collections of all 1-grams and ending with the collections of n-grams. The i-th element in pool stores the n-gram that should be mapped to coordinate ngram_indexes[i] in the output vector.
+
weights : list of floats
+
list of floats. This attribute stores the weight of each n-gram in pool. The i-th element in weights is the weight of the i-th n-gram in pool. Its length equals to the size of ngram_indexes. By default, weights is an all-one tensor.This attribute is used when mode is "IDF" or "TFIDF" to scale the associated word counts.
+
+ +#### Inputs + +
+
X (non-differentiable) : T
+
Input for n-gram extraction
+
+ +#### Outputs + +
+
Y (non-differentiable) : T1
+
Ngram results
+
+ +#### Type Constraints + +
+
T : tensor(string), tensor(int32), tensor(int64)
+
Input is either string UTF-8 or int32/int64
+
T1 : tensor(float)
+
1-D tensor of floats
+
+ + +#### Examples + +
+tf_batch_onlybigrams_skip0 + +```python +input = np.array([[1, 1, 3, 3, 3, 7], [8, 6, 7, 5, 6, 8]]).astype(np.int32) +output = np.array( + [[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0]] +).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=2, + max_gram_length=2, + max_skip_count=0, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_batch_onlybigrams_skip0", +) +``` + +
+ + +
+tf_batch_onlybigrams_skip5 + +```python +input = np.array([[1, 1, 3, 3, 3, 7], [8, 6, 7, 5, 6, 8]]).astype(np.int32) +output = np.array( + [[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0]] +).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=2, + max_gram_length=2, + max_skip_count=5, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_batch_onlybigrams_skip5", +) +``` + +
+ + +
+tf_batch_uniandbigrams_skip5 + +```python +input = np.array([[1, 1, 3, 3, 3, 7], [8, 6, 7, 5, 6, 8]]).astype(np.int32) +output = np.array( + [[0.0, 3.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0]] +).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=1, + max_gram_length=2, + max_skip_count=5, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_batch_uniandbigrams_skip5", +) +``` + +
+ + +
+tf_only_bigrams_skip0 + +```python +input = np.array([1, 1, 3, 3, 3, 7, 8, 6, 7, 5, 6, 8]).astype(np.int32) +output = np.array([0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0]).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=2, + max_gram_length=2, + max_skip_count=0, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_only_bigrams_skip0", +) +``` + +
+ + +
+tf_onlybigrams_levelempty + +```python +input = np.array([1, 1, 3, 3, 3, 7, 8, 6, 7, 5, 6, 8]).astype(np.int32) +output = np.array([1.0, 1.0, 1.0]).astype(np.float32) + +ngram_counts = np.array([0, 0]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2]).astype(np.int64) +pool_int64s = np.array([5, 6, 7, 8, 6, 7]).astype( # unigrams none + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=2, + max_gram_length=2, + max_skip_count=0, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_onlybigrams_levelempty", +) +``` + +
+ + +
+tf_onlybigrams_skip5 + +```python +input = np.array([1, 1, 3, 3, 3, 7, 8, 6, 7, 5, 6, 8]).astype(np.int32) +output = np.array([0.0, 0.0, 0.0, 0.0, 1.0, 3.0, 1.0]).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=2, + max_gram_length=2, + max_skip_count=5, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_onlybigrams_skip5", +) +``` + +
+ + +
+tf_uniandbigrams_skip5 + +```python +input = np.array([1, 1, 3, 3, 3, 7, 8, 6, 7, 5, 6, 8]).astype(np.int32) +output = np.array([0.0, 3.0, 1.0, 0.0, 1.0, 3.0, 1.0]).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=1, + max_gram_length=2, + max_skip_count=5, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_uniandbigrams_skip5", +) +``` + +
+ + +### **ThresholdedRelu** + + ThresholdedRelu takes one input data (Tensor) and produces one output data + (Tensor) where the rectified linear function, y = x for x > alpha, y = 0 otherwise, + is applied to the tensor elementwise. + +#### Version + +This version of the operator has been available since version 22 of the default ONNX operator set. + +Other versions of this operator: 10 + +#### Attributes + +
+
alpha : float (default is 1.0)
+
Threshold value
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Input tensor
+
+ +#### Outputs + +
+
Y (differentiable) : T
+
Output tensor
+
+ +#### Type Constraints + +
+
T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+default + +```python +default_alpha = 1.0 +node = onnx.helper.make_node("ThresholdedRelu", inputs=["x"], outputs=["y"]) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, default_alpha, np.inf) +y[y == default_alpha] = 0 + +expect(node, inputs=[x], outputs=[y], name="test_thresholdedrelu_default") +``` + +
+ + +
+thresholdedrelu + +```python +alpha = 2.0 +node = onnx.helper.make_node( + "ThresholdedRelu", inputs=["x"], outputs=["y"], alpha=alpha +) + +x = np.array([-1.5, 0.0, 1.2, 2.0, 2.2]).astype(np.float32) +y = np.clip(x, alpha, np.inf) # expected output [0., 0., 0., 0., 2.2] +y[y == alpha] = 0 + +expect(node, inputs=[x], outputs=[y], name="test_thresholdedrelu_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, alpha, np.inf) +y[y == alpha] = 0 + +expect(node, inputs=[x], outputs=[y], name="test_thresholdedrelu") +``` + +
+ + +### **Tile** + + Constructs a tensor by tiling a given tensor. + This is the same as function `tile` in Numpy, but no broadcast. + For example A = [[1, 2], [3, 4]], B = [1, 2], tile(A, B) = [[1, 2, 1, 2], [3, 4, 3, 4]] + +#### Version + +This version of the operator has been available since version 13 of the default ONNX operator set. + +Other versions of this operator: 1, 6 + +#### Inputs + +
+
input (differentiable) : T
+
Input tensor of any shape.
+
repeats (non-differentiable) : T1
+
1D int64 tensor of the same length as input's dimension number, includes numbers of repeated copies along input's dimensions.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor of the same dimensions and type as tensor input. output_dim[i] = input_dim[i] * repeats[i]
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
T1 : tensor(int64)
+
Constrain repeat's type to int64 tensors.
+
+ + +#### Examples + +
+tile + +```python +node = onnx.helper.make_node("Tile", inputs=["x", "y"], outputs=["z"]) + +x = np.random.rand(2, 3, 4, 5).astype(np.float32) + +repeats = np.random.randint(low=1, high=10, size=(np.ndim(x),)).astype(np.int64) + +z = np.tile(x, repeats) + +expect(node, inputs=[x, repeats], outputs=[z], name="test_tile") +``` + +
+ + +
+tile_precomputed + +```python +node = onnx.helper.make_node("Tile", inputs=["x", "y"], outputs=["z"]) + +x = np.array([[0, 1], [2, 3]], dtype=np.float32) + +repeats = np.array([2, 2], dtype=np.int64) + +z = np.array( + [[0, 1, 0, 1], [2, 3, 2, 3], [0, 1, 0, 1], [2, 3, 2, 3]], dtype=np.float32 +) + +expect(node, inputs=[x, repeats], outputs=[z], name="test_tile_precomputed") +``` + +
+ + +### **TopK** + + Retrieve the top-K largest or smallest elements along a specified axis. Given an input tensor of + shape [a_0, a_1, ..., a_{n-1}] and integer argument k, return two outputs: + + * Value tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] + which contains the values of the top k elements along the specified axis + * Index tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] which + contains the indices of the top k elements (original indices from the input + tensor). + + * If "largest" is 1 (the default value) then the k largest elements are returned. + * If "sorted" is 1 (the default value) then the resulting k elements will be sorted. + * If "sorted" is 0, order of returned 'Values' and 'Indices' are undefined. + + Given two equivalent values, this operator uses the indices along the axis as + a tiebreaker. That is, the element with the lower index will appear first. + +#### Version + +This version of the operator has been available since version 24 of the default ONNX operator set. + +Other versions of this operator: 1, 10, 11 + +#### Attributes + +
+
axis : int (default is -1)
+
Dimension on which to do the sort. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
largest : int (default is 1)
+
Whether to return the top-K largest or smallest elements.
+
sorted : int (default is 1)
+
Whether to return the elements in sorted order.
+
+ +#### Inputs + +
+
X (differentiable) : T
+
Tensor of shape [a_0, a_1, ..., a_{n-1}]
+
K (non-differentiable) : tensor(int64)
+
A 1-D tensor containing a single positive value corresponding to the number of top elements to retrieve
+
+ +#### Outputs + +
+
Values (differentiable) : T
+
Tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] containing top K values from the input tensor
+
Indices (non-differentiable) : I
+
Tensor of shape [a_0, a_1, ..., a_{axis-1}, k, a_{axis+1}, ... a_{n-1}] containing the corresponding input tensor indices for the top K values.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)
+
Constrain input and output types to numeric tensors.
+
I : tensor(int64)
+
Constrain index tensor to int64
+
+ + +#### Examples + +
+top_k + +```python +axis = 1 +largest = 1 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [ + [0, 1, 2, 3], + [4, 5, 6, 7], + [8, 9, 10, 11], + ], + dtype=np.float32, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [[ 3. 2. 1.] +# [ 7. 6. 5.] +# [11. 10. 9.]] +# print(indices_ref) +# [[3 2 1] +# [3 2 1] +# [3 2 1]] + +expect( + node, inputs=[X, K], outputs=[values_ref, indices_ref], name="test_top_k" +) +``` + +
+ + +
+top_k_negative_axis + +```python +axis = -1 +largest = 1 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [ + [0, 1, 2, 3], + [4, 5, 6, 7], + [8, 9, 10, 11], + ], + dtype=np.float32, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [[ 3. 2. 1.] +# [ 7. 6. 5.] +# [11. 10. 9.]] +# print(indices_ref) +# [[3 2 1] +# [3 2 1] +# [3 2 1]] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_negative_axis", +) +``` + +
+ + +
+top_k_same_values + +```python +axis = 0 +largest = 0 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [0, 0, 0, 0], + dtype=np.int64, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# (Pdb) print(values_ref) +# [0 0 0] +# (Pdb) print(indices_ref) +# [0 1 2] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_same_values", +) +``` + +
+ + +
+top_k_same_values_2d + +```python +axis = 1 +largest = 1 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [[0, 0, 0, 0], [1, 1, 1, 1], [2, 2, 1, 1]], + dtype=np.int64, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [[0 0 0] +# [1 1 1] +# [1 1 2]] +# print(indices_ref) +# [[0 1 2] +# [0 1 2] +# [2 3 0]] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_same_values_2d", +) +``` + +
+ + +
+top_k_same_values_largest + +```python +axis = 0 +largest = 1 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [0, 0, 0, 0], + dtype=np.int64, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [0 0 0] +# print(indices_ref) +# [0 1 2] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_same_values_largest", +) +``` + +
+ + +
+top_k_smallest + +```python +axis = 1 +largest = 0 +sorted_ = 1 +k = 3 + +node = onnx.helper.make_node( + "TopK", + inputs=["x", "k"], + outputs=["values", "indices"], + axis=axis, + largest=largest, + sorted=sorted_, +) + +X = np.array( + [ + [0, 1, 2, 3], + [4, 5, 6, 7], + [11, 10, 9, 8], + ], + dtype=np.float32, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [[ 0. 1. 2.] +# [ 4. 5. 6.] +# [ 8. 9. 10.]] +# print(indices_ref) +# [[0 1 2] +# [0 1 2] +# [3 2 1]] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_smallest", +) +``` + +
+ + +
+top_k_uint64 + +```python +axis = 1 +largest = 1 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [ + [0, 1, 2, 3], + [4, 5, 6, 7], + [8, 9, 10, 11], + ], + dtype=np.uint64, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [[ 3 2 1] +# [ 7 6 5] +# [11 10 9]] +# print(indices_ref) +# [[3 2 1] +# [3 2 1] +# [3 2 1]] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_uint64", +) +``` + +
+ + +### **Transpose** + + Returns a transpose of the input tensor. (Similar to `numpy.transpose`). + The optional attribute `perm` must be a permutation of the dimensions of + the input tensor. Axis `i` of the output tensor corresponds to the axis + `perm[i]` of the input tensor. + For example, when perm=(1, 0, 2), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 1, 3). + When perm=(1, 2, 0), given an input tensor of shape (1, 2, 3), + the output shape will be (2, 3, 1). + If the attribute `perm` is omitted, its default value is `(n-1, ..., 0)`, + where `n` is the rank of the input tensor. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 13, 21, 23, 24 + +#### Attributes + +
+
perm : list of ints
+
A list of integers. By default, reverse the dimensions, otherwise permute the axes according to the values given. Its length must be equal to the rank of the input.
+
+ +#### Inputs + +
+
data (differentiable) : T
+
An input tensor.
+
+ +#### Outputs + +
+
transposed (differentiable) : T
+
Transposed output.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types.
+
+ + +#### Examples + +
+all_permutations + +```python +shape = (2, 3, 4) +data = np.random.random_sample(shape).astype(np.float32) +permutations = list(itertools.permutations(np.arange(len(shape)))) + +for i, permutation in enumerate(permutations): + node = onnx.helper.make_node( + "Transpose", + inputs=["data"], + outputs=["transposed"], + perm=permutation, + ) + transposed = np.transpose(data, permutation) + expect( + node, + inputs=[data], + outputs=[transposed], + name=f"test_transpose_all_permutations_{i}", + ) +``` + +
+ + +
+default + +```python +shape = (2, 3, 4) +data = np.random.random_sample(shape).astype(np.float32) + +node = onnx.helper.make_node( + "Transpose", inputs=["data"], outputs=["transposed"] +) + +transposed = np.transpose(data) +expect(node, inputs=[data], outputs=[transposed], name="test_transpose_default") +``` + +
+ + +### **Trilu** + + Given a 2-D matrix or batches of 2-D matrices, returns the upper or lower triangular part of the tensor(s). + The attribute "upper" determines whether the upper or lower part is retained. If set to true, + the upper triangular matrix is retained. Lower triangular matrix is retained otherwise. + Default value for the "upper" attribute is true. + Trilu takes one input tensor of shape [*, N, M], where * is zero or more batch dimensions. The upper triangular part consists + of the elements on and above the given diagonal (k). The lower triangular part consists of elements on and below the diagonal. + All other elements in the matrix are set to zero. + If k = 0, the triangular part on and above/below the main diagonal is retained. + If upper is set to true, a positive k retains the upper triangular matrix excluding the main diagonal and (k-1) diagonals above it. + A negative k value retains the main diagonal and |k| diagonals below it. + If upper is set to false, a positive k retains the lower triangular matrix including the main diagonal and k diagonals above it. + A negative k value excludes the main diagonal and (|k|-1) diagonals below it. + +#### Version + +This version of the operator has been available since version 14 of the default ONNX operator set. + +#### Attributes + +
+
upper : int (default is 1)
+
Boolean. Indicates whether upper or lower part of matrix is retained. Default is true.
+
+ +#### Inputs (1 - 2) + +
+
input (differentiable) : T
+
Input tensor of rank 2 or higher.
+
k (optional, non-differentiable) : tensor(int64)
+
A 0-D tensor containing a single value corresponding to the number diagonals above or below the main diagonal to exclude or include. Default value is 0 if it's not specified.
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Output tensor of the same type and shape as the input tensor.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types.
+
+ + +#### Examples + +
+tril + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 0, 0, 0, 0], +# [1, 2, 0, 0, 0], +# [9, 4, 1, 0, 0], +# [4, 3, 4, 2, 0]] +y = tril_reference_implementation(x) +expect(node, inputs=[x], outputs=[y], name="test_tril") +``` + +
+ + +
+tril_neg + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(-1).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[0, 0, 0, 0, 0], +# [1, 0, 0, 0, 0], +# [9, 4, 0, 0, 0], +# [4, 3, 4, 0, 0]] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_neg") +``` + +
+ + +
+tril_one_row + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(3, 1, 5)).astype(np.int64) +# X: +# [[[6, 2, 4, 1, 6]], +# +# [[8, 3, 8, 7, 0]], +# +# [[2, 2, 9, 5, 9]]] +# expect result: +# [[[6, 0, 0, 0, 0]], +# +# [[8, 0, 0, 0, 0]], +# +# [[2, 0, 0, 0, 0]]] +y = tril_reference_implementation(x) +expect(node, inputs=[x], outputs=[y], name="test_tril_one_row_neg") +``` + +
+ + +
+tril_out_neg + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(-7).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0]] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_out_neg") +``` + +
+ + +
+tril_out_pos + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(6).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_out_pos") +``` + +
+ + +
+tril_pos + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(2).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 7, 3, 0, 0], +# [1, 2, 8, 6, 0], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_pos") +``` + +
+ + +
+tril_square + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(2, 3, 3)).astype(np.int64) +# X: +# [[[0, 4, 3], +# [2, 0, 9], +# [8, 2, 5]], +# +# [[2, 7, 2], +# [2, 6, 0], +# [2, 6, 5]]] +# expect result: +# [[[0, 0, 0], +# [2, 0, 0], +# [8, 2, 5]], +# +# [[2, 0, 0], +# [2, 6, 0], +# [2, 6, 5]]] +y = tril_reference_implementation(x) +expect(node, inputs=[x], outputs=[y], name="test_tril_square") +``` + +
+ + +
+tril_square_neg + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(2, 3, 3)).astype(np.int64) +k = np.array(-1).astype(np.int64) +# X: +# [[[0, 4, 3], +# [2, 0, 9], +# [8, 2, 5]], +# +# [[2, 7, 2], +# [2, 6, 0], +# [2, 6, 5]]] +# expect result: +# [[[0, 0, 0], +# [2, 0, 0], +# [8, 2, 0]], +# +# [[0, 0, 0], +# [2, 0, 0], +# [2, 6, 0]]] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_square_neg") +``` + +
+ + +
+tril_zero + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(3, 0, 5)).astype(np.int64) +k = np.array(6).astype(np.int64) +# X: +# [] +# expect result: +# [] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_zero") +``` + +
+ + +
+triu + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x"], + outputs=["y"], +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 7, 3, 7, 9], +# [0, 2, 8, 6, 9], +# [0, 0, 0, 8, 7], +# [0, 0, 0, 2, 4]] +y = triu_reference_implementation(x) +expect(node, inputs=[x], outputs=[y], name="test_triu") +``` + +
+ + +
+triu_neg + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(-1).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [0, 4, 0, 8, 7], +# [0, 0, 4, 2, 4]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_neg") +``` + +
+ + +
+triu_one_row + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(3, 1, 5)).astype(np.int64) +k = np.array(1).astype(np.int64) +# X: +# [[[1, 4, 9, 7, 1]], +# +# [[9, 2, 8, 8, 4]], +# +# [[3, 9, 7, 4, 2]]] +# expect result: +# [[[0, 4, 9, 7, 1]], +# +# [[0, 2, 8, 8, 4]], +# +# [[0, 9, 7, 4, 2]]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_one_row") +``` + +
+ + +
+triu_out_neg_out + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(-7).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_out_neg_out") +``` + +
+ + +
+triu_out_pos + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(6).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_out_pos") +``` + +
+ + +
+triu_pos + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(2).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[0, 0, 3, 7, 9], +# [0, 0, 0, 6, 9], +# [0, 0, 0, 0, 7], +# [0, 0, 0, 0, 0]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_pos") +``` + +
+ + +
+triu_square + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x"], + outputs=["y"], +) + +x = np.random.randint(10, size=(2, 3, 3)).astype(np.int64) +y = triu_reference_implementation(x) +# X: +# [[[4, 6, 9], +# [7, 5, 4], +# [8, 1, 2]], +# +# [[1, 4, 9], +# [9, 6, 3], +# [8, 9, 8]]] +# expect result: +# [[[4, 6, 9], +# [0, 5, 4], +# [0, 0, 2]], +# +# [[1, 4, 9], +# [0, 6, 3], +# [0, 0, 8]]] +expect(node, inputs=[x], outputs=[y], name="test_triu_square") +``` + +
+ + +
+triu_square_neg + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(2, 3, 3)).astype(np.int64) +k = np.array(-1).astype(np.int64) +# X: +# [[[4, 6, 9], +# [7, 5, 4], +# [8, 1, 2]], +# +# [[1, 4, 9], +# [9, 6, 3], +# [8, 9, 8]]] +# expect result: +# [[[4, 6, 9], +# [7, 5, 4], +# [0, 1, 2]], +# +# [[1, 4, 9], +# [9, 6, 3], +# [0, 9, 8]]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_square_neg") +``` + +
+ + +
+triu_zero + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(0, 5)).astype(np.int64) +k = np.array(6).astype(np.int64) +# X: +# [] +# expect result: +# [] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_zero") +``` + +
+ + +### **Unique** + + Find the unique elements of a tensor. When an optional attribute 'axis' is provided, unique subtensors sliced along the 'axis' are returned. + Otherwise the input tensor is flattened and unique values of the flattened tensor are returned. + + This operator returns the unique values or sliced unique subtensors of the input tensor and three optional outputs. + The first output tensor 'Y' contains all unique values or subtensors of the input. + The second optional output tensor 'indices' contains indices of 'Y' elements' first occurrence in 'X'. + The third optional output tensor 'inverse_indices' contains, for elements of 'X', its corresponding indices in 'Y'. + The fourth optional output tensor 'counts' contains the count of each element of 'Y' in the input. + + Outputs are either sorted in ascending order or optionally in the order of the first occurrence of the values in the input. + + https://docs.scipy.org/doc/numpy/reference/generated/numpy.unique.html + + Example 1: + ``` + input_X = [2, 1, 1, 3, 4, 3] + attribute_sorted = 0 + attribute_axis = None + output_Y = [2, 1, 3, 4] + output_indices = [0, 1, 3, 4] + output_inverse_indices = [0, 1, 1, 2, 3, 2] + output_counts = [1, 2, 2, 1] + ``` + + Example 2: + ``` + input_X = [[1, 3], [2, 3]] + attribute_sorted = 1 + attribute_axis = None + output_Y = [1, 2, 3] + output_indices = [0, 2, 1] + output_inverse_indices = [0, 2, 1, 2] + output_counts = [1, 1, 2] + ``` + + Example 3: + ``` + input_X = [[1, 0, 0], [1, 0, 0], [2, 3, 4]] + attribute_sorted = 1 + attribute_axis = 0 + output_Y = [[1, 0, 0], [2, 3, 4]] + output_indices = [0, 2] + output_inverse_indices = [0, 0, 1] + output_counts = [2, 1] + ``` + + Example 4: + ``` + input_x = [[[1., 1.], [0., 1.], [2., 1.], [0., 1.]], + [[1., 1.], [0., 1.], [2., 1.], [0., 1.]]] + attribute_sorted = 1 + attribute_axis = 1 + ``` + + intermediate data are presented below for better understanding: + there are 4 subtensors sliced along axis 1 of input_x (shape = (2, 4, 2)): + ``` + A: [[1, 1], [1, 1]], + [[0, 1], [0, 1]], + [[2, 1], [2, 1]], + [[0, 1], [0, 1]]. + ``` + + there are 3 unique subtensors: + ``` + [[1, 1], [1, 1]], + [[0, 1], [0, 1]], + [[2, 1], [2, 1]]. + ``` + + sorted unique subtensors: + ``` + B: [[0, 1], [0, 1]], + [[1, 1], [1, 1]], + [[2, 1], [2, 1]]. + ``` + + output_Y is constructed from B: + ``` + [[[0. 1.], [1. 1.], [2. 1.]], + [[0. 1.], [1. 1.], [2. 1.]]] + ``` + + output_indices is to map from B to A: + ``` + [1, 0, 2] + ``` + + output_inverse_indices is to map from A to B: + ``` + [1, 0, 2, 0] + ``` + + output_counts: + ``` + [2, 1, 1] + ``` + +#### Version + +This version of the operator has been available since version 11 of the default ONNX operator set. + +#### Attributes + +
+
axis : int
+
(Optional) The dimension to apply unique. If not specified, the unique elements of the flattened input are returned. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
+
sorted : int (default is 1)
+
(Optional) Whether to sort the unique elements in ascending order before returning as output. Must be one of 0, or 1 (default).
+
+ +#### Inputs + +
+
X (non-differentiable) : T
+
A N-D input tensor that is to be processed.
+
+ +#### Outputs (1 - 4) + +
+
Y (non-differentiable) : T
+
A tensor of the same type as 'X' containing all the unique values or subtensors sliced along a provided 'axis' in 'X', either sorted or maintained in the same order they occur in input 'X'
+
indices (optional, non-differentiable) : tensor(int64)
+
A 1-D INT64 tensor containing indices of 'Y' elements' first occurrence in 'X'. When 'axis' is provided, it contains indices to subtensors in input 'X' on the 'axis'. When 'axis' is not provided, it contains indices to values in the flattened input tensor.
+
inverse_indices (optional, non-differentiable) : tensor(int64)
+
A 1-D INT64 tensor containing, for elements of 'X', its corresponding indices in 'Y'. When 'axis' is provided, it contains indices to subtensors in output 'Y' on the 'axis'. When 'axis' is not provided, it contains indices to values in output 'Y'.
+
counts (optional, non-differentiable) : tensor(int64)
+
A 1-D INT64 tensor containing the count of each element of 'Y' in input 'X'
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Input can be of any tensor type.
+
+ + +#### Examples + +
+length_1 + +```python +node_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], + sorted=1, +) + +x = np.array([0], dtype=np.int64) +y, indices, inverse_indices, counts = np.unique(x, True, True, True) +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +# behavior changed with numpy >= 2.0 +inverse_indices = inverse_indices.reshape(-1) +# print(y) +# [0] +# print(indices) +# [0] +# print(inverse_indices) +# [0] +# print(counts) +# [1] + +expect( + node_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_length_1", +) +``` + +
+ + +
+not_sorted_without_axis + +```python +node_not_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], + sorted=0, +) +# numpy unique does not retain original order (it sorts the output unique values) +# https://github.com/numpy/numpy/issues/8621 +# we need to recover unsorted output and indices +x = np.array([2.0, 1.0, 1.0, 3.0, 4.0, 3.0], dtype=np.float32) +y, indices, inverse_indices, counts = np.unique(x, True, True, True) + +# prepare index mapping from sorted to unsorted +argsorted_indices = np.argsort(indices) +inverse_indices_map = dict( + zip(argsorted_indices, np.arange(len(argsorted_indices)), strict=True) +) + +indices = indices[argsorted_indices] +y = np.take(x, indices, axis=0) +inverse_indices = np.asarray( + [inverse_indices_map[i] for i in inverse_indices], dtype=np.int64 +) +counts = counts[argsorted_indices] +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +# print(y) +# [2.0, 1.0, 3.0, 4.0] +# print(indices) +# [0 1 3 4] +# print(inverse_indices) +# [0, 1, 1, 2, 3, 2] +# print(counts) +# [1, 2, 2, 1] + +expect( + node_not_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_not_sorted_without_axis", +) +``` + +
+ + +
+sorted_with_axis + +```python +node_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], + sorted=1, + axis=0, +) + +x = np.array([[1, 0, 0], [1, 0, 0], [2, 3, 4]], dtype=np.float32) +y, indices, inverse_indices, counts = np.unique(x, True, True, True, axis=0) +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +# behavior changed with numpy >= 2.0 +inverse_indices = inverse_indices.reshape(-1) +# print(y) +# [[1. 0. 0.] +# [2. 3. 4.]] +# print(indices) +# [0 2] +# print(inverse_indices) +# [0 0 1] +# print(counts) +# [2 1] + +expect( + node_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_sorted_with_axis", +) +``` + +
+ + +
+sorted_with_axis_3d + +```python +node_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], + sorted=1, + axis=1, +) + +x = np.array( + [ + [[1.0, 1.0], [0.0, 1.0], [2.0, 1.0], [0.0, 1.0]], + [[1.0, 1.0], [0.0, 1.0], [2.0, 1.0], [0.0, 1.0]], + ], + dtype=np.float32, +) +y, indices, inverse_indices, counts = np.unique(x, True, True, True, axis=1) +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +# behavior changed with numpy >= 2.0 +inverse_indices = inverse_indices.reshape(-1) +# print(y) +# [[[0. 1.] +# [1. 1.] +# [2. 1.]] +# [[0. 1.] +# [1. 1.] +# [2. 1.]]] +# print(indices) +# [1 0 2] +# print(inverse_indices) +# [1 0 2 0] +# print(counts) +# [2 1 1] +expect( + node_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_sorted_with_axis_3d", +) +``` + +
+ + +
+sorted_with_negative_axis + +```python +node_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], + sorted=1, + axis=-1, +) + +x = np.array([[1, 0, 0], [1, 0, 0], [2, 3, 3]], dtype=np.float32) +y, indices, inverse_indices, counts = np.unique(x, True, True, True, axis=-1) +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +# behavior changed with numpy >= 2.0 +inverse_indices = inverse_indices.reshape(-1) +# print(y) +# [[0. 1.] +# [0. 1.] +# [3. 2.]] +# print(indices) +# [1 0] +# print(inverse_indices) +# [1 0 0] +# print(counts) +# [2 1] + +expect( + node_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_sorted_with_negative_axis", +) +``` + +
+ + +
+sorted_without_axis + +```python +node_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], +) + +x = np.array([2.0, 1.0, 1.0, 3.0, 4.0, 3.0], dtype=np.float32) +y, indices, inverse_indices, counts = np.unique(x, True, True, True) +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +expect( + node_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_sorted_without_axis", +) +``` + +
+ + +### **Unsqueeze** + + Insert single-dimensional entries to the shape of an input tensor (`data`). + Takes one required input `axes` - which contains a list of dimension indices and this operator will insert a dimension of value `1` into the corresponding index of the output tensor (`expanded`). + + For example, given an input tensor (`data`) of shape [3, 4, 5], then + Unsqueeze(data, axes=[0, 4]) outputs a tensor (`expanded`) containing same data as `data` but with shape [1, 3, 4, 5, 1]. + + The input `axes` should not contain any duplicate entries. It is an error if it contains duplicates. + The rank of the output tensor (`output_rank`) is the rank of the input tensor (`data`) plus the number of values in `axes`. + Each value in `axes` should be within the (inclusive) range [-output_rank , output_rank - 1]. + The order of values in `axes` does not matter and can come in any order. + +#### Version + +This version of the operator has been available since version 25 of the default ONNX operator set. + +Other versions of this operator: 1, 11, 13, 21, 23, 24 + +#### Inputs + +
+
data (differentiable) : T
+
Original tensor
+
axes (non-differentiable) : tensor(int64)
+
1D tensor of integers indicating the dimensions to be inserted. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(expanded).
+
+ +#### Outputs + +
+
expanded (differentiable) : T
+
Reshaped tensor with same data as input.
+
+ +#### Type Constraints + +
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2)
+
Constrain input and output types to all tensor types up to IRv13.
+
+ + +#### Examples + +
+unsqueeze_negative_axes + +```python +node = onnx.helper.make_node( + "Unsqueeze", + inputs=["x", "axes"], + outputs=["y"], +) +x = np.random.randn(1, 3, 1, 5).astype(np.float32) +axes = np.array([-2]).astype(np.int64) +y = np.expand_dims(x, axis=-2) +expect(node, inputs=[x, axes], outputs=[y], name="test_unsqueeze_negative_axes") +``` + +
+ + +
+unsqueeze_one_axis + +```python +x = np.random.randn(3, 4, 5).astype(np.float32) + +for i in range(x.ndim): + axes = np.array([i]).astype(np.int64) + node = onnx.helper.make_node( + "Unsqueeze", + inputs=["x", "axes"], + outputs=["y"], + ) + y = np.expand_dims(x, axis=i) + + expect( + node, + inputs=[x, axes], + outputs=[y], + name="test_unsqueeze_axis_" + str(i), + ) +``` + +
+ + +
+unsqueeze_three_axes + +```python +x = np.random.randn(3, 4, 5).astype(np.float32) +axes = np.array([2, 4, 5]).astype(np.int64) + +node = onnx.helper.make_node( + "Unsqueeze", + inputs=["x", "axes"], + outputs=["y"], +) +y = np.expand_dims(x, axis=2) +y = np.expand_dims(y, axis=4) +y = np.expand_dims(y, axis=5) + +expect(node, inputs=[x, axes], outputs=[y], name="test_unsqueeze_three_axes") +``` + +
+ + +
+unsqueeze_two_axes + +```python +x = np.random.randn(3, 4, 5).astype(np.float32) +axes = np.array([1, 4]).astype(np.int64) + +node = onnx.helper.make_node( + "Unsqueeze", + inputs=["x", "axes"], + outputs=["y"], +) +y = np.expand_dims(x, axis=1) +y = np.expand_dims(y, axis=4) + +expect(node, inputs=[x, axes], outputs=[y], name="test_unsqueeze_two_axes") +``` + +
+ + +
+unsqueeze_unsorted_axes + +```python +x = np.random.randn(3, 4, 5).astype(np.float32) +axes = np.array([5, 4, 2]).astype(np.int64) + +node = onnx.helper.make_node( + "Unsqueeze", + inputs=["x", "axes"], + outputs=["y"], +) +y = np.expand_dims(x, axis=2) +y = np.expand_dims(y, axis=4) +y = np.expand_dims(y, axis=5) + +expect(node, inputs=[x, axes], outputs=[y], name="test_unsqueeze_unsorted_axes") +``` + +
+ + +### **Upsample** (deprecated) + + Upsample the input tensor. + Each dimension value of the output tensor is: + output_dimension = floor(input_dimension * scale). + +#### Version + +This version of the operator has been deprecated since version 10 of the default ONNX operator set. + +Other versions of this operator: 7, 9 + + +#### Examples + +
+nearest + +```python +node = onnx.helper.make_node( + "Upsample", + inputs=["X", "scales"], + outputs=["Y"], + mode="nearest", +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 3.0], dtype=np.float32) + +output = np.array( + [ + [ + [ + [1, 1, 1, 2, 2, 2], + [1, 1, 1, 2, 2, 2], + [3, 3, 3, 4, 4, 4], + [3, 3, 3, 4, 4, 4], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_upsample_nearest", + opset_imports=[helper.make_opsetid("", 9)], +) +``` + +
+ + +### **Where** + + Return elements, either from X or Y, depending on condition. + Where behaves like + [numpy.where](https://docs.scipy.org/doc/numpy/reference/generated/numpy.where.html) + with three parameters. + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 16 of the default ONNX operator set. + +Other versions of this operator: 9 + +#### Inputs + +
+
condition (non-differentiable) : B
+
When True (nonzero), yield X, otherwise yield Y
+
X (differentiable) : T
+
values selected at indices where condition is True
+
Y (differentiable) : T
+
values selected at indices where condition is False
+
+ +#### Outputs + +
+
output (differentiable) : T
+
Tensor of shape equal to the broadcasted shape of condition, X, and Y.
+
+ +#### Type Constraints + +
+
B : tensor(bool)
+
Constrain to boolean tensors.
+
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Constrain input and output types to all tensor types (including bfloat).
+
+ + +#### Examples + +
+long + +```python +node = onnx.helper.make_node( + "Where", + inputs=["condition", "x", "y"], + outputs=["z"], +) + +condition = np.array([[1, 0], [1, 1]], dtype=bool) +x = np.array([[1, 2], [3, 4]], dtype=np.int64) +y = np.array([[9, 8], [7, 6]], dtype=np.int64) +z = np.where(condition, x, y) # expected output [[1, 8], [3, 4]] +expect( + node, inputs=[condition, x, y], outputs=[z], name="test_where_long_example" +) +``` + +
+ + +
+where + +```python +node = onnx.helper.make_node( + "Where", + inputs=["condition", "x", "y"], + outputs=["z"], +) + +condition = np.array([[1, 0], [1, 1]], dtype=bool) +x = np.array([[1, 2], [3, 4]], dtype=np.float32) +y = np.array([[9, 8], [7, 6]], dtype=np.float32) +z = np.where(condition, x, y) # expected output [[1, 8], [3, 4]] +expect(node, inputs=[condition, x, y], outputs=[z], name="test_where_example") +``` + +
+ + +### **Xor** + + Returns the tensor resulted from performing the `xor` logical operation + elementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support). + + This operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md). + +#### Version + +This version of the operator has been available since version 7 of the default ONNX operator set. + +Other versions of this operator: 1 + +#### Inputs + +
+
A (non-differentiable) : T
+
First input operand for the logical operator.
+
B (non-differentiable) : T
+
Second input operand for the logical operator.
+
+ +#### Outputs + +
+
C (non-differentiable) : T1
+
Result tensor.
+
+ +#### Type Constraints + +
+
T : tensor(bool)
+
Constrain input to boolean tensor.
+
T1 : tensor(bool)
+
Constrain output to boolean tensor.
+
+ + +#### Examples + +
+xor + +```python +node = onnx.helper.make_node( + "Xor", + inputs=["x", "y"], + outputs=["xor"], +) + +# 2d +x = (np.random.randn(3, 4) > 0).astype(bool) +y = (np.random.randn(3, 4) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor2d") + +# 3d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(3, 4, 5) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor3d") + +# 4d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor4d") +``` + +
+ + +
+xor_broadcast + +```python +node = onnx.helper.make_node( + "Xor", + inputs=["x", "y"], + outputs=["xor"], +) + +# 3d vs 1d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(5) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor_bcast3v1d") + +# 3d vs 2d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(4, 5) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor_bcast3v2d") + +# 4d vs 2d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(5, 6) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor_bcast4v2d") + +# 4d vs 3d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(4, 5, 6) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor_bcast4v3d") + +# 4d vs 4d +x = (np.random.randn(1, 4, 1, 6) > 0).astype(bool) +y = (np.random.randn(3, 1, 5, 6) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor_bcast4v4d") +``` + +
+ + +## ai.onnx.preview +### experimental **ai.onnx.preview.FlexAttention** + + Computes scaled dot-product attention over rank-4 (batched, multi-head) inputs, + with optional user-provided customization subgraphs at two stages: + + 1. score_mod: Modify the attention score tensor after Q·K^T + 2. prob_mod: Modify the probability tensor after Softmax + + This operator mirrors the capabilities of PyTorch's flex_attention: + https://docs.pytorch.org/docs/stable/nn.attention.flex_attention.html + + Input Shapes (MUST be rank-4 tensors): + - Q: `(batch_size, q_num_heads, q_sequence_length, head_size)` + - K: `(batch_size, kv_num_heads, kv_sequence_length, head_size)` + - V: `(batch_size, kv_num_heads, kv_sequence_length, v_head_size)` + + Output Shape: + - Y: `(batch_size, q_num_heads, q_sequence_length, v_head_size)` + + FlexAttention Computation: + ``` + Scores = (Q @ K^T) * scale + Scores = score_mod(Scores) # if 'score_mod' is provided + Probs = Softmax(Scores, axis=-1) + Probs = prob_mod(Probs) # if 'prob_mod' is provided + Y = Probs @ V + ``` + + Grouped Query Attention (GQA): + When `q_num_heads != kv_num_heads`, each K/V head is shared by a contiguous + group of query heads in head-index order. Let + `group_size = q_num_heads / kv_num_heads`; then query head `h` uses K/V head + `floor(h / group_size)`. `q_num_heads` must be a multiple of + `kv_num_heads`. + + Modifier Subgraphs (score_mod, prob_mod): + Each modifier subgraph takes exactly one rank-4 tensor input and must produce + exactly one rank-4 tensor output of the same shape and element type. + - score_mod input/output shape: `(batch_size, q_num_heads, q_sequence_length, kv_sequence_length)` + - prob_mod input/output shape: `(batch_size, q_num_heads, q_sequence_length, kv_sequence_length)` + The element type is determined by softmax_precision (defaults to float32 for + non-double inputs, otherwise double). + + Masking can be expressed in score_mod by writing masked positions as -inf (or a + large negative value appropriate for the target precision). + +#### Version + +No versioning maintained for experimental ops. +#### Attributes + +
+
prob_mod : graph
+
Optional probability modifier subgraph with 1 rank-4 tensor input and 1 rank-4 tensor output of the same shape and element type: (probs) -> probs_out. probs has softmax_precision element type and shape (B, Hq, L, S). The output must preserve the input shape.
+
scale : float
+
Scaling factor for Q*K^T. Defaults to 1/sqrt(head_size).
+
score_mod : graph
+
Optional score modifier subgraph with 1 rank-4 tensor input and 1 rank-4 tensor output of the same shape and element type: (scores) -> scores_out. scores has softmax_precision element type and shape (B, Hq, L, S). The output must preserve the input shape.
+
softmax_precision : int
+
Floating-point precision for softmax computation. Defaults to float32 for non-double inputs, otherwise uses double. Must be explicitly specified for non-float types.
+
+ +#### Inputs + +
+
Q : T1
+
Query tensor with shape `(batch_size, q_num_heads, q_seq_len, head_size)`.
+
K : T1
+
Key tensor with shape `(batch_size, kv_num_heads, kv_seq_len, head_size)`.
+
V : T1
+
Value tensor with shape `(batch_size, kv_num_heads, kv_seq_len, v_head_size)`.
+
+ +#### Outputs + +
+
Y : T1
+
Output tensor with shape `(batch_size, q_num_heads, q_seq_len, v_head_size)`.
+
+ +#### Type Constraints + +
+
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double)
+
Constrain Q, K, V to float tensors.
+
+ + +#### Examples + +
+flexattention + +```python +"""Basic FlexAttention test with default settings.""" +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, L, E = 2, 4, 8, 16 +S, Ev = 6, 16 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +(Y,) = _compute_flex_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+flexattention_causal_mask + +```python +"""FlexAttention with causal masking score_mod (Qwen-3, Gemma-3, Llama-3 pattern).""" +score_mod_graph = _make_score_mod_causal_mask_graph(TensorProto.FLOAT) + +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) +score_mod_attr = helper.make_attribute("score_mod", score_mod_graph) +node.attribute.append(score_mod_attr) + +B, Hq, L, E = 1, 2, 4, 8 +S, Ev = 4, 8 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +# Manually compute expected output with causal masking +scale = 1.0 / np.sqrt(E) +scores = np.einsum("bhle,bhse->bhls", Q, K) * scale +# Apply causal mask: set future positions to -inf +q_idx = np.arange(L).reshape(1, 1, L, 1) +k_idx = np.arange(S).reshape(1, 1, 1, S) +mask = q_idx >= k_idx +scores = np.where(mask, scores, -np.inf) +probs = np.exp(scores - scores.max(axis=-1, keepdims=True)) +probs = probs / probs.sum(axis=-1, keepdims=True) +Y = np.einsum("bhls,bhsv->bhlv", probs, V).astype(np.float32) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_causal_mask", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+flexattention_diff_head_sizes + +```python +"""FlexAttention with different head sizes for Q/K vs V.""" +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, L, E = 2, 4, 8, 16 +S, Ev = 6, 32 # V has different head size + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +(Y,) = _compute_flex_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_diff_head_sizes", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+flexattention_double + +```python +"""FlexAttention with double precision inputs.""" +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, L, E = 2, 4, 8, 16 +S, Ev = 6, 16 + +Q = np.random.rand(B, Hq, L, E).astype(np.float64) +K = np.random.rand(B, Hq, S, E).astype(np.float64) +V = np.random.rand(B, Hq, S, Ev).astype(np.float64) + +(Y,) = _compute_flex_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_double", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+flexattention_fp16 + +```python +"""FlexAttention with float16 inputs.""" +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, L, E = 2, 4, 8, 16 +S, Ev = 6, 16 + +Q = np.random.rand(B, Hq, L, E).astype(np.float16) +K = np.random.rand(B, Hq, S, E).astype(np.float16) +V = np.random.rand(B, Hq, S, Ev).astype(np.float16) + +(Y,) = _compute_flex_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_fp16", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+flexattention_gqa + +```python +"""FlexAttention with Grouped Query Attention (GQA).""" +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, Hkv, L, S, E, Ev = 2, 8, 2, 4, 6, 16, 16 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hkv, S, E).astype(np.float32) +V = np.random.rand(B, Hkv, S, Ev).astype(np.float32) + +(Y,) = _compute_flex_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_gqa", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+flexattention_prob_mod + +```python +"""FlexAttention with prob_mod subgraph (scales probabilities).""" +scale_value = 0.5 +prob_mod_graph = _make_prob_mod_scale_graph(scale_value, TensorProto.FLOAT) + +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) +prob_mod_attr = helper.make_attribute("prob_mod", prob_mod_graph) +node.attribute.append(prob_mod_attr) + +B, Hq, L, E = 1, 2, 3, 4 +S, Ev = 3, 4 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +scale = 1.0 / np.sqrt(E) +scores = np.einsum("bhle,bhse->bhls", Q, K) * scale +probs = np.exp(scores - scores.max(axis=-1, keepdims=True)) +probs = probs / probs.sum(axis=-1, keepdims=True) +probs = probs * scale_value +Y = np.einsum("bhls,bhsv->bhlv", probs, V).astype(np.float32) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_prob_mod", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+flexattention_relative_positional + +```python +"""FlexAttention with relative positional bias score_mod.""" +score_mod_graph = _make_score_mod_relative_positional_graph(TensorProto.FLOAT) + +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) +score_mod_attr = helper.make_attribute("score_mod", score_mod_graph) +node.attribute.append(score_mod_attr) + +B, Hq, L, E = 1, 2, 4, 8 +S, Ev = 4, 8 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +# Manually compute expected output with relative positional bias +scale = 1.0 / np.sqrt(E) +scores = np.einsum("bhle,bhse->bhls", Q, K) * scale +q_idx = np.arange(L).reshape(-1, 1) +k_idx = np.arange(S).reshape(1, -1) +rel_pos = (q_idx - k_idx).astype(np.float32) +scores = scores + rel_pos +probs = np.exp(scores - scores.max(axis=-1, keepdims=True)) +probs = probs / probs.sum(axis=-1, keepdims=True) +Y = np.einsum("bhls,bhsv->bhlv", probs, V).astype(np.float32) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_relative_positional", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+flexattention_scaled + +```python +"""FlexAttention with explicit scale attribute.""" +scale = 0.1 +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, L, E = 2, 4, 8, 16 +S, Ev = 6, 16 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +(Y,) = _compute_flex_attention(Q, K, V, scale=scale) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_scaled", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+flexattention_score_mod + +```python +"""FlexAttention with score_mod subgraph (adds bias to scores).""" +bias_value = 0.5 +score_mod_graph = _make_score_mod_bias_graph(bias_value, TensorProto.FLOAT) + +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) +# Add score_mod as a graph attribute +score_mod_attr = helper.make_attribute("score_mod", score_mod_graph) +node.attribute.append(score_mod_attr) + +B, Hq, L, E = 1, 2, 3, 4 +S, Ev = 3, 4 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +scale = 1.0 / np.sqrt(E) +scores = np.einsum("bhle,bhse->bhls", Q, K) * scale +scores = scores + bias_value # score_mod: add bias +probs = np.exp(scores - scores.max(axis=-1, keepdims=True)) +probs = probs / probs.sum(axis=-1, keepdims=True) +Y = np.einsum("bhls,bhsv->bhlv", probs, V).astype(np.float32) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_score_mod", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+flexattention_soft_cap + +```python +"""FlexAttention with soft capping score_mod (Gemma-2 pattern).""" +cap_value = 20.0 +score_mod_graph = _make_score_mod_soft_cap_graph(cap_value, TensorProto.FLOAT) + +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) +score_mod_attr = helper.make_attribute("score_mod", score_mod_graph) +node.attribute.append(score_mod_attr) + +B, Hq, L, E = 1, 2, 4, 8 +S, Ev = 4, 8 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +# Manually compute expected output with soft capping +scale = 1.0 / np.sqrt(E) +scores = np.einsum("bhle,bhse->bhls", Q, K) * scale +scores = np.tanh(scores / cap_value) * cap_value +probs = np.exp(scores - scores.max(axis=-1, keepdims=True)) +probs = probs / probs.sum(axis=-1, keepdims=True) +Y = np.einsum("bhls,bhsv->bhlv", probs, V).astype(np.float32) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_soft_cap", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +## ai.onnx.preview.training +### **ai.onnx.preview.training.Adagrad** + + Compute one iteration of ADAGRAD, a stochastic gradient based optimization + algorithm. This operator can conduct the optimization of multiple tensor variables. + + Let's define the behavior of this operator. As you can imagine, ADAGRAD requires + some parameters: + + - The initial learning-rate "R". + - The update count "T". That is, the number of training iterations conducted. + - A L2-norm regularization coefficient "norm_coefficient". + - A learning-rate decay factor "decay_factor". + - A small constant "epsilon" to avoid dividing-by-zero. + + At each ADAGRAD iteration, the optimized tensors are moved along a direction + computed based on their estimated gradient and accumulated squared gradient. Assume + that only a single tensor "X" is updated by this operator. We need the value of "X", + its gradient "G", and its accumulated squared gradient "H". Therefore, variables in + this operator's input list are sequentially "R", "T", "X", "G", and "H". Other + parameters are given as attributes because they are usually constants. Also, the + corresponding output tensors are the new value of "X" (called "X_new"), and then + the new accumulated squared gradient (called "H_new"). Those outputs are computed + from the given inputs following the pseudo code below. + + Let "+", "-", "*", and "/" are all element-wise arithmetic operations with + numpy-style broadcasting support. The pseudo code to compute those outputs is: + + // Compute a scalar learning-rate factor. At the first update of X, T is generally + // 0 (0-based update index) or 1 (1-based update index). + r = R / (1 + T * decay_factor); + + // Add gradient of 0.5 * norm_coefficient * ||X||_2^2, where ||X||_2 is the 2-norm. + G_regularized = norm_coefficient * X + G; + + // Compute new accumulated squared gradient. + H_new = H + G_regularized * G_regularized; + + // Compute the adaptive part of per-coordinate learning rate. Note that Sqrt(...) + // computes element-wise square-root. + H_adaptive = Sqrt(H_new) + epsilon + + // Compute the new value of "X". + X_new = X - r * G_regularized / H_adaptive; + + If one assign this operators to optimize multiple inputs, for example, "X_1" and "X_2", the same + pseudo code may be extended to handle all tensors jointly. More specifically, we can view "X" as a + concatenation of "X_1" and "X_2" (of course, their gradient and accumulate gradient should + be concatenated too) and then just reuse the entire pseudo code. + + Note that ADAGRAD was first proposed in http://jmlr.org/papers/volume12/duchi11a/duchi11a.pdf. + In that reference paper, this operator is a special case of the Figure 1's composite mirror + descent update. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.preview.training' operator set. + +#### Attributes + +
+
decay_factor : float (default is 0.0)
+
The decay factor of learning rate after one update.The effective learning rate is computed by r = R / (1 + T * decay_factor). Default to 0 so that increasing update counts doesn't reduce the learning rate.
+
epsilon : float (default is 0.0)
+
Small scalar to avoid dividing by zero.
+
norm_coefficient : float (default is 0.0)
+
Regularization coefficient in 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization.
+
+ +#### Inputs (3 - ∞) + +
+
R : T1
+
The initial learning rate.
+
T : T2
+
The update count of "X". It should be a scalar.
+
inputs (variadic, heterogeneous) : T3
+
The current values of optimized tensors, followed by their respective gradients, followed by their respective accumulated squared gradients.For example, if two tensor "X_1" and "X_2" are optimized, The input list would be ["X_1", "X_2", gradient of "X_1", gradient of "X_2", accumulated squared gradient of "X_1", accumulated squared gradient of "X_2"].
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : T3
+
Updated values of optimized tensors, followed by their updated values of accumulated squared gradients. For example, if two tensor "X_1" and "X_2" are optimized, the output list would be [new value of "X_1," new value of "X_2" new accumulated squared gradient of "X_1", new accumulated squared gradient of "X_2"].
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double)
+
Constrain input types to float scalars.
+
T2 : tensor(int64)
+
Constrain input types to 64-bit integer scalars.
+
T3 : tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+adagrad + +```python +# Define operator attributes. +norm_coefficient = 0.001 +epsilon = 1e-5 +decay_factor = 0.1 + +# Create operator. +node = onnx.helper.make_node( + "Adagrad", + inputs=["R", "T", "X", "G", "H"], + outputs=["X_new", "H_new"], + norm_coefficient=norm_coefficient, + epsilon=epsilon, + decay_factor=decay_factor, + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar +x = np.array([1.0], dtype=np.float32) +g = np.array([-1.0], dtype=np.float32) +h = np.array([2.0], dtype=np.float32) + +# Compute expected outputs of Adagrad. +x_new, h_new = apply_adagrad( + r, t, x, g, h, norm_coefficient, epsilon, decay_factor +) + +# Check results. +expect( + node, + inputs=[r, t, x, g, h], + outputs=[x_new, h_new], + name="test_adagrad", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+ + +
+adagrad_multiple + +```python +# Define operator attributes. +norm_coefficient = 0.001 +epsilon = 1e-5 +decay_factor = 0.1 + +node = onnx.helper.make_node( + "Adagrad", + inputs=["R", "T", "X1", "X2", "G1", "G2", "H1", "H2"], + outputs=["X1_new", "X2_new", "H1_new", "H2_new"], + norm_coefficient=norm_coefficient, + epsilon=epsilon, + decay_factor=decay_factor, + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar + +x1 = np.array([1.0], dtype=np.float32) +g1 = np.array([-1.0], dtype=np.float32) +h1 = np.array([2.0], dtype=np.float32) + +x2 = np.array([1.0, 2.0], dtype=np.float32) +g2 = np.array([-1.0, -3.0], dtype=np.float32) +h2 = np.array([4.0, 1.0], dtype=np.float32) + +# Compute expected outputs of Adagrad. +x1_new, h1_new = apply_adagrad( + r, t, x1, g1, h1, norm_coefficient, epsilon, decay_factor +) +x2_new, h2_new = apply_adagrad( + r, t, x2, g2, h2, norm_coefficient, epsilon, decay_factor +) + +# Check results. +expect( + node, + inputs=[r, t, x1, x2, g1, g2, h1, h2], + outputs=[x1_new, x2_new, h1_new, h2_new], + name="test_adagrad_multiple", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+ + +### **ai.onnx.preview.training.Adam** + + Compute one iteration of Adam, a stochastic gradient based optimization + algorithm. This operator can conduct the optimization of multiple tensor variables. + + Let's define the behavior of this operator. First of all, Adam requires + some parameters: + + - The learning-rate "R". + - The update count "T". That is, the number of training iterations conducted. + - A L2-norm regularization coefficient "norm_coefficient". + - A small constant "epsilon" to avoid dividing-by-zero. + - Two coefficients, "alpha" and "beta". + + At each Adam iteration, the optimized tensors are moved along a direction + computed based on their exponentially-averaged historical gradient and + exponentially-averaged historical squared gradient. Assume that only a tensor + "X" is being optimized. The rest of required information is + + - the value of "X", + - "X"'s gradient (denoted by "G"), + - "X"'s exponentially-averaged historical gradient (denoted by "V"), and + - "X"'s exponentially-averaged historical squared gradient (denoted by "H"). + + Some of those parameters are passed into this operator as input tensors and others + are stored as this operator's attributes. Specifically, this operator's input tensor + list is ["R", "T", "X", "G", "V", "H"]. That is, "R" is the first input, "T" is + the second input, and so on. Other parameters are given as attributes because they + are constants. Moreover, the corresponding output tensors are + + - the new value of "X" (called "X_new"), + - the new exponentially-averaged historical gradient (denoted by "V_new"), and + - the new exponentially-averaged historical squared gradient (denoted by "H_new"). + + Those outputs are computed following the pseudo code below. + + Let "+", "-", "*", and "/" are all element-wise arithmetic operations with + numpy-style broadcasting support. The pseudo code to compute those outputs is: + + // Add gradient of 0.5 * norm_coefficient * ||X||_2^2, where ||X||_2 is the 2-norm. + G_regularized = norm_coefficient * X + G + + // Update exponentially-averaged historical gradient. + V_new = alpha * V + (1 - alpha) * G_regularized + + // Update exponentially-averaged historical squared gradient. + H_new = beta * H + (1 - beta) * G_regularized * G_regularized + + // Compute the element-wise square-root of H_new. V_new will be element-wisely + // divided by H_sqrt for a better update direction. + H_sqrt = Sqrt(H_new) + epsilon + + // Compute learning-rate. Note that "alpha**T"/"beta**T" is alpha's/beta's T-th power. + R_adjusted = T > 0 ? R * Sqrt(1 - beta**T) / (1 - alpha**T) : R + + // Compute new value of "X". + X_new = X - R_adjusted * V_new / H_sqrt + + // Post-update regularization. + X_final = (1 - norm_coefficient_post) * X_new + + If there are multiple inputs to be optimized, the pseudo code will be applied + independently to each of them. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.preview.training' operator set. + +#### Attributes + +
+
alpha : float (default is 0.9)
+
Coefficient of previously accumulated gradient in running average. Default to 0.9.
+
beta : float (default is 0.999)
+
Coefficient of previously accumulated squared-gradient in running average. Default to 0.999.
+
epsilon : float (default is 0.0)
+
Small scalar to avoid dividing by zero.
+
norm_coefficient : float (default is 0.0)
+
Regularization coefficient of 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization.
+
norm_coefficient_post : float (default is 0.0)
+
Regularization coefficient of 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization.
+
+ +#### Inputs (3 - ∞) + +
+
R : T1
+
The initial learning rate.
+
T : T2
+
The update count of "X". It should be a scalar.
+
inputs (variadic, heterogeneous) : T3
+
The tensors to be optimized, followed by their respective gradients, followed by their respective accumulated gradients (aka momentum), followed by their respective accumulated squared gradients. For example, to optimize tensors "X_1" and "X_2,", the input list would be ["X_1", "X_2", gradient of "X_1", gradient of "X_2", accumulated gradient of "X_1", accumulated gradient of "X_2", accumulated squared gradient of "X_1", accumulated squared gradient of "X_2"].
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : T3
+
New values of optimized tensors, followed by their respective new accumulated gradients, followed by their respective new accumulated squared gradients. For example, if two tensors "X_1" and "X_2" are optimized, the outputs list would be [new value of "X_1", new value of "X_2", new accumulated gradient of "X_1", new accumulated gradient of "X_2", new accumulated squared gradient of "X_1", new accumulated squared gradient of "X_2"].
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double)
+
Constrain input types to float scalars.
+
T2 : tensor(int64)
+
Constrain input types to 64-bit integer scalars.
+
T3 : tensor(float), tensor(double)
+
Constrain input and output types to float tensors.
+
+ + +#### Examples + +
+adam + +```python +# Define operator attributes. +norm_coefficient = 0.001 +alpha = 0.95 +beta = 0.1 +epsilon = 1e-7 + +# Create operator. +node = onnx.helper.make_node( + "Adam", + inputs=["R", "T", "X", "G", "V", "H"], + outputs=["X_new", "V_new", "H_new"], + norm_coefficient=norm_coefficient, + alpha=alpha, + beta=beta, + epsilon=epsilon, + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar +x = np.array([1.2, 2.8], dtype=np.float32) +g = np.array([-0.94, -2.5], dtype=np.float32) +v = np.array([1.7, 3.6], dtype=np.float32) +h = np.array([0.1, 0.1], dtype=np.float32) + +# Compute expected outputs of Adam. +x_new, v_new, h_new = apply_adam( + r, t, x, g, v, h, norm_coefficient, 0.0, alpha, beta, epsilon +) + +# Check results. +expect( + node, + inputs=[r, t, x, g, v, h], + outputs=[x_new, v_new, h_new], + name="test_adam", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+ + +
+adam_multiple + +```python +# Define operator attributes. +norm_coefficient = 0.001 +alpha = 0.95 +beta = 0.85 +epsilon = 1e-2 + +node = onnx.helper.make_node( + "Adam", + inputs=["R", "T", "X1", "X2", "G1", "G2", "V1", "V2", "H1", "H2"], + outputs=["X1_new", "X2_new", "V1_new", "V2_new", "H1_new", "H2_new"], + norm_coefficient=norm_coefficient, + alpha=alpha, + beta=beta, + epsilon=epsilon, + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar + +x1 = np.array([1.0], dtype=np.float32) +g1 = np.array([-1.0], dtype=np.float32) +v1 = np.array([2.0], dtype=np.float32) +h1 = np.array([0.5], dtype=np.float32) + +x2 = np.array([1.0, 2.0], dtype=np.float32) +g2 = np.array([-1.0, -3.0], dtype=np.float32) +v2 = np.array([4.0, 1.0], dtype=np.float32) +h2 = np.array([1.0, 10.0], dtype=np.float32) + +# Compute expected outputs of Adam. +x1_new, v1_new, h1_new = apply_adam( + r, t, x1, g1, v1, h1, norm_coefficient, 0.0, alpha, beta, epsilon +) +x2_new, v2_new, h2_new = apply_adam( + r, t, x2, g2, v2, h2, norm_coefficient, 0.0, alpha, beta, epsilon +) + +# Check results. +expect( + node, + inputs=[r, t, x1, x2, g1, g2, v1, v2, h1, h2], + outputs=[x1_new, x2_new, v1_new, v2_new, h1_new, h2_new], + name="test_adam_multiple", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+ + +### **ai.onnx.preview.training.Gradient** + + Gradient operator computes the partial derivatives of a specific tensor w.r.t. + some other tensors. This operator is widely used in gradient-based training + algorithms. To illustrate its use, let's consider a computation graph, + + ``` + X -----. + | + v + W --> Conv --> H --> Gemm --> Y + ^ + | + Z + ``` + + , where W and Z are trainable tensors. Note that operators' attributes are + omitted for the sake of simplicity. Let dY/dW (dY/dZ) be the gradient of + Y with respect to W (Z). The user can compute gradient by inserting Gradient + operator to form another graph shown below. + + ``` + W --> Conv --> H --> Gemm --> Y + | ^ ^ + | | | + | X Z + | | | + | | .----------' + | | | (W/Z/X is the 1st/2nd/3rd input of Gradient as shown in + | | | "xs" followed by "zs") + | v v + '---> Gradient(xs=["W", "Z"], zs=["X"], y="Y") + | | + | '-----------------------------------> dY/dW (1st output of Gradient) + | + '---------------------------------------> dY/dZ (2nd output of Gradient) + ``` + + By definition, the tensor "y" is a function of independent variables in "xs" + and "zs". Since we only compute the gradient of "y" w.r.t. the differentiable + variables in "xs", this Gradient only outputs dY/dW and dY/dZ. Note that "H" + cannot appear in "xs" and "zs". The reason is that "H" can be determined by + tensors "W" and "X" and therefore "H" is not an independent variable. + + All outputs are optional. If needed, for example, user can assign an empty + string to the 1st output name of that Gradient to skip the generation of dY/dW. + Note that the concept of optional outputs can also be found in ONNX's RNN, GRU, + and LSTM. + + Gradient operator can compute derivative against intermediate tensors. For + example, the gradient of Y with respect to H can be done via + + ``` + W --> Conv --> H --> Gemm --> Y + ^ | ^ + | | | + X | Z + .-------' | + | .----------' + | | (H/Z is the 1st/2nd input of Gradient as shown in "xs") + v v + Gradient(xs=["H", "Z"], y="Y") + | | + | '-----------------------------------> dY/dH (1st output of Gradient) + | + '---------------------------------------> dY/dZ (2nd output of Gradient) + ``` + + It is possible to represent high-order differentiation using Gradient operators. + For example, given the following linear model: + + ``` + W --> Gemm --> Y --> Loss --> O + ^ ^ + | | + X L + ``` + + To compute the 2nd order derivative of O with respect to W (denoted by + d^2O/dW^2), one can do + + ``` + W --> Gemm --> Y --> Loss --> O + | ^ ^ + | | | + | X .------------L + | | | | + | | | v + +------+-+> Gradient(xs=["X", "W"], zs=["L"], y="O") ---> dO/dX (1st output of Gradient) + | | | | + | | | '---> dO/dW (2nd output of Gradient) + | v v + '---> Gradient(xs=["X", "W"], zs=["L"], y="dO/dW") ---> d(dO/dW)dX (1st output of + | Gradient) + | + | + '---> d^2O/dW^2 (2nd output of Gradient) + ``` + + The tensors named in attributes "xs", "zs", and "y" define the differentiated + computation graph, and the inputs to Gradient node define the values at + which the gradient is computed. We can feed different tensors to the identified + graph. For example, one can compute the gradient of Y with respect to H at + a specific value of H, H_1, by providing that value as an input to the Gradient + node. + + ``` + W --> Conv --> H --> Gemm --> Y + ^ ^ + | | + X Z + + Z_1 (2nd input of Gradient) + | + v + H_1 --> Gradient(xs=["H", "Z"], y="Y") ---> dY/dH when H = H_1 and Y = Y_1. + | + '------------------------------> dY/dZ (2nd output of Gradient) + ``` + + When the inputs of Gradient are the tensors named in "xs" and "zs", the + computation can be optimized. More specifically, intermediate variables in + forward pass can be reused if the gradient is computed via reverse-mode + auto-differentiation. + + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.preview.training' operator set. + +#### Attributes + +
+
xs : list of strings (required)
+
Input tensor names of the differentiated sub-graph. It contains only the necessary differentiated inputs of a (sub-)graph. Variables (usually called intermediate variables) that can be generated from inputs cannot be included in this attribute.
+
y : string (required)
+
The targeted tensor. It can be viewed as the output of the differentiated function. The attribute "xs" and attribute "zs" are the minimal independent variable set that determines the value of "y".
+
zs : list of strings
+
Input tensor names of the differentiated sub-graph. It contains only the necessary non-differentiated inputs of a (sub-)graph. Variables (usually called intermediate variables) that can be generated from inputs cannot be included in this attribute.
+
+ +#### Inputs (1 - ∞) + +
+
Inputs (variadic, heterogeneous) : T1
+
The values fed into graph identified by the attributes. The i-th input is the value of the i-th tensor specified in the concatenated list of the attribute "xs" and the attribute "zs". For example, if xs=["A", "B"] and zs=["C"], the first input is used as the value of symbol "A" and the 3rd input is substituted for all the occurrences of "C".
+
+ +#### Outputs (1 - ∞) + +
+
Outputs (variadic, heterogeneous) : T2
+
The gradient of the tensor specified by the attribute "y" with respect to each of tensors specified in the attribute "xs". The i-th output is the gradient of "y" with respect to the i-th tensor specified in the attribute "xs".
+
+ +#### Type Constraints + +
+
T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)
+
Allow outputs to be any kind of tensor.
+
T2 : tensor(float16), tensor(float), tensor(double)
+
Allow inputs to be any kind of floating-point tensor.
+
+ + +#### Examples + +
+gradient_scalar_add + +```python +add_node = onnx.helper.make_node("Add", ["a", "b"], ["c"], name="my_add") +gradient_node = onnx.helper.make_node( + "Gradient", + ["a", "b"], + ["dc_da", "dc_db"], + name="my_gradient", + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, + xs=["a", "b"], + y="c", +) + +a = np.array(1.0).astype(np.float32) +b = np.array(2.0).astype(np.float32) +c = a + b +# dc / da = d(a+b) / da = 1 +dc_da = np.array(1).astype(np.float32) +# db / db = d(a+b) / db = 1 +dc_db = np.array(1).astype(np.float32) + +graph = onnx.helper.make_graph( + nodes=[add_node, gradient_node], + name="GradientOfAdd", + inputs=[ + onnx.helper.make_tensor_value_info("a", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("b", onnx.TensorProto.FLOAT, []), + ], + outputs=[ + onnx.helper.make_tensor_value_info("c", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("dc_da", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("dc_db", onnx.TensorProto.FLOAT, []), + ], +) +opsets = [ + onnx.helper.make_operatorsetid(ONNX_DOMAIN, 12), + onnx.helper.make_operatorsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1), +] +model = onnx.helper.make_model_gen_version( + graph, producer_name="backend-test", opset_imports=opsets +) +expect( + model, inputs=[a, b], outputs=[c, dc_da, dc_db], name="test_gradient_of_add" +) +``` + +
+ + +
+gradient_scalar_add_and_mul + +```python +add_node = onnx.helper.make_node("Add", ["a", "b"], ["c"], name="my_add") +mul_node = onnx.helper.make_node("Mul", ["c", "a"], ["d"], name="my_mul") +gradient_node = onnx.helper.make_node( + "Gradient", + ["a", "b"], + ["dd_da", "dd_db"], + name="my_gradient", + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, + xs=["a", "b"], + y="d", +) + +a = np.array(1.0).astype(np.float32) +b = np.array(2.0).astype(np.float32) +c = a + b +# d = a * c = a * (a + b) +d = a * c +# dd / da = d(a*a+a*b) / da = 2 * a + b +dd_da = (2 * a + b).astype(np.float32) +# dd / db = d(a*a+a*b) / db = a +dd_db = a + +graph = onnx.helper.make_graph( + nodes=[add_node, mul_node, gradient_node], + name="GradientOfTwoOperators", + inputs=[ + onnx.helper.make_tensor_value_info("a", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("b", onnx.TensorProto.FLOAT, []), + ], + outputs=[ + onnx.helper.make_tensor_value_info("d", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("dd_da", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("dd_db", onnx.TensorProto.FLOAT, []), + ], +) + +opsets = [ + onnx.helper.make_operatorsetid(ONNX_DOMAIN, 12), + onnx.helper.make_operatorsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1), +] +model = onnx.helper.make_model_gen_version( + graph, producer_name="backend-test", opset_imports=opsets +) +expect( + model, + inputs=[a, b], + outputs=[d, dd_da, dd_db], + name="test_gradient_of_add_and_mul", +) +``` + +
+ + +### **ai.onnx.preview.training.Momentum** + + Compute one iteration of stochastic gradient update with momentum. + This operator can conduct the optimization of multiple tensor variables. + + Let's define the behavior of this operator. As you can imagine, SG with momentum requires + several parameters: + + - The learning-rate "R". + - The update count "T". That is, the number of conducted training iterations. It should + be zero in the first training iteration. + - A L2-norm regularization coefficient "norm_coefficient". + - A decay coefficient of previous accumulated gradient (i.e., momentum) "alpha". + - The scaling coefficient of current gradient "beta". + - An attribute to choose either standard momentum or Nesterov's momentum "mode" should + be used. + + For the sake of simplicity, assume that there is only one tensor (called "X") to be optimized. + Other necessary inputs are "X"'s gradient (called "G") and "X"'s momentum (called "V"). This + Momentum operator maps all these inputs to the new value of "X" (called "X_new") and its new + momentum (called "V_new"). + + This operator supports two different momentum algorithms. Set the attribute "mode" to + "nesterov" if Nesterov's momentum is desired. Otherwise, set the attribute "model" to + "standard" to use standard momentum. Computation details are described subsequently. + + Let "+", "-", "*", and "/" are all element-wise operations with numpy-style broadcasting. + + Pseudo code for SG with standard momentum: + + // Add gradient of 0.5 * norm_coefficient * ||X||^2, where ||X|| is the sum of squared + // values of all elements in X. + G_regularized = norm_coefficient * X + G + + // In the first training iteration, beta should always be 1. + beta_adjusted = T > 0 ? beta : 1 + + // Compute the current momentum based on previous momentum and the current gradient. + V_new = alpha * V + beta_adjusted * G_regularized + + // Update X. + X_new = X - R * V_new + + Pseudo code for SG with Nesterov's momentum: + + // Add gradient of 0.5 * norm_coefficient * ||X||^2, where ||X|| is the sum of squared + // values of all elements in X. + G_regularized = norm_coefficient * X + G; + + // In the first training iteration, beta should always be 1. + beta_adjusted = T > 0 ? beta : 1 + + // Compute the current momentum based on previous momentum and the current gradient. + V_new = alpha * V + beta_adjusted * G_regularized; + + // Compute final update direction and then update X. + X_new = X - R * (G_regularized + alpha * V_new) + + If one assign this operators to optimize multiple inputs, for example, "X_1" and "X_2". The same + pseudo code would be extended to handle all tensors jointly. More specifically, we can view "X" as a + concatenation of "X_1" and "X_2" (of course, their gradient and accumulate gradient should + be concatenated too) and then our pseudo code becomes applicable. + +#### Version + +This version of the operator has been available since version 1 of the 'ai.onnx.preview.training' operator set. + +#### Attributes + +
+
alpha : float (required)
+
The decay factor of momentum. It should be a scalar.
+
beta : float (required)
+
The coefficient of gradient in computing new momentum. It should be a scalar.
+
mode : string (required)
+
Its value should be either "nesterov" or "standard". The value "nesterov" leads to the use of Nesterov's momentum while "standard" invokes stochastic gradient method using standard momentum
+
norm_coefficient : float (required)
+
Coefficient of 0.5 * norm_coefficient * ||X||^2.
+
+ +#### Inputs (3 - ∞) + +
+
R : T1
+
The learning rate.
+
T : T2
+
Update count of "X". It should be a scalar.
+
inputs (variadic, heterogeneous) : T3
+
It sequentially contains the current values of optimized tensors, then their gradient tensors, and finally their momentum tensors. For example, if two tensors "X_1" and "X_2" are optimized, The expected input list would be ["X_1", "X_2", gradient of "X_1", gradient of "X_2", momentum of "X_1", momentum of "X_2"].
+
+ +#### Outputs (1 - ∞) + +
+
outputs (variadic, heterogeneous) : T3
+
It sequentially contains the new values of optimized tensors and then the new values of their momentum tensors. For example, if two tensors "X_1" and "X_2" are optimized, the output list would be [new value of "X_1," new value of "X_2" new momentum of "X_1", new momentum of "X_2"].
+
+ +#### Type Constraints + +
+
T1 : tensor(float), tensor(double)
+
Constrain input types to float scalars.
+
T2 : tensor(int64)
+
Constrain input types to 64-bit integer scalars.
+
T3 : tensor(float), tensor(double)
+
Constrain input types to float tensors.
+
+ + +#### Examples + +
+momentum + +```python +# Define operator attributes. +norm_coefficient = 0.001 +alpha = 0.95 +beta = 0.1 + +# Create operator. +node = onnx.helper.make_node( + "Momentum", + inputs=["R", "T", "X", "G", "V"], + outputs=["X_new", "V_new"], + norm_coefficient=norm_coefficient, + alpha=alpha, + beta=beta, + mode="standard", + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar +x = np.array([1.2, 2.8], dtype=np.float32) +g = np.array([-0.94, -2.5], dtype=np.float32) +v = np.array([1.7, 3.6], dtype=np.float32) + +# Compute expected outputs of Momentum. +x_new, v_new = apply_momentum(r, t, x, g, v, norm_coefficient, alpha, beta) + +# Check results. +expect( + node, + inputs=[r, t, x, g, v], + outputs=[x_new, v_new], + name="test_momentum", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+ + +
+momentum_multiple + +```python +# Define operator attributes. +norm_coefficient = 0.001 +alpha = 0.95 +beta = 0.85 + +node = onnx.helper.make_node( + "Momentum", + inputs=["R", "T", "X1", "X2", "G1", "G2", "H1", "H2"], + outputs=["X1_new", "X2_new", "V1_new", "V2_new"], + norm_coefficient=norm_coefficient, + alpha=alpha, + beta=beta, + mode="standard", + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar + +x1 = np.array([1.0], dtype=np.float32) +g1 = np.array([-1.0], dtype=np.float32) +v1 = np.array([2.0], dtype=np.float32) + +x2 = np.array([1.0, 2.0], dtype=np.float32) +g2 = np.array([-1.0, -3.0], dtype=np.float32) +v2 = np.array([4.0, 1.0], dtype=np.float32) + +# Compute expected outputs of Momentum. +x1_new, v1_new = apply_momentum(r, t, x1, g1, v1, norm_coefficient, alpha, beta) +x2_new, v2_new = apply_momentum(r, t, x2, g2, v2, norm_coefficient, alpha, beta) + +# Check results. +expect( + node, + inputs=[r, t, x1, x2, g1, g2, v1, v2], + outputs=[x1_new, x2_new, v1_new, v2_new], + name="test_momentum_multiple", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+ + +
+nesterov_momentum + +```python +# Define operator attributes. +norm_coefficient = 0.01 +alpha = 0.95 +beta = 1.0 + +# Create operator. +node = onnx.helper.make_node( + "Momentum", + inputs=["R", "T", "X", "G", "V"], + outputs=["X_new", "V_new"], + norm_coefficient=norm_coefficient, + alpha=alpha, + beta=beta, + mode="nesterov", + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar +x = np.array([1.2, 2.8], dtype=np.float32) +g = np.array([-0.94, -2.5], dtype=np.float32) +v = np.array([1.7, 3.6], dtype=np.float32) + +# Compute expected outputs of Momentum. +x_new, v_new = apply_nesterov(r, t, x, g, v, norm_coefficient, alpha, beta) + +# Check results. +expect( + node, + inputs=[r, t, x, g, v], + outputs=[x_new, v_new], + name="test_nesterov_momentum", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+ + diff --git a/docs/Overview.md b/docs/Overview.md new file mode 100644 index 0000000..bbd6b9b --- /dev/null +++ b/docs/Overview.md @@ -0,0 +1,23 @@ + + +# Overview + +Deep learning with neural networks is accomplished through computation over dataflow graphs. Some frameworks (such as CNTK, Caffe2, Theano, and TensorFlow) make use of static graphs, while others (such as PyTorch and Chainer) use dynamic graphs. However, they all provide interfaces that make it simple for developers to construct computation graphs and runtimes that process the graphs in an optimized way. The graph serves as an Intermediate Representation (IR) that captures the specific intent of the developer's source code, and is conducive for optimization and translation to run on specific devices (CPU, GPU, FPGA, etc.). + +## Why a common IR? + +Today, each framework has its own proprietary representation of the graph, though they all provide similar capabilities – meaning each framework is a siloed stack of API, graph, and runtime. Furthermore, frameworks are typically optimized for some characteristic, such as fast training, supporting complicated network architectures, inference on mobile devices, etc. It's up to the developer to select a framework that is optimized for one of these characteristics. Additionally, these optimizations may be better suited for particular stages of development. This leads to significant delays between research and production due to the necessity of conversion. + +With the goal of democratizing AI, we envision empowering developers to select the framework that works best for their project, at any stage of development or deployment. The Open Neural Network Exchange (ONNX) format is a common IR to help establish this powerful ecosystem. + +By providing a common representation of the computation graph, ONNX helps developers choose the right framework for their task, allows authors to focus on innovative enhancements, and enables hardware vendors to streamline optimizations for their platforms. + +ONNX is designed to be an open format. We welcome contributions from the community and encourage everyone to adopt ONNX in their ecosystem. + +## Why two variants? + +The base definition of ONNX includes the necessary support for machine learning algorithms based on neural network technologies. ONNX-ML includes additional types and standard operators commonly used in classical machine learning algorithms. The two variants were created in order to explicitly recognize the desire for some frameworks to go beyond neural network algorithms in a standardized fashion, while allowing other frameworks to support only neural networks. diff --git a/docs/PythonAPIOverview.md b/docs/PythonAPIOverview.md new file mode 100644 index 0000000..ec0b46f --- /dev/null +++ b/docs/PythonAPIOverview.md @@ -0,0 +1,503 @@ + + +# Python API Overview + +The full API is described at [API Reference](https://onnx.ai/onnx/api). + +## Loading an ONNX Model + +```python +import onnx + +# onnx_model is an in-memory ModelProto +onnx_model = onnx.load("path/to/the/model.onnx") +``` + +## Loading an ONNX Model with External Data + +* If the external data is under the same directory of the model, simply use `onnx.load()` + +```python +import onnx + +onnx_model = onnx.load("path/to/the/model.onnx") +``` + +* If the external data is under another directory, use `load_external_data_for_model()` to specify the directory path and load after using `onnx.load()` + +```python +import onnx +from onnx.external_data_helper import load_external_data_for_model + +onnx_model = onnx.load("path/to/the/model.onnx", load_external_data=False) +load_external_data_for_model(onnx_model, "data/directory/path/") +# Then the onnx_model has loaded the external data from the specific directory +``` + +## Converting an ONNX Model to External Data + +```python +from onnx.external_data_helper import convert_model_to_external_data + +# onnx_model is an in-memory ModelProto +onnx_model = ... +convert_model_to_external_data(onnx_model, all_tensors_to_one_file=True, location="filename", size_threshold=1024, convert_attribute=False) +# Then the onnx_model has converted raw data as external data +# Must be followed by save +``` + +## Saving an ONNX Model + +```python +import onnx + +# onnx_model is an in-memory ModelProto +onnx_model = ... + +# Save the ONNX model +onnx.save(onnx_model, "path/to/the/model.onnx") +``` + +## Converting and Saving an ONNX Model to External Data + +```python +import onnx + +# onnx_model is an in-memory ModelProto +onnx_model = ... +onnx.save_model(onnx_model, "path/to/save/the/model.onnx", save_as_external_data=True, all_tensors_to_one_file=True, location="filename", size_threshold=1024, convert_attribute=False) +# Then the onnx_model has converted raw data as external data and saved to specific directory +``` + +## Manipulating TensorProto and Numpy Array + +```{eval-rst} +.. exec_code:: + + import numpy + import onnx + from onnx import numpy_helper + + # Preprocessing: create a Numpy array + numpy_array = numpy.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]], dtype=float) + print(f"Original Numpy array:\n{numpy_array}\n") + + # Convert the Numpy array to a TensorProto + tensor = numpy_helper.from_array(numpy_array) + print(f"TensorProto:\n{tensor}") + + # Convert the TensorProto to a Numpy array + new_array = numpy_helper.to_array(tensor) + print(f"After round trip, Numpy array:\n{new_array}\n") + + # Save the TensorProto + with open("tensor.pb", "wb") as f: + f.write(tensor.SerializeToString()) + + # Load a TensorProto + new_tensor = onnx.TensorProto() + with open("tensor.pb", "rb") as f: + new_tensor.ParseFromString(f.read()) + print(f"After saving and loading, new TensorProto:\n{new_tensor}") + + from onnx import TensorProto, helper + + # Conversion utilities for mapping attributes in ONNX IR + # The functions below are available after ONNX 1.13 + np_dtype = helper.tensor_dtype_to_np_dtype(TensorProto.FLOAT) + print(f"The converted numpy dtype for {helper.tensor_dtype_to_string(TensorProto.FLOAT)} is {np_dtype}.") + storage_dtype = helper.tensor_dtype_to_storage_tensor_dtype(TensorProto.FLOAT) + print(f"The storage dtype for {helper.tensor_dtype_to_string(TensorProto.FLOAT)} is {helper.tensor_dtype_to_string(storage_dtype)}.") + field_name = helper.tensor_dtype_to_field(TensorProto.FLOAT) + print(f"The field name for {helper.tensor_dtype_to_string(TensorProto.FLOAT)} is {field_name}.") + tensor_dtype = helper.np_dtype_to_tensor_dtype(np_dtype) + print(f"The tensor data type for numpy dtype: {np_dtype} is {helper.tensor_dtype_to_string(tensor_dtype)}.") + + for tensor_dtype in helper.get_all_tensor_dtypes(): + print(helper.tensor_dtype_to_string(tensor_dtype)) +``` + +## Creating an ONNX Model Using Helper Functions + +```{eval-rst} +.. exec_code:: + + import onnx + from onnx import helper, TensorProto + + # Create inputs and output value info + X = helper.make_tensor_value_info("X", TensorProto.FLOAT, [3, 2]) + pads = helper.make_tensor_value_info("pads", TensorProto.INT64, [4]) + Y = helper.make_tensor_value_info("Y", TensorProto.FLOAT, [5, 4]) + + # Create Pad node + node_def = helper.make_node( + "Pad", + inputs=["X", "pads"], + outputs=["Y"], + mode="constant", + ) + + # Build graph and model + graph_def = helper.make_graph( + [node_def], + "test-model", + [X, pads], + [Y], + ) + model_def = helper.make_model( + graph_def, + producer_name="onnx-example", + opset_imports=[helper.make_opsetid("", 11)] + ) + + # Validate the model + onnx.checker.check_model(model_def) + print("Model is valid!") +``` + +## Conversion utilities for mapping attributes in ONNX IR + +```{eval-rst} +.. exec_code:: + + from onnx import TensorProto, helper + + np_dtype = helper.tensor_dtype_to_np_dtype(TensorProto.FLOAT) + print(f"The converted numpy dtype for {helper.tensor_dtype_to_string(TensorProto.FLOAT)} is {np_dtype}.") + + field_name = helper.tensor_dtype_to_field(TensorProto.FLOAT) + print(f"The field name for {helper.tensor_dtype_to_string(TensorProto.FLOAT)} is {field_name}.") + + # There are other useful conversion utilities. Please check onnx.helper +``` + +## Checking an ONNX Model + +```python +import onnx + +# Preprocessing: load the ONNX model +model_path = "path/to/the/model.onnx" +onnx_model = onnx.load(model_path) + +print(f"The model is:\n{onnx_model}") + +# Check the model +try: + onnx.checker.check_model(onnx_model) +except onnx.checker.ValidationError as e: + print(f"The model is invalid: {e}") +else: + print("The model is valid!") +``` + +### Checking a Large ONNX Model >2GB + +Current checker supports checking models with external data, but for those models larger than 2GB, please use the model path for onnx.checker and the external data needs to be under the same directory. + +```python +import onnx + +onnx.checker.check_model("path/to/the/model.onnx") +# onnx.checker.check_model(loaded_onnx_model) will fail if given >2GB model +``` + +## Running Shape Inference on an ONNX Model + +```{eval-rst} +.. exec_code:: + + import onnx + from onnx import helper, shape_inference + from onnx import TensorProto + + # Preprocessing: create a model with two nodes, Y's shape is unknown + node1 = helper.make_node("Transpose", ["X"], ["Y"], perm=[1, 0, 2]) + node2 = helper.make_node("Transpose", ["Y"], ["Z"], perm=[1, 0, 2]) + + graph = helper.make_graph( + [node1, node2], + "two-transposes", + [helper.make_tensor_value_info("X", TensorProto.FLOAT, (2, 3, 4))], + [helper.make_tensor_value_info("Z", TensorProto.FLOAT, (2, 3, 4))], + ) + + original_model = helper.make_model(graph, producer_name="onnx-examples") + + # Check the model and print Y's shape information + onnx.checker.check_model(original_model) + print(f"Before shape inference, the shape info of Y is:\n{original_model.graph.value_info}") + + # Apply shape inference on the model + inferred_model = shape_inference.infer_shapes(original_model) + + # Check the model and print Y's shape information + onnx.checker.check_model(inferred_model) + print(f"After shape inference, the shape info of Y is:\n{inferred_model.graph.value_info}") +``` + +### Shape inference a Large ONNX Model >2GB + +Current shape_inference supports models with external data, but for those models larger than 2GB, please use the model path for onnx.shape_inference.infer_shapes_path and the external data needs to be under the same directory. You can specify the output path for saving the inferred model; otherwise, the default output path is same as the original model path. + +```python +import onnx + +# output the inferred model to the original model path +onnx.shape_inference.infer_shapes_path("path/to/the/model.onnx") + +# output the inferred model to the specified model path +onnx.shape_inference.infer_shapes_path("path/to/the/model.onnx", "output/inferred/model.onnx") + +# inferred_model = onnx.shape_inference.infer_shapes(loaded_onnx_model) will fail if given >2GB model +``` + +## Running Type Inference on an ONNX Function + +```{eval-rst} +.. exec_code:: + + import onnx + import onnx.helper + import onnx.parser + import onnx.shape_inference + + function_text = """ + + CastTo (x) => (y) { + y = Cast (x) + } + """ + function = onnx.parser.parse_function(function_text) + + # The function above has one input-parameter x, and one attribute-parameter dtype. + # To apply type-and-shape-inference to this function, we must supply the type of + # input-parameter and an attribute value for the attribute-parameter as below: + + float_type_ = onnx.helper.make_tensor_type_proto(1, None) + dtype_6 = onnx.helper.make_attribute("dtype", 6) + result = onnx.shape_inference.infer_function_output_types( + function, [float_type_], [dtype_6] + ) + print(result) # a list containing the (single) output type +``` + +## Converting Version of an ONNX Model within Default Domain (""/"ai.onnx") + +```python +import onnx +from onnx import version_converter, helper + +# Preprocessing: load the model to be converted. +model_path = "path/to/the/model.onnx" +original_model = onnx.load(model_path) + +print(f"The model before conversion:\n{original_model}") + +# A full list of supported adapters can be found here: +# https://github.com/onnx/onnx/blob/main/onnx/version_converter.py#L21 +# Apply the version conversion on the original model +converted_model = version_converter.convert_version(original_model, ) + +print(f"The model after conversion:\n{converted_model}") +``` + +## Utility Functions + +### Extracting Sub-model with Inputs Outputs Tensor Names + +Function `extract_model()` extracts sub-model from an ONNX model. +The sub-model is defined by the names of the input and output tensors *exactly*. + +```python +import onnx + +input_path = "path/to/the/original/model.onnx" +output_path = "path/to/save/the/extracted/model.onnx" +input_names = ["input_0", "input_1", "input_2"] +output_names = ["output_0", "output_1"] + +onnx.utils.extract_model(input_path, output_path, input_names, output_names) +``` + +Note: For control-flow operators, e.g. If and Loop, the *boundary of sub-model*, +which is defined by the input and output tensors, should not *cut through* the +subgraph that is connected to the *main graph* as attributes of these operators. + +### ONNX Compose + +`onnx.compose` module provides tools to create combined models. + +`onnx.compose.merge_models` can be used to merge two models, by connecting some of the outputs +from the first model with inputs from the second model. By default, inputs/outputs not present in the +`io_map` argument will remain as inputs/outputs of the combined model. + +In this example we merge two models by connecting each output of the first model to an input in the second. The resulting model will have the same inputs as the first model and the same outputs as the second: + +```python +import onnx + +model1 = onnx.load("path/to/model1.onnx") +# agraph (float[N] A, float[N] B) => (float[N] C, float[N] D) +# { +# C = Add(A, B) +# D = Sub(A, B) +# } + +model2 = onnx.load("path/to/model2.onnx") +# agraph (float[N] X, float[N] Y) => (float[N] Z) +# { +# Z = Mul(X, Y) +# } + +combined_model = onnx.compose.merge_models( + model1, model2, + io_map=[("C", "X"), ("D", "Y")] +) +``` + +Additionally, a user can specify a list of `inputs`/`outputs` to be included in the combined model, +effectively dropping the part of the graph that does't contribute to the combined model outputs. +In the following example, we are connecting only one of the two outputs in the first model +to both inputs in the second. By specifying the outputs of the combined model explicitly, we are dropping the output not consumed from the first model, and the relevant part of the graph: + +```python +import onnx + +# Default case. Include all outputs in the combined model +combined_model = onnx.compose.merge_models( + model1, model2, + io_map=[("C", "X"), ("C", "Y")], +) # outputs: "D", "Z" + +# Explicit outputs. "Y" output and the Sub node are not present in the combined model +combined_model = onnx.compose.merge_models( + model1, model2, + io_map=[("C", "X"), ("C", "Y")], + outputs=["Z"], +) # outputs: "Z" +``` + +`onnx.compose.add_prefix` allows you to add a prefix to names in the model, to avoid a name collision +when merging them. By default, it renames all names in the graph: inputs, outputs, edges, nodes, +initializers, sparse initializers and value infos. + +```python +import onnx + +model = onnx.load("path/to/the/model.onnx") +# model - outputs: ["out0", "out1"], inputs: ["in0", "in1"] + +new_model = onnx.compose.add_prefix(model, prefix="m1/") +# new_model - outputs: ["m1/out0", "m1/out1"], inputs: ["m1/in0", "m1/in1"] + +# Can also be run in-place +onnx.compose.add_prefix(model, prefix="m1/", inplace=True) +``` + +`onnx.compose.expand_out_dim` can be used to connect models that expect a different number + of dimensions by inserting dimensions with extent one. This can be useful, when combining a + model producing samples with a model that works with batches of samples. + +```python +import onnx + +# outputs: "out0", shape=[200, 200, 3] +model1 = onnx.load("path/to/the/model1.onnx") + +# outputs: "in0", shape=[N, 200, 200, 3] +model2 = onnx.load("path/to/the/model2.onnx") + +# outputs: "out0", shape=[1, 200, 200, 3] +new_model1 = onnx.compose.expand_out_dims(model1, dim_idx=0) + +# Models can now be merged +combined_model = onnx.compose.merge_models( + new_model1, model2, io_map=[("out0", "in0")] +) + +# Can also be run in-place +onnx.compose.expand_out_dims(model1, dim_idx=0, inplace=True) +``` + +## Tools + +### Updating Model"s Inputs Outputs Dimension Sizes with Variable Length + +Function `update_inputs_outputs_dims` updates the dimension of the inputs and outputs of the model, +to the provided values in the parameter. You could provide both static and dynamic dimension size, +by using dim_param. For more information on static and dynamic dimension size, checkout [Tensor Shapes](IR.md#tensor-shapes). + +The function runs model checker after the input/output sizes are updated. + +```python +import onnx +from onnx.tools import update_model_dims + +model = onnx.load("path/to/the/model.onnx") +# Here both "seq", "batch" and -1 are dynamic using dim_param. +variable_length_model = update_model_dims.update_inputs_outputs_dims(model, {"input_name": ["seq", "batch", 3, -1]}, {"output_name": ["seq", "batch", 1, -1]}) +``` + +## ONNX Parser + +Functions `onnx.parser.parse_model` and `onnx.parser.parse_graph` can be used to create an ONNX model +or graph from a textual representation as shown below. See [Language Syntax](Syntax.md) for more details +about the language syntax. + +```{eval-rst} +.. exec_code:: + + import onnx + + input = """ + agraph (float[N, 128] X, float[128, 10] W, float[10] B) => (float[N, 10] C) + { + T = MatMul(X, W) + S = Add(T, B) + C = Softmax(S) + } + """ + graph = onnx.parser.parse_graph(input) + print(f"Graph name: {graph.name}") + print(f"Graph nodes: {[n.op_type for n in graph.node]}") + + input = """ + < + ir_version: 7, + opset_import: ["" : 10] + > + agraph (float[N, 128] X, float[128, 10] W, float[10] B) => (float[N, 10] C) + { + T = MatMul(X, W) + S = Add(T, B) + C = Softmax(S) + } + """ + model = onnx.parser.parse_model(input) + print(f"Model IR version: {model.ir_version}") + print(f"Model graph: {model.graph.name}") +``` + +## ONNX Inliner + +Functions `onnx.inliner.inline_local_functions` and `inline_selected_functions` can be used +to inline model-local functions in an ONNX model. In particular, `inline_local_functions` can +be used to produce a function-free model (suitable for backends that do not handle or support +functions). On the other hand, `inline_selected_functions` can be used to inline selected +functions. There is no support yet for inlining ONNX standard ops that are functions (also known +as schema-defined functions). + +```python +import onnx +import onnx.inliner + +model = onnx.load("path/to/the/model.onnx") +inlined = onnx.inliner.inline_local_functions(model) +onnx.save("path/to/the/inlinedmodel.onnx") +``` diff --git a/docs/ReleaseVerification.md b/docs/ReleaseVerification.md new file mode 100644 index 0000000..b37acfc --- /dev/null +++ b/docs/ReleaseVerification.md @@ -0,0 +1,39 @@ + + +# Verifying ONNX PyPI Releases with Sigstore Attestations + +ONNX PyPI releases include **Sigstore attestations** compliant with **PEP 740**, enabling cryptographic verification of **integrity, provenance, and publisher identity**. + +## Security Guarantees + +Verification confirms that: + +- the artifact **has not been modified**, +- it was **built and published by ONNX CI**, +- the signature is **publicly auditable** in Sigstore’s transparency log, +- the publisher identity matches **`onnx/onnx`**. + +## Verify a Release + +```bash +pip install pypi-attestations + +pypi-attestations verify pypi \ + --repository https://github.com/onnx/onnx \ + pypi:onnx-1.20.1-cp313-cp313t-win_amd64.whl +``` + +## References + +- PEP 740 – Digital Attestations for Python Packages + https://peps.python.org/pep-0740/ + +- Sigstore + https://www.sigstore.dev/ + +- PyPI Attestations + https://pypi.org/project/pypi-attestations/ diff --git a/docs/Relicensing.md b/docs/Relicensing.md new file mode 100644 index 0000000..09fa854 --- /dev/null +++ b/docs/Relicensing.md @@ -0,0 +1,15 @@ + + +# Relicensing MIT to Apache-2.0 + +The following copyright holders agree that all of their contributions originally submitted to this project under the MIT license are hereby relicensed to Apache-2.0, and are submitted pursuant to the Developer Certificate of Origin, version 1.1: + +Intel Corporation +Microsoft Corporation +NVIDIA Corporation +IBM Corporation +Facebook Inc. diff --git a/docs/ShapeInference.md b/docs/ShapeInference.md new file mode 100644 index 0000000..5720716 --- /dev/null +++ b/docs/ShapeInference.md @@ -0,0 +1,197 @@ + + +# ONNX Shape Inference + +ONNX provides an optional implementation of shape inference on ONNX +graphs. This implementation covers each of the core operators, as well +as provides an interface for extensibility. Therefore, you may choose +to invoke the existing shape inference functionality on your graphs, +or to define shape inference implementations to go along with your +custom operators (or both!). Shape inference functions are stored as a +member of the OpSchema objects. + +In ONNX 1.10 release, symbol generation and propagation along with shape +data propagation was added to ONNX graph level shape inference. +Detailed proposal is [here](proposals/0005-SymbolicShapeInfProposal.md) + +## Background + +Please see this [section](IR.md#static-tensor-shapes) of IR.md for a review of static tensor shapes. +In particular, a static tensor shape (represented by a `TensorShapeProto`) is distinct from +a runtime tensor shape. This feature is commonly used when the exact runtime tensor shape is +not known statically (that is, at compile time). + +* A `Tensor` with an undefined `shape` field is used to represent a tensor of unknown rank. +* A `Tensor` with a defined `shape` represents a tensor of known rank. +* Each `Dimension` of a `TensorShapeProto` can have a known integer value +(represented by the `dim_value` field) or it can have an unknown value +represented by a symbolic identified (the `dim_param` field) or it +may have neither field defined (in which case it represents an anonymous +unknown value). + +## Invoking Shape Inference + +Shape inference can be invoked either via C++ or Python. The Python +API is described, with example, +[here](PythonAPIOverview.md#running-shape-inference-on-an-onnx-model). + +The C++ API consists of a single function + +```cpp +shape_inference::InferShapes( + ModelProto& m, + const ISchemaRegistry* schema_registry); +``` + +The first argument is a `ModelProto` to perform shape inference on, +which is annotated in-place with shape information. The second +argument is optional. + +## Limitations + +Shape inference is not guaranteed to be complete. In particular, some +dynamic behaviors block the flow of shape inference, for example a +Reshape to a dynamically-provide shape. Also, all operators are not +required to have a shape inference implementation. + +Shape inference works only with constants and simple variables. It +does not support arithmetic expressions containing variables. For +example, `Concat` on tensors of shapes `(5, 2)` and `(7, 2)` can be +inferred to produce a result of shape `(12, 2)`, but `Concat` on +tensors of shapes `(5, 2)` and `(N, 2)` will simply produce `(M, 2)`, +rather than containing a representation of `N+5`. Note that differing +unknown symbolic values will be propagated, so the `M` here represents +an unknown quantity that is the same as other occurrences of `M`. + +These limitations are a property of the current implementation, not +fundamental constraints - if you are in need of something more +advanced, do let us know! + +## Type Inference vs. Shape Inference + +**Type inference** (determining the element type of outputs) is typically handled +automatically by the schema's type constraints. When a type constraint variable +(e.g., `"T"`) is shared between an input and an output in the schema definition, +the framework propagates the element type from the input to the output without +any explicit inference code. + +However, many existing ops still explicitly call `propagateElemTypeFromInputToOutput` +as a best practice for robustness. This is harmless when type constraints already +cover the case, and ensures correct behavior regardless of how shape inference +is invoked. + +Explicit type inference logic in `TypeAndShapeInferenceFunction` is only needed when: +- The output type is determined by an **attribute** rather than an input type + (e.g., `Cast`, where the `to` attribute specifies the output element type) +- The output type differs from all input types in a way that cannot be expressed + via shared type constraint variables +- The operator uses **heterogeneous** variadic inputs/outputs (see below) + +### Homogeneous vs. Heterogeneous variadic inputs/outputs + +The homogeneous/heterogeneous flag applies only to variadic (repeated) inputs or +outputs in the schema definition: + +- **Homogeneous** (the default): All repeated arguments must have the same type. + The type constraint variable constrains them to be identical, and the framework + enforces and propagates this automatically. +- **Heterogeneous**: Each repeated argument may have a distinct type. The type + constraint variable only describes the set of *allowed* types — it does not + constrain different arguments to have the same type. This is used by operators + like `Loop` and `Scan`, whose carried state variables can have mixed types. + +When using heterogeneous variadic arguments, the operator's +`TypeAndShapeInferenceFunction` must explicitly propagate types for each +individual argument, since the framework cannot do it automatically. + +**Shape inference**, on the other hand, almost always requires explicit logic, +since output shapes typically depend on input shapes, attributes, or both. + +## Implementing Shape Inference For Operators + +You can add a shape inference function to your operator's Schema with + +```cpp +OpSchema& Opschema::TypeAndShapeInferenceFunction(InferenceFunction inferenceFunction); +``` + +`InferenceFunction` is defined in +[shape_inference.h](/onnx/defs/shape_inference.h), along with the core +interface struct `InferenceContext` and an assortment of helper +methods. `InferenceContext` is the core struct which is provided to +your inference function. It allows accessing information about the +operator's inputs, and also allows writing out inferred information. + +To see numerous examples, search for occurrences of +`TypeAndShapeInferenceFunction` in the codebase. One that is +relatively involved is the implementation for `Concat`, in +onnx/defs/tensor/defs.cc. + +Please note the following points when implementing the shape-inference method for +operators to avoid common errors: + +* Before accessing the `shape` of any input, the code must check that +the shape is available. If unavailable, it should be treated as a dynamic +tensor whose rank is unknown and handled appropriately. Usually, the +shape-inference logic is guarded by a call to `hasInputShape` or +`hasNInputShapes`. + +* Before accessing the `dim_value` or `dim_param` of any dimension, the +code must check if these fields have a value. In particular, the code must +handle the possibility that the dimension may not have a statically +known value. + +There are several utility functions in [shape_inference.h](/onnx/defs/shape_inference.h) +to handle various common situations. + +* Use `checkInputRank` for inputs that must have a fixed rank. (See the +inference for `RoiAlign` as an example.) + +* `unifyInputDim` and `unifyDim` and `updateOutputShape` can be used +when multiple input dims are expected to be the same, and when input +dimensions are propagated to specific output dimensions. (See the inference +for `RoiAlign` for an example.) + +* `unifyInputShape` and `unifyInputShapePrefix` are higher-level utilities +built on `unifyInputDim`. They unify all dimensions (or a prefix of dimensions) +of an input in one call, enabling a more declarative style. `unifyInputDim` +remains useful for more complex scenarios where individual dimensions are +accessed selectively. + +* Overloaded operators `*` and `/` can be used on symbolic dimensions when output +dimensions are computed from input dimensions using arithmetic. (See the inference +for `SpaceToDepth` for an example.) + +These utilities handle missing shapes and dimensions safely. + +_Example_: Consider a simple matrix-multiplication op that expects inputs of shape +`[M,K]` and `[K,N]` and returns an output of shape `[M,N]`. This can be coded +up as below: + +```cpp + // Check that input 0 has rank 2 (if its rank is known). + checkInputRank(ctx, 0, 2); + // Check that input 1 has rank 2 (if its rank is known). + checkInputRank(ctx, 1, 2); + Dim M, K, N; + // Check various dimensions, handling missing dimensions/shapes safely. + unifyInputDim(ctx, 0, 0, M); + unifyInputDim(ctx, 0, 1, K); + unifyInputDim(ctx, 1, 0, K); + unifyInputDim(ctx, 1, 1, N); + updateOutputShape(ctx, 0, {M, N}); +``` + +The same example can be written more concisely using `unifyInputShape`, which +checks rank and unifies all dimensions in one call: + +```cpp + Dim M, K, N; + unifyInputShape(ctx, 0, {M, K}); + unifyInputShape(ctx, 1, {K, N}); + updateOutputShape(ctx, 0, {M, N}); +``` diff --git a/docs/Syntax.md b/docs/Syntax.md new file mode 100644 index 0000000..0dc8da5 --- /dev/null +++ b/docs/Syntax.md @@ -0,0 +1,97 @@ + + +# ONNX Textual Syntax + +## Overview + +This document describes a textual syntax for ONNX models, which is currently an experimental feature. +The syntax enables a compact and readable representation of ONNX models. It is motivated by a couple +of use-cases. One is to enable compact description of test-cases and its use in CI (both in the ONNX +repo as well as in other dependent repos such as ONNX-MLIR). The second is to help simplify the +definition of ONNX functions. Several of the existing function-definitions are verbose, and the +use of this syntax will lead to more compact, readable, and easier-to-maintain function definitions. +Efficient representation and efficient parsing of very large tensor-constants is *not* a goal. +Alternative methods should be used for that. + +## The API + +The key parser methods are the ```OnnxParser::Parse``` methods, used as below. + +```cpp + const char* code = R"ONNX( +< + ir_version: 7, + opset_import: [ "" : 10 ] +> +agraph (float[N, 128] X, float[128, 10] W, float[10] B) => (float[N, 10] C) +{ + T = MatMul(X, W) + S = Add(T, B) + C = Softmax(S) +} +)ONNX"; + + ModelProto model; + OnnxParser::Parse(model, code); + + checker::check_model(model); +``` + +See the [test-cases](../onnx/test/cpp/parser_test.cc) for more examples illustrating the API and syntax. + +## The Syntax + +The grammar below describes the syntax: + +```bnf + id-list ::= id (',' id)* + quotable-id-list ::= quotable-id (',' quotable-id)* + tensor-dim ::= '?' | quotable-id | int-constant + tensor-dims ::= tensor-dim (',' tensor-dim)* + tensor-type ::= prim-type | prim-type '[' ']' | prim-type '[' tensor-dims ']' + type ::= tensor-type | 'seq' '(' type ')' | 'map' '(' prim-type ',' type ')' + | 'optional' '(' type ')' | 'sparse_tensor' '(' tensor-type ')' + value-info ::= type quotable-id + value-infos ::= value-info (',' value-info)* + value-info-list ::= '(' value-infos? ') + id-or-value-info ::= type? quotable-id + id-or-value-infos ::= id-or-value-info (',' id-or-value-info)* + quoted-str :== '"' ([^"])* '"' + quotable-id :== id | quoted-str + str-str :== quoted-str ':' quoted-str + str-str-list :== '[' str-str (',' str-str)* ']' + internal-data ::= '{' prim-constants '}' + external-data ::= str-str-list + constant-data ::= internal-data | external-data + value-info-or-initializer ::= type quotable-id [ '=' constant-data ] + value-info-or-initializers ::= value-info-or-initializer (',' value-info-or-initializer)* + input-list ::= '(' value-info-or-initializers? ')' + output-list ::= '(' value-infos? ')' + initializer-list ::= '<' value-info-or-initializers? '>' + prim-constants ::= prim-constant (',' prim-constant)* + tensor-constant ::= tensor-type (quotable-id)? ('=')? '{' prim-constants '}' + attr-ref ::= '@' id + single-attr-value ::= tensor-constant | graph | prim-constant | attr-ref + attr-value-list ::= '[' single-attr-value (',' single-attr-value)* ']' + attr-value ::= single-attr-value | attr-value-list + attr-type ::= ':' id + attr ::= id attr-type? '=' attr-value + attr-list ::= '<' attr (',' attr)* '>' + node-label ::= '[' quotable-id ']' + node ::= node-label? quotable-id-list? '=' qualified-id attr-list? '(' quotable-id-list? ')' + | node-label? quotable-id-list? '=' qualified-id '(' quotable-id-list? ')' attr-list + node-list ::= '{' node* '}' + graph ::= quotable-id input-list '=>' output-list initializer-list node-list + other-data ::= id ':' value + other-data-list ::= '<' other-data (',' other-data)* '>' + fun-attr-list ::= '<' id | attr (',' id | attr)* '>' + fun-input-list ::= '(' id-or-value-infos ')' + fun-output-list ::= '(' id-or-value-infos ')' + fun-value-infos ::= ( '<' value-infos '>' )? + function ::= other-data-list? id fun-attr-list? quotable-id fun-input-list '=>' fun-output-list fun-value-infos node-list + model ::= other-data-list? graph function* +``` diff --git a/docs/TestCoverage-ml.md b/docs/TestCoverage-ml.md new file mode 100644 index 0000000..89b6771 --- /dev/null +++ b/docs/TestCoverage-ml.md @@ -0,0 +1,350 @@ + +# Test Coverage Report (ONNX-ML Operators) +## Outlines +* [Node Test Coverage](#node-test-coverage) +* [Model Test Coverage](#model-test-coverage) +* [Overall Test Coverage](#overall-test-coverage) +# Node Test Coverage +## Summary +Node tests have covered 4/19 (21.05%, 0 generators excluded) common operators. + +Node tests have covered 0/0 (N/A) experimental operators. + +* [Covered Common Operators](#covered-common-operators) +* [No Cover Common Operators](#no-cover-common-operators) +* [Covered Experimental Operators](#covered-experimental-operators) +* [No Cover Experimental Operators](#no-cover-experimental-operators) + +## 💚Covered Common Operators +### ArrayFeatureExtractor +There are 1 test cases, listed as following: +
+arrayfeatureextractor + +```python +node = onnx.helper.make_node( + "ArrayFeatureExtractor", + inputs=["x", "y"], + outputs=["z"], + domain="ai.onnx.ml", +) + +x = np.arange(12).reshape((3, 4)).astype(np.float32) +y = np.array([0, 1], dtype=np.int64) +z = np.array([[0, 4, 8], [1, 5, 9]], dtype=np.float32).T +expect( + node, + inputs=[x, y], + outputs=[z], + name="test_ai_onnx_ml_array_feature_extractor", +) +``` + +
+ + +### Binarizer +There are 1 test cases, listed as following: +
+binarizer + +```python +threshold = 1.0 +node = onnx.helper.make_node( + "Binarizer", + inputs=["X"], + outputs=["Y"], + threshold=threshold, + domain="ai.onnx.ml", +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = compute_binarizer(x, threshold)[0] + +expect(node, inputs=[x], outputs=[y], name="test_ai_onnx_ml_binarizer") +``` + +
+ + +### LabelEncoder +There are 2 test cases, listed as following: +
+string_int_label_encoder + +```python +node = onnx.helper.make_node( + "LabelEncoder", + inputs=["X"], + outputs=["Y"], + domain="ai.onnx.ml", + keys_strings=["a", "b", "c"], + values_int64s=[0, 1, 2], + default_int64=42, +) +x = np.array(["a", "b", "d", "c", "g"]).astype(object) +y = np.array([0, 1, 42, 2, 42]).astype(np.int64) +expect( + node, + inputs=[x], + outputs=[y], + name="test_ai_onnx_ml_label_encoder_string_int", +) + +node = onnx.helper.make_node( + "LabelEncoder", + inputs=["X"], + outputs=["Y"], + domain="ai.onnx.ml", + keys_strings=["a", "b", "c"], + values_int64s=[0, 1, 2], +) +x = np.array(["a", "b", "d", "c", "g"]).astype(object) +y = np.array([0, 1, -1, 2, -1]).astype(np.int64) +expect( + node, + inputs=[x], + outputs=[y], + name="test_ai_onnx_ml_label_encoder_string_int_no_default", +) +``` + +
+
+tensor_based_label_encoder + +```python +tensor_keys = make_tensor( + "keys_tensor", onnx.TensorProto.STRING, (3,), ["a", "b", "c"] +) +repeated_string_keys = ["a", "b", "c"] +x = np.array(["a", "b", "d", "c", "g"]).astype(object) +y = np.array([0, 1, 42, 2, 42]).astype(np.int16) + +node = onnx.helper.make_node( + "LabelEncoder", + inputs=["X"], + outputs=["Y"], + domain="ai.onnx.ml", + keys_tensor=tensor_keys, + values_tensor=make_tensor( + "values_tensor", onnx.TensorProto.INT16, (3,), [0, 1, 2] + ), + default_tensor=make_tensor( + "default_tensor", onnx.TensorProto.INT16, (1,), [42] + ), +) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_ai_onnx_ml_label_encoder_tensor_mapping", +) + +node = onnx.helper.make_node( + "LabelEncoder", + inputs=["X"], + outputs=["Y"], + domain="ai.onnx.ml", + keys_strings=repeated_string_keys, + values_tensor=make_tensor( + "values_tensor", onnx.TensorProto.INT16, (3,), [0, 1, 2] + ), + default_tensor=make_tensor( + "default_tensor", onnx.TensorProto.INT16, (1,), [42] + ), +) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_ai_onnx_ml_label_encoder_tensor_value_only_mapping", +) +``` + +
+ + +### TreeEnsemble +There are 2 test cases, listed as following: +
+tree_ensemble_set_membership + +```python +node = onnx.helper.make_node( + "TreeEnsemble", + ["X"], + ["Y"], + domain="ai.onnx.ml", + n_targets=4, + aggregate_function=1, + membership_values=make_tensor( + "membership_values", + onnx.TensorProto.FLOAT, + (8,), + [1.2, 3.7, 8, 9, np.nan, 12, 7, np.nan], + ), + nodes_missing_value_tracks_true=None, + nodes_hitrates=None, + post_transform=0, + tree_roots=[0], + nodes_modes=make_tensor( + "nodes_modes", + onnx.TensorProto.UINT8, + (3,), + np.array([0, 6, 6], dtype=np.uint8), + ), + nodes_featureids=[0, 0, 0], + nodes_splits=make_tensor( + "nodes_splits", + onnx.TensorProto.FLOAT, + (3,), + np.array([11, 232344.0, np.nan], dtype=np.float32), + ), + nodes_trueleafs=[0, 1, 1], + nodes_truenodeids=[1, 0, 1], + nodes_falseleafs=[1, 0, 1], + nodes_falsenodeids=[2, 2, 3], + leaf_targetids=[0, 1, 2, 3], + leaf_weights=make_tensor( + "leaf_weights", onnx.TensorProto.FLOAT, (4,), [1, 10, 1000, 100] + ), +) + +x = np.array([1.2, 3.4, -0.12, np.nan, 12, 7], np.float32).reshape(-1, 1) +expected = np.array( + [ + [1, 0, 0, 0], + [0, 0, 0, 100], + [0, 0, 0, 100], + [0, 0, 1000, 0], + [0, 0, 1000, 0], + [0, 10, 0, 0], + ], + dtype=np.float32, +) +expect( + node, + inputs=[x], + outputs=[expected], + name="test_ai_onnx_ml_tree_ensemble_set_membership", +) +``` + +
+
+tree_ensemble_single_tree + +```python +node = onnx.helper.make_node( + "TreeEnsemble", + ["X"], + ["Y"], + domain="ai.onnx.ml", + n_targets=2, + membership_values=None, + nodes_missing_value_tracks_true=None, + nodes_hitrates=None, + aggregate_function=1, + post_transform=0, + tree_roots=[0], + nodes_modes=make_tensor( + "nodes_modes", + onnx.TensorProto.UINT8, + (3,), + np.array([0, 0, 0], dtype=np.uint8), + ), + nodes_featureids=[0, 0, 0], + nodes_splits=make_tensor( + "nodes_splits", + onnx.TensorProto.DOUBLE, + (3,), + np.array([3.14, 1.2, 4.2], dtype=np.float64), + ), + nodes_truenodeids=[1, 0, 1], + nodes_trueleafs=[0, 1, 1], + nodes_falsenodeids=[2, 2, 3], + nodes_falseleafs=[0, 1, 1], + leaf_targetids=[0, 1, 0, 1], + leaf_weights=make_tensor( + "leaf_weights", + onnx.TensorProto.DOUBLE, + (4,), + np.array([5.23, 12.12, -12.23, 7.21], dtype=np.float64), + ), +) + +x = np.array([1.2, 3.4, -0.12, 1.66, 4.14, 1.77], np.float64).reshape(3, 2) +y = np.array([[5.23, 0], [5.23, 0], [0, 12.12]], dtype=np.float64) +expect( + node, + inputs=[x], + outputs=[y], + name="test_ai_onnx_ml_tree_ensemble_single_tree", +) +``` + +
+ + +
+ +## 💔No Cover Common Operators +### CastMap (call for test cases) + + +### CategoryMapper (call for test cases) + + +### DictVectorizer (call for test cases) + + +### FeatureVectorizer (call for test cases) + + +### Imputer (call for test cases) + + +### LinearClassifier (call for test cases) + + +### LinearRegressor (call for test cases) + + +### Normalizer (call for test cases) + + +### OneHotEncoder (call for test cases) + + +### SVMClassifier (call for test cases) + + +### SVMRegressor (call for test cases) + + +### Scaler (call for test cases) + + +### TreeEnsembleClassifier (call for test cases) + + +### TreeEnsembleRegressor (call for test cases) + + +### ZipMap (call for test cases) + + +
+ +## 💚Covered Experimental Operators +
+ +## 💔No Cover Experimental Operators +
+ +# Model Test Coverage +No model tests present for selected domain +# Overall Test Coverage +## To be filled. diff --git a/docs/TestCoverage.md b/docs/TestCoverage.md new file mode 100644 index 0000000..2931494 --- /dev/null +++ b/docs/TestCoverage.md @@ -0,0 +1,32001 @@ + +# Test Coverage Report (ONNX Core Operators) +## Outlines +* [Node Test Coverage](#node-test-coverage) +* [Model Test Coverage](#model-test-coverage) +* [Overall Test Coverage](#overall-test-coverage) +# Node Test Coverage +## Summary +Node tests have covered 189/201 (94.03%, 5 generators excluded) common operators. + +Node tests have covered 1/1 (100.00%, 0 generators excluded) experimental operators. + +* [Covered Common Operators](#covered-common-operators) +* [No Cover Common Operators](#no-cover-common-operators) +* [Covered Experimental Operators](#covered-experimental-operators) +* [No Cover Experimental Operators](#no-cover-experimental-operators) + +## 💚Covered Common Operators +### Abs +There are 1 test cases, listed as following: +
+abs + +```python +node = onnx.helper.make_node( + "Abs", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.abs(x) + +expect(node, inputs=[x], outputs=[y], name="test_abs") +``` + +
+ + +### Acos +There are 1 test cases, listed as following: +
+acos + +```python +node = onnx.helper.make_node( + "Acos", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-0.5, 0, 0.5]).astype(np.float32) +y = np.arccos(x) +expect(node, inputs=[x], outputs=[y], name="test_acos_example") + +x = np.random.rand(3, 4, 5).astype(np.float32) +y = np.arccos(x) +expect(node, inputs=[x], outputs=[y], name="test_acos") +``` + +
+ + +### Acosh +There are 1 test cases, listed as following: +
+acosh + +```python +node = onnx.helper.make_node( + "Acosh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([10, np.e, 1]).astype(np.float32) +y = np.arccosh(x) # expected output [2.99322295, 1.65745449, 0.] +expect(node, inputs=[x], outputs=[y], name="test_acosh_example") + +x = np.random.uniform(1.0, 10.0, (3, 4, 5)).astype(np.float32) +y = np.arccosh(x) +expect(node, inputs=[x], outputs=[y], name="test_acosh") +``` + +
+ + +### Adagrad +There are 2 test cases, listed as following: +
+adagrad + +```python +# Define operator attributes. +norm_coefficient = 0.001 +epsilon = 1e-5 +decay_factor = 0.1 + +# Create operator. +node = onnx.helper.make_node( + "Adagrad", + inputs=["R", "T", "X", "G", "H"], + outputs=["X_new", "H_new"], + norm_coefficient=norm_coefficient, + epsilon=epsilon, + decay_factor=decay_factor, + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar +x = np.array([1.0], dtype=np.float32) +g = np.array([-1.0], dtype=np.float32) +h = np.array([2.0], dtype=np.float32) + +# Compute expected outputs of Adagrad. +x_new, h_new = apply_adagrad( + r, t, x, g, h, norm_coefficient, epsilon, decay_factor +) + +# Check results. +expect( + node, + inputs=[r, t, x, g, h], + outputs=[x_new, h_new], + name="test_adagrad", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+
+adagrad_multiple + +```python +# Define operator attributes. +norm_coefficient = 0.001 +epsilon = 1e-5 +decay_factor = 0.1 + +node = onnx.helper.make_node( + "Adagrad", + inputs=["R", "T", "X1", "X2", "G1", "G2", "H1", "H2"], + outputs=["X1_new", "X2_new", "H1_new", "H2_new"], + norm_coefficient=norm_coefficient, + epsilon=epsilon, + decay_factor=decay_factor, + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar + +x1 = np.array([1.0], dtype=np.float32) +g1 = np.array([-1.0], dtype=np.float32) +h1 = np.array([2.0], dtype=np.float32) + +x2 = np.array([1.0, 2.0], dtype=np.float32) +g2 = np.array([-1.0, -3.0], dtype=np.float32) +h2 = np.array([4.0, 1.0], dtype=np.float32) + +# Compute expected outputs of Adagrad. +x1_new, h1_new = apply_adagrad( + r, t, x1, g1, h1, norm_coefficient, epsilon, decay_factor +) +x2_new, h2_new = apply_adagrad( + r, t, x2, g2, h2, norm_coefficient, epsilon, decay_factor +) + +# Check results. +expect( + node, + inputs=[r, t, x1, x2, g1, g2, h1, h2], + outputs=[x1_new, x2_new, h1_new, h2_new], + name="test_adagrad_multiple", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+ + +### Adam +There are 2 test cases, listed as following: +
+adam + +```python +# Define operator attributes. +norm_coefficient = 0.001 +alpha = 0.95 +beta = 0.1 +epsilon = 1e-7 + +# Create operator. +node = onnx.helper.make_node( + "Adam", + inputs=["R", "T", "X", "G", "V", "H"], + outputs=["X_new", "V_new", "H_new"], + norm_coefficient=norm_coefficient, + alpha=alpha, + beta=beta, + epsilon=epsilon, + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar +x = np.array([1.2, 2.8], dtype=np.float32) +g = np.array([-0.94, -2.5], dtype=np.float32) +v = np.array([1.7, 3.6], dtype=np.float32) +h = np.array([0.1, 0.1], dtype=np.float32) + +# Compute expected outputs of Adam. +x_new, v_new, h_new = apply_adam( + r, t, x, g, v, h, norm_coefficient, 0.0, alpha, beta, epsilon +) + +# Check results. +expect( + node, + inputs=[r, t, x, g, v, h], + outputs=[x_new, v_new, h_new], + name="test_adam", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+
+adam_multiple + +```python +# Define operator attributes. +norm_coefficient = 0.001 +alpha = 0.95 +beta = 0.85 +epsilon = 1e-2 + +node = onnx.helper.make_node( + "Adam", + inputs=["R", "T", "X1", "X2", "G1", "G2", "V1", "V2", "H1", "H2"], + outputs=["X1_new", "X2_new", "V1_new", "V2_new", "H1_new", "H2_new"], + norm_coefficient=norm_coefficient, + alpha=alpha, + beta=beta, + epsilon=epsilon, + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar + +x1 = np.array([1.0], dtype=np.float32) +g1 = np.array([-1.0], dtype=np.float32) +v1 = np.array([2.0], dtype=np.float32) +h1 = np.array([0.5], dtype=np.float32) + +x2 = np.array([1.0, 2.0], dtype=np.float32) +g2 = np.array([-1.0, -3.0], dtype=np.float32) +v2 = np.array([4.0, 1.0], dtype=np.float32) +h2 = np.array([1.0, 10.0], dtype=np.float32) + +# Compute expected outputs of Adam. +x1_new, v1_new, h1_new = apply_adam( + r, t, x1, g1, v1, h1, norm_coefficient, 0.0, alpha, beta, epsilon +) +x2_new, v2_new, h2_new = apply_adam( + r, t, x2, g2, v2, h2, norm_coefficient, 0.0, alpha, beta, epsilon +) + +# Check results. +expect( + node, + inputs=[r, t, x1, x2, g1, g2, v1, v2, h1, h2], + outputs=[x1_new, x2_new, v1_new, v2_new, h1_new, h2_new], + name="test_adam_multiple", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+ + +### Add +There are 2 test cases, listed as following: +
+add + +```python +node = onnx.helper.make_node( + "Add", + inputs=["x", "y"], + outputs=["sum"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.int8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int8) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_int8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.int16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int16) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_uint64") +``` + +
+
+add_broadcast + +```python +node = onnx.helper.make_node( + "Add", + inputs=["x", "y"], + outputs=["sum"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +expect(node, inputs=[x, y], outputs=[x + y], name="test_add_bcast") +``` + +
+ + +### AffineGrid +There are 2 test cases, listed as following: +
+2d_no_reference_evaluator + +```python +theta_2d = create_theta_2d() +N, C, H, W = len(theta_2d), 3, 5, 6 +data_size = (H, W) +for align_corners in (0, 1): + node = onnx.helper.make_node( + "AffineGrid", + inputs=["theta", "size"], + outputs=["grid"], + align_corners=align_corners, + ) + + original_grid = construct_original_grid(data_size, align_corners) + grid = apply_affine_transform(theta_2d, original_grid) + + test_name = "test_affine_grid_2d" + if align_corners == 1: + test_name += "_align_corners" + expect( + node, + inputs=[theta_2d, np.array([N, C, H, W], dtype=np.int64)], + outputs=[grid], + name=test_name, + ) +``` + +
+
+3d_no_reference_evaluator + +```python +theta_3d = create_theta_3d() +N, C, D, H, W = len(theta_3d), 3, 4, 5, 6 +data_size = (D, H, W) +for align_corners in (0, 1): + node = onnx.helper.make_node( + "AffineGrid", + inputs=["theta", "size"], + outputs=["grid"], + align_corners=align_corners, + ) + + original_grid = construct_original_grid(data_size, align_corners) + grid = apply_affine_transform(theta_3d, original_grid) + + test_name = "test_affine_grid_3d" + if align_corners == 1: + test_name += "_align_corners" + expect( + node, + inputs=[theta_3d, np.array([N, C, D, H, W], dtype=np.int64)], + outputs=[grid], + name=test_name, + ) +``` + +
+ + +### And +There are 2 test cases, listed as following: +
+and + +```python +node = onnx.helper.make_node( + "And", + inputs=["x", "y"], + outputs=["and"], +) + +# 2d +x = (np.random.randn(3, 4) > 0).astype(bool) +y = (np.random.randn(3, 4) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and2d") + +# 3d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(3, 4, 5) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and3d") + +# 4d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and4d") +``` + +
+
+and_broadcast + +```python +node = onnx.helper.make_node( + "And", + inputs=["x", "y"], + outputs=["and"], +) + +# 3d vs 1d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(5) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and_bcast3v1d") + +# 3d vs 2d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(4, 5) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and_bcast3v2d") + +# 4d vs 2d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(5, 6) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and_bcast4v2d") + +# 4d vs 3d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(4, 5, 6) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and_bcast4v3d") + +# 4d vs 4d +x = (np.random.randn(1, 4, 1, 6) > 0).astype(bool) +y = (np.random.randn(3, 1, 5, 6) > 0).astype(bool) +z = np.logical_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_and_bcast4v4d") +``` + +
+ + +### ArgMax +There are 8 test cases, listed as following: +
+default_axes_keepdims + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", inputs=["data"], outputs=["result"], keepdims=keepdims +) + +# result: [[1, 1]] +result = argmax_use_numpy(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_default_axis_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [1, 3, 4] +result = argmax_use_numpy(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_default_axis_random", +) +``` + +
+
+default_axes_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", + inputs=["data"], + outputs=["result"], + keepdims=keepdims, + select_last_index=True, +) + +# result: [[1, 1]] +result = argmax_use_numpy_select_last_index(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_default_axis_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [1, 3, 4] +result = argmax_use_numpy_select_last_index(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_default_axis_random_select_last_index", +) +``` + +
+
+keepdims + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# result: [[0], [1]] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmax_keepdims_example" +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 1, 4] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmax_keepdims_random" +) +``` + +
+
+keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [[1], [1]] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 1, 4] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_keepdims_random_select_last_index", +) +``` + +
+
+negative_axis_keepdims + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = -1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# result: [[0], [1]] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_negative_axis_keepdims_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 3, 1] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_negative_axis_keepdims_random", +) +``` + +
+
+negative_axis_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = -1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMax", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [[1], [1]] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_negative_axis_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 3, 1] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_negative_axis_keepdims_random_select_last_index", +) +``` + +
+
+no_keepdims + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 0 +node = onnx.helper.make_node( + "ArgMax", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# result: [0, 1] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_no_keepdims_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 4] +result = argmax_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmax_no_keepdims_random" +) +``` + +
+
+no_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 0 +node = onnx.helper.make_node( + "ArgMax", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [1, 1] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_no_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 4] +result = argmax_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmax_no_keepdims_random_select_last_index", +) +``` + +
+ + +### ArgMin +There are 8 test cases, listed as following: +
+default_axes_keepdims + +```python +data = np.array([[2, 1], [3, 10]], dtype=np.float32) +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", inputs=["data"], outputs=["result"], keepdims=keepdims +) + +# The content of result is : [[0], [0]] +result = argmin_use_numpy(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_default_axis_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [1, 3, 4] +result = argmin_use_numpy(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_default_axis_random", +) +``` + +
+
+default_axes_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", + inputs=["data"], + outputs=["result"], + keepdims=keepdims, + select_last_index=True, +) + +# result: [[0, 0]] +result = argmin_use_numpy_select_last_index(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_default_axis_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [1, 3, 4] +result = argmin_use_numpy_select_last_index(data, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_default_axis_random_select_last_index", +) +``` + +
+
+keepdims + +```python +data = np.array([[2, 1], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# The content of result is : [[1], [0]] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmin_keepdims_example" +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 1, 4] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmin_keepdims_random" +) +``` + +
+
+keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [[1], [0]] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 1, 4] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_keepdims_random_select_last_index", +) +``` + +
+
+negative_axis_keepdims + +```python +data = np.array([[2, 1], [3, 10]], dtype=np.float32) +axis = -1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# The content of result is : [[1], [0]] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_negative_axis_keepdims_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 3, 1] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_negative_axis_keepdims_random", +) +``` + +
+
+negative_axis_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = -1 +keepdims = 1 +node = onnx.helper.make_node( + "ArgMin", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [[1], [0]] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_negative_axis_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 3, 1] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_negative_axis_keepdims_random_select_last_index", +) +``` + +
+
+no_keepdims + +```python +data = np.array([[2, 1], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 0 +node = onnx.helper.make_node( + "ArgMin", inputs=["data"], outputs=["result"], axis=axis, keepdims=keepdims +) +# The content of result is : [[1, 0]] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_no_keepdims_example", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 4] +result = argmin_use_numpy(data, axis=axis, keepdims=keepdims) +expect( + node, inputs=[data], outputs=[result], name="test_argmin_no_keepdims_random" +) +``` + +
+
+no_keepdims_select_last_index + +```python +data = np.array([[2, 2], [3, 10]], dtype=np.float32) +axis = 1 +keepdims = 0 +node = onnx.helper.make_node( + "ArgMin", + inputs=["data"], + outputs=["result"], + axis=axis, + keepdims=keepdims, + select_last_index=True, +) +# result: [[1, 0]] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_no_keepdims_example_select_last_index", +) + +data = np.random.uniform(-10, 10, [2, 3, 4]).astype(np.float32) +# result's shape: [2, 4] +result = argmin_use_numpy_select_last_index(data, axis=axis, keepdims=keepdims) +expect( + node, + inputs=[data], + outputs=[result], + name="test_argmin_no_keepdims_random_select_last_index", +) +``` + +
+ + +### Asin +There are 1 test cases, listed as following: +
+asin + +```python +node = onnx.helper.make_node( + "Asin", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-0.5, 0, 0.5]).astype(np.float32) +y = np.arcsin(x) +expect(node, inputs=[x], outputs=[y], name="test_asin_example") + +x = np.random.rand(3, 4, 5).astype(np.float32) +y = np.arcsin(x) +expect(node, inputs=[x], outputs=[y], name="test_asin") +``` + +
+ + +### Asinh +There are 1 test cases, listed as following: +
+asinh + +```python +node = onnx.helper.make_node( + "Asinh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.arcsinh(x) # expected output [-0.88137358, 0., 0.88137358] +expect(node, inputs=[x], outputs=[y], name="test_asinh_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.arcsinh(x) +expect(node, inputs=[x], outputs=[y], name="test_asinh") +``` + +
+ + +### Atan +There are 1 test cases, listed as following: +
+atan + +```python +node = onnx.helper.make_node( + "Atan", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.arctan(x) +expect(node, inputs=[x], outputs=[y], name="test_atan_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.arctan(x) +expect(node, inputs=[x], outputs=[y], name="test_atan") +``` + +
+ + +### Atanh +There are 1 test cases, listed as following: +
+atanh + +```python +node = onnx.helper.make_node( + "Atanh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-0.5, 0, 0.5]).astype(np.float32) +y = np.arctanh(x) # expected output [-0.54930615, 0., 0.54930615] +expect(node, inputs=[x], outputs=[y], name="test_atanh_example") + +x = np.random.uniform(0.0, 1.0, (3, 4, 5)).astype(np.float32) +y = np.arctanh(x) +expect(node, inputs=[x], outputs=[y], name="test_atanh") +``` + +
+ + +### Attention +There are 76 test cases, listed as following: +
+attention + +```python +node = onnx.helper.make_node("Attention", inputs=["Q", "K", "V"], outputs=["Y"]) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_23_boolmask_fullymasked_row_nan_robustness + +```python +"""Opset-23 fully-masked boolean ``attn_mask`` row -> zero (not ``NaN``). + +This locks the opset-23 / ``old.cc`` function-body fully-masked-row guard +against future regressions. In opset 23 the only in-contract fully-masked +row comes from an all-``False`` boolean ``attn_mask`` row (``is_causal`` is +not set here): every key for that query is disallowed, so ``softmax`` over an +all-``-inf`` bias row is ``NaN``. The guard zeros that row's probabilities +before the ``P @ V`` contraction so the output row is exactly ``0``, while +rows with at least one allowed key are unchanged. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(4) +B, H, S, D = 1, 2, 2, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(B, H, S, D).astype(np.float32) +K = np.random.rand(B, H, S, D).astype(np.float32) +V = np.random.rand(B, H, S, D).astype(np.float32) +# Row 0: no key allowed -> fully masked (Bug-2 empty row). Row 1: both keys +# allowed -> finite, unchanged by the guard. +attn_mask = np.array([[False, False], [True, True]], dtype=np.bool_) + +Y, _, _, _ = _compute_attention(Q, K, V, attn_mask=attn_mask) + +# Fully-masked row 0 is exactly zero (not NaN); every other cell is finite. +assert np.all(np.isfinite(Y)), "non-masked rows must be finite" +assert np.array_equal(Y[:, :, 0, :], np.zeros_like(Y[:, :, 0, :])), ( + "fully-masked row must be zero (Bug-2)" +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_23_boolmask_fullymasked_row_nan_robustness", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_23_fullymasked_qk_matmul_output_mode3_zero + +```python +"""Opset-23 ``qk_matmul_output_mode=3`` fully-masked row is a zero row. + +Mode ``3`` exposes the post-softmax matrix as the optional +``qk_matmul_output``. For a fully-masked query row (all-``False`` boolean +``attn_mask`` row), the fully-masked-row guard is applied before this output +is produced, so the mode-3 row is zeroed, consistent with the primary output +``Y`` row (both are ``0``). This pins the mandated agreement between the +guarded primary output and the mode-3 output at opset 23. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(13) +B, H, S, D = 1, 2, 2, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + qk_matmul_output_mode=3, +) + +Q = np.random.rand(B, H, S, D).astype(np.float32) +K = np.random.rand(B, H, S, D).astype(np.float32) +V = np.random.rand(B, H, S, D).astype(np.float32) +# Row 0: no key allowed -> fully masked. Row 1: both keys allowed -> finite. +attn_mask = np.array([[False, False], [True, True]], dtype=np.bool_) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + qk_matmul_output_mode=3, +) + +# Primary output row 0 and the mode-3 row 0 are both guarded to zero. +assert np.array_equal(Y[:, :, 0, :], np.zeros_like(Y[:, :, 0, :])), ( + "fully-masked primary output row must be zero" +) +assert np.array_equal( + qk_matmul_output[:, :, 0, :], np.zeros_like(qk_matmul_output[:, :, 0, :]) +), "mode-3 output row for a fully-masked query must be zero (consistent with Y)" +assert np.all(np.isfinite(qk_matmul_output)), ( + "all mode-3 rows are finite (the fully-masked row is guarded to 0.0)" +) +assert np.all(np.isfinite(Y)), ( + "all Y rows are finite (the fully-masked row is guarded to 0.0)" +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_23_fullymasked_qk_matmul_output_mode3_zero", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_24_fullymasked_qk_matmul_output_mode3_zero + +```python +"""Opset-24 ``qk_matmul_output_mode=3`` fully-masked row is a zero row. + +The opset-24 twin of +``export_attention_23_fullymasked_qk_matmul_output_mode3_zero``. Mode ``3`` +exposes the post-softmax matrix as the optional ``qk_matmul_output``. For a +fully-masked query row (all-``False`` boolean ``attn_mask`` row), the +fully-masked-row guard is applied before this output is produced, so the +mode-3 row is zeroed, consistent with the primary output ``Y`` row (both are +``0``). This pins the mandated agreement between the guarded primary output +and the mode-3 output at opset 24. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(13) +B, H, S, D = 1, 2, 2, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + qk_matmul_output_mode=3, +) + +Q = np.random.rand(B, H, S, D).astype(np.float32) +K = np.random.rand(B, H, S, D).astype(np.float32) +V = np.random.rand(B, H, S, D).astype(np.float32) +# Row 0: no key allowed -> fully masked. Row 1: both keys allowed -> finite. +attn_mask = np.array([[False, False], [True, True]], dtype=np.bool_) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + qk_matmul_output_mode=3, +) + +# Primary output row 0 and the mode-3 row 0 are both guarded to zero. +assert np.array_equal(Y[:, :, 0, :], np.zeros_like(Y[:, :, 0, :])), ( + "fully-masked primary output row must be zero" +) +assert np.array_equal( + qk_matmul_output[:, :, 0, :], np.zeros_like(qk_matmul_output[:, :, 0, :]) +), "mode-3 output row for a fully-masked query must be zero (consistent with Y)" +assert np.all(np.isfinite(qk_matmul_output)), ( + "all mode-3 rows are finite (the fully-masked row is guarded to 0.0)" +) +assert np.all(np.isfinite(Y)), ( + "all Y rows are finite (the fully-masked row is guarded to 0.0)" +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_24_fullymasked_qk_matmul_output_mode3_zero", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_24_qk_matmul_output_mode3_softmax_precision + +```python +"""Mode-3 ``qk_matmul_output`` is emitted at the output precision ``T1``. + +``qk_matmul_output_mode=3`` exposes the post-softmax probabilities. When +``softmax_precision`` differs from the operator's output type ``T1`` (here +``T1 = float16`` with softmax computed in ``float32``), the mode-3 output is +cast back to ``T1`` -- matching the reference implementation, which casts the +exposed matrix to ``Q.dtype``. This locks both the dtype contract and the +fully-masked-row zeroing under a non-default ``softmax_precision``. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(17) +B, H, S, D = 1, 2, 2, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + qk_matmul_output_mode=3, + softmax_precision=int(onnx.TensorProto.FLOAT), +) + +# T1 = float16; softmax runs in float32, so the mode-3 output is cast back to +# float16 on emission. +Q = np.random.rand(B, H, S, D).astype(np.float16) +K = np.random.rand(B, H, S, D).astype(np.float16) +V = np.random.rand(B, H, S, D).astype(np.float16) +# Row 0: fully masked. Row 1: both keys allowed -> finite. +attn_mask = np.array([[False, False], [True, True]], dtype=np.bool_) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + qk_matmul_output_mode=3, + softmax_precision=int(onnx.TensorProto.FLOAT), +) + +# The mode-3 output is emitted at T1 (float16), not the float32 softmax +# precision, matching the operator's output type. +assert qk_matmul_output.dtype == np.float16, ( + "mode-3 qk_matmul_output must be emitted at the output precision T1 (float16)" +) +# The fully-masked row is still guarded to zero, consistent with Y. +assert np.array_equal( + qk_matmul_output[:, :, 0, :], np.zeros_like(qk_matmul_output[:, :, 0, :]) +), "mode-3 output row for a fully-masked query must be zero (consistent with Y)" +assert np.all(np.isfinite(qk_matmul_output)), ( + "all mode-3 rows are finite (the fully-masked row is guarded to 0.0)" +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_24_qk_matmul_output_mode3_softmax_precision", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_3d + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_attn_mask + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_3d_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_causal + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_diff_head_sizes + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_diff_heads_sizes", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_diff_head_sizes_attn_mask + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_3d_diff_heads_sizes_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_diff_head_sizes_causal + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_diff_heads_sizes_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_diff_head_sizes_scaled + +```python +scale = 1e-2 +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_diff_heads_sizes_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_diff_head_sizes_softcap + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_diff_heads_sizes_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_diff_head_sizes_with_past_and_present + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 30).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 10).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_3d_diff_heads_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_gqa + +```python +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_gqa", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_gqa_attn_mask + +```python +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_3d_gqa_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_gqa_causal + +```python +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + is_causal=1, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_gqa_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_gqa_scaled + +```python +scale = 1e-2 +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_gqa_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_gqa_softcap + +```python +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_gqa_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_gqa_with_past_and_present + +```python +q_num_heads, kv_num_heads = 9, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 72).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_3d_gqa_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_scaled + +```python +scale = 1e-2 +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + scale=scale, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_softcap + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + softcap=3.0, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_transpose_verification + +```python +"""Test case to verify correct 3D to 4D transpose behavior. + +This test verifies that 3D inputs are correctly reshaped and transposed +according to the ONNX specification: +[batch_size, seq_length, hidden_size] -> +[batch_size, seq_length, num_heads, head_size] -> +[batch_size, num_heads, seq_length, head_size] +""" +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +# Test inputs that will clearly demonstrate the transpose behavior +batch_size = 1 +q_seq_length = 2 +kv_seq_length = 2 +head_size = 4 +q_hidden_size = q_num_heads * head_size # 3 * 4 = 12 +kv_hidden_size = kv_num_heads * head_size # 3 * 4 = 12 + +# Create structured inputs to verify correct transpose behavior +# Q has a pattern where each position in hidden dimension has a specific value +Q = np.zeros((batch_size, q_seq_length, q_hidden_size), dtype=np.float32) +# Fill Q with pattern: head0=[1,1,1,1], head1=[2,2,2,2], head2=[3,3,3,3] +for head in range(q_num_heads): + start_idx = head * head_size + end_idx = start_idx + head_size + Q[0, :, start_idx:end_idx] = float(head + 1) + +K = np.ones((batch_size, kv_seq_length, kv_hidden_size), dtype=np.float32) * 0.1 +V = np.ones((batch_size, kv_seq_length, kv_hidden_size), dtype=np.float32) * 0.1 + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_3d_transpose_verification", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_with_past_and_present + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_3d_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_with_past_and_present_qk_matmul + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_3d_with_past_and_present_qk_matmul", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_with_past_and_present_qk_matmul_bias + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + qk_matmul_output_mode=2, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + qk_matmul_output_mode=2, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_3d_with_past_and_present_qk_matmul_bias", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_with_past_and_present_qk_matmul_softcap + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + softcap=2.0, + qk_matmul_output_mode=1, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + softcap=2.0, + qk_matmul_output_mode=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_3d_with_past_and_present_qk_matmul_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_3d_with_past_and_present_qk_matmul_softmax + +```python +q_num_heads, kv_num_heads = 3, 3 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + qk_matmul_output_mode=3, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 4, 24).astype(np.float32) +K = np.random.rand(2, 6, 24).astype(np.float32) +V = np.random.rand(2, 6, 24).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + q_num_heads=q_num_heads, + kv_num_heads=kv_num_heads, + qk_matmul_output_mode=3, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_3d_with_past_and_present_qk_matmul_softmax", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_4d_causal_nonpad_attn_mask_composition + +```python +"""Compose ``is_causal`` + ``nonpad_kv_seqlen`` + boolean ``attn_mask``. + +The existing nonpad tests use no ``attn_mask`` and the existing mask tests +use no ``nonpad_kv_seqlen``; this is the first to activate all three +constraints together on the external-cache path with ``batch > 1``. The +three biases are summed additively and a key is attended only if allowed by +all three. Crucially the inputs are designed so that **each constraint is +independently necessary** -- removing any one changes the golden -- to avoid a +degenerate test that a backend ignoring ``is_causal`` and/or +``nonpad_kv_seqlen`` could still pass: + +* **``is_causal`` binds.** Each batch has a key that the boolean mask allows + (``True``) but the bottom-right causal frontier disallows + (``j > i + offset``); only ``is_causal`` masks it (batch 0 row 0 key 2, + batch 1 row 0 key 3). +* **``attn_mask`` binds.** Each batch has a key the causal frontier and the + padding bound both allow but the boolean mask sets ``False`` (batch 0 row 2 + key 1, batch 1 row 2 key 2); only the mask masks it. +* **``nonpad_kv_seqlen`` binds.** ``nonpad_kv_seqlen`` sets the per-batch + causal *offset* (``offset = nonpad_kv_seqlen - q_sequence_length``), so + dropping it collapses the frontier to top-left (``offset = 0``) and shifts + which keys are attended. (Under ``is_causal=1`` the causal frontier already + subsumes the ``j < nonpad`` padding bound, so ``nonpad_kv_seqlen`` binds + through the offset it induces rather than through a redundant padding cut.) + +The mask is chosen to leave at least one allowed key on every query row, so +this exercises the *intersection* of the three constraints with finite outputs +(the fully-masked-row guard is covered by +``test_attention_4d_causal_nonpad_negative_offset_structural_empty`` and +``test_attention_24_fullymasked_qk_matmul_output_mode3_zero``). + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(11) +B, H, L, D = 2, 2, 6, 8 +S_q = 3 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, L, D).astype(np.float32) +V = np.random.rand(B, H, L, D).astype(np.float32) +nonpad_kv_seqlen = np.array([4, 5], dtype=np.int64) # offsets [1, 2] +# Per-batch (B, 1, S_q, L) bool mask. Each batch is laid out so all three +# constraints uniquely bind (see the docstring): a causal-only-masked key +# (mask True, j > i + offset), a mask-only-masked key (mask False, causal + +# nonpad allow it), and >=1 allowed key per row. +attn_mask = np.array( + [ + [ + [ + [True, True, True, False, False, False], + [True, True, True, False, False, False], + [True, False, True, True, False, False], + ] + ], + [ + [ + [True, True, True, True, False, False], + [True, True, True, True, False, False], + [True, True, False, True, True, False], + ] + ], + ], + dtype=np.bool_, +) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +# The chosen mask leaves >=1 allowed key per row, so the composition stays +# finite (no fully-masked row in this case). +assert np.all(np.isfinite(Y)), "composed-constraint output must be finite" + +# Non-degeneracy: each constraint is independently necessary. Removing any one +# of the three (is_causal, attn_mask, nonpad_kv_seqlen) must change the result, +# so a backend that ignores is_causal or nonpad_kv_seqlen cannot reproduce the +# golden by applying only the most restrictive mask. +y_no_causal, _, _, _ = _compute_attention( + Q, K, V, attn_mask=attn_mask, nonpad_kv_seqlen=nonpad_kv_seqlen, is_causal=0 +) +y_no_mask, _, _, _ = _compute_attention( + Q, K, V, nonpad_kv_seqlen=nonpad_kv_seqlen, is_causal=1 +) +y_no_nonpad, _, _, _ = _compute_attention( + Q, K, V, attn_mask=attn_mask, is_causal=1 +) +assert not np.allclose(Y, y_no_causal, equal_nan=True), ( + "is_causal must bind: dropping it changes the result" +) +assert not np.allclose(Y, y_no_mask, equal_nan=True), ( + "attn_mask must bind: dropping it changes the result" +) +assert not np.allclose(Y, y_no_nonpad, equal_nan=True), ( + "nonpad_kv_seqlen must bind (via the causal offset): dropping it changes the result" +) + +expect( + node, + inputs=[Q, K, V, attn_mask, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_causal_nonpad_attn_mask_composition", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_4d_causal_nonpad_batch_prefill + +```python +"""Batch>1 continued prefill with distinct per-batch bottom-right offsets. + +The batched generalization of the ``batch == 1`` continued-prefill case: with +``nonpad_kv_seqlen = [4, 5, 6]`` and ``S_q = 2`` the per-batch bottom-right +offsets are ``[2, 3, 4]`` (all ``>= 0``), so each batch realigns its causal +frontier to its own valid-key prefix. This pins that the per-batch offset is +applied independently across the batch dimension. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(12) +B, H, L, D = 3, 2, 6, 8 +S_q = 2 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, L, D).astype(np.float32) +V = np.random.rand(B, H, L, D).astype(np.float32) +nonpad_kv_seqlen = np.array([4, 5, 6], dtype=np.int64) # offsets [2, 3, 4] + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +assert np.all(np.isfinite(Y)), "per-batch prefill output must be finite" + +expect( + node, + inputs=[Q, K, V, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_causal_nonpad_batch_prefill", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_4d_causal_nonpad_continued_prefill + +```python +"""Continued / chunked prefill (S_q=2) into a partially-filled static cache. + +With ``nonpad_kv_seqlen = [4]`` and ``S_q = 2`` the bottom-right offset is +``4 - 2 = 2``: query 0 attends keys ``{0,1,2}`` and query 1 attends +``{0,1,2,3}``. The old top-left alignment would mask everything past the +diagonal (``{0}`` and ``{0,1}``), so this test fails pre-fix. +""" +np.random.seed(1) +B, H, L, D = 1, 2, 4, 8 +S_q = 2 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, L, D).astype(np.float32) +V = np.random.rand(B, H, L, D).astype(np.float32) +nonpad_kv_seqlen = np.array([4], dtype=np.int64) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_causal_nonpad_continued_prefill", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_4d_causal_nonpad_negative_offset_structural_empty + +```python +"""Negative bottom-right offset: structurally-empty early query rows -> zero. + +This is the onnx-node twin of the ORT gtest +``Attention_Causal_NonPadKVSeqLen_StructuralEmptyRow_Zero`` / +``StructuralEmptyRows_Zero_CUDA``. With ``nonpad_kv_seqlen = [2]`` and +``S_q = 4`` the bottom-right offset is ``2 - 4 = -2``: query row ``sq`` +attends keys ``0..(sq - 2)``, so rows 0 and 1 have an empty key set. Their +``softmax`` over an all-``-inf`` bias row is ``NaN``; the fully-masked-row +guard zeros those rows before the ``P @ V`` contraction so the output rows are +exactly ``0``, while rows 2 and 3 (attending keys ``{0}`` and ``{0,1}``) stay finite +and nonzero. A ``nonpad_kv_seqlen[b] < q_sequence_length`` input is out of +the contract's intended use, but its result is still well-defined (zeroed +rows) rather than ``NaN``; this test pins that defined behavior. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(7) +B, H, L, D = 1, 2, 4, 8 +S_q = 4 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, L, D).astype(np.float32) +V = np.random.rand(B, H, L, D).astype(np.float32) +# offset = nonpad - S_q = 2 - 4 = -2 -> rows 0,1 structurally empty. +nonpad_kv_seqlen = np.array([2], dtype=np.int64) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +# Structurally-empty early rows are exactly zero (not NaN); later rows finite. +assert np.all(np.isfinite(Y)), "all output rows must be finite" +assert np.array_equal(Y[:, :, 0, :], np.zeros_like(Y[:, :, 0, :])), ( + "structurally-empty row 0 must be zero" +) +assert np.array_equal(Y[:, :, 1, :], np.zeros_like(Y[:, :, 1, :])), ( + "structurally-empty row 1 must be zero" +) +assert np.any(Y[:, :, 2, :] != 0) and np.any(Y[:, :, 3, :] != 0), ( + "rows with a non-empty key set must be nonzero" +) + +expect( + node, + inputs=[Q, K, V, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_causal_nonpad_negative_offset_structural_empty", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_4d_causal_with_past_and_present + +```python +"""Regression guard: internal (past_key) cache + is_causal. + +This exercises the unchanged scalar bottom-right path (offset = +past_sequence_length). Its golden output must remain identical to the +pre-fix behavior, proving the external-cache change does not touch the +past_key path. +""" +np.random.seed(2) +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], + is_causal=1, +) + +past_sequence_length = 3 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 4, 8).astype(np.float32) +V = np.random.rand(2, 3, 4, 8).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + past_key=past_key, + past_value=past_value, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_causal_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_4d_diff_heads_mask4d_padded_kv + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 4).astype(np.float32) +nonpad_kv_seqlen = np.array([3, 4], dtype=np.int64) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + nonpad_kv_seqlen=nonpad_kv_seqlen, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_diff_heads_mask4d_padded_kv", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_4d_gqa_causal_nonpad_decode + +```python +"""External/static-cache decode (S_q=1) with per-batch valid lengths. + +K/V are the full static cache buffer; ``nonpad_kv_seqlen`` marks how many +leading keys are valid per batch. With bottom-right (offset-aware) causal +masking the single decode query attends keys ``0..nonpad[b]-1``. Under the +old top-left alignment it would attend only key 0, so this test fails +pre-fix and passes post-fix. +""" +np.random.seed(0) +B, H_q, H_kv, L, D = 2, 4, 2, 8, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H_q, 1, D).astype(np.float32) +K = np.random.rand(B, H_kv, L, D).astype(np.float32) +V = np.random.rand(B, H_kv, L, D).astype(np.float32) +# Batch 0 has all 8 keys valid, batch 1 only the first 5. +nonpad_kv_seqlen = np.array([8, 5], dtype=np.int64) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_gqa_causal_nonpad_decode", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_4d_gqa_causal_nonpad_decode_fp16 + +```python +"""fp16 variant of the external-cache decode case (locks -inf dtype handling).""" +np.random.seed(0) +B, H_q, H_kv, L, D = 2, 4, 2, 8, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "", "", "", "nonpad_kv_seqlen"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H_q, 1, D).astype(np.float16) +K = np.random.rand(B, H_kv, L, D).astype(np.float16) +V = np.random.rand(B, H_kv, L, D).astype(np.float16) +nonpad_kv_seqlen = np.array([8, 5], dtype=np.int64) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + nonpad_kv_seqlen=nonpad_kv_seqlen, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, nonpad_kv_seqlen], + outputs=[Y], + name="test_attention_4d_gqa_causal_nonpad_decode_fp16", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_attn_3d_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 1, 4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_3d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_attn_3d_mask_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 1, 4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_3d_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_attn_4d_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_4d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_attn_4d_mask_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_4d_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_attn_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_attn_mask_bool + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(bool) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_bool", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_attn_mask_bool_4d + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6).astype(bool) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_attn_mask_bool_4d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, is_causal=1) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_causal_boolmask_nan_robustness + +```python +"""Composed ``is_causal`` + boolean ``attn_mask`` NaN-robustness. + +The causal frontier (lower-triangular here, offset 0) and the boolean +``attn_mask`` are intersected: a key is attended only if allowed by both. +This exercises two pre-fix NaN sources on the same forward pass: + +* **Bug-1 (allowed cells stay finite).** Query 0 is allowed key 0 by both + the causal frontier (``{0}``) and the mask (``True`` at key 0). The old + ``(1 - attn_mask) * -inf`` conversion computes ``0 * -inf = NaN`` at that + allowed cell, poisoning the row. The select conversion + ``where(attn_mask, 0, -inf)`` keeps it finite. +* **Bug-2 (fully-masked row -> 0).** Query 1 is allowed keys ``{0, 1}`` by + the causal frontier but the mask is ``False`` at both, so the combined + constraint allows no key. ``softmax`` of an all-``-inf`` row is ``NaN``; + the fully-masked-row guard zeros it before the ``P @ V`` contraction so + the output row is ``0``. + +4D Q/K/V is used so ``q_num_heads``/``kv_num_heads`` are omitted (passing +them would make the function body treat the input as 3D). +""" +np.random.seed(3) +B, H, S, D = 1, 2, 2, 8 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(B, H, S, D).astype(np.float32) +K = np.random.rand(B, H, S, D).astype(np.float32) +V = np.random.rand(B, H, S, D).astype(np.float32) +# Row 0: key 0 allowed (Bug-1 allowed cell). Row 1: no key allowed -> fully +# masked once intersected with the causal frontier (Bug-2 empty row). +attn_mask = np.array([[True, False], [False, False]], dtype=np.bool_) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + is_causal=1, +) + +# Bug-1: allowed cells are finite (no NaN anywhere). Bug-2: the fully-masked +# query row is exactly zero, not NaN. +assert np.all(np.isfinite(Y)), "allowed cells must be finite (Bug-1)" +assert np.array_equal(Y[:, :, 1, :], np.zeros_like(Y[:, :, 1, :])), ( + "fully-masked row must be zero (Bug-2)" +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_causal_boolmask_nan_robustness", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+
+attention_diff_head_sizes + +```python +node = onnx.helper.make_node("Attention", inputs=["Q", "K", "V"], outputs=["Y"]) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_diff_heads_sizes", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_diff_head_sizes_attn_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_diff_heads_sizes_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_diff_head_sizes_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_diff_heads_sizes_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_diff_head_sizes_scaled + +```python +scale = 1e-2 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, scale=scale) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_diff_heads_sizes_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_diff_head_sizes_softcap + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=2.0, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + softcap=2.0, +) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_diff_heads_sizes_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_diff_head_sizes_with_past_and_present + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 10).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_diff_heads_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_diff_head_sizes_with_past_and_present_mask3D + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) +attn_mask = np.random.rand(2, 1, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 10).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_diff_heads_with_past_and_present_mask3d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_diff_head_sizes_with_past_and_present_mask4D + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 10).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 10).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_diff_heads_with_past_and_present_mask4d", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_fp16 + +```python +node = onnx.helper.make_node("Attention", inputs=["Q", "K", "V"], outputs=["Y"]) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float16) +K = np.random.rand(2, 3, 6, 8).astype(np.float16) +V = np.random.rand(2, 3, 6, 8).astype(np.float16) + +Y, _, _, _ = _compute_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_fp16", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_gqa + +```python +node = onnx.helper.make_node("Attention", inputs=["Q", "K", "V"], outputs=["Y"]) + +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_gqa", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_gqa_attn_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], +) + +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_gqa_attn_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_gqa_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + is_causal=1, +) + +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, is_causal=1) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_gqa_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_gqa_scaled + +```python +scale = 1e-2 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, +) + +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, scale=scale) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_gqa_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_gqa_softcap + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=2.0, +) + +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, softcap=2.0) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_gqa_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_gqa_with_past_and_present + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 9, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_gqa_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_gqa_with_past_and_present_fp16 + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 9, 4, 8).astype(np.float16) +K = np.random.rand(2, 3, 6, 8).astype(np.float16) +V = np.random.rand(2, 3, 6, 8).astype(np.float16) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float16) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float16) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float16) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_gqa_with_past_and_present_fp16", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_scaled + +```python +scale = 1e-2 +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, scale=scale) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_scaled", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_softcap + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y"], + softcap=2.0, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, _ = _compute_attention(Q, K, V, softcap=2.0) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_attention_4d_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_softcap_with_neginf_mask + +```python +"""Softcap + -inf mask: verifies softcap is applied BEFORE mask/bias. + +If ordering were wrong (mask then softcap), tanh(-inf/softcap) = -1, +so softcap * tanh(-inf/softcap) = -softcap (finite). That leaks +probability to masked positions. With correct ordering (softcap then +mask), the -inf mask values survive to softmax and yield zero weight. +""" +np.random.seed(42) +B, H, S_q, S_kv, D = 1, 1, 4, 6, 8 + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, S_kv, D).astype(np.float32) +V = np.random.rand(B, H, S_kv, D).astype(np.float32) + +# All Q positions are blocked from KV positions 4 and 5. +attn_mask = np.zeros((S_q, S_kv), dtype=np.float32) +attn_mask[:, 4:] = -np.inf + +softcap = 0.5 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + softcap=softcap, +) + +Y, _, _, _ = _compute_attention(Q, K, V, attn_mask=attn_mask, softcap=softcap) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_softcap_neginf_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_softcap_with_neginf_mask_poison + +```python +"""Softcap + -inf mask + poison values at masked KV positions. + +V has value 1000 at the masked positions (4 and 5). With correct +ordering the output stays in [0, 1] because the mask zeros out those +positions. With wrong ordering the output explodes (> 50), making +the failure obvious even with loose tolerances. +""" +np.random.seed(42) +B, H, S_q, S_kv, D = 1, 1, 4, 6, 8 + +Q = np.random.rand(B, H, S_q, D).astype(np.float32) +K = np.random.rand(B, H, S_kv, D).astype(np.float32) +V = np.random.rand(B, H, S_kv, D).astype(np.float32) + +# Block all Q positions from KV positions 4 and 5. +attn_mask = np.zeros((S_q, S_kv), dtype=np.float32) +attn_mask[:, 4:] = -np.inf + +# Poison: if attention leaks to masked positions, output >> 1. +V[:, :, 4:, :] = 1000.0 + +softcap = 0.5 + +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y"], + softcap=softcap, +) + +Y, _, _, _ = _compute_attention(Q, K, V, attn_mask=attn_mask, softcap=softcap) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y], + name="test_attention_4d_softcap_neginf_mask_poison", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_past_and_present + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, _ = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value], + name="test_attention_4d_with_past_and_present", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_past_and_present_qk_matmul + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_past_and_present_qk_matmul_bias + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + qk_matmul_output_mode=2, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + qk_matmul_output_mode=2, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul_bias", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_past_and_present_qk_matmul_bias_3d_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + qk_matmul_output_mode=2, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 1, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + qk_matmul_output_mode=2, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_past_and_present_qk_matmul_bias_3d_mask_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + qk_matmul_output_mode=2, + is_causal=1, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 1, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + qk_matmul_output_mode=2, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_past_and_present_qk_matmul_bias_4d_mask + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + qk_matmul_output_mode=2, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + qk_matmul_output_mode=2, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_past_and_present_qk_matmul_bias_4d_mask_causal + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask", "past_key", "past_value"], + outputs=["Y", "present_key", "present_value", "qk_matmul_output"], + qk_matmul_output_mode=2, + is_causal=1, +) + +past_sequence_length = 12 +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(2, 3, 4, 6 + past_sequence_length).astype(np.float32) +past_key = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) +past_value = np.random.rand(2, 3, past_sequence_length, 8).astype(np.float32) + +Y, present_key, present_value, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + past_key=past_key, + past_value=past_value, + qk_matmul_output_mode=2, + is_causal=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask, past_key, past_value], + outputs=[Y, present_key, present_value, qk_matmul_output], + name="test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_qk_matmul + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V"], + outputs=["Y", "", "", "qk_matmul_output"], +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) + +Y, _, _, qk_matmul_output = _compute_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y, qk_matmul_output], + name="test_attention_4d_with_qk_matmul", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_qk_matmul_bias + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + qk_matmul_output_mode=2, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + qk_matmul_output_mode=2, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_4d_with_qk_matmul_bias", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_qk_matmul_softcap + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + softcap=2.0, + qk_matmul_output_mode=1, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + softcap=2.0, + qk_matmul_output_mode=1, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_4d_with_qk_matmul_softcap", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+
+attention_with_qk_matmul_softmax + +```python +node = onnx.helper.make_node( + "Attention", + inputs=["Q", "K", "V", "attn_mask"], + outputs=["Y", "", "", "qk_matmul_output"], + qk_matmul_output_mode=3, +) + +Q = np.random.rand(2, 3, 4, 8).astype(np.float32) +K = np.random.rand(2, 3, 6, 8).astype(np.float32) +V = np.random.rand(2, 3, 6, 8).astype(np.float32) +attn_mask = np.random.rand(4, 6).astype(np.float32) + +Y, _, _, qk_matmul_output = _compute_attention( + Q, + K, + V, + attn_mask=attn_mask, + qk_matmul_output_mode=3, +) + +expect( + node, + inputs=[Q, K, V, attn_mask], + outputs=[Y, qk_matmul_output], + name="test_attention_4d_with_qk_matmul_softmax", + opset_imports=[onnx.helper.make_opsetid("", 23)], +) +``` + +
+ + +### AveragePool +There are 17 test cases, listed as following: +
+averagepool_1d_default + +```python +"""input_shape: [1, 3, 32] +output_shape: [1, 3, 31] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2], +) +x = np.random.randn(1, 3, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = [2] +strides = [1] +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "AVG") + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_1d_default") +``` + +
+
+averagepool_2d_ceil + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + strides=[2, 2], + ceil_mode=True, +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[6, 7.5], [12, 13.5]]]]).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_ceil") +``` + +
+
+averagepool_2d_ceil_last_window_starts_on_pad + +```python +"""input_shape: [1, 3, 2, 2] +output_shape: [1, 3, 1, 1] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + strides=[3, 3], + pads=[1, 1, 1, 1], + ceil_mode=True, + count_include_pad=1, +) +x = np.array( + [ + [ + [[0.8580, 0.0786], [0.2692, 0.1537]], + [[0.8816, 0.4353], [0.5772, 0.6623]], + [[0.9067, 0.9483], [0.5970, 0.7630]], + ] + ] +).astype(np.float32) +y = np.array([[[[0.1511]], [[0.2841]], [[0.3572]]]]).astype(np.float32) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_averagepool_2d_ceil_last_window_starts_on_pad", +) +``` + +
+
+averagepool_2d_default + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 31, 31] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = (2, 2) +strides = (1, 1) +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "AVG") + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_default") +``` + +
+
+averagepool_2d_dilations + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[1, 1], + dilations=[2, 2], + ceil_mode=True, +) + +# input shape: [1, 1, 4, 4] +x = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ] +).astype(np.float32) + +y = np.array([[[[6, 7], [10, 11]]]]).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_dilations") +``` + +
+
+averagepool_2d_pads + +```python +"""input_shape: [1, 3, 28, 28] +output_shape: [1, 3, 30, 30] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[2, 2, 2, 2], +) +x = np.random.randn(1, 3, 28, 28).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (3, 3) +strides = (1, 1) +pad_bottom = 2 +pad_top = 2 +pad_right = 2 +pad_left = 2 +pads = [pad_top, pad_left, pad_bottom, pad_right] +out_shape, extra_pads = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides, ceil_mode=False +) +padded = np.pad( + x, + ( + (0, 0), + (0, 0), + (extra_pads[0], extra_pads[2]), + (extra_pads[1], extra_pads[3]), + ), + mode="constant", + constant_values=np.nan, +) +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=extra_pads, + pads=pads, +) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_pads") +``` + +
+
+averagepool_2d_pads_count_include_pad + +```python +"""input_shape: [1, 3, 28, 28] +output_shape: [1, 3, 30, 30] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[2, 2, 2, 2], + count_include_pad=1, +) +x = np.random.randn(1, 3, 28, 28).astype(np.float32) +x_shape = np.shape(x) +dilations = (1, 1) +kernel_shape = (3, 3) +strides = (1, 1) +pad_bottom = 2 +pad_top = 2 +pad_right = 2 +pad_left = 2 +pads = [pad_top, pad_left, pad_bottom, pad_right] +out_shape, extra_pads = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides, dilations, ceil_mode=False +) +padded = np.pad( + x, + ( + (0, 0), + (0, 0), + (extra_pads[0], extra_pads[2]), + (extra_pads[1], extra_pads[3]), + ), + mode="constant", + constant_values=0, +) +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=extra_pads, + pads=pads, + count_include_pad=1, +) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_averagepool_2d_pads_count_include_pad", +) +``` + +
+
+averagepool_2d_precomputed_pads + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 5, 5] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + pads=[2, 2, 2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array( + [ + [ + [ + [7, 7.5, 8, 8.5, 9], + [9.5, 10, 10.5, 11, 11.5], + [12, 12.5, 13, 13.5, 14], + [14.5, 15, 15.5, 16, 16.5], + [17, 17.5, 18, 18.5, 19], + ] + ] + ] +).astype(np.float32) + +expect( + node, inputs=[x], outputs=[y], name="test_averagepool_2d_precomputed_pads" +) +``` + +
+
+averagepool_2d_precomputed_pads_count_include_pad + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 5, 5] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + pads=[2, 2, 2, 2], + count_include_pad=1, +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array( + [ + [ + [ + [2.5200, 3.6000, 4.8000, 4.0800, 3.2400], + [4.5600, 6.4000, 8.4000, 7.0400, 5.5200], + [7.2000, 10.0000, 13.0000, 10.8000, 8.4000], + [6.9600, 9.6000, 12.4000, 10.2400, 7.9200], + [6.1200, 8.4000, 10.8000, 8.8800, 6.8400], + ] + ] + ] +).astype(np.float32) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_averagepool_2d_precomputed_pads_count_include_pad", +) +``` + +
+
+averagepool_2d_precomputed_same_upper + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 3, 3] +pad_shape: [2, 2] -> [1, 1, 1, 1] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + strides=[2, 2], + auto_pad="SAME_UPPER", +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[4, 5.5, 7], [11.5, 13, 14.5], [19, 20.5, 22]]]]).astype( + np.float32 +) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_averagepool_2d_precomputed_same_upper", +) +``` + +
+
+averagepool_2d_precomputed_strides + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[4, 6], [14, 16]]]]).astype(np.float32) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_averagepool_2d_precomputed_strides", +) +``` + +
+
+averagepool_2d_same_lower + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [1, 0, 1, 0] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_LOWER", +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_LOWER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_LOWER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_bottom = pad_shape[0] // 2 +pad_top = pad_shape[0] - pad_bottom +pad_right = pad_shape[1] // 2 +pad_left = pad_shape[1] - pad_right +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=np.nan, +) +pads = (pad_top, pad_left, pad_bottom, pad_right) +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=pads, + pads=pads, +) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_same_lower") +``` + +
+
+averagepool_2d_same_upper + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [0, 1, 0, 1] by axis +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_UPPER", +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_UPPER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_UPPER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_top = pad_shape[0] // 2 +pad_bottom = pad_shape[0] - pad_top +pad_left = pad_shape[1] // 2 +pad_right = pad_shape[1] - pad_left +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=np.nan, +) +pads = (pad_top, pad_left, pad_bottom, pad_right) +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=pads, + pads=pads, +) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_same_upper") +``` + +
+
+averagepool_2d_strides + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 10, 10] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + strides=[3, 3], +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (5, 5) +strides = (3, 3) +out_shape, pads = get_output_shape_explicit_padding( + None, x_shape[2:], kernel_shape, strides, ceil_mode=False +) +padded = x +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=pads, + pads=None, +) + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_strides") +``` + +
+
+averagepool_3d_default + +```python +"""input_shape: [1, 3, 32, 32, 32] +output_shape: [1, 3, 31, 31, 31] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], +) +x = np.random.randn(1, 3, 32, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = [2, 2, 2] +strides = [1, 1, 1] +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "AVG") + +expect(node, inputs=[x], outputs=[y], name="test_averagepool_3d_default") +``` + +
+
+averagepool_3d_dilations + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], + strides=[1, 1, 1], + dilations=[2, 2, 2], + ceil_mode=True, +) + +# input shape: [1, 1, 4, 4, 4] +x = np.array( + [ + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + ] + ] + ] +).astype(np.float32) + +y = np.array([[[[[6, 7], [10, 11]], [[6, 7], [10, 11]]]]]).astype(np.float32) + +expect( + node, inputs=[x], outputs=[y], name="test_averagepool_3d_dilations_small" +) +``` + +
+
+averagepool_3d_dilations_large + +```python +x_shape = (32, 32, 32) +dilations = (2, 2, 2) +kernel_shape = (5, 5, 5) +strides = (3, 3, 3) +count_include_pad = 0 + +for count_include_pad in (0, 1): + for ceil_mode in (True, False): + node = onnx.helper.make_node( + "AveragePool", + inputs=["x"], + outputs=["y"], + kernel_shape=kernel_shape, + strides=strides, + dilations=dilations, + count_include_pad=count_include_pad, + ceil_mode=ceil_mode, + ) + + x = np.random.randn(1, 1, *x_shape).astype(np.float32) + out_shape, extra_pads = get_output_shape_explicit_padding( + None, + x_shape, + kernel_shape, + strides, + dilations=dilations, + ceil_mode=ceil_mode, + ) + padded = np.pad( + x, + ( + (0, 0), + (0, 0), + (extra_pads[0], extra_pads[3]), + (extra_pads[1], extra_pads[4]), + (extra_pads[2], extra_pads[5]), + ), + mode="constant", + constant_values=0 if count_include_pad == 1 else np.nan, + ) + y = pool( + padded, + (1, 1, *x_shape), + kernel_shape, + strides, + out_shape, + "AVG", + pads_required=extra_pads, + pads=None, + dilations=dilations, + count_include_pad=count_include_pad, + ) + + test_name = f"test_averagepool_3d_dilations_large_count_include_pad_is_{count_include_pad}_ceil_mode_is_{ceil_mode}" + expect(node, inputs=[x], outputs=[y], name=test_name) +``` + +
+ + +### BatchNormalization +There are 2 test cases, listed as following: +
+batchnormalization + +```python +# input size: (2, 3, 4, 5) +x = np.random.randn(2, 3, 4, 5).astype(np.float32) +s = np.random.randn(3).astype(np.float32) +bias = np.random.randn(3).astype(np.float32) +mean = np.random.randn(3).astype(np.float32) +var = np.random.rand(3).astype(np.float32) +y = _batchnorm_test_mode(x, s, bias, mean, var).astype(np.float32) + +node = onnx.helper.make_node( + "BatchNormalization", + inputs=["x", "s", "bias", "mean", "var"], + outputs=["y"], +) + +# output size: (2, 3, 4, 5) +expect( + node, + inputs=[x, s, bias, mean, var], + outputs=[y], + name="test_batchnorm_example", +) + +# input size: (2, 3, 4, 5) +x = np.random.randn(2, 3, 4, 5).astype(np.float32) +s = np.random.randn(3).astype(np.float32) +bias = np.random.randn(3).astype(np.float32) +mean = np.random.randn(3).astype(np.float32) +var = np.random.rand(3).astype(np.float32) +epsilon = 1e-2 +y = _batchnorm_test_mode(x, s, bias, mean, var, epsilon).astype(np.float32) + +node = onnx.helper.make_node( + "BatchNormalization", + inputs=["x", "s", "bias", "mean", "var"], + outputs=["y"], + epsilon=epsilon, +) + +# output size: (2, 3, 4, 5) +expect( + node, + inputs=[x, s, bias, mean, var], + outputs=[y], + name="test_batchnorm_epsilon", +) +``` + +
+
+train + +```python +# input size: (2, 3, 4, 5) +x = np.random.randn(2, 3, 4, 5).astype(np.float32) +s = np.random.randn(3).astype(np.float32) +bias = np.random.randn(3).astype(np.float32) +mean = np.random.randn(3).astype(np.float32) +var = np.random.rand(3).astype(np.float32) +# using np.bool(1) while generating test data with "'bool' object has no attribute 'dtype'" +# working around by using np.byte(1).astype(bool) +training_mode = 1 +y, output_mean, output_var = _batchnorm_training_mode(x, s, bias, mean, var) + +node = onnx.helper.make_node( + "BatchNormalization", + inputs=["x", "s", "bias", "mean", "var"], + outputs=["y", "output_mean", "output_var"], + training_mode=training_mode, +) + +# output size: (2, 3, 4, 5) +expect( + node, + inputs=[x, s, bias, mean, var], + outputs=[y, output_mean, output_var], + name="test_batchnorm_example_training_mode", +) + +# input size: (2, 3, 4, 5) +x = np.random.randn(2, 3, 4, 5).astype(np.float32) +s = np.random.randn(3).astype(np.float32) +bias = np.random.randn(3).astype(np.float32) +mean = np.random.randn(3).astype(np.float32) +var = np.random.rand(3).astype(np.float32) +training_mode = 1 +momentum = 0.9 +epsilon = 1e-2 +y, output_mean, output_var = _batchnorm_training_mode( + x, s, bias, mean, var, momentum, epsilon +) + +node = onnx.helper.make_node( + "BatchNormalization", + inputs=["x", "s", "bias", "mean", "var"], + outputs=["y", "output_mean", "output_var"], + epsilon=epsilon, + training_mode=training_mode, +) + +# output size: (2, 3, 4, 5) +expect( + node, + inputs=[x, s, bias, mean, var], + outputs=[y, output_mean, output_var], + name="test_batchnorm_epsilon_training_mode", +) +``` + +
+ + +### Bernoulli +There are 3 test cases, listed as following: +
+bernoulli_with_dtype + +```python +node = onnx.helper.make_node( + "Bernoulli", + inputs=["x"], + outputs=["y"], + dtype=onnx.TensorProto.DOUBLE, +) + +x = np.random.uniform(0.0, 1.0, 10).astype(np.float32) +y = bernoulli_reference_implementation(x, float) +expect(node, inputs=[x], outputs=[y], name="test_bernoulli_double") +``` + +
+
+bernoulli_with_seed + +```python +seed = float(0) +node = onnx.helper.make_node( + "Bernoulli", + inputs=["x"], + outputs=["y"], + seed=seed, +) + +x = np.random.uniform(0.0, 1.0, 10).astype(np.float32) +y = bernoulli_reference_implementation(x, np.float32) +expect(node, inputs=[x], outputs=[y], name="test_bernoulli_seed") +``` + +
+
+bernoulli_without_dtype + +```python +node = onnx.helper.make_node( + "Bernoulli", + inputs=["x"], + outputs=["y"], +) + +x = np.random.uniform(0.0, 1.0, 10).astype(float) +y = bernoulli_reference_implementation(x, float) +expect(node, inputs=[x], outputs=[y], name="test_bernoulli") +``` + +
+ + +### BitCast +There are 10 test cases, listed as following: +
+bitcast_2d_float32_to_int32 + +```python +"""Test bitcasting 2D array from float32 to int32.""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT32, +) +x = np.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]], dtype=np.float32) +y = x.view(np.int32) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_2d_float32_to_int32") +``` + +
+
+bitcast_bool_to_uint8 + +```python +"""Test bitcasting from bool to uint8 (same size).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.UINT8, +) +x = np.array([True, False, True, False], dtype=np.bool_) +y = x.view(np.uint8) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_bool_to_uint8") +``` + +
+
+bitcast_float32_to_int32 + +```python +"""Test bitcasting from float32 to int32 (same size).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT32, +) +x = np.array([1.0, -2.5, 3.75], dtype=np.float32) +y = x.view(np.int32) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_float32_to_int32") +``` + +
+
+bitcast_float64_to_int64 + +```python +"""Test bitcasting from float64 to int64 (same size).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT64, +) +x = np.array([1.0, -2.5, 3.75], dtype=np.float64) +y = x.view(np.int64) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_float64_to_int64") +``` + +
+
+bitcast_int32_to_float32 + +```python +"""Test bitcasting from int32 to float32 (same size).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.FLOAT, +) +x = np.array([1065353216, -1071644672, 1081081856], dtype=np.int32) +y = x.view(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_int32_to_float32") +``` + +
+
+bitcast_int64_to_float64 + +```python +"""Test bitcasting from int64 to float64 (same size).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.DOUBLE, +) +x = np.array( + [4607182418800017408, -4611686018427387904, 4614256656552045184], + dtype=np.int64, +) +y = x.view(np.float64) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_int64_to_float64") +``` + +
+
+bitcast_int8_to_uint8 + +```python +"""Test bitcasting from int8 to uint8 (same size, different signedness).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.UINT8, +) +x = np.array([-1, -128, 127, 0], dtype=np.int8) +y = x.view(np.uint8) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_int8_to_uint8") +``` + +
+
+bitcast_scalar_float32_to_int32 + +```python +"""Test bitcasting scalar from float32 to int32.""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT32, +) +x = np.array(1.0, dtype=np.float32) +y = x.view(np.int32) +expect( + node, inputs=[x], outputs=[y], name="test_bitcast_scalar_float32_to_int32" +) +``` + +
+
+bitcast_uint16_to_int16 + +```python +"""Test bitcasting from uint16 to int16 (same size, different signedness).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT16, +) +x = np.array([1, 32768, 65535], dtype=np.uint16) +y = x.view(np.int16) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_uint16_to_int16") +``` + +
+
+bitcast_uint32_to_int32 + +```python +"""Test bitcasting from uint32 to int32 (same size, different signedness).""" +node = onnx.helper.make_node( + "BitCast", + inputs=["x"], + outputs=["y"], + to=onnx.TensorProto.INT32, +) +x = np.array([4294967295, 2147483648, 2147483647], dtype=np.uint32) +y = x.view(np.int32) +expect(node, inputs=[x], outputs=[y], name="test_bitcast_uint32_to_int32") +``` + +
+ + +### BitShift +There are 8 test cases, listed as following: +
+left_unit16 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="LEFT" +) + +x = np.array([16, 4, 1]).astype(np.uint16) +y = np.array([1, 2, 3]).astype(np.uint16) +z = x << y # expected output [32, 16, 8] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_left_uint16") +``` + +
+
+left_unit32 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="LEFT" +) + +x = np.array([16, 4, 1]).astype(np.uint32) +y = np.array([1, 2, 3]).astype(np.uint32) +z = x << y # expected output [32, 16, 8] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_left_uint32") +``` + +
+
+left_unit64 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="LEFT" +) + +x = np.array([16, 4, 1]).astype(np.uint64) +y = np.array([1, 2, 3]).astype(np.uint64) +z = x << y # expected output [32, 16, 8] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_left_uint64") +``` + +
+
+left_unit8 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="LEFT" +) + +x = np.array([16, 4, 1]).astype(np.uint8) +y = np.array([1, 2, 3]).astype(np.uint8) +z = x << y # expected output [32, 16, 8] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_left_uint8") +``` + +
+
+right_unit16 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="RIGHT" +) + +x = np.array([16, 4, 1]).astype(np.uint16) +y = np.array([1, 2, 3]).astype(np.uint16) +z = x >> y # expected output [8, 1, 0] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_right_uint16") +``` + +
+
+right_unit32 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="RIGHT" +) + +x = np.array([16, 4, 1]).astype(np.uint32) +y = np.array([1, 2, 3]).astype(np.uint32) +z = x >> y # expected output [8, 1, 0] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_right_uint32") +``` + +
+
+right_unit64 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="RIGHT" +) + +x = np.array([16, 4, 1]).astype(np.uint64) +y = np.array([1, 2, 3]).astype(np.uint64) +z = x >> y # expected output [8, 1, 0] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_right_uint64") +``` + +
+
+right_unit8 + +```python +node = onnx.helper.make_node( + "BitShift", inputs=["x", "y"], outputs=["z"], direction="RIGHT" +) + +x = np.array([16, 4, 1]).astype(np.uint8) +y = np.array([1, 2, 3]).astype(np.uint8) +z = x >> y # expected output [8, 1, 0] +expect(node, inputs=[x, y], outputs=[z], name="test_bitshift_right_uint8") +``` + +
+ + +### BitwiseAnd +There are 2 test cases, listed as following: +
+bitwiseand + +```python +node = onnx.helper.make_node( + "BitwiseAnd", + inputs=["x", "y"], + outputs=["bitwiseand"], +) + +# 2d +x = create_random_int((3, 4), np.int32) +y = create_random_int((3, 4), np.int32) +z = np.bitwise_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_and_i32_2d") + +# 3d +x = create_random_int((3, 4, 5), np.int16) +y = create_random_int((3, 4, 5), np.int16) +z = np.bitwise_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_and_i16_3d") +``` + +
+
+bitwiseand_broadcast + +```python +node = onnx.helper.make_node( + "BitwiseAnd", + inputs=["x", "y"], + outputs=["bitwiseand"], +) + +# 3d vs 1d +x = create_random_int((3, 4, 5), np.uint64) +y = create_random_int((5,), np.uint64) +z = np.bitwise_and(x, y) +expect( + node, inputs=[x, y], outputs=[z], name="test_bitwise_and_ui64_bcast_3v1d" +) + +# 4d vs 3d +x = create_random_int((3, 4, 5, 6), np.uint8) +y = create_random_int((4, 5, 6), np.uint8) +z = np.bitwise_and(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_and_ui8_bcast_4v3d") +``` + +
+ + +### BitwiseNot +There are 1 test cases, listed as following: +
+bitwisenot + +```python +node = onnx.helper.make_node( + "BitwiseNot", + inputs=["x"], + outputs=["bitwise_not"], +) + +# 2d +x = create_random_int((3, 4), np.int32) +y = np.bitwise_not(x) +expect(node, inputs=[x], outputs=[y], name="test_bitwise_not_2d") + +# 3d +x = create_random_int((3, 4, 5), np.uint16) +y = np.bitwise_not(x) +expect(node, inputs=[x], outputs=[y], name="test_bitwise_not_3d") + +# 4d +x = create_random_int((3, 4, 5, 6), np.uint8) +y = np.bitwise_not(x) +expect(node, inputs=[x], outputs=[y], name="test_bitwise_not_4d") +``` + +
+ + +### BitwiseOr +There are 2 test cases, listed as following: +
+bitwiseor + +```python +node = onnx.helper.make_node( + "BitwiseOr", + inputs=["x", "y"], + outputs=["bitwiseor"], +) +# 2d +x = create_random_int((3, 4), np.int32) +y = create_random_int((3, 4), np.int32) +z = np.bitwise_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_or_i32_2d") + +# 4d +x = create_random_int((3, 4, 5, 6), np.int8) +y = create_random_int((3, 4, 5, 6), np.int8) +z = np.bitwise_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_or_i16_4d") +``` + +
+
+bitwiseor_broadcast + +```python +node = onnx.helper.make_node( + "BitwiseOr", + inputs=["x", "y"], + outputs=["bitwiseor"], +) + +# 3d vs 1d +x = create_random_int((3, 4, 5), np.uint64) +y = create_random_int((5,), np.uint64) +z = np.bitwise_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_or_ui64_bcast_3v1d") + +# 4d vs 3d +x = create_random_int((3, 4, 5, 6), np.uint8) +y = create_random_int((4, 5, 6), np.uint8) +z = np.bitwise_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_or_ui8_bcast_4v3d") +``` + +
+ + +### BitwiseXor +There are 2 test cases, listed as following: +
+bitwiseor_broadcast + +```python +node = onnx.helper.make_node( + "BitwiseXor", + inputs=["x", "y"], + outputs=["bitwisexor"], +) + +# 3d vs 1d +x = create_random_int((3, 4, 5), np.uint64) +y = create_random_int((5,), np.uint64) +z = np.bitwise_xor(x, y) +expect( + node, inputs=[x, y], outputs=[z], name="test_bitwise_xor_ui64_bcast_3v1d" +) + +# 4d vs 3d +x = create_random_int((3, 4, 5, 6), np.uint8) +y = create_random_int((4, 5, 6), np.uint8) +z = np.bitwise_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_xor_ui8_bcast_4v3d") +``` + +
+
+bitwisexor + +```python +node = onnx.helper.make_node( + "BitwiseXor", + inputs=["x", "y"], + outputs=["bitwisexor"], +) + +# 2d +x = create_random_int((3, 4), np.int32) +y = create_random_int((3, 4), np.int32) +z = np.bitwise_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_xor_i32_2d") + +# 3d +x = create_random_int((3, 4, 5), np.int16) +y = create_random_int((3, 4, 5), np.int16) +z = np.bitwise_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_bitwise_xor_i16_3d") +``` + +
+ + +### BlackmanWindow +There are 1 test cases, listed as following: +
+blackmanwindow + +```python +# Test periodic window +node = onnx.helper.make_node( + "BlackmanWindow", + inputs=["x"], + outputs=["y"], +) +size = np.int32(10) +a0 = 0.42 +a1 = -0.5 +a2 = 0.08 +y = a0 +y += a1 * np.cos(2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / size) +y += a2 * np.cos(4 * np.pi * np.arange(0, size, 1, dtype=np.float32) / size) +expect( + node, + inputs=[size], + outputs=[y.astype(np.float32)], + name="test_blackmanwindow", +) + +# Test symmetric window +node = onnx.helper.make_node( + "BlackmanWindow", inputs=["x"], outputs=["y"], periodic=0 +) +size = np.int32(10) +a0 = 0.42 +a1 = -0.5 +a2 = 0.08 +y = a0 +y += a1 * np.cos( + 2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / (size - 1) +) +y += a2 * np.cos( + 4 * np.pi * np.arange(0, size, 1, dtype=np.float32) / (size - 1) +) +expect( + node, + inputs=[size], + outputs=[y.astype(np.float32)], + name="test_blackmanwindow_symmetric", +) +``` + +
+ + +### Cast +There are 3 test cases, listed as following: +
+cast + +```python +test_cases = [ + ("FLOAT", "FLOAT16"), + ("FLOAT", "DOUBLE"), + ("FLOAT16", "FLOAT"), + ("FLOAT16", "DOUBLE"), + ("DOUBLE", "FLOAT"), + ("DOUBLE", "FLOAT16"), + ("FLOAT", "BFLOAT16"), + ("BFLOAT16", "FLOAT"), + ("FLOAT", "FLOAT8E4M3FN"), + ("FLOAT16", "FLOAT8E4M3FN"), + ("FLOAT", "FLOAT8E4M3FNUZ"), + ("FLOAT16", "FLOAT8E4M3FNUZ"), + ("FLOAT8E4M3FN", "FLOAT"), + ("FLOAT8E4M3FN", "FLOAT16"), + ("FLOAT8E4M3FNUZ", "FLOAT"), + ("FLOAT8E4M3FNUZ", "FLOAT16"), + ("FLOAT", "FLOAT8E5M2"), + ("FLOAT16", "FLOAT8E5M2"), + ("FLOAT", "FLOAT8E5M2FNUZ"), + ("FLOAT16", "FLOAT8E5M2FNUZ"), + ("FLOAT8E5M2", "FLOAT"), + ("FLOAT8E5M2", "FLOAT16"), + ("FLOAT8E5M2FNUZ", "FLOAT"), + ("FLOAT8E5M2FNUZ", "FLOAT16"), + ("FLOAT", "UINT4"), + ("FLOAT16", "UINT4"), + ("FLOAT", "INT4"), + ("FLOAT16", "INT4"), + ("UINT4", "FLOAT"), + ("UINT4", "FLOAT16"), + ("UINT4", "UINT8"), + ("INT4", "FLOAT"), + ("INT4", "FLOAT16"), + ("INT4", "INT8"), + ("FLOAT4E2M1", "FLOAT"), + ("FLOAT4E2M1", "FLOAT16"), + ("FLOAT", "FLOAT4E2M1"), + ("FLOAT16", "FLOAT4E2M1"), + ("FLOAT", "UINT2"), + ("FLOAT16", "UINT2"), + ("FLOAT", "INT2"), + ("FLOAT16", "INT2"), + ("UINT2", "FLOAT"), + ("UINT2", "FLOAT16"), + ("UINT2", "UINT8"), + ("INT2", "FLOAT"), + ("INT2", "FLOAT16"), + ("INT2", "INT8"), +] + +for from_type, to_type in test_cases: + if from_type == to_type: + # Skip cases where from_type and to_type are the same + continue + from_dtype = getattr(TensorProto, from_type) + to_dtype = getattr(TensorProto, to_type) + from_np_dtype = tensor_dtype_to_np_dtype(from_dtype) + to_np_dtype = tensor_dtype_to_np_dtype(to_dtype) + + if from_type == "BFLOAT16" or to_type == "BFLOAT16": + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.816468", + "0.21087195", + "0.7229038", + "NaN", + "INF", + "+INF", + "-INF", + ], + dtype=np.float32, + ) + input_shape = (3, 4) + + elif from_type in F8_TYPES or to_type in F8_TYPES: + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.7229038", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-0.0000001", + "0.0000001", + "-1000000", + ], + dtype=np.float32, + ) + input_shape = (3, 5) + elif from_type in ("UINT4", "INT4") or to_type in ("UINT4", "INT4"): + np_fp32 = np.arange(-9, 16).astype(np.float32) + input_shape = (5, 5) + elif from_type in ("UINT2", "INT2") or to_type in ("UINT2", "INT2"): + np_fp32 = np.arange(-3, 4).astype(np.float32) + input_shape = (7, 1) + elif from_type == "FLOAT4E2M1" or to_type == "FLOAT4E2M1": + np_fp32 = np.array( + [ + "0.48", + "0.25", + "1.05", + "-3.5", + "-8", + "9", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-4", + "0.01", + "-0.0", + ], + dtype=np.float32, + ) + input_shape = (3, 5) + + else: + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.816468", + "0.21087195", + "0.7229038", + "NaN", + "INF", + "+INF", + "-INF", + ], + dtype=np.float32, + ).reshape([3, 4]) + input_shape = (3, 4) + + if from_type in F8_TYPES: + np_from = onnx.numpy_helper.saturate_cast(np_fp32, from_np_dtype) + input = make_tensor( + "input", + from_dtype, + input_shape, + vals=np_from, + raw=True, + ) + elif from_type in FOUR_BIT_TYPES: + np_from = np_fp32.astype(from_np_dtype) + packed = onnx.numpy_helper._pack_4bitx2(np_from) + # No byteswap needed on big-endian machines as _pack_4bitx2() + # returns a numpy array with uint8 datatype. + input = make_tensor( + "input", from_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + elif from_type in TWO_BIT_TYPES: + np_from = np_fp32.astype(from_np_dtype) + packed = onnx.numpy_helper._pack_2bitx4(np_from) + input = make_tensor( + "input", from_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + else: + np_from = np_fp32.astype(from_np_dtype) + input = make_tensor( + "input", from_dtype, input_shape, vals=np_from, raw=True + ) + + if to_type in F8_TYPES: + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=onnx.numpy_helper.saturate_cast(np_from, to_np_dtype), + raw=True, + ) + elif to_type in FOUR_BIT_TYPES: + packed = onnx.numpy_helper._pack_4bitx2(np_from.astype(to_np_dtype)) + # No byteswap needed on big-endian machines as _pack_4bitx2() + # returns a numpy array with uint8 datatype. + output = make_tensor( + "output", to_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + elif to_type in TWO_BIT_TYPES: + packed = onnx.numpy_helper._pack_2bitx4(np_from.astype(to_np_dtype)) + output = make_tensor( + "output", to_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + else: + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=np_from.astype(to_np_dtype), + raw=True, + ) + + node = onnx.helper.make_node( + "Cast", + inputs=["input"], + outputs=["output"], + to=to_dtype, + ) + expect( + node, + inputs=[input], + outputs=[output], + name="test_cast_" + from_type + "_to_" + to_type, + ) +``` + +
+
+e8m0 + +```python +np_fp32 = np.array( + [ + "0.0", + "0.124", + "0.25", + "0.5", + "1.1", + "2.0", + "4.0", + "8.0", + ], + dtype=np.float32, +) +test_cases = [ + ("FLOAT", "FLOAT8E8M0"), + ("FLOAT16", "FLOAT8E8M0"), + ("FLOAT8E8M0", "FLOAT"), + ("FLOAT8E8M0", "FLOAT16"), +] +for from_type, to_type in test_cases: + if from_type == "FLOAT": + input_np = np_fp32 + output_np = to_float8e8m0(np_fp32) + elif from_type == "FLOAT16": + input_np = np_fp32.astype(np.float16) + output_np = to_float8e8m0(input_np) + elif from_type == "FLOAT8E8M0": + input_np = to_float8e8m0(np_fp32) + if to_type == "FLOAT": + output_np = input_np.astype(np.float32) + elif to_type == "FLOAT16": + output_np = input_np.astype(np.float16) + else: + raise ValueError( + f"Conversion from {from_type} to {to_type} is not tested." + ) + else: + raise ValueError( + f"Conversion from {from_type} to {to_type} is not tested." + ) + input = make_tensor( + "input", + getattr(TensorProto, from_type), + [2, 4], + input_np, + raw=True, + ) + output = make_tensor( + "output", + getattr(TensorProto, to_type), + [2, 4], + output_np, + raw=True, + ) + if to_type == "FLOAT8E8M0": + node = onnx.helper.make_node( + "Cast", + inputs=["input"], + outputs=["output"], + to=getattr(TensorProto, to_type), + saturate=1, + round_mode="up", + ) + else: + node = onnx.helper.make_node( + "Cast", + inputs=["input"], + outputs=["output"], + to=getattr(TensorProto, to_type), + ) + + expect( + node, + inputs=[input], + outputs=[output], + name="test_cast_e8m0_" + from_type + "_to_" + to_type, + ) +``` + +
+
+saturate_false + +```python +test_cases = itertools.product( + [ + "FLOAT", + "FLOAT16", + ], + [ + "FLOAT8E4M3FN", + "FLOAT8E4M3FNUZ", + "FLOAT8E5M2", + "FLOAT8E5M2FNUZ", + ], +) +input_shape = (3, 5) +for from_type, to_type in test_cases: + from_dtype = getattr(TensorProto, from_type) + to_dtype = getattr(TensorProto, to_type) + from_np_dtype = tensor_dtype_to_np_dtype(from_dtype) + to_np_dtype = tensor_dtype_to_np_dtype(to_dtype) + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.7229038", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-0.0000001", + "0.0000001", + "-1000000", + ], + dtype=np.float32, + ) + + input = make_tensor( + "input", + from_dtype, + input_shape, + vals=np_fp32.astype(from_np_dtype), + raw=True, + ) + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=np_fp32.astype(from_np_dtype).astype(to_np_dtype), + raw=True, + ) + + node = onnx.helper.make_node( + "Cast", + inputs=["input"], + outputs=["output"], + to=to_dtype, + saturate=0, + ) + expect( + node, + inputs=[input], + outputs=[output], + name="test_cast_no_saturate_" + from_type + "_to_" + to_type, + ) +``` + +
+ + +### CastLike +There are 2 test cases, listed as following: +
+castlike + +```python +test_cases = [ + ("FLOAT", "FLOAT16"), + ("FLOAT", "DOUBLE"), + ("FLOAT16", "FLOAT"), + ("FLOAT16", "DOUBLE"), + ("DOUBLE", "FLOAT"), + ("DOUBLE", "FLOAT16"), + ("FLOAT", "BFLOAT16"), + ("BFLOAT16", "FLOAT"), + ("FLOAT", "FLOAT8E4M3FN"), + ("FLOAT16", "FLOAT8E4M3FN"), + ("FLOAT", "FLOAT8E4M3FNUZ"), + ("FLOAT16", "FLOAT8E4M3FNUZ"), + ("FLOAT8E4M3FN", "FLOAT"), + ("FLOAT8E4M3FN", "FLOAT16"), + ("FLOAT8E4M3FNUZ", "FLOAT"), + ("FLOAT8E4M3FNUZ", "FLOAT16"), + ("FLOAT", "FLOAT8E5M2"), + ("FLOAT16", "FLOAT8E5M2"), + ("FLOAT", "FLOAT8E5M2FNUZ"), + ("FLOAT16", "FLOAT8E5M2FNUZ"), + ("FLOAT8E5M2", "FLOAT"), + ("FLOAT8E5M2", "FLOAT16"), + ("FLOAT8E5M2FNUZ", "FLOAT"), + ("FLOAT8E5M2FNUZ", "FLOAT16"), + ("FLOAT", "UINT4"), + ("FLOAT16", "UINT4"), + ("FLOAT", "INT4"), + ("FLOAT16", "INT4"), + ("UINT4", "FLOAT"), + ("UINT4", "FLOAT16"), + ("UINT4", "UINT8"), + ("INT4", "FLOAT"), + ("INT4", "FLOAT16"), + ("INT4", "INT8"), + ("FLOAT4E2M1", "FLOAT"), + ("FLOAT4E2M1", "FLOAT16"), + ("FLOAT", "FLOAT4E2M1"), + ("FLOAT16", "FLOAT4E2M1"), + ("FLOAT", "UINT2"), + ("FLOAT16", "UINT2"), + ("FLOAT", "INT2"), + ("FLOAT16", "INT2"), + ("UINT2", "FLOAT"), + ("UINT2", "FLOAT16"), + ("UINT2", "UINT8"), + ("INT2", "FLOAT"), + ("INT2", "FLOAT16"), + ("INT2", "INT8"), +] + +f8_types = {"FLOAT8E4M3FN", "FLOAT8E4M3FNUZ", "FLOAT8E5M2", "FLOAT8E5M2FNUZ"} + +for from_type, to_type in test_cases: + if from_type == to_type: + # Skip cases where from_type and to_type are the same + continue + from_dtype = getattr(TensorProto, from_type) + to_dtype = getattr(TensorProto, to_type) + from_np_dtype = tensor_dtype_to_np_dtype(from_dtype) + to_np_dtype = tensor_dtype_to_np_dtype(to_dtype) + + if from_type == "BFLOAT16" or to_type == "BFLOAT16": + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.816468", + "0.21087195", + "0.7229038", + "NaN", + "INF", + "+INF", + "-INF", + ], + dtype=np.float32, + ) + input_shape = (3, 4) + + elif from_type in f8_types or to_type in f8_types: + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.7229038", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-0.0000001", + "0.0000001", + "-1000000", + ], + dtype=np.float32, + ) + input_shape = (3, 5) + elif from_type in ("UINT4", "INT4") or to_type in ("UINT4", "INT4"): + np_fp32 = np.arange(-9, 16).astype(np.float32) + input_shape = (5, 5) + elif from_type in ("UINT2", "INT2") or to_type in ("UINT2", "INT2"): + np_fp32 = np.arange(-3, 4).astype(np.float32) + input_shape = (7, 1) + elif from_type == "FLOAT4E2M1" or to_type == "FLOAT4E2M1": + np_fp32 = np.array( + [ + "0.48", + "0.25", + "1.05", + "-3.5", + "-8", + "9", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-4", + "0.01", + "-0.0", + ], + dtype=np.float32, + ) + input_shape = (3, 5) + + else: + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.816468", + "0.21087195", + "0.7229038", + "NaN", + "INF", + "+INF", + "-INF", + ], + dtype=np.float32, + ).reshape([3, 4]) + input_shape = (3, 4) + + if from_type in F8_TYPES: + np_from = onnx.numpy_helper.saturate_cast(np_fp32, from_np_dtype) + input = make_tensor( + "input", + from_dtype, + input_shape, + vals=np_from, + raw=True, + ) + elif from_type in FOUR_BIT_TYPES: + np_from = np_fp32.astype(from_np_dtype) + packed = onnx.numpy_helper._pack_4bitx2(np_from) + # No byteswap needed on big-endian machines as _pack_4bitx2() + # returns a numpy array with uint8 datatype. + input = make_tensor( + "input", from_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + elif from_type in TWO_BIT_TYPES: + np_from = np_fp32.astype(from_np_dtype) + packed = onnx.numpy_helper._pack_2bitx4(np_from) + # No byteswap needed on big-endian machines as _pack_2bitx4() + # returns a numpy array with uint8 datatype. + input = make_tensor( + "input", from_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + else: + np_from = np_fp32.astype(from_np_dtype) + input = make_tensor( + "input", from_dtype, input_shape, vals=np_from, raw=True + ) + + if to_type in F8_TYPES: + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=onnx.numpy_helper.saturate_cast(np_from, to_np_dtype), + raw=True, + ) + elif to_type in FOUR_BIT_TYPES: + packed = onnx.numpy_helper._pack_4bitx2(np_from.astype(to_np_dtype)) + # No byteswap needed on big-endian machines as _pack_4bitx2() + # returns a numpy array with uint8 datatype. + output = make_tensor( + "output", to_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + elif to_type in TWO_BIT_TYPES: + packed = onnx.numpy_helper._pack_2bitx4(np_from.astype(to_np_dtype)) + # No byteswap needed on big-endian machines as _pack_2bitx4() + # returns a numpy array with uint8 datatype. + output = make_tensor( + "output", to_dtype, input_shape, vals=packed.tobytes(), raw=True + ) + else: + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=np_from.astype(to_np_dtype), + raw=True, + ) + + like = make_tensor("like", to_dtype, (0,), vals=[]) + + node = onnx.helper.make_node( + "CastLike", + inputs=["input", "like"], + outputs=["output"], + ) + + expect( + node, + inputs=[input, like], + outputs=[output], + name="test_castlike_" + from_type + "_to_" + to_type, + ) +``` + +
+
+saturate_false + +```python +test_cases = itertools.product( + [ + "FLOAT", + "FLOAT16", + ], + [ + "FLOAT8E4M3FN", + "FLOAT8E4M3FNUZ", + "FLOAT8E5M2", + "FLOAT8E5M2FNUZ", + ], +) +input_shape = (3, 5) +for from_type, to_type in test_cases: + from_dtype = getattr(TensorProto, from_type) + to_dtype = getattr(TensorProto, to_type) + from_np_dtype = tensor_dtype_to_np_dtype(from_dtype) + to_np_dtype = tensor_dtype_to_np_dtype(to_dtype) + np_fp32 = np.array( + [ + "0.47892547", + "0.48033667", + "0.49968487", + "0.81910545", + "0.47031248", + "0.7229038", + "1000000", + "1e-7", + "NaN", + "INF", + "+INF", + "-INF", + "-0.0000001", + "0.0000001", + "-1000000", + ], + dtype=np.float32, + ) + + input = make_tensor( + "input", + from_dtype, + input_shape, + vals=np_fp32.astype(from_np_dtype), + raw=True, + ) + output = make_tensor( + "output", + to_dtype, + input_shape, + vals=np_fp32.astype(from_np_dtype).astype(to_np_dtype), + raw=True, + ) + + like = make_tensor("like", to_dtype, (0,), vals=[]) + + node = onnx.helper.make_node( + "CastLike", + inputs=["input", "like"], + outputs=["output"], + saturate=0, + ) + + expect( + node, + inputs=[input, like], + outputs=[output], + name="test_castlike_no_saturate_" + from_type + "_to_" + to_type, + ) +``` + +
+ + +### CausalConvWithState +There are 13 test cases, listed as following: +
+b1_c1_degenerate + +```python +# Mamba/GDN inner-head edge case: B=1, C=1. +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 1, 1, 6, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight) + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_b1_c1_degenerate", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+basic + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight) + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_basic", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+decode_step + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight", "bias", "past_state"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 1, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) +bias = np.random.randn(channels).astype(np.float32) +past_state = np.random.randn(batch_size, channels, k - 1).astype(np.float32) + +output, present_state = _compute( + input_, weight, bias=bias, past_state=past_state +) + +expect( + node, + inputs=[input_, weight, bias, past_state], + outputs=[output, present_state], + name="test_causal_conv_with_state_decode_step", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+fp16 + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.rand(batch_size, channels, length).astype(np.float16) +weight = np.random.rand(channels, 1, k).astype(np.float16) + +output, present_state = _compute(input_, weight) + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_fp16", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+kernel_size_one + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 1 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight) + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_kernel_size_one", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+short_input_no_past_state + +```python +# L < k-1 with no past_state: zero-pad is wider than the input. +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 2, 5 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight) + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_short_input_no_past_state", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+silu + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], + activation="silu", +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight, activation="silu") + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_silu", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+silu_fp16 + +```python +# fp16 + SiLU: the reference upcasts Sigmoid/Mul to float32, so the +# function-body expansion must do the same to stay numerically faithful. +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], + activation="silu", +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.rand(batch_size, channels, length).astype(np.float16) +weight = np.random.rand(channels, 1, k).astype(np.float16) + +output, present_state = _compute(input_, weight, activation="silu") + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_silu_fp16", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+silu_with_past_state + +```python +# Fused activation combined with concat-from-past variant of PaddedInput. +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight", "", "past_state"], + outputs=["output", "present_state"], + activation="silu", +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) +past_state = np.random.randn(batch_size, channels, k - 1).astype(np.float32) + +output, present_state = _compute( + input_, weight, past_state=past_state, activation="silu" +) + +expect( + node, + inputs=[input_, weight, past_state], + outputs=[output, present_state], + name="test_causal_conv_with_state_silu_with_past_state", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+swish_alias + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight"], + outputs=["output", "present_state"], + activation="swish", +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) + +output, present_state = _compute(input_, weight, activation="swish") + +expect( + node, + inputs=[input_, weight], + outputs=[output, present_state], + name="test_causal_conv_with_state_swish_alias", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+with_bias + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight", "bias"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) +bias = np.random.randn(channels).astype(np.float32) + +output, present_state = _compute(input_, weight, bias=bias) + +expect( + node, + inputs=[input_, weight, bias], + outputs=[output, present_state], + name="test_causal_conv_with_state_with_bias", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+with_bias_and_past_state + +```python +# Multi-token (T>1) path through Concat(past, input) -> Conv(+bias). +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight", "bias", "past_state"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) +bias = np.random.randn(channels).astype(np.float32) +past_state = np.random.randn(batch_size, channels, k - 1).astype(np.float32) + +output, present_state = _compute( + input_, weight, bias=bias, past_state=past_state +) + +expect( + node, + inputs=[input_, weight, bias, past_state], + outputs=[output, present_state], + name="test_causal_conv_with_state_with_bias_and_past_state", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+
+with_past_state + +```python +node = onnx.helper.make_node( + "CausalConvWithState", + inputs=["input", "weight", "", "past_state"], + outputs=["output", "present_state"], +) + +batch_size, channels, length, k = 2, 4, 8, 4 +input_ = np.random.randn(batch_size, channels, length).astype(np.float32) +weight = np.random.randn(channels, 1, k).astype(np.float32) +past_state = np.random.randn(batch_size, channels, k - 1).astype(np.float32) + +output, present_state = _compute(input_, weight, past_state=past_state) + +expect( + node, + inputs=[input_, weight, past_state], + outputs=[output, present_state], + name="test_causal_conv_with_state_with_past_state", + opset_imports=[onnx.helper.make_opsetid("", 27)], +) +``` + +
+ + +### Ceil +There are 1 test cases, listed as following: +
+ceil + +```python +node = onnx.helper.make_node( + "Ceil", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.5, 1.2]).astype(np.float32) +y = np.ceil(x) # expected output [-1., 2.] +expect(node, inputs=[x], outputs=[y], name="test_ceil_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.ceil(x) +expect(node, inputs=[x], outputs=[y], name="test_ceil") +``` + +
+ + +### Celu +There are 3 test cases, listed as following: +
+celu + +```python +alpha = 2.0 +node = onnx.helper.make_node( + "Celu", + inputs=["X"], + outputs=["Y"], + alpha=alpha, +) + +input_data = np.array( + [ + [ + [[0.8439683], [0.5665144], [0.05836735]], + [[0.02916367], [0.12964272], [0.5060197]], + [[0.79538304], [0.9411346], [0.9546573]], + ], + [ + [[0.17730942], [0.46192095], [0.26480448]], + [[0.6746842], [0.01665257], [0.62473077]], + [[0.9240844], [0.9722341], [0.11965699]], + ], + [ + [[0.41356155], [0.9129373], [0.59330076]], + [[0.81929934], [0.7862604], [0.11799799]], + [[0.69248444], [0.54119414], [0.07513223]], + ], + ], + dtype=np.float32, +) + +# Calculate expected output data +positive_input = np.maximum(0, input_data) +negative_input = np.minimum(0, alpha * (np.exp(input_data / alpha) - 1)) +expected_output = positive_input + negative_input + +expect(node, inputs=[input_data], outputs=[expected_output], name="test_celu") +``` + +
+
+celu_bfloat16 + +```python +alpha = 2.0 +node = onnx.helper.make_node( + "Celu", + inputs=["X"], + outputs=["Y"], + alpha=alpha, +) + +input_data = np.array([-3.0, -0.5, 0.0, 0.5, 3.0], dtype=ml_dtypes.bfloat16) + +positive_input = np.maximum(0, input_data) +negative_input = np.minimum(0, alpha * (np.exp(input_data / alpha) - 1)) +expected_output = (positive_input + negative_input).astype(ml_dtypes.bfloat16) + +expect( + node, + inputs=[input_data], + outputs=[expected_output], + name="test_celu_bfloat16", +) +``` + +
+
+celu_float16 + +```python +alpha = 2.0 +node = onnx.helper.make_node( + "Celu", + inputs=["X"], + outputs=["Y"], + alpha=alpha, +) + +input_data = np.array([-3.0, -0.5, 0.0, 0.5, 3.0], dtype=np.float16) + +positive_input = np.maximum(0, input_data) +negative_input = np.minimum(0, alpha * (np.exp(input_data / alpha) - 1)) +expected_output = (positive_input + negative_input).astype(np.float16) + +expect( + node, + inputs=[input_data], + outputs=[expected_output], + name="test_celu_float16", +) +``` + +
+ + +### CenterCropPad +There are 6 test cases, listed as following: +
+center_crop_pad_crop + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], +) + +# First dim is even diff, second is uneven +x = np.random.randn(20, 10, 3).astype(np.float32) +shape = np.array([10, 7, 3], dtype=np.int64) +y = x[5:15, 1:8, :] + +expect(node, inputs=[x, shape], outputs=[y], name="test_center_crop_pad_crop") +``` + +
+
+center_crop_pad_crop_and_pad + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], +) + +# Cropping on first dim, padding on second, third stays the same +x = np.random.randn(20, 8, 3).astype(np.float32) +shape = np.array([10, 10, 3], dtype=np.int64) +y = np.zeros([10, 10, 3], dtype=np.float32) +y[:, 1:9, :] = x[5:15, :, :] + +expect( + node, + inputs=[x, shape], + outputs=[y], + name="test_center_crop_pad_crop_and_pad", +) +``` + +
+
+center_crop_pad_crop_axes_chw + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], + axes=[1, 2], +) + +# Cropping on second dim, padding on third, first stays the same +x = np.random.randn(3, 20, 8).astype(np.float32) +shape = np.array([10, 9], dtype=np.int64) +y = np.zeros([3, 10, 9], dtype=np.float32) +y[:, :, :8] = x[:, 5:15, :] + +expect( + node, + inputs=[x, shape], + outputs=[y], + name="test_center_crop_pad_crop_axes_chw", +) +``` + +
+
+center_crop_pad_crop_axes_hwc + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], + axes=[0, 1], +) + +# Cropping on first dim, padding on second, third stays the same +x = np.random.randn(20, 8, 3).astype(np.float32) +shape = np.array([10, 9], dtype=np.int64) +y = np.zeros([10, 9, 3], dtype=np.float32) +y[:, :8, :] = x[5:15, :, :] + +expect( + node, + inputs=[x, shape], + outputs=[y], + name="test_center_crop_pad_crop_axes_hwc", +) +``` + +
+
+center_crop_pad_crop_negative_axes_hwc + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], + axes=[-3, -2], +) + +# Cropping on first dim, padding on second, third stays the same +x = np.random.randn(20, 8, 3).astype(np.float32) +shape = np.array([10, 9], dtype=np.int64) +y = np.zeros([10, 9, 3], dtype=np.float32) +y[:, :8, :] = x[5:15, :, :] + +expect( + node, + inputs=[x, shape], + outputs=[y], + name="test_center_crop_pad_crop_negative_axes_hwc", +) +``` + +
+
+center_crop_pad_pad + +```python +node = onnx.helper.make_node( + "CenterCropPad", + inputs=["x", "shape"], + outputs=["y"], +) + +# First dim is even diff, second is uneven +x = np.random.randn(10, 7, 3).astype(np.float32) +shape = np.array([20, 10, 3], dtype=np.int64) +y = np.zeros([20, 10, 3], dtype=np.float32) +y[5:15, 1:8, :] = x + +expect(node, inputs=[x, shape], outputs=[y], name="test_center_crop_pad_pad") +``` + +
+ + +### Clip +There are 3 test cases, listed as following: +
+clip + +```python +node = onnx.helper.make_node( + "Clip", + inputs=["x", "min", "max"], + outputs=["y"], +) + +x = np.array([-2, 0, 2]).astype(np.float32) +min_val = np.float32(-1) +max_val = np.float32(1) +y = np.clip(x, min_val, max_val) # expected output [-1., 0., 1.] +expect( + node, inputs=[x, min_val, max_val], outputs=[y], name="test_clip_example" +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, min_val, max_val) +expect(node, inputs=[x, min_val, max_val], outputs=[y], name="test_clip") +node = onnx.helper.make_node( + "Clip", + inputs=["x", "min", "max"], + outputs=["y"], +) + +min_val = np.float32(-5) +max_val = np.float32(5) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.array([-1, 0, 1]).astype(np.float32) +expect( + node, inputs=[x, min_val, max_val], outputs=[y], name="test_clip_inbounds" +) + +x = np.array([-6, 0, 6]).astype(np.float32) +y = np.array([-5, 0, 5]).astype(np.float32) +expect( + node, inputs=[x, min_val, max_val], outputs=[y], name="test_clip_outbounds" +) + +x = np.array([-1, 0, 6]).astype(np.float32) +y = np.array([-1, 0, 5]).astype(np.float32) +expect( + node, + inputs=[x, min_val, max_val], + outputs=[y], + name="test_clip_splitbounds", +) + +x = np.array([-2, 0, 6]).astype(np.float32) +y = np.array([1, 1, 1]).astype(np.float32) +min_val = np.float32(2) +max_val = np.float32(1) +expect( + node, + inputs=[x, min_val, max_val], + outputs=[y], + name="test_clip_min_greater_than_max", +) +``` + +
+
+clip_default + +```python +node = onnx.helper.make_node( + "Clip", + inputs=["x", "min"], + outputs=["y"], +) +min_val = np.float32(0) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, min_val, np.inf) +expect(node, inputs=[x, min_val], outputs=[y], name="test_clip_default_min") + +no_min = "" # optional input, not supplied +node = onnx.helper.make_node( + "Clip", + inputs=["x", no_min, "max"], + outputs=["y"], +) +max_val = np.float32(0) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, -np.inf, max_val) +expect(node, inputs=[x, max_val], outputs=[y], name="test_clip_default_max") + +no_max = "" # optional input, not supplied +node = onnx.helper.make_node( + "Clip", + inputs=["x", no_min, no_max], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.array([-1, 0, 1]).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_clip_default_inbounds") +``` + +
+
+clip_default_int8 + +```python +node = onnx.helper.make_node( + "Clip", + inputs=["x", "min"], + outputs=["y"], +) +min_val = np.int8(0) +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.clip(x, min_val, np.iinfo(np.int8).max) +expect( + node, inputs=[x, min_val], outputs=[y], name="test_clip_default_int8_min" +) + +no_min = "" # optional input, not supplied +node = onnx.helper.make_node( + "Clip", + inputs=["x", no_min, "max"], + outputs=["y"], +) +max_val = np.int8(0) +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.clip(x, np.iinfo(np.int8).min, max_val) +expect( + node, inputs=[x, max_val], outputs=[y], name="test_clip_default_int8_max" +) + +no_max = "" # optional input, not supplied +node = onnx.helper.make_node( + "Clip", + inputs=["x", no_min, no_max], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.int8) +y = np.array([-1, 0, 1]).astype(np.int8) +expect(node, inputs=[x], outputs=[y], name="test_clip_default_int8_inbounds") +``` + +
+ + +### Col2Im +There are 5 test cases, listed as following: +
+col2im + +```python +input = np.array( + [ + [ + [1.0, 6.0, 11.0, 16.0, 21.0], # (1, 5, 5) + [2.0, 7.0, 12.0, 17.0, 22.0], + [3.0, 8.0, 13.0, 18.0, 23.0], + [4.0, 9.0, 14.0, 19.0, 24.0], + [5.0, 0.0, 15.0, 20.0, 25.0], + ] + ] +).astype(np.float32) + +image_shape = np.array([5, 5]).astype(np.int64) +block_shape = np.array([1, 5]).astype(np.int64) +node = onnx.helper.make_node( + "Col2Im", ["input", "image_shape", "block_shape"], ["output"] +) + +output = np.array( + [ + [ + [ + [1.0, 2.0, 3.0, 4.0, 5.0], # (1, 1, 5, 5) + [6.0, 7.0, 8.0, 9.0, 0.0], + [11.0, 12.0, 13.0, 14.0, 15.0], + [16.0, 17.0, 18.0, 19.0, 20.0], + [21.0, 22.0, 23.0, 24.0, 25.0], + ] + ] + ] +).astype(np.float32) + +expect( + node, + inputs=[input, image_shape, block_shape], + outputs=[output], + name="test_col2im", +) +``` + +
+
+col2im_5d + +```python +input = np.array( + [ + [ + [1, 6, 11, 16, 21, 26, 31, 36, 41, 46, 51, 56], # (1, 10, 12) + [2, 7, 12, 17, 22, 27, 32, 37, 42, 47, 52, 57], + [3, 8, 13, 18, 23, 28, 33, 38, 43, 48, 53, 58], + [4, 9, 14, 19, 24, 29, 34, 39, 44, 49, 54, 59], + [5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60], + [61, 66, 71, 76, 81, 86, 91, 96, 101, 106, 111, 116], + [62, 67, 72, 77, 82, 87, 92, 97, 102, 107, 112, 117], + [63, 68, 73, 78, 83, 88, 93, 98, 103, 108, 113, 118], + [64, 69, 74, 79, 84, 89, 94, 99, 104, 109, 114, 119], + [65, 70, 75, 80, 85, 90, 95, 100, 105, 110, 115, 120], + ] + ] +).astype(np.float32) +image_shape = np.array([3, 4, 5]).astype(np.int64) +block_shape = np.array([1, 1, 5]).astype(np.int64) + +output = np.array( + [ + [ + [ + [ + [1, 2, 3, 4, 5], # (1, 2, 3, 4, 5) + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + ], + [ + [21, 22, 23, 24, 25], + [26, 27, 28, 29, 30], + [31, 32, 33, 34, 35], + [36, 37, 38, 39, 40], + ], + [ + [41, 42, 43, 44, 45], + [46, 47, 48, 49, 50], + [51, 52, 53, 54, 55], + [56, 57, 58, 59, 60], + ], + ], + [ + [ + [61, 62, 63, 64, 65], + [66, 67, 68, 69, 70], + [71, 72, 73, 74, 75], + [76, 77, 78, 79, 80], + ], + [ + [81, 82, 83, 84, 85], + [86, 87, 88, 89, 90], + [91, 92, 93, 94, 95], + [96, 97, 98, 99, 100], + ], + [ + [101, 102, 103, 104, 105], + [106, 107, 108, 109, 110], + [111, 112, 113, 114, 115], + [116, 117, 118, 119, 120], + ], + ], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "Col2Im", ["input", "image_shape", "block_shape"], ["output"] +) +expect( + node, + inputs=[input, image_shape, block_shape], + outputs=[output], + name="test_col2im_5d", +) +``` + +
+
+col2im_dilations + +```python +input = np.array( + [ + [ + [1.0, 5.0, 9.0, 13.0, 17], # (1, 4, 5) + [2.0, 6.0, 10.0, 14.0, 18], + [3.0, 7.0, 11.0, 15.0, 19], + [4.0, 8.0, 12.0, 16.0, 20], + ] + ] +).astype(np.float32) +image_shape = np.array([6, 6]).astype(np.int64) +block_shape = np.array([2, 2]).astype(np.int64) + +output = np.array( + [ + [ + [ + [1.0, 0.0, 0.0, 0.0, 0.0, 2.0], # (1, 1, 6, 6) + [8.0, 0.0, 0.0, 0.0, 0.0, 10.0], + [16.0, 0.0, 0.0, 0.0, 0.0, 18.0], + [24.0, 0.0, 0.0, 0.0, 0.0, 26.0], + [32.0, 0.0, 0.0, 0.0, 0.0, 34.0], + [19.0, 0.0, 0.0, 0.0, 0.0, 20.0], + ] + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "Col2Im", + ["input", "image_shape", "block_shape"], + ["output"], + dilations=[1, 5], +) +expect( + node, + inputs=[input, image_shape, block_shape], + outputs=[output], + name="test_col2im_dilations", +) +``` + +
+
+col2im_pads + +```python +input = np.array( + [ + [ + [ + 1.0, + 6.0, + 11.0, + 16.0, + 21.0, + 26, + 31, + 36, + 41, + 46, + 51, + 56, + 61, + 66, + 71, + ], # (1, 5, 15) + [ + 2.0, + 7.0, + 12.0, + 17.0, + 22.0, + 27, + 32, + 37, + 42, + 47, + 52, + 57, + 62, + 67, + 72, + ], + [ + 3.0, + 8.0, + 13.0, + 18.0, + 23.0, + 28, + 33, + 38, + 43, + 48, + 53, + 58, + 63, + 68, + 73, + ], + [ + 4.0, + 9.0, + 14.0, + 19.0, + 24.0, + 29, + 34, + 39, + 44, + 49, + 54, + 59, + 64, + 69, + 74, + ], + [ + 5.0, + 10.0, + 15.0, + 20.0, + 25.0, + 30, + 35, + 40, + 45, + 50, + 55, + 60, + 65, + 70, + 75, + ], + ] + ] +).astype(np.float32) +image_shape = np.array([5, 5]).astype(np.int64) +block_shape = np.array([1, 5]).astype(np.int64) + +output = np.array( + [ + [ + [ + [8.0, 21.0, 24.0, 27.0, 24.0], # (1, 1, 5, 5) + [38.0, 66.0, 69.0, 72.0, 54.0], + [68.0, 111.0, 114.0, 117.0, 84.0], + [98.0, 156.0, 159.0, 162.0, 114.0], + [128.0, 201.0, 204.0, 207.0, 144.0], + ] + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "Col2Im", + ["input", "image_shape", "block_shape"], + ["output"], + pads=[0, 1, 0, 1], +) +expect( + node, + inputs=[input, image_shape, block_shape], + outputs=[output], + name="test_col2im_pads", +) +``` + +
+
+col2im_strides + +```python +input = np.array( + [ + [ + [0.0, 0.0, 0.0, 0.0], # (1, 9, 4) + [1.0, 1.0, 1.0, 1.0], + [1.0, 1.0, 1.0, 1.0], + [1.0, 1.0, 1.0, 1.0], + [0.0, 0.0, 0.0, 0.0], + [0.0, 0.0, 0.0, 0.0], + [0.0, 0.0, 0.0, 0.0], + [1.0, 1.0, 1.0, 1.0], + [0.0, 0.0, 0.0, 0.0], + ] + ] +).astype(np.float32) +image_shape = np.array([5, 5]).astype(np.int64) +block_shape = np.array([3, 3]).astype(np.int64) + +output = np.array( + [ + [ + [ + [0.0, 1.0, 1.0, 1.0, 1.0], # (1, 1, 5, 5) + [1.0, 0.0, 1.0, 0.0, 0.0], + [0.0, 2.0, 1.0, 2.0, 1.0], + [1.0, 0.0, 1.0, 0.0, 0.0], + [0.0, 1.0, 0.0, 1.0, 0.0], + ] + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "Col2Im", + ["input", "image_shape", "block_shape"], + ["output"], + strides=[2, 2], +) +expect( + node, + inputs=[input, image_shape, block_shape], + outputs=[output], + name="test_col2im_strides", +) +``` + +
+ + +### Compress +There are 4 test cases, listed as following: +
+compress_0 + +```python +node = onnx.helper.make_node( + "Compress", + inputs=["input", "condition"], + outputs=["output"], + axis=0, +) +input = np.array([[1, 2], [3, 4], [5, 6]]).astype(np.float32) +condition = np.array([0, 1, 1]) +output = np.compress(condition, input, axis=0) +# print(output) +# [[ 3. 4.] +# [ 5. 6.]] + +expect( + node, + inputs=[input, condition.astype(bool)], + outputs=[output], + name="test_compress_0", +) +``` + +
+
+compress_1 + +```python +node = onnx.helper.make_node( + "Compress", + inputs=["input", "condition"], + outputs=["output"], + axis=1, +) +input = np.array([[1, 2], [3, 4], [5, 6]]).astype(np.float32) +condition = np.array([0, 1]) +output = np.compress(condition, input, axis=1) +# print(output) +# [[ 2.] +# [ 4.] +# [ 6.]] + +expect( + node, + inputs=[input, condition.astype(bool)], + outputs=[output], + name="test_compress_1", +) +``` + +
+
+compress_default_axis + +```python +node = onnx.helper.make_node( + "Compress", + inputs=["input", "condition"], + outputs=["output"], +) +input = np.array([[1, 2], [3, 4], [5, 6]]).astype(np.float32) +condition = np.array([0, 1, 0, 0, 1]) +output = np.compress(condition, input) +# print(output) +# [ 2., 5.] + +expect( + node, + inputs=[input, condition.astype(bool)], + outputs=[output], + name="test_compress_default_axis", +) +``` + +
+
+compress_negative_axis + +```python +node = onnx.helper.make_node( + "Compress", + inputs=["input", "condition"], + outputs=["output"], + axis=-1, +) +input = np.array([[1, 2], [3, 4], [5, 6]]).astype(np.float32) +condition = np.array([0, 1]) +output = np.compress(condition, input, axis=-1) +# print(output) +# [[ 2.] +# [ 4.] +# [ 6.]] +expect( + node, + inputs=[input, condition.astype(bool)], + outputs=[output], + name="test_compress_negative_axis", +) +``` + +
+ + +### Concat +There are 1 test cases, listed as following: +
+concat + +```python +test_cases: dict[str, Sequence[Any]] = { + "1d": ([1, 2], [3, 4]), + "2d": ([[1, 2], [3, 4]], [[5, 6], [7, 8]]), + "3d": ( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]]], + [[[9, 10], [11, 12]], [[13, 14], [15, 16]]], + ), +} + +for test_case, values_ in test_cases.items(): + values = [np.asarray(v, dtype=np.float32) for v in values_] + for i in range(len(values[0].shape)): + in_args = ["value" + str(k) for k in range(len(values))] + node = onnx.helper.make_node( + "Concat", inputs=list(in_args), outputs=["output"], axis=i + ) + output = np.concatenate(values, i) + expect( + node, + inputs=list(values), + outputs=[output], + name="test_concat_" + test_case + "_axis_" + str(i), + ) + + for i in range(-len(values[0].shape), 0): + in_args = ["value" + str(k) for k in range(len(values))] + node = onnx.helper.make_node( + "Concat", inputs=list(in_args), outputs=["output"], axis=i + ) + output = np.concatenate(values, i) + expect( + node, + inputs=list(values), + outputs=[output], + name="test_concat_" + test_case + "_axis_negative_" + str(abs(i)), + ) +``` + +
+ + +### Constant +There are 1 test cases, listed as following: +
+constant + +```python +values = np.random.randn(5, 5).astype(np.float32) +node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["values"], + value=onnx.helper.make_tensor( + name="const_tensor", + data_type=onnx.TensorProto.FLOAT, + dims=values.shape, + vals=values.flatten().astype(float), + ), +) + +expect(node, inputs=[], outputs=[values], name="test_constant") +``` + +
+ + +### ConstantOfShape +There are 3 test cases, listed as following: +
+float_ones + +```python +x = np.array([4, 3, 2]).astype(np.int64) +tensor_value = onnx.helper.make_tensor( + "value", onnx.TensorProto.FLOAT, [1], [1] +) +node = onnx.helper.make_node( + "ConstantOfShape", + inputs=["x"], + outputs=["y"], + value=tensor_value, +) + +y = np.ones(x, dtype=np.float32) +expect(node, inputs=[x], outputs=[y], name="test_constantofshape_float_ones") +``` + +
+
+int32_shape_zero + +```python +x = np.array( + [ + 0, + ] +).astype(np.int64) +tensor_value = onnx.helper.make_tensor( + "value", onnx.TensorProto.INT32, [1], [0] +) +node = onnx.helper.make_node( + "ConstantOfShape", + inputs=["x"], + outputs=["y"], + value=tensor_value, +) +y = np.zeros(x, dtype=np.int32) +expect( + node, inputs=[x], outputs=[y], name="test_constantofshape_int_shape_zero" +) +``` + +
+
+int32_zeros + +```python +x = np.array([10, 6]).astype(np.int64) +tensor_value = onnx.helper.make_tensor( + "value", onnx.TensorProto.INT32, [1], [0] +) +node = onnx.helper.make_node( + "ConstantOfShape", + inputs=["x"], + outputs=["y"], + value=tensor_value, +) +y = np.zeros(x, dtype=np.int32) +expect(node, inputs=[x], outputs=[y], name="test_constantofshape_int_zeros") +``` + +
+ + +### Conv +There are 3 test cases, listed as following: +
+conv + +```python +x = np.array( + [ + [ + [ + [0.0, 1.0, 2.0, 3.0, 4.0], # (1, 1, 5, 5) input tensor + [5.0, 6.0, 7.0, 8.0, 9.0], + [10.0, 11.0, 12.0, 13.0, 14.0], + [15.0, 16.0, 17.0, 18.0, 19.0], + [20.0, 21.0, 22.0, 23.0, 24.0], + ] + ] + ] +).astype(np.float32) +W = np.array( + [ + [ + [ + [1.0, 1.0, 1.0], # (1, 1, 3, 3) tensor for convolution weights + [1.0, 1.0, 1.0], + [1.0, 1.0, 1.0], + ] + ] + ] +).astype(np.float32) + +# Convolution with padding +node_with_padding = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + kernel_shape=[3, 3], + # Default values for other attributes: strides=[1, 1], dilations=[1, 1], groups=1 + pads=[1, 1, 1, 1], +) +y_with_padding = np.array( + [ + [ + [ + [12.0, 21.0, 27.0, 33.0, 24.0], # (1, 1, 5, 5) output tensor + [33.0, 54.0, 63.0, 72.0, 51.0], + [63.0, 99.0, 108.0, 117.0, 81.0], + [93.0, 144.0, 153.0, 162.0, 111.0], + [72.0, 111.0, 117.0, 123.0, 84.0], + ] + ] + ] +).astype(np.float32) +expect( + node_with_padding, + inputs=[x, W], + outputs=[y_with_padding], + name="test_basic_conv_with_padding", +) + +# Convolution without padding +node_without_padding = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + kernel_shape=[3, 3], + # Default values for other attributes: strides=[1, 1], dilations=[1, 1], groups=1 + pads=[0, 0, 0, 0], +) +y_without_padding = np.array( + [ + [ + [ + [54.0, 63.0, 72.0], # (1, 1, 3, 3) output tensor + [99.0, 108.0, 117.0], + [144.0, 153.0, 162.0], + ] + ] + ] +).astype(np.float32) +expect( + node_without_padding, + inputs=[x, W], + outputs=[y_without_padding], + name="test_basic_conv_without_padding", +) +``` + +
+
+conv_with_autopad_same + +```python +x = np.array( + [ + [ + [ + [0.0, 1.0, 2.0, 3.0, 4.0], # (1, 1, 5, 5) input tensor + [5.0, 6.0, 7.0, 8.0, 9.0], + [10.0, 11.0, 12.0, 13.0, 14.0], + [15.0, 16.0, 17.0, 18.0, 19.0], + [20.0, 21.0, 22.0, 23.0, 24.0], + ] + ] + ] +).astype(np.float32) +W = np.array( + [ + [ + [ + [1.0, 1.0, 1.0], # (1, 1, 3, 3) tensor for convolution weights + [1.0, 1.0, 1.0], + [1.0, 1.0, 1.0], + ] + ] + ] +).astype(np.float32) + +# Convolution with auto_pad='SAME_LOWER' and strides=2 +node = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + auto_pad="SAME_LOWER", + kernel_shape=[3, 3], + strides=[2, 2], +) +y = np.array( + [[[[12.0, 27.0, 24.0], [63.0, 108.0, 81.0], [72.0, 117.0, 84.0]]]] +).astype(np.float32) +expect(node, inputs=[x, W], outputs=[y], name="test_conv_with_autopad_same") +``` + +
+
+conv_with_strides + +```python +x = np.array( + [ + [ + [ + [0.0, 1.0, 2.0, 3.0, 4.0], # (1, 1, 7, 5) input tensor + [5.0, 6.0, 7.0, 8.0, 9.0], + [10.0, 11.0, 12.0, 13.0, 14.0], + [15.0, 16.0, 17.0, 18.0, 19.0], + [20.0, 21.0, 22.0, 23.0, 24.0], + [25.0, 26.0, 27.0, 28.0, 29.0], + [30.0, 31.0, 32.0, 33.0, 34.0], + ] + ] + ] +).astype(np.float32) +W = np.array( + [ + [ + [ + [1.0, 1.0, 1.0], # (1, 1, 3, 3) tensor for convolution weights + [1.0, 1.0, 1.0], + [1.0, 1.0, 1.0], + ] + ] + ] +).astype(np.float32) + +# Convolution with strides=2 and padding +node_with_padding = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[1, 1, 1, 1], + strides=[ + 2, + 2, + ], # Default values for other attributes: dilations=[1, 1], groups=1 +) +y_with_padding = np.array( + [ + [ + [ + [12.0, 27.0, 24.0], # (1, 1, 4, 3) output tensor + [63.0, 108.0, 81.0], + [123.0, 198.0, 141.0], + [112.0, 177.0, 124.0], + ] + ] + ] +).astype(np.float32) +expect( + node_with_padding, + inputs=[x, W], + outputs=[y_with_padding], + name="test_conv_with_strides_padding", +) + +# Convolution with strides=2 and no padding +node_without_padding = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[0, 0, 0, 0], + strides=[ + 2, + 2, + ], # Default values for other attributes: dilations=[1, 1], groups=1 +) +y_without_padding = np.array( + [ + [ + [ + [54.0, 72.0], # (1, 1, 3, 2) output tensor + [144.0, 162.0], + [234.0, 252.0], + ] + ] + ] +).astype(np.float32) +expect( + node_without_padding, + inputs=[x, W], + outputs=[y_without_padding], + name="test_conv_with_strides_no_padding", +) + +# Convolution with strides=2 and padding only along one dimension (the H dimension in NxCxHxW tensor) +node_with_asymmetric_padding = onnx.helper.make_node( + "Conv", + inputs=["x", "W"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[1, 0, 1, 0], + strides=[ + 2, + 2, + ], # Default values for other attributes: dilations=[1, 1], groups=1 +) +y_with_asymmetric_padding = np.array( + [ + [ + [ + [21.0, 33.0], # (1, 1, 4, 2) output tensor + [99.0, 117.0], + [189.0, 207.0], + [171.0, 183.0], + ] + ] + ] +).astype(np.float32) +expect( + node_with_asymmetric_padding, + inputs=[x, W], + outputs=[y_with_asymmetric_padding], + name="test_conv_with_strides_and_asymmetric_padding", +) +``` + +
+ + +### ConvInteger +There are 2 test cases, listed as following: +
+with_padding + +```python +x = ( + np.array([2, 3, 4, 5, 6, 7, 8, 9, 10]) + .astype(np.uint8) + .reshape((1, 1, 3, 3)) +) +x_zero_point = np.uint8(1) +w_zero_points = np.array([0, 1], dtype=np.uint8) +w = np.array([1, 1, 1, 1, 1, 1, 1, 1]).astype(np.uint8).reshape((2, 1, 2, 2)) + +y = ( + np.array( + [ + 1, + 3, + 5, + 3, + 5, + 12, + 16, + 9, + 11, + 24, + 28, + 15, + 7, + 15, + 17, + 9, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + ] + ) + .astype(np.int32) + .reshape((1, 2, 4, 4)) +) + +# ConvInteger with padding +convinteger_node_with_padding = onnx.helper.make_node( + "ConvInteger", + inputs=["x", "w", "x_zero_point", "w_zero_points"], + outputs=["y"], + pads=[1, 1, 1, 1], +) + +expect( + convinteger_node_with_padding, + inputs=[x, w, x_zero_point, w_zero_points], + outputs=[y], + name="test_convinteger_with_padding", +) +``` + +
+
+without_padding + +```python +x = ( + np.array([2, 3, 4, 5, 6, 7, 8, 9, 10]) + .astype(np.uint8) + .reshape((1, 1, 3, 3)) +) +x_zero_point = np.uint8(1) +w = np.array([1, 1, 1, 1]).astype(np.uint8).reshape((1, 1, 2, 2)) + +y = np.array([12, 16, 24, 28]).astype(np.int32).reshape(1, 1, 2, 2) + +# ConvInteger without padding +convinteger_node = onnx.helper.make_node( + "ConvInteger", inputs=["x", "w", "x_zero_point"], outputs=["y"] +) + +expect( + convinteger_node, + inputs=[x, w, x_zero_point], + outputs=[y], + name="test_convinteger_without_padding", +) +``` + +
+ + +### ConvTranspose +There are 9 test cases, listed as following: +
+convtranspose + +```python +x = np.array( + [[[[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]]]] # (1, 1, 3, 3) +).astype(np.float32) + +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], # (1, 2, 3, 3) + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node("ConvTranspose", ["X", "W"], ["Y"]) + +y = np.array( + [ + [ + [ + [0.0, 1.0, 3.0, 3.0, 2.0], # (1, 2, 5, 5) + [3.0, 8.0, 15.0, 12.0, 7.0], + [9.0, 21.0, 36.0, 27.0, 15.0], + [9.0, 20.0, 33.0, 24.0, 13.0], + [6.0, 13.0, 21.0, 15.0, 8.0], + ], + [ + [0.0, 1.0, 3.0, 3.0, 2.0], + [3.0, 8.0, 15.0, 12.0, 7.0], + [9.0, 21.0, 36.0, 27.0, 15.0], + [9.0, 20.0, 33.0, 24.0, 13.0], + [6.0, 13.0, 21.0, 15.0, 8.0], + ], + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose") +``` + +
+
+convtranspose_1d + +```python +x = np.array([[[0.0, 1.0, 2.0]]]).astype(np.float32) # (1, 1, 3) + +W = np.array([[[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]]]).astype( # (1, 2, 3) + np.float32 +) + +node = onnx.helper.make_node("ConvTranspose", ["X", "W"], ["Y"]) + +y = np.array( + [[[0.0, 1.0, 3.0, 3.0, 2.0], [0.0, 1.0, 3.0, 3.0, 2.0]]] # (1, 2, 5) +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_1d") +``` + +
+
+convtranspose_3d + +```python +x = np.array( + [ + [ + [ + [ + [0.0, 1.0, 2.0, 3.0, 4.0], # (1, 1, 3, 4, 5) + [5.0, 6.0, 7.0, 8.0, 9.0], + [10.0, 11.0, 12.0, 13.0, 14.0], + [15.0, 16.0, 17.0, 18.0, 19.0], + ], + [ + [20.0, 21.0, 22.0, 23.0, 24.0], + [25.0, 26.0, 27.0, 28.0, 29.0], + [30.0, 31.0, 32.0, 33.0, 34.0], + [35.0, 36.0, 37.0, 38.0, 39.0], + ], + [ + [40.0, 41.0, 42.0, 43.0, 44.0], + [45.0, 46.0, 47.0, 48.0, 49.0], + [50.0, 51.0, 52.0, 53.0, 54.0], + [55.0, 56.0, 57.0, 58.0, 59.0], + ], + ] + ] + ] +).astype(np.float32) + +W = np.array( + [ + [ + [ + [ + [1.0, 1.0, 1.0], # (1, 2, 3, 3, 3) + [1.0, 1.0, 1.0], + [1.0, 1.0, 1.0], + ], + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node("ConvTranspose", ["X", "W"], ["Y"]) + +y = np.array( + [ + [ + [ + [ + [0.0, 1.0, 3.0, 6.0, 9.0, 7.0, 4.0], # (1, 2, 5, 6, 7) + [5.0, 12.0, 21.0, 27.0, 33.0, 24.0, 13.0], + [15.0, 33.0, 54.0, 63.0, 72.0, 51.0, 27.0], + [30.0, 63.0, 99.0, 108.0, 117.0, 81.0, 42.0], + [25.0, 52.0, 81.0, 87.0, 93.0, 64.0, 33.0], + [15.0, 31.0, 48.0, 51.0, 54.0, 37.0, 19.0], + ], + [ + [20.0, 42.0, 66.0, 72.0, 78.0, 54.0, 28.0], + [50.0, 104.0, 162.0, 174.0, 186.0, 128.0, 66.0], + [90.0, 186.0, 288.0, 306.0, 324.0, 222.0, 114.0], + [120.0, 246.0, 378.0, 396.0, 414.0, 282.0, 144.0], + [90.0, 184.0, 282.0, 294.0, 306.0, 208.0, 106.0], + [50.0, 102.0, 156.0, 162.0, 168.0, 114.0, 58.0], + ], + [ + [60.0, 123.0, 189.0, 198.0, 207.0, 141.0, 72.0], + [135.0, 276.0, 423.0, 441.0, 459.0, 312.0, 159.0], + [225.0, 459.0, 702.0, 729.0, 756.0, 513.0, 261.0], + [270.0, 549.0, 837.0, 864.0, 891.0, 603.0, 306.0], + [195.0, 396.0, 603.0, 621.0, 639.0, 432.0, 219.0], + [105.0, 213.0, 324.0, 333.0, 342.0, 231.0, 117.0], + ], + [ + [60.0, 122.0, 186.0, 192.0, 198.0, 134.0, 68.0], + [130.0, 264.0, 402.0, 414.0, 426.0, 288.0, 146.0], + [210.0, 426.0, 648.0, 666.0, 684.0, 462.0, 234.0], + [240.0, 486.0, 738.0, 756.0, 774.0, 522.0, 264.0], + [170.0, 344.0, 522.0, 534.0, 546.0, 368.0, 186.0], + [90.0, 182.0, 276.0, 282.0, 288.0, 194.0, 98.0], + ], + [ + [40.0, 81.0, 123.0, 126.0, 129.0, 87.0, 44.0], + [85.0, 172.0, 261.0, 267.0, 273.0, 184.0, 93.0], + [135.0, 273.0, 414.0, 423.0, 432.0, 291.0, 147.0], + [150.0, 303.0, 459.0, 468.0, 477.0, 321.0, 162.0], + [105.0, 212.0, 321.0, 327.0, 333.0, 224.0, 113.0], + [55.0, 111.0, 168.0, 171.0, 174.0, 117.0, 59.0], + ], + ], + [ + [ + [0.0, 1.0, 3.0, 6.0, 9.0, 7.0, 4.0], + [5.0, 12.0, 21.0, 27.0, 33.0, 24.0, 13.0], + [15.0, 33.0, 54.0, 63.0, 72.0, 51.0, 27.0], + [30.0, 63.0, 99.0, 108.0, 117.0, 81.0, 42.0], + [25.0, 52.0, 81.0, 87.0, 93.0, 64.0, 33.0], + [15.0, 31.0, 48.0, 51.0, 54.0, 37.0, 19.0], + ], + [ + [20.0, 42.0, 66.0, 72.0, 78.0, 54.0, 28.0], + [50.0, 104.0, 162.0, 174.0, 186.0, 128.0, 66.0], + [90.0, 186.0, 288.0, 306.0, 324.0, 222.0, 114.0], + [120.0, 246.0, 378.0, 396.0, 414.0, 282.0, 144.0], + [90.0, 184.0, 282.0, 294.0, 306.0, 208.0, 106.0], + [50.0, 102.0, 156.0, 162.0, 168.0, 114.0, 58.0], + ], + [ + [60.0, 123.0, 189.0, 198.0, 207.0, 141.0, 72.0], + [135.0, 276.0, 423.0, 441.0, 459.0, 312.0, 159.0], + [225.0, 459.0, 702.0, 729.0, 756.0, 513.0, 261.0], + [270.0, 549.0, 837.0, 864.0, 891.0, 603.0, 306.0], + [195.0, 396.0, 603.0, 621.0, 639.0, 432.0, 219.0], + [105.0, 213.0, 324.0, 333.0, 342.0, 231.0, 117.0], + ], + [ + [60.0, 122.0, 186.0, 192.0, 198.0, 134.0, 68.0], + [130.0, 264.0, 402.0, 414.0, 426.0, 288.0, 146.0], + [210.0, 426.0, 648.0, 666.0, 684.0, 462.0, 234.0], + [240.0, 486.0, 738.0, 756.0, 774.0, 522.0, 264.0], + [170.0, 344.0, 522.0, 534.0, 546.0, 368.0, 186.0], + [90.0, 182.0, 276.0, 282.0, 288.0, 194.0, 98.0], + ], + [ + [40.0, 81.0, 123.0, 126.0, 129.0, 87.0, 44.0], + [85.0, 172.0, 261.0, 267.0, 273.0, 184.0, 93.0], + [135.0, 273.0, 414.0, 423.0, 432.0, 291.0, 147.0], + [150.0, 303.0, 459.0, 468.0, 477.0, 321.0, 162.0], + [105.0, 212.0, 321.0, 327.0, 333.0, 224.0, 113.0], + [55.0, 111.0, 168.0, 171.0, 174.0, 117.0, 59.0], + ], + ], + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_3d") +``` + +
+
+convtranspose_attributes + +```python +x = np.array( + [[[[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]]]] # (1, 1, 3, 3) +).astype(np.float32) + +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], # (1, 2, 3, 3) + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ] + ] +).astype(np.float32) + +y = np.array( + [ + [ + [ + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], # (1, 2, 10, 8) + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], + ], + [ + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0, 2.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0, 5.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0, 8.0, 0.0], + [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], + ], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "ConvTranspose", ["X", "W"], ["Y"], strides=[3, 2], output_shape=[10, 8] +) +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_output_shape") + +node = onnx.helper.make_node( + "ConvTranspose", ["X", "W"], ["Y"], strides=[3, 2], output_padding=[1, 1] +) +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_pad") + +node = onnx.helper.make_node( + "ConvTranspose", + ["X", "W"], + ["Y"], + name="test", + strides=[3, 2], + output_shape=[10, 8], + kernel_shape=[3, 3], + output_padding=[1, 1], +) +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_kernel_shape") +``` + +
+
+convtranspose_autopad_same + +```python +x = np.array( + [[[[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]]]] # (1, 1, 3, 3) +).astype(np.float32) + +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], # (1, 2, 3, 3) + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "ConvTranspose", ["X", "W"], ["Y"], auto_pad="SAME_UPPER", strides=[2, 2] +) + +y = np.array( + [ + [ + [ + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0], + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0], + [3.0, 3.0, 8.0, 5.0, 12.0, 7.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0], + [9.0, 9.0, 20.0, 11.0, 24.0, 13.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0], + ], + [ + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0], + [0.0, 0.0, 1.0, 1.0, 3.0, 2.0], + [3.0, 3.0, 8.0, 5.0, 12.0, 7.0], + [3.0, 3.0, 7.0, 4.0, 9.0, 5.0], + [9.0, 9.0, 20.0, 11.0, 24.0, 13.0], + [6.0, 6.0, 13.0, 7.0, 15.0, 8.0], + ], + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_autopad_same") +``` + +
+
+convtranspose_dilations + +```python +x = np.array( + [[[[3.0, 8.0, 1.0], [9.0, 5.0, 7.0], [3.0, 2.0, 6.0]]]] # (1, 1, 3, 3) +).astype(np.float32) +W = np.array([[[[7.0, 2.0], [1.0, 9.0]]]]).astype(np.float32) # (1, 1, 2, 2) + +node = onnx.helper.make_node( + "ConvTranspose", ["X", "W"], ["Y"], dilations=[2, 2] +) + +y = np.array( + [ + [ + [ + [21.0, 56.0, 13.0, 16.0, 2.0], # [1, 1, 5, 5] + [63.0, 35.0, 67.0, 10.0, 14.0], + [24.0, 22.0, 76.0, 76.0, 21.0], + [9.0, 5.0, 88.0, 45.0, 63.0], + [3.0, 2.0, 33.0, 18.0, 54.0], + ] + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_dilations") +``` + +
+
+convtranspose_group_2 + +```python +x = np.array( + [ + [ + [[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0], [15.0, 16.0, 17.0]], + ] + ] +).astype(np.float32) +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + ] +).astype(np.float32) + +node = onnx.helper.make_node("ConvTranspose", ["X", "W"], ["Y"], group=2) + +y = np.array( + [ + [ + [ + [0.0, 1.0, 3.0, 3.0, 2.0], + [3.0, 8.0, 15.0, 12.0, 7.0], + [9.0, 21.0, 36.0, 27.0, 15.0], + [9.0, 20.0, 33.0, 24.0, 13.0], + [6.0, 13.0, 21.0, 15.0, 8.0], + ], + [ + [9.0, 19.0, 30.0, 21.0, 11.0], + [21.0, 44.0, 69.0, 48.0, 25.0], + [36.0, 75.0, 117.0, 81.0, 42.0], + [27.0, 56.0, 87.0, 60.0, 31.0], + [15.0, 31.0, 48.0, 33.0, 17.0], + ], + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_group_2") +``` + +
+
+convtranspose_group_2_image_3 + +```python +x = np.array( + [ + [ + [[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0], [15.0, 16.0, 17.0]], + ], + [ + [[18.0, 19.0, 20.0], [21.0, 22.0, 23.0], [24.0, 25.0, 26.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0], [15.0, 16.0, 17.0]], + ], + [ + [[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0], [15.0, 16.0, 17.0]], + ], + ] +).astype(np.float32) +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ], + ] +).astype(np.float32) + +node = onnx.helper.make_node("ConvTranspose", ["X", "W"], ["Y"], group=2) + +y = np.array( + [ + [ + [ + [0.0, 1.0, 3.0, 3.0, 2.0], + [3.0, 8.0, 15.0, 12.0, 7.0], + [9.0, 21.0, 36.0, 27.0, 15.0], + [9.0, 20.0, 33.0, 24.0, 13.0], + [6.0, 13.0, 21.0, 15.0, 8.0], + ], + [ + [9.0, 19.0, 30.0, 21.0, 11.0], + [21.0, 44.0, 69.0, 48.0, 25.0], + [36.0, 75.0, 117.0, 81.0, 42.0], + [27.0, 56.0, 87.0, 60.0, 31.0], + [15.0, 31.0, 48.0, 33.0, 17.0], + ], + ], + [ + [ + [18.0, 37.0, 57.0, 39.0, 20.0], + [39.0, 80.0, 123.0, 84.0, 43.0], + [63.0, 129.0, 198.0, 135.0, 69.0], + [45.0, 92.0, 141.0, 96.0, 49.0], + [24.0, 49.0, 75.0, 51.0, 26.0], + ], + [ + [9.0, 19.0, 30.0, 21.0, 11.0], + [21.0, 44.0, 69.0, 48.0, 25.0], + [36.0, 75.0, 117.0, 81.0, 42.0], + [27.0, 56.0, 87.0, 60.0, 31.0], + [15.0, 31.0, 48.0, 33.0, 17.0], + ], + ], + [ + [ + [0.0, 1.0, 3.0, 3.0, 2.0], + [3.0, 8.0, 15.0, 12.0, 7.0], + [9.0, 21.0, 36.0, 27.0, 15.0], + [9.0, 20.0, 33.0, 24.0, 13.0], + [6.0, 13.0, 21.0, 15.0, 8.0], + ], + [ + [9.0, 19.0, 30.0, 21.0, 11.0], + [21.0, 44.0, 69.0, 48.0, 25.0], + [36.0, 75.0, 117.0, 81.0, 42.0], + [27.0, 56.0, 87.0, 60.0, 31.0], + [15.0, 31.0, 48.0, 33.0, 17.0], + ], + ], + ] +).astype(np.float32) + +expect( + node, inputs=[x, W], outputs=[y], name="test_convtranspose_group_2_image_3" +) +``` + +
+
+convtranspose_pads + +```python +x = np.array( + [[[[0.0, 1.0, 2.0], [3.0, 4.0, 5.0], [6.0, 7.0, 8.0]]]] # (1, 1, 3, 3) +).astype(np.float32) + +W = np.array( + [ + [ + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], # (1, 2, 3, 3) + [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], + ] + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "ConvTranspose", ["X", "W"], ["Y"], strides=[3, 2], pads=[1, 2, 1, 2] +) + +y = np.array( + [ + [ + [ + [1.0, 1.0, 3.0], # (1, 2, 7, 3) + [1.0, 1.0, 3.0], + [7.0, 4.0, 9.0], + [7.0, 4.0, 9.0], + [7.0, 4.0, 9.0], + [13.0, 7.0, 15.0], + [13.0, 7.0, 15.0], + ], + [ + [1.0, 1.0, 3.0], + [1.0, 1.0, 3.0], + [7.0, 4.0, 9.0], + [7.0, 4.0, 9.0], + [7.0, 4.0, 9.0], + [13.0, 7.0, 15.0], + [13.0, 7.0, 15.0], + ], + ] + ] +).astype(np.float32) + +expect(node, inputs=[x, W], outputs=[y], name="test_convtranspose_pads") +``` + +
+ + +### Cos +There are 1 test cases, listed as following: +
+cos + +```python +node = onnx.helper.make_node( + "Cos", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.cos(x) +expect(node, inputs=[x], outputs=[y], name="test_cos_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.cos(x) +expect(node, inputs=[x], outputs=[y], name="test_cos") +``` + +
+ + +### Cosh +There are 1 test cases, listed as following: +
+cosh + +```python +node = onnx.helper.make_node( + "Cosh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.cosh(x) # expected output [1.54308069, 1., 1.54308069] +expect(node, inputs=[x], outputs=[y], name="test_cosh_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.cosh(x) +expect(node, inputs=[x], outputs=[y], name="test_cosh") +``` + +
+ + +### CumProd +There are 9 test cases, listed as following: +
+cumprod_1d + +```python +node = onnx.helper.make_node("CumProd", inputs=["x", "axis"], outputs=["y"]) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.array(0, dtype=np.int32) +y = np.array([1.0, 2.0, 6.0, 24.0, 120.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_1d") +``` + +
+
+cumprod_1d_exclusive + +```python +node = onnx.helper.make_node( + "CumProd", inputs=["x", "axis"], outputs=["y"], exclusive=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.array(0, dtype=np.int32) +y = np.array([1.0, 1.0, 2.0, 6.0, 24.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_1d_exclusive") +``` + +
+
+cumprod_1d_int32_exclusive + +```python +node = onnx.helper.make_node( + "CumProd", inputs=["x", "axis"], outputs=["y"], exclusive=1 +) +x = np.array([1, 2, 3, 4, 5]).astype(np.int32) +axis = np.array(0, dtype=np.int32) +y = np.array([1, 1, 2, 6, 24]).astype(np.int32) +expect( + node, inputs=[x, axis], outputs=[y], name="test_cumprod_1d_int32_exclusive" +) +``` + +
+
+cumprod_1d_reverse + +```python +node = onnx.helper.make_node( + "CumProd", inputs=["x", "axis"], outputs=["y"], reverse=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.array(0, dtype=np.int32) +y = np.array([120.0, 120.0, 60.0, 20.0, 5.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_1d_reverse") +``` + +
+
+cumprod_1d_reverse_exclusive + +```python +node = onnx.helper.make_node( + "CumProd", inputs=["x", "axis"], outputs=["y"], reverse=1, exclusive=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.array(0, dtype=np.int32) +y = np.array([120.0, 60.0, 20.0, 5.0, 1.0]).astype(np.float64) +expect( + node, + inputs=[x, axis], + outputs=[y], + name="test_cumprod_1d_reverse_exclusive", +) +``` + +
+
+cumprod_2d_axis_0 + +```python +node = onnx.helper.make_node( + "CumProd", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.array(0, dtype=np.int32) +y = ( + np.array([1.0, 2.0, 3.0, 4.0, 10.0, 18.0]) + .astype(np.float64) + .reshape((2, 3)) +) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_2d_axis_0") +``` + +
+
+cumprod_2d_axis_1 + +```python +node = onnx.helper.make_node( + "CumProd", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.array(1, dtype=np.int32) +y = ( + np.array([1.0, 2.0, 6.0, 4.0, 20.0, 120.0]) + .astype(np.float64) + .reshape((2, 3)) +) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_2d_axis_1") +``` + +
+
+cumprod_2d_int32 + +```python +node = onnx.helper.make_node( + "CumProd", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1, 2, 3, 4, 5, 6]).astype(np.int32).reshape((2, 3)) +axis = np.array(0, dtype=np.int32) +y = np.array([1, 2, 3, 4, 10, 18]).astype(np.int32).reshape((2, 3)) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumprod_2d_int32") +``` + +
+
+cumprod_2d_negative_axis + +```python +node = onnx.helper.make_node( + "CumProd", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.array(-1, dtype=np.int32) +y = ( + np.array([1.0, 2.0, 6.0, 4.0, 20.0, 120.0]) + .astype(np.float64) + .reshape((2, 3)) +) +expect( + node, inputs=[x, axis], outputs=[y], name="test_cumprod_2d_negative_axis" +) +``` + +
+ + +### CumSum +There are 9 test cases, listed as following: +
+cumsum_1d + +```python +node = onnx.helper.make_node("CumSum", inputs=["x", "axis"], outputs=["y"]) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.int32(0) +y = np.array([1.0, 3.0, 6.0, 10.0, 15.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_1d") +``` + +
+
+cumsum_1d_exclusive + +```python +node = onnx.helper.make_node( + "CumSum", inputs=["x", "axis"], outputs=["y"], exclusive=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.int32(0) +y = np.array([0.0, 1.0, 3.0, 6.0, 10.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_1d_exclusive") +``` + +
+
+cumsum_1d_int32_exclusive + +```python +node = onnx.helper.make_node( + "CumSum", inputs=["x", "axis"], outputs=["y"], exclusive=1 +) +x = np.array([1, 2, 3, 4, 5]).astype(np.int32) +axis = np.int32(0) +y = np.array([0, 1, 3, 6, 10]).astype(np.int32) +expect( + node, inputs=[x, axis], outputs=[y], name="test_cumsum_1d_int32_exclusive" +) +``` + +
+
+cumsum_1d_reverse + +```python +node = onnx.helper.make_node( + "CumSum", inputs=["x", "axis"], outputs=["y"], reverse=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.int32(0) +y = np.array([15.0, 14.0, 12.0, 9.0, 5.0]).astype(np.float64) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_1d_reverse") +``` + +
+
+cumsum_1d_reverse_exclusive + +```python +node = onnx.helper.make_node( + "CumSum", inputs=["x", "axis"], outputs=["y"], reverse=1, exclusive=1 +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0]).astype(np.float64) +axis = np.int32(0) +y = np.array([14.0, 12.0, 9.0, 5.0, 0.0]).astype(np.float64) +expect( + node, inputs=[x, axis], outputs=[y], name="test_cumsum_1d_reverse_exclusive" +) +``` + +
+
+cumsum_2d_axis_0 + +```python +node = onnx.helper.make_node( + "CumSum", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.int32(0) +y = np.array([1.0, 2.0, 3.0, 5.0, 7.0, 9.0]).astype(np.float64).reshape((2, 3)) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_2d_axis_0") +``` + +
+
+cumsum_2d_axis_1 + +```python +node = onnx.helper.make_node( + "CumSum", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.int32(1) +y = np.array([1.0, 3.0, 6.0, 4.0, 9.0, 15.0]).astype(np.float64).reshape((2, 3)) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_2d_axis_1") +``` + +
+
+cumsum_2d_int32 + +```python +node = onnx.helper.make_node( + "CumSum", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1, 2, 3, 4, 5, 6]).astype(np.int32).reshape((2, 3)) +axis = np.int32(0) +y = np.array([1, 2, 3, 5, 7, 9]).astype(np.int32).reshape((2, 3)) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_2d_int32") +``` + +
+
+cumsum_2d_negative_axis + +```python +node = onnx.helper.make_node( + "CumSum", + inputs=["x", "axis"], + outputs=["y"], +) +x = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float64).reshape((2, 3)) +axis = np.int32(-1) +y = np.array([1.0, 3.0, 6.0, 4.0, 9.0, 15.0]).astype(np.float64).reshape((2, 3)) +expect(node, inputs=[x, axis], outputs=[y], name="test_cumsum_2d_negative_axis") +``` + +
+ + +### DFT +There are 2 test cases, listed as following: +
+dft + +```python +node = onnx.helper.make_node("DFT", inputs=["x", "", "axis"], outputs=["y"]) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +axis = np.array(1, dtype=np.int64) +y = np.fft.fft(x, axis=0) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect(node, inputs=[x, axis], outputs=[y], name="test_dft") + +node = onnx.helper.make_node("DFT", inputs=["x", "", "axis"], outputs=["y"]) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +axis = np.array(2, dtype=np.int64) +y = np.fft.fft(x, axis=1) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect(node, inputs=[x, axis], outputs=[y], name="test_dft_axis") + +node = onnx.helper.make_node( + "DFT", inputs=["x", "", "axis"], outputs=["y"], inverse=1 +) +x = np.arange(0, 100, dtype=np.complex64).reshape(10, 10) +axis = np.array(1, dtype=np.int64) +y = np.fft.ifft(x, axis=0) + +x = np.stack((x.real, x.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect(node, inputs=[x, axis], outputs=[y], name="test_dft_inverse") + +# Test RFFT (Real FFT): real input -> one-sided complex output +node = onnx.helper.make_node( + "DFT", inputs=["x", "", "axis"], outputs=["y"], onesided=1 +) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +axis = np.array(1, dtype=np.int64) +y = np.fft.rfft(x, axis=0) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 6, 10, 2) +expect(node, inputs=[x, axis], outputs=[y], name="test_dft_rfft") + +# Test IRFFT (Inverse Real FFT): one-sided complex input -> real output +node = onnx.helper.make_node( + "DFT", inputs=["x", "", "axis"], outputs=["y"], onesided=1, inverse=1 +) +# Create one-sided complex input (6 bins for signal length 10) +x = np.fft.rfft(np.arange(0, 100).reshape(10, 10), axis=0).astype(np.complex64) +axis = np.array(1, dtype=np.int64) +y = np.fft.irfft(x, n=10, axis=0) + +x = np.stack((x.real, x.imag), axis=2).astype(np.float32).reshape(1, 6, 10, 2) +y = y.reshape(1, 10, 10, 1).astype(np.float32) +expect(node, inputs=[x, axis], outputs=[y], name="test_dft_irfft") +``` + +
+
+opset19 + +```python +node = onnx.helper.make_node("DFT", inputs=["x"], outputs=["y"], axis=1) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +y = np.fft.fft(x, axis=0) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect( + node, + inputs=[x], + outputs=[y], + name="test_dft_opset19", + opset_imports=[onnx.helper.make_opsetid("", 19)], +) + +node = onnx.helper.make_node("DFT", inputs=["x"], outputs=["y"], axis=2) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +y = np.fft.fft(x, axis=1) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect( + node, + inputs=[x], + outputs=[y], + name="test_dft_axis_opset19", + opset_imports=[onnx.helper.make_opsetid("", 19)], +) + +node = onnx.helper.make_node( + "DFT", inputs=["x"], outputs=["y"], inverse=1, axis=1 +) +x = np.arange(0, 100, dtype=np.complex64).reshape( + 10, + 10, +) +y = np.fft.ifft(x, axis=0) + +x = np.stack((x.real, x.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 10, 10, 2) +expect( + node, + inputs=[x], + outputs=[y], + name="test_dft_inverse_opset19", + opset_imports=[onnx.helper.make_opsetid("", 19)], +) + +# Test RFFT (Real FFT): real input -> one-sided complex output +node = onnx.helper.make_node( + "DFT", inputs=["x"], outputs=["y"], onesided=1, axis=1 +) +x = np.arange(0, 100).reshape(10, 10).astype(np.float32) +y = np.fft.rfft(x, axis=0) + +x = x.reshape(1, 10, 10, 1) +y = np.stack((y.real, y.imag), axis=2).astype(np.float32).reshape(1, 6, 10, 2) +expect( + node, + inputs=[x], + outputs=[y], + name="test_dft_rfft_opset19", + opset_imports=[onnx.helper.make_opsetid("", 19)], +) + +# Test IRFFT (Inverse Real FFT): one-sided complex input -> real output +node = onnx.helper.make_node( + "DFT", inputs=["x"], outputs=["y"], onesided=1, inverse=1, axis=1 +) +# Create one-sided complex input (6 bins for signal length 10) +x = np.fft.rfft(np.arange(0, 100).reshape(10, 10), axis=0).astype(np.complex64) +y = np.fft.irfft(x, n=10, axis=0) + +x = np.stack((x.real, x.imag), axis=2).astype(np.float32).reshape(1, 6, 10, 2) +y = y.reshape(1, 10, 10, 1).astype(np.float32) +expect( + node, + inputs=[x], + outputs=[y], + name="test_dft_irfft_opset19", + opset_imports=[onnx.helper.make_opsetid("", 19)], +) +``` + +
+ + +### DeformConv +There are 3 test cases, listed as following: +
+deformconv + +```python +X = np.arange(9).astype(np.float32) +X.shape = (1, 1, 3, 3) +W = np.ones((1, 1, 2, 2), dtype=np.float32) + +# Convolution with padding +offset_with_padding = np.zeros((1, 8, 4, 4), dtype=np.float32) +# h-coord of [0, 0] element of kernel, at output position [0, 0] +offset_with_padding[0, 0, 0, 0] = 0.5 +# w-coord of [1, 0] element of kernel, at output position [1, 2] +offset_with_padding[0, 5, 1, 2] = -0.1 + +node_with_padding = onnx.helper.make_node( + "DeformConv", + inputs=["X", "W", "offset_with_padding"], + outputs=["Y_with_padding"], + kernel_shape=[2, 2], + pads=[1, 1, 1, 1], +) +Y_with_padding = np.array( + [ + [ + [ + [0.0, 1.0, 3.0, 2.0], # (1, 1, 4, 4) output tensor + [3.0, 8.0, 11.9, 7.0], + [9.0, 20.0, 24.0, 13.0], + [6.0, 13.0, 15.0, 8.0], + ] + ] + ] +).astype(np.float32) +expect( + node_with_padding, + inputs=[X, W, offset_with_padding], + outputs=[Y_with_padding], + name="test_basic_deform_conv_with_padding", +) + +# Convolution without padding +offset_without_padding = np.zeros((1, 8, 2, 2), dtype=np.float32) +# h-coord of [0, 0] element of kernel, at output position [0, 0] +offset_without_padding[0, 0, 0, 0] = 0.5 +# w-coord of [1, 0] element of kernel, at output position [0, 1] +offset_without_padding[0, 5, 0, 1] = -0.1 + +node_without_padding = onnx.helper.make_node( + "DeformConv", + inputs=["X", "W", "offset_without_padding"], + outputs=["Y_without_padding"], + kernel_shape=[2, 2], + pads=[0, 0, 0, 0], +) +Y_without_padding = np.array( + [ + [ + [ + [9.5, 11.9], # (1, 1, 2, 2) output tensor + [20.0, 24.0], + ] + ] + ] +).astype(np.float32) +expect( + node_without_padding, + inputs=[X, W, offset_without_padding], + outputs=[Y_without_padding], + name="test_basic_deform_conv_without_padding", +) +``` + +
+
+deformconv_with_mask_bias + +```python +X = np.arange(9).astype(np.float32) +X.shape = (1, 1, 3, 3) +W = np.ones((1, 1, 2, 2), dtype=np.float32) +B = np.ones((1,), dtype=np.float32) + +offset = np.zeros((1, 8, 2, 2), dtype=np.float32) +# h-coord of [0, 0] element of kernel, at output position [0, 0] +offset[0, 0, 0, 0] = 0.5 +# w-coord of [1, 0] element of kernel, at output position [0, 1] +offset[0, 5, 0, 1] = -0.1 + +mask = np.ones((1, 4, 2, 2), dtype=np.float32) +mask[0, 2, 1, 1] = 0.2 # [1, 0] element of kernel at output position [1, 1] + +node = onnx.helper.make_node( + "DeformConv", + inputs=["X", "W", "offset", "B", "mask"], + outputs=["Y"], + kernel_shape=[2, 2], + pads=[0, 0, 0, 0], +) +Y = np.array( + [ + [ + [ + [10.5, 12.9], # (1, 1, 2, 2) output tensor + [21.0, 19.4], + ] + ] + ] +).astype(np.float32) +expect( + node, + inputs=[X, W, offset, B, mask], + outputs=[Y], + name="test_deform_conv_with_mask_bias", +) +``` + +
+
+deformconv_with_multiple_offset_groups + +```python +X = np.zeros((1, 2, 3, 3), dtype=np.float32) +X[0, 0] = np.reshape(np.arange(9).astype(np.float32), (3, 3)) +X[0, 1] = np.reshape(np.arange(8, -1, -1).astype(np.float32), (3, 3)) +X.shape = (1, 2, 3, 3) +W = np.ones((1, 2, 2, 2), dtype=np.float32) + +offset = np.zeros((1, 16, 2, 2), dtype=np.float32) +# h-coord of [0, 0] element of kernel in channel 0, at output position [0, 0] +offset[0, 0, 0, 0] = 0.5 +# w-coord of [1, 0] element of kernel in channel 1, at output position [0, 1] +offset[0, 13, 0, 1] = -0.1 + +node = onnx.helper.make_node( + "DeformConv", + inputs=["X", "W", "offset"], + outputs=["Y"], + kernel_shape=[2, 2], + pads=[0, 0, 0, 0], + offset_group=2, +) +Y = np.array( + [ + [ + [ + [33.5, 32.1], # (1, 1, 2, 2) output tensor + [32.0, 32.0], + ] + ] + ] +).astype(np.float32) +expect( + node, + inputs=[X, W, offset], + outputs=[Y], + name="test_deform_conv_with_multiple_offset_groups", +) +``` + +
+ + +### DepthToSpace +There are 2 test cases, listed as following: +
+crd_mode_example + +```python +node = onnx.helper.make_node( + "DepthToSpace", inputs=["x"], outputs=["y"], blocksize=2, mode="CRD" +) + +# (1, 8, 2, 3) input tensor +x = np.array( + [ + [ + [[0.0, 1.0, 2.0], [3.0, 4.0, 5.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0]], + [[18.0, 19.0, 20.0], [21.0, 22.0, 23.0]], + [[27.0, 28.0, 29.0], [30.0, 31.0, 32.0]], + [[36.0, 37.0, 38.0], [39.0, 40.0, 41.0]], + [[45.0, 46.0, 47.0], [48.0, 49.0, 50.0]], + [[54.0, 55.0, 56.0], [57.0, 58.0, 59.0]], + [[63.0, 64.0, 65.0], [66.0, 67.0, 68.0]], + ] + ] +).astype(np.float32) + +# (1, 2, 4, 6) output tensor +y = np.array( + [ + [ + [ + [0.0, 9.0, 1.0, 10.0, 2.0, 11.0], + [18.0, 27.0, 19.0, 28.0, 20.0, 29.0], + [3.0, 12.0, 4.0, 13.0, 5.0, 14.0], + [21.0, 30.0, 22.0, 31.0, 23.0, 32.0], + ], + [ + [36.0, 45.0, 37.0, 46.0, 38.0, 47.0], + [54.0, 63.0, 55.0, 64.0, 56.0, 65.0], + [39.0, 48.0, 40.0, 49.0, 41.0, 50.0], + [57.0, 66.0, 58.0, 67.0, 59.0, 68.0], + ], + ] + ] +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_depthtospace_crd_mode_example") +``` + +
+
+default_mode_example + +```python +node = onnx.helper.make_node( + "DepthToSpace", inputs=["x"], outputs=["y"], blocksize=2, mode="DCR" +) + +# (1, 8, 2, 3) input tensor +x = np.array( + [ + [ + [[0.0, 1.0, 2.0], [3.0, 4.0, 5.0]], + [[9.0, 10.0, 11.0], [12.0, 13.0, 14.0]], + [[18.0, 19.0, 20.0], [21.0, 22.0, 23.0]], + [[27.0, 28.0, 29.0], [30.0, 31.0, 32.0]], + [[36.0, 37.0, 38.0], [39.0, 40.0, 41.0]], + [[45.0, 46.0, 47.0], [48.0, 49.0, 50.0]], + [[54.0, 55.0, 56.0], [57.0, 58.0, 59.0]], + [[63.0, 64.0, 65.0], [66.0, 67.0, 68.0]], + ] + ] +).astype(np.float32) + +# (1, 2, 4, 6) output tensor +y = np.array( + [ + [ + [ + [0.0, 18.0, 1.0, 19.0, 2.0, 20.0], + [36.0, 54.0, 37.0, 55.0, 38.0, 56.0], + [3.0, 21.0, 4.0, 22.0, 5.0, 23.0], + [39.0, 57.0, 40.0, 58.0, 41.0, 59.0], + ], + [ + [9.0, 27.0, 10.0, 28.0, 11.0, 29.0], + [45.0, 63.0, 46.0, 64.0, 47.0, 65.0], + [12.0, 30.0, 13.0, 31.0, 14.0, 32.0], + [48.0, 66.0, 49.0, 67.0, 50.0, 68.0], + ], + ] + ] +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_depthtospace_example") +``` + +
+ + +### DequantizeLinear +There are 14 test cases, listed as following: +
+axis + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], +) + +# 1-D tensor zero point and scale of size equal to axis 1 of the input tensor +x = np.array( + [ + [ + [[3, 89], [34, 200], [74, 59]], + [[5, 24], [24, 87], [32, 13]], + [[245, 99], [4, 142], [121, 102]], + ], + ], + dtype=np.uint8, +) +x_scale = np.array([2, 4, 5], dtype=np.float32) +x_zero_point = np.array([84, 24, 196], dtype=np.uint8) +y = ( + x.astype(np.float32) - x_zero_point.reshape(1, 3, 1, 1).astype(np.float32) +) * x_scale.reshape(1, 3, 1, 1) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_axis", +) +``` + +
+
+blocked + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=1, + block_size=2, +) + +x = np.array( + [ + [ + [[3, 89], [34, 200], [74, 59]], + [[5, 24], [24, 87], [32, 13]], + [[5, 12], [12, 33], [65, 42]], + [[245, 99], [4, 142], [121, 102]], + ], + ], + dtype=np.uint8, +) + +x_scale = np.array( + [ + [ + [[3.0, 2.0], [4.0, 1.0], [2.0, 2.0]], + [[5.0, 2.0], [4.0, 3.0], [5.0, 2.0]], + ], + ], + dtype=np.float32, +) +x_zero_point = np.array( + [ + [ + [[1, 0], [0, 1], [2, 20]], + [[3, 2], [4, 3], [15, 2]], + ], + ], + dtype=np.uint8, +) + +# x.shape = (1, 4, 3, 2) +# x_scale.shape = (1, 2, 3, 2) +assert x_scale.shape == x_zero_point.shape +block_axis = 1 +# The block shape is [x.shape[i] // x_scale.shape[i] for i in range(len(x.shape))] = (1, 2, 1, 1) +assert all( + x.shape[i] == x_scale.shape[i] + for i in range(len(x.shape)) + if i != block_axis +) +assert x.shape[block_axis] % x_scale.shape[block_axis] == 0 +repeats = x.shape[block_axis] // x_scale.shape[block_axis] + +# Create element-wise scale and zero point +x_scale_elementwise = np.repeat(x_scale, repeats=repeats, axis=block_axis) +x_zero_point_elementwise = np.repeat( + x_zero_point, repeats=repeats, axis=block_axis +) + +y = ( + x.astype(np.float32) - x_zero_point_elementwise.astype(np.float32) +) * x_scale_elementwise + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_blocked", +) +``` + +
+
+dequantizelinear + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], +) + +# scalar zero point and scale +x = np.array([0, 3, 128, 255]).astype(np.uint8) +x_scale = np.float32(2) +x_zero_point = np.uint8(128) +y = np.array([-256, -250, 0, 254], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear", +) +``` + +
+
+e4m3fn + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.FLOAT8E4M3FN, [5], [0, 0.5, 1, 448, -104]) +x_scale = np.float32(2) +y = np.array([0.0, 1.0, 2.0, 896.0, -208.0], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale], + outputs=[y], + name="test_dequantizelinear_e4m3fn", +) +``` + +
+
+e4m3fn_float16 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.FLOAT8E4M3FN, [5], [0, 0.5, 1, 448, -104]) +x_scale = np.float16(2) +y = np.array([0.0, 1.0, 2.0, 896.0, -208.0], dtype=np.float16) + +expect( + node, + inputs=[x, x_scale], + outputs=[y], + name="test_dequantizelinear_e4m3fn_float16", +) +``` + +
+
+e4m3fn_zero_point + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.FLOAT8E4M3FN, [5], [0, 0.5, 1, 448, -104]) +zero_point = make_tensor("zero_point", TensorProto.FLOAT8E4M3FN, [1], [0]) +x_scale = np.float32(2) +y = np.array([0.0, 1.0, 2.0, 896.0, -208.0], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, zero_point], + outputs=[y], + name="test_dequantizelinear_e4m3fn_zero_point", +) +``` + +
+
+e5m2 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.FLOAT8E5M2, [5], [0, 0.5, 1, 49152, -96]) +x_scale = np.float32(2) +y = np.array([0.0, 1.0, 2.0, 98304.0, -192.0], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale], + outputs=[y], + name="test_dequantizelinear_e5m2", +) +``` + +
+
+float4e2m1 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.FLOAT4E2M1, [5], [0, 1, -1, 1.5, -4]) +x_scale = np.float32(2) +x_zero_point = make_tensor("x_zero_point", TensorProto.FLOAT4E2M1, (1,), [0]) +y = np.array([0, 2, -2, 3, -8], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_float4e2m1", +) +``` + +
+
+int16 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], +) + +x = np.array([-300, -30, -1025, 1270]).astype(np.int16) +x_scale = np.float32(2) +x_zero_point = np.int16(-1024) +y = np.array([1448.0, 1988.0, -2.0, 4588.0], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_int16", +) +``` + +
+
+int2 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.INT2, [4], [0, 1, -1, -2]) +x_scale = np.float32(2) +x_zero_point = make_tensor("x_zero_point", TensorProto.INT2, (1,), [1]) +y = np.array([-2, 0, -4, -6], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_int2", +) +``` + +
+
+int4 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.INT4, [5], [0, 1, 7, -4, -8]) +x_scale = np.float32(2) +x_zero_point = make_tensor("x_zero_point", TensorProto.INT4, (1,), [1]) +y = np.array([-2, 0, 12, -10, -18], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_int4", +) +``` + +
+
+uint16 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], +) + +x = np.array([30000, 31000, 32768, 33000]).astype(np.uint16) +x_scale = np.float32(2) +x_zero_point = np.uint16(32767) +y = np.array([-5534.0, -3534.0, 2.0, 466.0], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_uint16", +) +``` + +
+
+uint2 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.UINT2, [4], [0, 1, 2, 3]) +x_scale = np.float32(2) +x_zero_point = make_tensor("x_zero_point", TensorProto.UINT2, (1,), [1]) +y = np.array([-2, 0, 2, 4], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_uint2", +) +``` + +
+
+uint4 + +```python +node = onnx.helper.make_node( + "DequantizeLinear", + inputs=["x", "x_scale", "x_zero_point"], + outputs=["y"], + axis=0, +) + +# scalar zero point and scale +x = make_tensor("x", TensorProto.UINT4, [5], [0, 1, 7, 10, 15]) +x_scale = np.float32(2) +x_zero_point = make_tensor("x_zero_point", TensorProto.UINT4, (1,), [1]) +y = np.array([-2, 0, 12, 18, 28], dtype=np.float32) + +expect( + node, + inputs=[x, x_scale, x_zero_point], + outputs=[y], + name="test_dequantizelinear_uint4", +) +``` + +
+ + +### Det +There are 2 test cases, listed as following: +
+2d + +```python +node = onnx.helper.make_node( + "Det", + inputs=["x"], + outputs=["y"], +) + +x = np.arange(4).reshape(2, 2).astype(np.float32) +y = np.linalg.det(x) # expect -2 +expect(node, inputs=[x], outputs=[y], name="test_det_2d") +``` + +
+
+nd + +```python +node = onnx.helper.make_node( + "Det", + inputs=["x"], + outputs=["y"], +) + +x = np.array([[[1, 2], [3, 4]], [[1, 2], [2, 1]], [[1, 3], [3, 1]]]).astype( + np.float32 +) +y = np.linalg.det(x) # expect array([-2., -3., -8.]) +expect(node, inputs=[x], outputs=[y], name="test_det_nd") +``` + +
+ + +### Div +There are 2 test cases, listed as following: +
+div + +```python +node = onnx.helper.make_node( + "Div", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([3, 4]).astype(np.float32) +y = np.array([1, 2]).astype(np.float32) +z = x / y # expected output [3., 2.] +expect(node, inputs=[x, y], outputs=[z], name="test_div_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.rand(3, 4, 5).astype(np.float32) + 1.0 +z = x / y +expect(node, inputs=[x, y], outputs=[z], name="test_div") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.int8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int8) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_int8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.int16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int16) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_int16") + +x = np.array([-3, 3, -3, 3], dtype=np.int32) +y = np.array([2, 2, -2, -2], dtype=np.int32) +z = np.array([-1, 1, 1, -1], dtype=np.int32) +expect(node, inputs=[x, y], outputs=[z], name="test_div_int32_trunc") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) + 1 +z = x // y +expect(node, inputs=[x, y], outputs=[z], name="test_div_uint64") +``` + +
+
+div_broadcast + +```python +node = onnx.helper.make_node( + "Div", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.rand(5).astype(np.float32) + 1.0 +z = x / y +expect(node, inputs=[x, y], outputs=[z], name="test_div_bcast") +``` + +
+ + +### Dropout +There are 12 test cases, listed as following: +
+default + +```python +seed = np.int64(0) +node = onnx.helper.make_node("Dropout", inputs=["x"], outputs=["y"], seed=seed) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = dropout(x) +expect(node, inputs=[x], outputs=[y], name="test_dropout_default") +``` + +
+
+default_mask + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x"], outputs=["y", "z"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y, z = dropout(x, return_mask=True) +expect(node, inputs=[x], outputs=[y, z], name="test_dropout_default_mask") +``` + +
+
+default_mask_ratio + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r"], outputs=["y", "z"], seed=seed +) + +r = np.float32(0.1) +x = np.random.randn(3, 4, 5).astype(np.float32) +y, z = dropout(x, r, return_mask=True) +expect( + node, inputs=[x, r], outputs=[y, z], name="test_dropout_default_mask_ratio" +) +``` + +
+
+default_old + +```python +node = onnx.helper.make_node( + "Dropout", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = x +expect( + node, + inputs=[x], + outputs=[y], + name="test_dropout_default_old", + opset_imports=[helper.make_opsetid("", 11)], +) +``` + +
+
+default_ratio + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r"], outputs=["y"], seed=seed +) + +r = np.float32(0.1) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = dropout(x, r) +expect(node, inputs=[x, r], outputs=[y], name="test_dropout_default_ratio") +``` + +
+
+random_old + +```python +node = onnx.helper.make_node( + "Dropout", + inputs=["x"], + outputs=["y"], + ratio=0.2, +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = x +expect( + node, + inputs=[x], + outputs=[y], + name="test_dropout_random_old", + opset_imports=[helper.make_opsetid("", 11)], +) +``` + +
+
+training + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.75) +t = np.bool_(True) +y = dropout(x, r, training_mode=t) +expect(node, inputs=[x, r, t], outputs=[y], name="test_training_dropout") +``` + +
+
+training_default + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.5) +t = np.bool_(True) +y = dropout(x, r, training_mode=t) +expect( + node, inputs=[x, r, t], outputs=[y], name="test_training_dropout_default" +) +``` + +
+
+training_default_ratio_mask + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y", "z"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.5) +t = np.bool_(True) +y, z = dropout(x, r, training_mode=t, return_mask=True) +expect( + node, + inputs=[x, r, t], + outputs=[y, z], + name="test_training_dropout_default_mask", +) +``` + +
+
+training_default_zero_ratio + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.0) +t = np.bool_(True) +y = dropout(x, r, training_mode=t) +expect( + node, inputs=[x, r, t], outputs=[y], name="test_training_dropout_zero_ratio" +) +``` + +
+
+training_default_zero_ratio_mask + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y", "z"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.0) +t = np.bool_(True) +y, z = dropout(x, r, training_mode=t, return_mask=True) +expect( + node, + inputs=[x, r, t], + outputs=[y, z], + name="test_training_dropout_zero_ratio_mask", +) +``` + +
+
+training_ratio_mask + +```python +seed = np.int64(0) +node = onnx.helper.make_node( + "Dropout", inputs=["x", "r", "t"], outputs=["y", "z"], seed=seed +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +r = np.float32(0.75) +t = np.bool_(True) +y, z = dropout(x, r, training_mode=t, return_mask=True) +expect( + node, inputs=[x, r, t], outputs=[y, z], name="test_training_dropout_mask" +) +``` + +
+ + +### DynamicQuantizeLinear +There are 1 test cases, listed as following: +
+dynamicquantizelinear + +```python +node = onnx.helper.make_node( + "DynamicQuantizeLinear", + inputs=["x"], + outputs=["y", "y_scale", "y_zero_point"], +) + +# expected scale 0.0196078438 and zero point 153 +X = np.array([0, 2, -3, -2.5, 1.34, 0.5]).astype(np.float32) +x_min = np.minimum(0, np.min(X)) +x_max = np.maximum(0, np.max(X)) +Y_Scale = np.float32((x_max - x_min) / (255 - 0)) # uint8 -> [0, 255] +Y_ZeroPoint = np.clip(round((0 - x_min) / Y_Scale), 0, 255).astype(np.uint8) +Y = np.clip(np.round(X / Y_Scale) + Y_ZeroPoint, 0, 255).astype(np.uint8) + +expect( + node, + inputs=[X], + outputs=[Y, Y_Scale, Y_ZeroPoint], + name="test_dynamicquantizelinear", +) + +# expected scale 0.0156862754 and zero point 255 +X = np.array([-1.0, -2.1, -1.3, -2.5, -3.34, -4.0]).astype(np.float32) +x_min = np.minimum(0, np.min(X)) +x_max = np.maximum(0, np.max(X)) +Y_Scale = np.float32((x_max - x_min) / (255 - 0)) # uint8 -> [0, 255] +Y_ZeroPoint = np.clip(round((0 - x_min) / Y_Scale), 0, 255).astype(np.uint8) +Y = np.clip(np.round(X / Y_Scale) + Y_ZeroPoint, 0, 255).astype(np.uint8) + +expect( + node, + inputs=[X], + outputs=[Y, Y_Scale, Y_ZeroPoint], + name="test_dynamicquantizelinear_max_adjusted", +) + +X = ( + np.array([1, 2.1, 1.3, 2.5, 3.34, 4.0, 1.5, 2.6, 3.9, 4.0, 3.0, 2.345]) + .astype(np.float32) + .reshape((3, 4)) +) + +# expected scale 0.0156862754 and zero point 0 +x_min = np.minimum(0, np.min(X)) +x_max = np.maximum(0, np.max(X)) +Y_Scale = np.float32((x_max - x_min) / (255 - 0)) # uint8 -> [0, 255] +Y_ZeroPoint = np.clip(round((0 - x_min) / Y_Scale), 0, 255).astype(np.uint8) +Y = np.clip(np.round(X / Y_Scale) + Y_ZeroPoint, 0, 255).astype(np.uint8) + +expect( + node, + inputs=[X], + outputs=[Y, Y_Scale, Y_ZeroPoint], + name="test_dynamicquantizelinear_min_adjusted", +) +``` + +
+ + +### Einsum +There are 6 test cases, listed as following: +
+einsum_batch_diagonal + +```python +Eqn = "...ii ->...i" +node = onnx.helper.make_node( + "Einsum", inputs=["x"], outputs=["y"], equation=Eqn +) + +X = np.random.randn(3, 5, 5) +Z = einsum_reference_implementation(Eqn, (X,)) + +expect(node, inputs=[X], outputs=[Z], name="test_einsum_batch_diagonal") +``` + +
+
+einsum_batch_matmul + +```python +Eqn = "bij, bjk -> bik" +node = onnx.helper.make_node( + "Einsum", inputs=["x", "y"], outputs=["z"], equation=Eqn +) + +X = np.random.randn(5, 2, 3) +Y = np.random.randn(5, 3, 4) +Z = einsum_reference_implementation(Eqn, (X, Y)) + +expect(node, inputs=[X, Y], outputs=[Z], name="test_einsum_batch_matmul") +``` + +
+
+einsum_inner_prod + +```python +Eqn = "i,i" +node = onnx.helper.make_node( + "Einsum", inputs=["x", "y"], outputs=["z"], equation=Eqn +) + +X = np.random.randn(5) +Y = np.random.randn(5) +Z = einsum_reference_implementation(Eqn, (X, Y)) + +expect(node, inputs=[X, Y], outputs=[Z], name="test_einsum_inner_prod") +``` + +
+
+einsum_scalar + +```python +Eqn = "->" +node = onnx.helper.make_node( + "Einsum", inputs=["x"], outputs=["y"], equation=Eqn +) + +X = np.array(5.0) # scalar input +Z = einsum_reference_implementation(Eqn, (X,)) + +expect(node, inputs=[X], outputs=[Z], name="test_einsum_scalar") +``` + +
+
+einsum_sum + +```python +Eqn = "ij->i" +node = onnx.helper.make_node( + "Einsum", inputs=["x"], outputs=["y"], equation=Eqn +) + +X = np.random.randn(3, 4) +Z = einsum_reference_implementation(Eqn, (X,)) + +expect(node, inputs=[X], outputs=[Z], name="test_einsum_sum") +``` + +
+
+einsum_transpose + +```python +Eqn = "ij->ji" +node = onnx.helper.make_node( + "Einsum", inputs=["x"], outputs=["y"], equation=Eqn +) + +X = np.random.randn(3, 4) +Y = einsum_reference_implementation(Eqn, (X,)) + +expect(node, inputs=[X], outputs=[Y], name="test_einsum_transpose") +``` + +
+ + +### Elu +There are 2 test cases, listed as following: +
+elu + +```python +node = onnx.helper.make_node("Elu", inputs=["x"], outputs=["y"], alpha=2.0) + +x = np.array([-1, 0, 1]).astype(np.float32) +# expected output [-1.2642411, 0., 1.] +y = np.clip(x, 0, np.inf) + (np.exp(np.clip(x, -np.inf, 0)) - 1) * 2.0 +expect(node, inputs=[x], outputs=[y], name="test_elu_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + (np.exp(np.clip(x, -np.inf, 0)) - 1) * 2.0 +expect(node, inputs=[x], outputs=[y], name="test_elu") +``` + +
+
+elu_default + +```python +default_alpha = 1.0 +node = onnx.helper.make_node( + "Elu", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + (np.exp(np.clip(x, -np.inf, 0)) - 1) * default_alpha +expect(node, inputs=[x], outputs=[y], name="test_elu_default") +``` + +
+ + +### Equal +There are 4 test cases, listed as following: +
+equal + +```python +node = onnx.helper.make_node( + "Equal", + inputs=["x", "y"], + outputs=["z"], +) + +x = (np.random.randn(3, 4, 5) * 10).astype(np.int32) +y = (np.random.randn(3, 4, 5) * 10).astype(np.int32) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal") + +x = (np.random.randn(3, 4, 5) * 10).astype(np.int8) +y = (np.random.randn(3, 4, 5) * 10).astype(np.int8) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_int8") + +x = (np.random.randn(3, 4, 5) * 10).astype(np.int16) +y = (np.random.randn(3, 4, 5) * 10).astype(np.int16) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_uint64") +``` + +
+
+equal_broadcast + +```python +node = onnx.helper.make_node( + "Equal", + inputs=["x", "y"], + outputs=["z"], +) + +x = (np.random.randn(3, 4, 5) * 10).astype(np.int32) +y = (np.random.randn(5) * 10).astype(np.int32) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_bcast") +``` + +
+
+equal_string + +```python +node = onnx.helper.make_node( + "Equal", + inputs=["x", "y"], + outputs=["z"], +) +x = np.array(["string1", "string2"], dtype=np.dtype(object)) +y = np.array(["string1", "string3"], dtype=np.dtype(object)) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_string") +``` + +
+
+equal_string_broadcast + +```python +node = onnx.helper.make_node( + "Equal", + inputs=["x", "y"], + outputs=["z"], +) +x = np.array(["string1", "string2"], dtype=np.dtype(object)) +y = np.array(["string1"], dtype=np.dtype(object)) +z = np.equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_equal_string_broadcast") +``` + +
+ + +### Erf +There are 1 test cases, listed as following: +
+erf + +```python +node = onnx.helper.make_node( + "Erf", + inputs=["x"], + outputs=["y"], +) + +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +y = np.vectorize(math.erf)(x).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_erf") +``` + +
+ + +### Exp +There are 1 test cases, listed as following: +
+exp + +```python +node = onnx.helper.make_node( + "Exp", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.exp(x) # expected output [0.36787945, 1., 2.71828175] +expect(node, inputs=[x], outputs=[y], name="test_exp_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.exp(x) +expect(node, inputs=[x], outputs=[y], name="test_exp") +``` + +
+ + +### Expand +There are 2 test cases, listed as following: +
+dim_changed + +```python +node = onnx.helper.make_node( + "Expand", + inputs=["data", "new_shape"], + outputs=["expanded"], +) +shape = [3, 1] +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[1.], [2.], [3.]] +new_shape = [2, 1, 6] +expanded = data * np.ones(new_shape, dtype=np.float32) +# print(expanded) +# [[[1., 1., 1., 1., 1., 1.], +# [2., 2., 2., 2., 2., 2.], +# [3., 3., 3., 3., 3., 3.]], +# +# [[1., 1., 1., 1., 1., 1.], +# [2., 2., 2., 2., 2., 2.], +# [3., 3., 3., 3., 3., 3.]]] +new_shape = np.array(new_shape, dtype=np.int64) +expect( + node, + inputs=[data, new_shape], + outputs=[expanded], + name="test_expand_dim_changed", +) +``` + +
+
+dim_unchanged + +```python +node = onnx.helper.make_node( + "Expand", + inputs=["data", "new_shape"], + outputs=["expanded"], +) +shape = [3, 1] +new_shape = [3, 4] +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[1.], [2.], [3.]] +expanded = np.tile(data, 4) +# print(expanded) +# [[1., 1., 1., 1.], +# [2., 2., 2., 2.], +# [3., 3., 3., 3.]] +new_shape = np.array(new_shape, dtype=np.int64) +expect( + node, + inputs=[data, new_shape], + outputs=[expanded], + name="test_expand_dim_unchanged", +) +``` + +
+ + +### EyeLike +There are 3 test cases, listed as following: +
+populate_off_main_diagonal + +```python +shape = (4, 5) +off_diagonal_offset = 1 +node = onnx.helper.make_node( + "EyeLike", + inputs=["x"], + outputs=["y"], + k=off_diagonal_offset, + dtype=onnx.TensorProto.FLOAT, +) + +x = np.random.randint(0, 100, size=shape, dtype=np.int32) +y = np.eye(shape[0], shape[1], k=off_diagonal_offset, dtype=np.float32) +expect( + node, + inputs=[x], + outputs=[y], + name="test_eyelike_populate_off_main_diagonal", +) +``` + +
+
+with_dtype + +```python +shape = (3, 4) +node = onnx.helper.make_node( + "EyeLike", + inputs=["x"], + outputs=["y"], + dtype=onnx.TensorProto.DOUBLE, +) + +x = np.random.randint(0, 100, size=shape, dtype=np.int32) +y = np.eye(shape[0], shape[1], dtype=np.float64) +expect(node, inputs=[x], outputs=[y], name="test_eyelike_with_dtype") +``` + +
+
+without_dtype + +```python +shape = (4, 4) +node = onnx.helper.make_node( + "EyeLike", + inputs=["x"], + outputs=["y"], +) + +x = np.random.randint(0, 100, size=shape, dtype=np.int32) +y = np.eye(shape[0], shape[1], dtype=np.int32) +expect(node, inputs=[x], outputs=[y], name="test_eyelike_without_dtype") +``` + +
+ + +### Flatten +There are 3 test cases, listed as following: +
+flatten + +```python +shape = (2, 3, 4, 5) +a = np.random.random_sample(shape).astype(np.float32) + +for i in range(len(shape)): + node = onnx.helper.make_node( + "Flatten", + inputs=["a"], + outputs=["b"], + axis=i, + ) + + new_shape = (1, -1) if i == 0 else (np.prod(shape[0:i]).astype(int), -1) + b = np.reshape(a, new_shape) + expect(node, inputs=[a], outputs=[b], name="test_flatten_axis" + str(i)) +``` + +
+
+flatten_negative_axis + +```python +shape = (2, 3, 4, 5) +a = np.random.random_sample(shape).astype(np.float32) + +for i in range(-len(shape), 0): + node = onnx.helper.make_node( + "Flatten", + inputs=["a"], + outputs=["b"], + axis=i, + ) + + new_shape = (np.prod(shape[0:i]).astype(int), -1) + b = np.reshape(a, new_shape) + expect( + node, + inputs=[a], + outputs=[b], + name="test_flatten_negative_axis" + str(abs(i)), + ) +``` + +
+
+flatten_with_default_axis + +```python +node = onnx.helper.make_node( + "Flatten", + inputs=["a"], + outputs=["b"], # Default value for axis: axis=1 +) + +shape = (5, 4, 3, 2) +a = np.random.random_sample(shape).astype(np.float32) +new_shape = (5, 24) +b = np.reshape(a, new_shape) +expect(node, inputs=[a], outputs=[b], name="test_flatten_default_axis") +``` + +
+ + +### Floor +There are 1 test cases, listed as following: +
+floor + +```python +node = onnx.helper.make_node( + "Floor", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.5, 1.2, 2]).astype(np.float32) +y = np.floor(x) # expected output [-2., 1., 2.] +expect(node, inputs=[x], outputs=[y], name="test_floor_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.floor(x) +expect(node, inputs=[x], outputs=[y], name="test_floor") +``` + +
+ + +### GRU +There are 4 test cases, listed as following: +
+batchwise + +```python +input = np.array([[[1.0, 2.0]], [[3.0, 4.0]], [[5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 6 +number_of_gates = 3 +weight_scale = 0.2 +layout = 1 + +node = onnx.helper.make_node( + "GRU", + inputs=["X", "W", "R"], + outputs=["Y", "Y_h"], + hidden_size=hidden_size, + layout=layout, +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +gru = GRUHelper(X=input, W=W, R=R, layout=layout) +Y, Y_h = gru.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y.astype(np.float32), Y_h.astype(np.float32)], + name="test_gru_batchwise", +) +``` + +
+
+defaults + +```python +input = np.array([[[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 5 +weight_scale = 0.1 +number_of_gates = 3 + +node = onnx.helper.make_node( + "GRU", inputs=["X", "W", "R"], outputs=["", "Y_h"], hidden_size=hidden_size +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +gru = GRUHelper(X=input, W=W, R=R) +_, Y_h = gru.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y_h.astype(np.float32)], + name="test_gru_defaults", +) +``` + +
+
+initial_bias + +```python +input = np.array([[[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]]]).astype( + np.float32 +) + +input_size = 3 +hidden_size = 3 +weight_scale = 0.1 +custom_bias = 0.1 +number_of_gates = 3 + +node = onnx.helper.make_node( + "GRU", + inputs=["X", "W", "R", "B"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +# Adding custom bias +W_B = custom_bias * np.ones((1, number_of_gates * hidden_size)).astype( + np.float32 +) +R_B = np.zeros((1, number_of_gates * hidden_size)).astype(np.float32) +B = np.concatenate((W_B, R_B), axis=1) + +gru = GRUHelper(X=input, W=W, R=R, B=B) +_, Y_h = gru.step() +expect( + node, + inputs=[input, W, R, B], + outputs=[Y_h.astype(np.float32)], + name="test_gru_with_initial_bias", +) +``` + +
+
+seq_length + +```python +input = np.array( + [ + [[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]], + [[10.0, 11.0, 12.0], [13.0, 14.0, 15.0], [16.0, 17.0, 18.0]], + ] +).astype(np.float32) + +input_size = 3 +hidden_size = 5 +number_of_gates = 3 + +node = onnx.helper.make_node( + "GRU", + inputs=["X", "W", "R", "B"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +W = np.random.randn(1, number_of_gates * hidden_size, input_size).astype( + np.float32 +) +R = np.random.randn(1, number_of_gates * hidden_size, hidden_size).astype( + np.float32 +) + +# Adding custom bias +W_B = np.random.randn(1, number_of_gates * hidden_size).astype(np.float32) +R_B = np.random.randn(1, number_of_gates * hidden_size).astype(np.float32) +B = np.concatenate((W_B, R_B), axis=1) + +gru = GRUHelper(X=input, W=W, R=R, B=B) +_, Y_h = gru.step() +expect( + node, + inputs=[input, W, R, B], + outputs=[Y_h.astype(np.float32)], + name="test_gru_seq_length", +) +``` + +
+ + +### Gather +There are 4 test cases, listed as following: +
+gather_0 + +```python +node = onnx.helper.make_node( + "Gather", + inputs=["data", "indices"], + outputs=["y"], + axis=0, +) +data = np.random.randn(5, 4, 3, 2).astype(np.float32) +indices = np.array([0, 1, 3]) +y = np.take(data, indices, axis=0) + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_0", +) +``` + +
+
+gather_1 + +```python +node = onnx.helper.make_node( + "Gather", + inputs=["data", "indices"], + outputs=["y"], + axis=1, +) +data = np.random.randn(5, 4, 3, 2).astype(np.float32) +indices = np.array([0, 1, 3]) +y = np.take(data, indices, axis=1) + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_1", +) +``` + +
+
+gather_2d_indices + +```python +node = onnx.helper.make_node( + "Gather", + inputs=["data", "indices"], + outputs=["y"], + axis=1, +) +data = np.random.randn(3, 3).astype(np.float32) +indices = np.array([[0, 2]]) +y = np.take(data, indices, axis=1) + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_2d_indices", +) +``` + +
+
+gather_negative_indices + +```python +node = onnx.helper.make_node( + "Gather", + inputs=["data", "indices"], + outputs=["y"], + axis=0, +) +data = np.arange(10).astype(np.float32) +indices = np.array([0, -9, -10]) +y = np.take(data, indices, axis=0) + +# print(y) +# [0. 1. 0.] + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_negative_indices", +) +``` + +
+ + +### GatherElements +There are 3 test cases, listed as following: +
+gather_elements_0 + +```python +axis = 1 +node = onnx.helper.make_node( + "GatherElements", + inputs=["data", "indices"], + outputs=["y"], + axis=axis, +) +data = np.array([[1, 2], [3, 4]], dtype=np.float32) +indices = np.array([[0, 0], [1, 0]], dtype=np.int32) + +y = gather_elements(data, indices, axis) +# print(y) produces +# [[1, 1], +# [4, 3]] + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_elements_0", +) +``` + +
+
+gather_elements_1 + +```python +axis = 0 +node = onnx.helper.make_node( + "GatherElements", + inputs=["data", "indices"], + outputs=["y"], + axis=axis, +) +data = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype=np.float32) +indices = np.array([[1, 2, 0], [2, 0, 0]], dtype=np.int32) + +y = gather_elements(data, indices, axis) +# print(y) produces +# [[4, 8, 3], +# [7, 2, 3]] + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_elements_1", +) +``` + +
+
+gather_elements_negative_indices + +```python +axis = 0 +node = onnx.helper.make_node( + "GatherElements", + inputs=["data", "indices"], + outputs=["y"], + axis=axis, +) +data = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype=np.float32) +indices = np.array([[-1, -2, 0], [-2, 0, 0]], dtype=np.int32) + +y = gather_elements(data, indices, axis) +# print(y) produces +# [[7, 5, 3], +# [4, 2, 3]] + +expect( + node, + inputs=[data, indices.astype(np.int64)], + outputs=[y], + name="test_gather_elements_negative_indices", +) +``` + +
+ + +### GatherND +There are 3 test cases, listed as following: +
+float32 + +```python +node = onnx.helper.make_node( + "GatherND", + inputs=["data", "indices"], + outputs=["output"], +) + +data = np.array([[[0, 1], [2, 3]], [[4, 5], [6, 7]]], dtype=np.float32) +indices = np.array([[[0, 1]], [[1, 0]]], dtype=np.int64) +output = gather_nd_impl(data, indices, 0) +expected_output = np.array([[[2, 3]], [[4, 5]]], dtype=np.float32) +assert np.array_equal(output, expected_output) +expect( + node, + inputs=[data, indices], + outputs=[output], + name="test_gathernd_example_float32", +) +``` + +
+
+int32 + +```python +node = onnx.helper.make_node( + "GatherND", + inputs=["data", "indices"], + outputs=["output"], +) + +data = np.array([[0, 1], [2, 3]], dtype=np.int32) +indices = np.array([[0, 0], [1, 1]], dtype=np.int64) +output = gather_nd_impl(data, indices, 0) +expected_output = np.array([0, 3], dtype=np.int32) +assert np.array_equal(output, expected_output) +expect( + node, + inputs=[data, indices], + outputs=[output], + name="test_gathernd_example_int32", +) +``` + +
+
+int32_batchdim_1 + +```python +node = onnx.helper.make_node( + "GatherND", + inputs=["data", "indices"], + outputs=["output"], + batch_dims=1, +) + +data = np.array([[[0, 1], [2, 3]], [[4, 5], [6, 7]]], dtype=np.int32) +indices = np.array([[1], [0]], dtype=np.int64) +output = gather_nd_impl(data, indices, 1) +expected_output = np.array([[2, 3], [4, 5]], dtype=np.int32) +assert np.array_equal(output, expected_output) +expect( + node, + inputs=[data, indices], + outputs=[output], + name="test_gathernd_example_int32_batch_dim1", +) +``` + +
+ + +### Gelu +There are 2 test cases, listed as following: +
+gelu_default + +```python +node = onnx.helper.make_node("Gelu", inputs=["x"], outputs=["y"]) + +x = np.array([-1, 0, 1]).astype(np.float32) +# expected output [-0.15865526, 0., 0.84134474] +y = (0.5 * x * (1 + np.vectorize(math.erf)(x / np.sqrt(2)))).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_gelu_default_1") + +x = np.random.randn(3, 4, 5).astype(np.float32) +# expected output [2.99595031, 3.99987331, 4.99999857] +y = (0.5 * x * (1 + np.vectorize(math.erf)(x / np.sqrt(2)))).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_gelu_default_2") +``` + +
+
+gelu_tanh + +```python +node = onnx.helper.make_node( + "Gelu", inputs=["x"], outputs=["y"], approximate="tanh" +) + +x = np.array([-1, 0, 1]).astype(np.float32) +# expected output [-0.158808, 0., 0.841192] +y = ( + 0.5 + * x + * (1 + np.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * np.power(x, 3)))) +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_gelu_tanh_1") + +x = np.random.randn(3, 4, 5).astype(np.float32) +# expected output [2.9963627, 3.99993, 4.9999995] +y = ( + 0.5 + * x + * (1 + np.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * np.power(x, 3)))) +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_gelu_tanh_2") +``` + +
+ + +### Gemm +There are 11 test cases, listed as following: +
+all_attributes + +```python +node = onnx.helper.make_node( + "Gemm", + inputs=["a", "b", "c"], + outputs=["y"], + alpha=0.25, + beta=0.35, + transA=1, + transB=1, +) +a = np.random.ranf([4, 3]).astype(np.float32) +b = np.random.ranf([5, 4]).astype(np.float32) +c = np.random.ranf([1, 5]).astype(np.float32) +y = gemm_reference_implementation( + a, b, c, transA=1, transB=1, alpha=0.25, beta=0.35 +) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_all_attributes") +``` + +
+
+alpha + +```python +node = onnx.helper.make_node( + "Gemm", inputs=["a", "b", "c"], outputs=["y"], alpha=0.5 +) +a = np.random.ranf([3, 5]).astype(np.float32) +b = np.random.ranf([5, 4]).astype(np.float32) +c = np.zeros([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c, alpha=0.5) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_alpha") +``` + +
+
+beta + +```python +node = onnx.helper.make_node( + "Gemm", inputs=["a", "b", "c"], outputs=["y"], beta=0.5 +) +a = np.random.ranf([2, 7]).astype(np.float32) +b = np.random.ranf([7, 4]).astype(np.float32) +c = np.random.ranf([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c, beta=0.5) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_beta") +``` + +
+
+default_matrix_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b", "c"], outputs=["y"]) +a = np.random.ranf([3, 6]).astype(np.float32) +b = np.random.ranf([6, 4]).astype(np.float32) +c = np.random.ranf([3, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c) +expect( + node, inputs=[a, b, c], outputs=[y], name="test_gemm_default_matrix_bias" +) +``` + +
+
+default_no_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b"], outputs=["y"]) +a = np.random.ranf([2, 10]).astype(np.float32) +b = np.random.ranf([10, 3]).astype(np.float32) +y = gemm_reference_implementation(a, b) +expect(node, inputs=[a, b], outputs=[y], name="test_gemm_default_no_bias") +``` + +
+
+default_scalar_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b", "c"], outputs=["y"]) +a = np.random.ranf([2, 3]).astype(np.float32) +b = np.random.ranf([3, 4]).astype(np.float32) +c = np.array(3.14).astype(np.float32) +y = gemm_reference_implementation(a, b, c) +expect( + node, inputs=[a, b, c], outputs=[y], name="test_gemm_default_scalar_bias" +) +``` + +
+
+default_single_elem_vector_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b", "c"], outputs=["y"]) +a = np.random.ranf([3, 7]).astype(np.float32) +b = np.random.ranf([7, 3]).astype(np.float32) +c = np.random.ranf([1]).astype(np.float32) +y = gemm_reference_implementation(a, b, c) +expect( + node, + inputs=[a, b, c], + outputs=[y], + name="test_gemm_default_single_elem_vector_bias", +) +``` + +
+
+default_vector_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b", "c"], outputs=["y"]) +a = np.random.ranf([2, 7]).astype(np.float32) +b = np.random.ranf([7, 4]).astype(np.float32) +c = np.random.ranf([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c) +expect( + node, inputs=[a, b, c], outputs=[y], name="test_gemm_default_vector_bias" +) +``` + +
+
+default_zero_bias + +```python +node = onnx.helper.make_node("Gemm", inputs=["a", "b", "c"], outputs=["y"]) +a = np.random.ranf([3, 5]).astype(np.float32) +b = np.random.ranf([5, 4]).astype(np.float32) +c = np.zeros([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_default_zero_bias") +``` + +
+
+transposeA + +```python +node = onnx.helper.make_node( + "Gemm", inputs=["a", "b", "c"], outputs=["y"], transA=1 +) +a = np.random.ranf([6, 3]).astype(np.float32) +b = np.random.ranf([6, 4]).astype(np.float32) +c = np.zeros([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c, transA=1) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_transposeA") +``` + +
+
+transposeB + +```python +node = onnx.helper.make_node( + "Gemm", inputs=["a", "b", "c"], outputs=["y"], transB=1 +) +a = np.random.ranf([3, 6]).astype(np.float32) +b = np.random.ranf([4, 6]).astype(np.float32) +c = np.zeros([1, 4]).astype(np.float32) +y = gemm_reference_implementation(a, b, c, transB=1) +expect(node, inputs=[a, b, c], outputs=[y], name="test_gemm_transposeB") +``` + +
+ + +### GlobalAveragePool +There are 2 test cases, listed as following: +
+globalaveragepool + +```python +node = onnx.helper.make_node( + "GlobalAveragePool", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(1, 3, 5, 5).astype(np.float32) +y = np.mean(x, axis=tuple(range(2, np.ndim(x))), keepdims=True) +expect(node, inputs=[x], outputs=[y], name="test_globalaveragepool") +``` + +
+
+globalaveragepool_precomputed + +```python +node = onnx.helper.make_node( + "GlobalAveragePool", + inputs=["x"], + outputs=["y"], +) +x = np.array( + [ + [ + [ + [1, 2, 3], + [4, 5, 6], + [7, 8, 9], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[5]]]]).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_globalaveragepool_precomputed") +``` + +
+ + +### GlobalMaxPool +There are 2 test cases, listed as following: +
+globalmaxpool + +```python +node = onnx.helper.make_node( + "GlobalMaxPool", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(1, 3, 5, 5).astype(np.float32) +y = np.max(x, axis=tuple(range(2, np.ndim(x))), keepdims=True) +expect(node, inputs=[x], outputs=[y], name="test_globalmaxpool") +``` + +
+
+globalmaxpool_precomputed + +```python +node = onnx.helper.make_node( + "GlobalMaxPool", + inputs=["x"], + outputs=["y"], +) +x = np.array( + [ + [ + [ + [1, 2, 3], + [4, 5, 6], + [7, 8, 9], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[9]]]]).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_globalmaxpool_precomputed") +``` + +
+ + +### Gradient +There are 2 test cases, listed as following: +
+gradient_scalar_add + +```python +add_node = onnx.helper.make_node("Add", ["a", "b"], ["c"], name="my_add") +gradient_node = onnx.helper.make_node( + "Gradient", + ["a", "b"], + ["dc_da", "dc_db"], + name="my_gradient", + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, + xs=["a", "b"], + y="c", +) + +a = np.array(1.0).astype(np.float32) +b = np.array(2.0).astype(np.float32) +c = a + b +# dc / da = d(a+b) / da = 1 +dc_da = np.array(1).astype(np.float32) +# db / db = d(a+b) / db = 1 +dc_db = np.array(1).astype(np.float32) + +graph = onnx.helper.make_graph( + nodes=[add_node, gradient_node], + name="GradientOfAdd", + inputs=[ + onnx.helper.make_tensor_value_info("a", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("b", onnx.TensorProto.FLOAT, []), + ], + outputs=[ + onnx.helper.make_tensor_value_info("c", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("dc_da", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("dc_db", onnx.TensorProto.FLOAT, []), + ], +) +opsets = [ + onnx.helper.make_operatorsetid(ONNX_DOMAIN, 12), + onnx.helper.make_operatorsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1), +] +model = onnx.helper.make_model_gen_version( + graph, producer_name="backend-test", opset_imports=opsets +) +expect( + model, inputs=[a, b], outputs=[c, dc_da, dc_db], name="test_gradient_of_add" +) +``` + +
+
+gradient_scalar_add_and_mul + +```python +add_node = onnx.helper.make_node("Add", ["a", "b"], ["c"], name="my_add") +mul_node = onnx.helper.make_node("Mul", ["c", "a"], ["d"], name="my_mul") +gradient_node = onnx.helper.make_node( + "Gradient", + ["a", "b"], + ["dd_da", "dd_db"], + name="my_gradient", + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, + xs=["a", "b"], + y="d", +) + +a = np.array(1.0).astype(np.float32) +b = np.array(2.0).astype(np.float32) +c = a + b +# d = a * c = a * (a + b) +d = a * c +# dd / da = d(a*a+a*b) / da = 2 * a + b +dd_da = (2 * a + b).astype(np.float32) +# dd / db = d(a*a+a*b) / db = a +dd_db = a + +graph = onnx.helper.make_graph( + nodes=[add_node, mul_node, gradient_node], + name="GradientOfTwoOperators", + inputs=[ + onnx.helper.make_tensor_value_info("a", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("b", onnx.TensorProto.FLOAT, []), + ], + outputs=[ + onnx.helper.make_tensor_value_info("d", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("dd_da", onnx.TensorProto.FLOAT, []), + onnx.helper.make_tensor_value_info("dd_db", onnx.TensorProto.FLOAT, []), + ], +) + +opsets = [ + onnx.helper.make_operatorsetid(ONNX_DOMAIN, 12), + onnx.helper.make_operatorsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1), +] +model = onnx.helper.make_model_gen_version( + graph, producer_name="backend-test", opset_imports=opsets +) +expect( + model, + inputs=[a, b], + outputs=[d, dd_da, dd_db], + name="test_gradient_of_add_and_mul", +) +``` + +
+ + +### Greater +There are 4 test cases, listed as following: +
+greater + +```python +node = onnx.helper.make_node( + "Greater", + inputs=["x", "y"], + outputs=["greater"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater") + +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.random.randn(3, 4, 5).astype(np.int8) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_int8") + +x = np.random.randn(3, 4, 5).astype(np.int16) +y = np.random.randn(3, 4, 5).astype(np.int16) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_uint64") +``` + +
+
+greater + +```python +node = onnx.helper.make_node( + "GreaterOrEqual", + inputs=["x", "y"], + outputs=["greater_equal"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal") + +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.random.randn(3, 4, 5).astype(np.int8) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_int8") + +x = np.random.randn(3, 4, 5).astype(np.int16) +y = np.random.randn(3, 4, 5).astype(np.int16) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_uint64") +``` + +
+
+greater_broadcast + +```python +node = onnx.helper.make_node( + "Greater", + inputs=["x", "y"], + outputs=["greater"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = np.greater(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_bcast") +``` + +
+
+greater_broadcast + +```python +node = onnx.helper.make_node( + "GreaterOrEqual", + inputs=["x", "y"], + outputs=["greater_equal"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = np.greater_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_greater_equal_bcast") +``` + +
+ + +### GridSample +There are 4 test cases, listed as following: +
+gridsample + +```python +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + padding_mode="zeros", + align_corners=0, +) +# X shape, [N, C, H, W] - [1, 1, 4, 4] +X = np.array( + [ + [ + [ + [0.0, 1.0, 2.0, 3.0], + [4.0, 5.0, 6.0, 7.0], + [8.0, 9.0, 10.0, 11.0], + [12.0, 13.0, 14.0, 15.0], + ] + ] + ], + dtype=np.float32, +) +# Grid shape, [N, H_out, W_out, 2] - [1, 6, 6, 2] +Grid = np.array( + [ + [ + [ + [-1.0000, -1.0000], + [-0.6000, -1.0000], + [-0.2000, -1.0000], + [0.2000, -1.0000], + [0.6000, -1.0000], + [1.0000, -1.0000], + ], + [ + [-1.0000, -0.6000], + [-0.6000, -0.6000], + [-0.2000, -0.6000], + [0.2000, -0.6000], + [0.6000, -0.6000], + [1.0000, -0.6000], + ], + [ + [-1.0000, -0.2000], + [-0.6000, -0.2000], + [-0.2000, -0.2000], + [0.2000, -0.2000], + [0.6000, -0.2000], + [1.0000, -0.2000], + ], + [ + [-1.0000, 0.2000], + [-0.6000, 0.2000], + [-0.2000, 0.2000], + [0.2000, 0.2000], + [0.6000, 0.2000], + [1.0000, 0.2000], + ], + [ + [-1.0000, 0.6000], + [-0.6000, 0.6000], + [-0.2000, 0.6000], + [0.2000, 0.6000], + [0.6000, 0.6000], + [1.0000, 0.6000], + ], + [ + [-1.0000, 1.0000], + [-0.6000, 1.0000], + [-0.2000, 1.0000], + [0.2000, 1.0000], + [0.6000, 1.0000], + [1.0000, 1.0000], + ], + ] + ], + dtype=np.float32, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 6, 6] +Y = np.array( + [ + [ + [ + [0.0000, 0.1500, 0.5500, 0.9500, 1.3500, 0.7500], + [0.6000, 1.5000, 2.3000, 3.1000, 3.9000, 2.1000], + [2.2000, 4.7000, 5.5000, 6.3000, 7.1000, 3.7000], + [3.8000, 7.9000, 8.7000, 9.5000, 10.3000, 5.3000], + [5.4000, 11.1000, 11.9000, 12.7000, 13.5000, 6.9000], + [3.0000, 6.1500, 6.5500, 6.9500, 7.3500, 3.7500], + ] + ] + ], + dtype=np.float32, +) +expect(node, inputs=[X, Grid], outputs=[Y], name="test_gridsample") +``` + +
+
+gridsample_mode_aligncorners + +```python +# X shape, [N, C, H, W] - [1, 1, 3, 2] +X = np.array( + [[[[0.0, 1.0], [2.0, 3.0], [4.0, 5.0]]]], + dtype=np.float32, +) +# Grid shape, [N, H_out, W_out, 2] - [1, 2, 4, 2] +Grid = np.array( + [ + [ + [ + [-1.0000, -1.0000], + [-0.5000, -0.5000], + [-0.2000, -0.2000], + [0.0000, 0.0000], + ], + [ + [0.0000, 0.0000], + [-0.2000, -0.2000], + [0.5000, 0.5000], + [1.0000, 1.0000], + ], + ] + ], + dtype=np.float32, +) + +# setting mode = 'bilinear', default align_corners = 0 +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bilinear = np.array( + [[[[0.0000, 0.5000, 1.7000, 2.5000], [2.5000, 1.7000, 4.5000, 1.2500]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bilinear], + name="test_gridsample_bilinear", +) + +# setting mode = 'bilinear', align_corners = 1 +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_align_corners = np.array( + [[[[0.0000, 1.2500, 2.0000, 2.5000], [2.5000, 2.0000, 3.7500, 5.0000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_align_corners], + name="test_gridsample_aligncorners_true", +) + +# setting mode = 'nearest' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="nearest", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_nearest = np.array( + [[[[0.0, 0.0, 2.0, 2.0], [2.0, 2.0, 5.0, 0.0]]]], + dtype=np.float32, +) + +expect( + node, inputs=[X, Grid], outputs=[Y_nearest], name="test_gridsample_nearest" +) + +# setting mode = 'bicubic' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="cubic", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bicubic = np.array( + [[[[-0.1406, 0.3828, 1.7556, 2.9688], [2.9688, 1.7556, 5.1445, 1.3906]]]], + dtype=np.float32, +) + +expect( + node, inputs=[X, Grid], outputs=[Y_bicubic], name="test_gridsample_bicubic" +) + +# ============================================================================ +# Additional tests +# The reference output tensors were generated using PyTorch 2.0. +Grid = np.array( + [ + [ + [[-1.0, -0.8], [-0.6, -0.5], [-0.1, -0.2], [0.7, 0.0]], + [[0.0, 0.4], [0.2, -0.2], [-0.3, 0.5], [-1.0, 1.0]], + ] + ], + dtype=np.float32, +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="nearest", + align_corners=0, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_nearest = np.array( + [[[[0.0, 0.0, 2.0, 3.0], [4.0, 3.0, 4.0, 4.0]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_nearest], + name="test_gridsample_nearest_align_corners_0_additional_1", +) + +# setting mode = 'nearest' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="nearest", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_nearest = np.array( + [[[[0.0, 0.0, 2.0, 3.0], [2.0, 3.0, 4.0, 4.0]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_nearest], + name="test_gridsample_nearest_align_corners_1_additional_1", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + align_corners=0, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bilinear = np.array( + [[[[0.0000, 0.4500, 1.8000, 2.4000], [3.7000, 2.1000, 3.7000, 1.0000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bilinear], + name="test_gridsample_bilinear_align_corners_0_additional_1", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bilinear = np.array( + [[[[0.4000, 1.2000, 2.0500, 2.8500], [3.3000, 2.2000, 3.3500, 4.0000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bilinear], + name="test_gridsample_bilinear_align_corners_1_additional_1", +) + +# These two new bicubic tests produces slightly higher error ~5e-5 +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="cubic", + align_corners=0, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bicubic = np.array( + [ + [ + [ + [-0.173250, 0.284265, 1.923106, 2.568000], + [5.170375, 2.284414, 4.744844, 1.046875], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bicubic], + name="test_gridsample_bicubic_align_corners_0_additional_1", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="cubic", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bicubic = np.array( + [ + [ + [ + [0.304001, 1.128750, 2.266270, 3.144844], + [4.531500, 2.455360, 4.599819, 4.000000], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bicubic], + name="test_gridsample_bicubic_align_corners_1_additional_1", +) +``` + +
+
+gridsample_paddingmode + +```python +# X shape, [N, C, H, W] - [1, 1, 3, 2] +X = np.array( + [[[[0.0, 1.0], [2.0, 3.0], [4.0, 5.0]]]], + dtype=np.float32, +) +# Grid shape, [N, H_out, W_out, 2] - [1, 2, 4, 2] +Grid = np.array( + [ + [ + [ + [-10.0000, -10.0000], + [-5.0000, -5.0000], + [-0.2000, -0.2000], + [10.0000, 10.0000], + ], + [ + [10.0000, 10.0000], + [-0.2000, -0.2000], + [5.0000, 5.0000], + [10.0000, 10.0000], + ], + ] + ], + dtype=np.float32, +) + +# setting padding_mode = 'zeros' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + padding_mode="zeros", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_zeros = np.array( + [[[[0.0000, 0.0000, 1.7000, 0.0000], [0.0000, 1.7000, 0.0000, 0.0000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_zeros], + name="test_gridsample_zeros_padding", +) + +# setting padding_mode = 'border' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + padding_mode="border", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_border = np.array( + [[[[0.0000, 0.0000, 1.7000, 5.0000], [5.0000, 1.7000, 5.0000, 5.0000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_border], + name="test_gridsample_border_padding", +) + +# setting padding_mode = 'reflection' +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + padding_mode="reflection", +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_reflection = np.array( + [[[[2.5000, 0.0000, 1.7000, 2.5000], [2.5000, 1.7000, 5.0000, 2.5000]]]], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_reflection], + name="test_gridsample_reflection_padding", +) +``` + +
+
+volumeetric_gridsample_mode_aligncorners + +```python +X = np.array( + [ + [ + [ + [[1.0, 2.0], [3.0, 4.0]], + [[5.0, 6.0], [7.0, 8.0]], + [[9.0, 10.0], [11.0, 12.0]], + ] + ] + ], + dtype=np.float32, +) + +Grid = np.array( + [ + [ + [ + [[-1.0, -1.0, -1.0], [-1.0, -0.5, 0.3]], + [[-0.5, -0.5, -0.5], [1.0, -0.6, -1.0]], + [[-0.2, -0.2, -0.2], [0.4, 0.2, 0.6]], + [[0.0, 0.0, 0.0], [-1.0, 0.0, 0.0]], + ], + [ + [[0.0, 0.0, 0.0], [-1.0, 1.0, 0.0]], + [[-0.2, -0.2, -0.2], [1.0, 0.4, -0.2]], + [[0.5, 0.5, 0.5], [-1.0, -0.8, 0.8]], + [[1.0, 1.0, 1.0], [0.4, 0.6, -0.3]], + ], + ] + ], + dtype=np.float32, +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="nearest", + align_corners=0, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_nearest = np.array( + [ + [ + [ + [[1.0, 5.0], [1.0, 0.0], [5.0, 12.0], [5.0, 5.0]], + [[5.0, 0.0], [5.0, 0.0], [12.0, 9.0], [0.0, 8.0]], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_nearest], + name="test_gridsample_volumetric_nearest_align_corners_0", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="nearest", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_nearest = np.array( + [ + [ + [ + [[1.0, 5.0], [1.0, 2.0], [5.0, 12.0], [5.0, 5.0]], + [[5.0, 7.0], [5.0, 8.0], [12.0, 9.0], [12.0, 8.0]], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_nearest], + name="test_gridsample_volumetric_nearest_align_corners_1", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + align_corners=0, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bilinear = np.array( + [ + [ + [ + [ + [0.1250, 3.4000], + [2.0000, 0.4500], + [4.7000, 10.9000], + [6.5000, 3.0000], + ], + [ + [6.5000, 1.7500], + [4.7000, 3.3000], + [11.0000, 2.5200], + [1.5000, 5.4900], + ], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bilinear], + name="test_gridsample_volumetric_bilinear_align_corners_0", +) + +node = onnx.helper.make_node( + "GridSample", + inputs=["X", "Grid"], + outputs=["Y"], + mode="linear", + align_corners=1, +) +# Y shape, [N, C, H_out, W_out] - [1, 1, 2, 4] +Y_bilinear = np.array( + [ + [ + [ + [ + [1.0000, 6.7000], + [3.7500, 2.4000], + [5.4000, 9.3000], + [6.5000, 6.0000], + ], + [ + [6.5000, 7.0000], + [5.4000, 6.6000], + [9.2500, 8.4000], + [12.0000, 6.1000], + ], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, Grid], + outputs=[Y_bilinear], + name="test_gridsample_volumetric_bilinear_align_corners_1", +) +``` + +
+ + +### GroupNormalization +There are 2 test cases, listed as following: +
+epsilon + +```python +c = 4 +num_groups = 2 +x = np.random.randn(3, c, 2, 2).astype(np.float32) +scale = np.random.randn(c).astype(np.float32) +bias = np.random.randn(c).astype(np.float32) +epsilon = 1e-2 +y = _group_normalization(x, num_groups, scale, bias, epsilon).astype(np.float32) + +node = onnx.helper.make_node( + "GroupNormalization", + inputs=["x", "scale", "bias"], + outputs=["y"], + epsilon=epsilon, + num_groups=num_groups, +) + +expect( + node, + inputs=[x, scale, bias], + outputs=[y], + name="test_group_normalization_epsilon", +) +``` + +
+
+groupnormalization + +```python +c = 4 +num_groups = 2 +x = np.random.randn(3, c, 2, 2).astype(np.float32) +scale = np.random.randn(c).astype(np.float32) +bias = np.random.randn(c).astype(np.float32) +y = _group_normalization(x, num_groups, scale, bias).astype(np.float32) + +node = onnx.helper.make_node( + "GroupNormalization", + inputs=["x", "scale", "bias"], + outputs=["y"], + num_groups=num_groups, +) + +expect( + node, + inputs=[x, scale, bias], + outputs=[y], + name="test_group_normalization_example", +) +``` + +
+ + +### HammingWindow +There are 1 test cases, listed as following: +
+hammingwindow + +```python +# Test periodic window +node = onnx.helper.make_node( + "HammingWindow", + inputs=["x"], + outputs=["y"], +) +size = np.int32(10) +a0 = 25 / 46 +a1 = 1 - a0 +y = a0 - a1 * np.cos(2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / size) +expect( + node, + inputs=[size], + outputs=[y.astype(np.float32)], + name="test_hammingwindow", +) + +# Test symmetric window +node = onnx.helper.make_node( + "HammingWindow", inputs=["x"], outputs=["y"], periodic=0 +) +size = np.int32(10) +a0 = 25 / 46 +a1 = 1 - a0 +y = a0 - a1 * np.cos( + 2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / (size - 1) +) +expect( + node, + inputs=[size], + outputs=[y.astype(np.float32)], + name="test_hammingwindow_symmetric", +) +``` + +
+ + +### HannWindow +There are 1 test cases, listed as following: +
+hannwindow + +```python +# Test periodic window +node = onnx.helper.make_node( + "HannWindow", + inputs=["x"], + outputs=["y"], +) +size = np.int32(10) +a0 = 0.5 +a1 = 0.5 +y = a0 - a1 * np.cos(2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / size) +expect( + node, inputs=[size], outputs=[y.astype(np.float32)], name="test_hannwindow" +) + +# Test symmetric window +node = onnx.helper.make_node( + "HannWindow", inputs=["x"], outputs=["y"], periodic=0 +) +size = np.int32(10) +a0 = 0.5 +a1 = 0.5 +y = a0 - a1 * np.cos( + 2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / (size - 1) +) +expect( + node, + inputs=[size], + outputs=[y.astype(np.float32)], + name="test_hannwindow_symmetric", +) +``` + +
+ + +### HardSigmoid +There are 2 test cases, listed as following: +
+hardsigmoid + +```python +node = onnx.helper.make_node( + "HardSigmoid", inputs=["x"], outputs=["y"], alpha=0.5, beta=0.6 +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.clip(x * 0.5 + 0.6, 0, 1) # expected output [0.1, 0.6, 1.] +expect(node, inputs=[x], outputs=[y], name="test_hardsigmoid_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x * 0.5 + 0.6, 0, 1) +expect(node, inputs=[x], outputs=[y], name="test_hardsigmoid") +``` + +
+
+hardsigmoid_default + +```python +default_alpha = 0.2 +default_beta = 0.5 +node = onnx.helper.make_node( + "HardSigmoid", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x * default_alpha + default_beta, 0, 1) +expect(node, inputs=[x], outputs=[y], name="test_hardsigmoid_default") +``` + +
+ + +### HardSwish +There are 1 test cases, listed as following: +
+hardswish + +```python +node = onnx.helper.make_node( + "HardSwish", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = hardswish(x) + +expect(node, inputs=[x], outputs=[y], name="test_hardswish") +``` + +
+ + +### Hardmax +There are 2 test cases, listed as following: +
+hardmax + +```python +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], +) + +x = np.array([[3, 0, 1, 2], [2, 5, 1, 0], [0, 1, 3, 2], [0, 1, 2, 3]]).astype( + np.float32 +) +# expect result: +# [[1. 0. 0. 0.] +# [0. 1. 0. 0.] +# [0. 0. 1. 0.] +# [0. 0. 0. 1.]] +y = hardmax(x) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_example") + +# For multiple occurrences of the maximal values, the first occurrence is selected for one-hot output +x = np.array([[3, 3, 3, 1]]).astype(np.float32) +# expect result: +# [[1, 0, 0, 0]] +y = hardmax(x) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_one_hot") +``` + +
+
+hardmax_axis + +```python +x = np.random.randn(3, 4, 5).astype(np.float32) +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], + axis=0, +) +y = hardmax(x, axis=0) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_axis_0") + +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], + axis=1, +) +y = hardmax(x, axis=1) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_axis_1") + +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], + axis=2, +) +y = hardmax(x, axis=2) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_axis_2") + +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], + axis=-1, +) +y = hardmax(x, axis=-1) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_negative_axis") + +# default axis is -1 +node = onnx.helper.make_node( + "Hardmax", + inputs=["x"], + outputs=["y"], +) +expect(node, inputs=[x], outputs=[y], name="test_hardmax_default_axis") +``` + +
+ + +### Identity +There are 3 test cases, listed as following: +
+identity + +```python +node = onnx.helper.make_node( + "Identity", + inputs=["x"], + outputs=["y"], +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +expect(node, inputs=[data], outputs=[data], name="test_identity") +``` + +
+
+identity_opt + +```python +ten_in_tp = onnx.helper.make_tensor_type_proto( + onnx.TensorProto.FLOAT, shape=[5] +) +seq_in_tp = onnx.helper.make_sequence_type_proto(ten_in_tp) +opt_in_tp = onnx.helper.make_optional_type_proto(seq_in_tp) + +identity_node = onnx.helper.make_node( + "Identity", inputs=["opt_in"], outputs=["opt_out"] +) + +x = [np.array([1, 2, 3, 4, 5]).astype(np.float32)] + +expect( + identity_node, + inputs=[x], + outputs=[x], + name="test_identity_opt", + opset_imports=[onnx.helper.make_opsetid("", 16)], + input_type_protos=[opt_in_tp], + output_type_protos=[opt_in_tp], +) +``` + +
+
+sequence + +```python +node = onnx.helper.make_node( + "Identity", + inputs=["x"], + outputs=["y"], +) + +data = [ + np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, + ), + np.array( + [ + [ + [ + [2, 3], + [1, 5], + ] + ] + ], + dtype=np.float32, + ), +] + +expect(node, inputs=[data], outputs=[data], name="test_identity_sequence") +``` + +
+ + +### If +There are 3 test cases, listed as following: +
+if + +```python +# Given a bool scalar input cond. +# return constant tensor x if cond is True, otherwise return constant tensor y. + +then_out = onnx.helper.make_tensor_value_info( + "then_out", onnx.TensorProto.FLOAT, [5] +) +else_out = onnx.helper.make_tensor_value_info( + "else_out", onnx.TensorProto.FLOAT, [5] +) + +x = np.array([1, 2, 3, 4, 5]).astype(np.float32) +y = np.array([5, 4, 3, 2, 1]).astype(np.float32) + +then_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["then_out"], + value=onnx.numpy_helper.from_array(x), +) + +else_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["else_out"], + value=onnx.numpy_helper.from_array(y), +) + +then_body = onnx.helper.make_graph( + [then_const_node], "then_body", [], [then_out] +) + +else_body = onnx.helper.make_graph( + [else_const_node], "else_body", [], [else_out] +) + +if_node = onnx.helper.make_node( + "If", + inputs=["cond"], + outputs=["res"], + then_branch=then_body, + else_branch=else_body, +) + +cond = np.array(1).astype(bool) +res = x if cond else y +expect( + if_node, + inputs=[cond], + outputs=[res], + name="test_if", + opset_imports=[onnx.helper.make_opsetid("", 11)], +) +``` + +
+
+if_optional + +```python +# Given a bool scalar input cond, return an empty optional sequence of +# tensor if True, return an optional sequence with value x +# (the input optional sequence) otherwise. + +ten_in_tp = onnx.helper.make_tensor_type_proto( + onnx.TensorProto.FLOAT, shape=[5] +) +seq_in_tp = onnx.helper.make_sequence_type_proto(ten_in_tp) + +then_out_tensor_tp = onnx.helper.make_tensor_type_proto( + onnx.TensorProto.FLOAT, shape=[5] +) +then_out_seq_tp = onnx.helper.make_sequence_type_proto(then_out_tensor_tp) +then_out_opt_tp = onnx.helper.make_optional_type_proto(then_out_seq_tp) +then_out = onnx.helper.make_value_info("optional_empty", then_out_opt_tp) + +else_out_tensor_tp = onnx.helper.make_tensor_type_proto( + onnx.TensorProto.FLOAT, shape=[5] +) +else_out_seq_tp = onnx.helper.make_sequence_type_proto(else_out_tensor_tp) +else_out_opt_tp = onnx.helper.make_optional_type_proto(else_out_seq_tp) +else_out = onnx.helper.make_value_info("else_opt", else_out_opt_tp) + +x = [np.array([1, 2, 3, 4, 5]).astype(np.float32)] +cond = np.array(0).astype(bool) +res = compute_if_outputs(x, cond) + +opt_empty_in = onnx.helper.make_node( + "Optional", inputs=[], outputs=["optional_empty"], type=seq_in_tp +) + +then_body = onnx.helper.make_graph([opt_empty_in], "then_body", [], [then_out]) + +else_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["x"], + value=onnx.numpy_helper.from_array(x[0]), +) + +else_seq_node = onnx.helper.make_node( + "SequenceConstruct", inputs=["x"], outputs=["else_seq"] +) + +else_optional_seq_node = onnx.helper.make_node( + "Optional", inputs=["else_seq"], outputs=["else_opt"] +) + +else_body = onnx.helper.make_graph( + [else_const_node, else_seq_node, else_optional_seq_node], + "else_body", + [], + [else_out], +) + +if_node = onnx.helper.make_node( + "If", + inputs=["cond"], + outputs=["sequence"], + then_branch=then_body, + else_branch=else_body, +) + +expect( + if_node, + inputs=[cond], + outputs=[res], + name="test_if_opt", + output_type_protos=[else_out_opt_tp], + opset_imports=[onnx.helper.make_opsetid("", 16)], +) +``` + +
+
+if_seq + +```python +# Given a bool scalar input cond. +# return constant sequence x if cond is True, otherwise return constant sequence y. + +then_out = onnx.helper.make_tensor_sequence_value_info( + "then_out", onnx.TensorProto.FLOAT, shape=[5] +) +else_out = onnx.helper.make_tensor_sequence_value_info( + "else_out", onnx.TensorProto.FLOAT, shape=[5] +) + +x = [np.array([1, 2, 3, 4, 5]).astype(np.float32)] +y = [np.array([5, 4, 3, 2, 1]).astype(np.float32)] + +then_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["x"], + value=onnx.numpy_helper.from_array(x[0]), +) + +then_seq_node = onnx.helper.make_node( + "SequenceConstruct", inputs=["x"], outputs=["then_out"] +) + +else_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["y"], + value=onnx.numpy_helper.from_array(y[0]), +) + +else_seq_node = onnx.helper.make_node( + "SequenceConstruct", inputs=["y"], outputs=["else_out"] +) + +then_body = onnx.helper.make_graph( + [then_const_node, then_seq_node], "then_body", [], [then_out] +) + +else_body = onnx.helper.make_graph( + [else_const_node, else_seq_node], "else_body", [], [else_out] +) + +if_node = onnx.helper.make_node( + "If", + inputs=["cond"], + outputs=["res"], + then_branch=then_body, + else_branch=else_body, +) + +cond = np.array(1).astype(bool) +res = x if cond else y +expect( + if_node, + inputs=[cond], + outputs=[res], + name="test_if_seq", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) +``` + +
+ + +### ImageDecoder +There are 9 test cases, listed as following: +
+image_decoder_decode_bmp_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "bmp", _image_decoder_data.image_decoder_decode_bmp_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_bmp_rgb", +) +``` + +
+
+image_decoder_decode_jpeg2k_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "jpeg2000", _image_decoder_data.image_decoder_decode_jpeg2k_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_jpeg2k_rgb", +) +``` + +
+
+image_decoder_decode_jpeg_bgr + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="BGR", +) + +data, output = _generate_test_data( + "jpeg", _image_decoder_data.image_decoder_decode_jpeg_bgr, "BGR" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_jpeg_bgr", +) +``` + +
+
+image_decoder_decode_jpeg_grayscale + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="Grayscale", +) + +data, output = _generate_test_data( + "jpeg", _image_decoder_data.image_decoder_decode_jpeg_grayscale, "Grayscale" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_jpeg_grayscale", +) +``` + +
+
+image_decoder_decode_jpeg_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "jpeg", _image_decoder_data.image_decoder_decode_jpeg_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_jpeg_rgb", +) +``` + +
+
+image_decoder_decode_png_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "png", _image_decoder_data.image_decoder_decode_png_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_png_rgb", +) +``` + +
+
+image_decoder_decode_pnm_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "ppm", _image_decoder_data.image_decoder_decode_pnm_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_pnm_rgb", +) +``` + +
+
+image_decoder_decode_tiff_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "tiff", _image_decoder_data.image_decoder_decode_tiff_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_tiff_rgb", +) +``` + +
+
+image_decoder_decode_webp_rgb + +```python +node = onnx.helper.make_node( + "ImageDecoder", + inputs=["data"], + outputs=["output"], + pixel_format="RGB", +) + +data, output = _generate_test_data( + "webp", _image_decoder_data.image_decoder_decode_webp_rgb, "RGB" +) +expect( + node, + inputs=[data], + outputs=[output], + name="test_image_decoder_decode_webp_rgb", +) +``` + +
+ + +### InstanceNormalization +There are 1 test cases, listed as following: +
+instancenormalization + +```python +def _instancenorm_test_mode( + x: np.ndarray, s: np.ndarray, bias: np.ndarray, epsilon: float = 1e-5 +) -> np.ndarray: + dims_x = len(x.shape) + axis = tuple(range(2, dims_x)) + mean = np.mean(x, axis=axis, keepdims=True) + var = np.var(x, axis=axis, keepdims=True) + dim_ones = (1,) * (dims_x - 2) + s = s.reshape(-1, *dim_ones) + bias = bias.reshape(-1, *dim_ones) + return s * (x - mean) / np.sqrt(var + epsilon) + bias + +# input size: (1, 2, 1, 3) +x = np.array([[[[-1, 0, 1]], [[2, 3, 4]]]]).astype(np.float32) +s = np.array([1.0, 1.5]).astype(np.float32) +bias = np.array([0, 1]).astype(np.float32) +y = _instancenorm_test_mode(x, s, bias).astype(np.float32) + +node = onnx.helper.make_node( + "InstanceNormalization", + inputs=["x", "s", "bias"], + outputs=["y"], +) + +# output size: (1, 2, 1, 3) +expect(node, inputs=[x, s, bias], outputs=[y], name="test_instancenorm_example") + +# input size: (2, 3, 4, 5) +x = np.random.randn(2, 3, 4, 5).astype(np.float32) +s = np.random.randn(3).astype(np.float32) +bias = np.random.randn(3).astype(np.float32) +epsilon = 1e-2 +y = _instancenorm_test_mode(x, s, bias, epsilon).astype(np.float32) + +node = onnx.helper.make_node( + "InstanceNormalization", + inputs=["x", "s", "bias"], + outputs=["y"], + epsilon=epsilon, +) + +# output size: (2, 3, 4, 5) +expect(node, inputs=[x, s, bias], outputs=[y], name="test_instancenorm_epsilon") +``` + +
+ + +### IsInf +There are 4 test cases, listed as following: +
+infinity + +```python +node = onnx.helper.make_node( + "IsInf", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.2, np.nan, np.inf, 2.8, -np.inf, np.inf], dtype=np.float32) +y = np.isinf(x) +expect(node, inputs=[x], outputs=[y], name="test_isinf") +``` + +
+
+infinity_float16 + +```python +node = onnx.helper.make_node( + "IsInf", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.2, np.nan, np.inf, 2.8, -np.inf, np.inf], dtype=np.float16) +y = np.isinf(x) +expect(node, inputs=[x], outputs=[y], name="test_isinf_float16") +``` + +
+
+negative_infinity_only + +```python +node = onnx.helper.make_node( + "IsInf", inputs=["x"], outputs=["y"], detect_positive=0 +) + +x = np.array([-1.7, np.nan, np.inf, -3.6, -np.inf, np.inf], dtype=np.float32) +y = np.isneginf(x) +expect(node, inputs=[x], outputs=[y], name="test_isinf_negative") +``` + +
+
+positive_infinity_only + +```python +node = onnx.helper.make_node( + "IsInf", inputs=["x"], outputs=["y"], detect_negative=0 +) + +x = np.array([-1.7, np.nan, np.inf, 3.6, -np.inf, np.inf], dtype=np.float32) +y = np.isposinf(x) +expect(node, inputs=[x], outputs=[y], name="test_isinf_positive") +``` + +
+ + +### IsNaN +There are 2 test cases, listed as following: +
+float16 + +```python +node = onnx.helper.make_node( + "IsNaN", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.2, np.nan, np.inf, 2.8, -np.inf, np.inf], dtype=np.float16) +y = np.isnan(x) +expect(node, inputs=[x], outputs=[y], name="test_isnan_float16") +``` + +
+
+isnan + +```python +node = onnx.helper.make_node( + "IsNaN", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1.2, np.nan, np.inf, 2.8, -np.inf, np.inf], dtype=np.float32) +y = np.isnan(x) +expect(node, inputs=[x], outputs=[y], name="test_isnan") +``` + +
+ + +### LRN +There are 2 test cases, listed as following: +
+default + +```python +alpha = 0.0001 +beta = 0.75 +bias = 1.0 +nsize = 3 +node = onnx.helper.make_node("LRN", inputs=["x"], outputs=["y"], size=3) +x = np.random.randn(5, 5, 5, 5).astype(np.float32) +square_sum = np.zeros((5, 5, 5, 5)).astype(np.float32) +for n, c, h, w in np.ndindex(x.shape): + square_sum[n, c, h, w] = sum( + x[ + n, + max(0, c - math.floor((nsize - 1) / 2)) : min( + 5, c + math.ceil((nsize - 1) / 2) + 1 + ), + h, + w, + ] + ** 2 + ) +y = x / ((bias + (alpha / nsize) * square_sum) ** beta) +expect(node, inputs=[x], outputs=[y], name="test_lrn_default") +``` + +
+
+lrn + +```python +alpha = 0.0002 +beta = 0.5 +bias = 2.0 +nsize = 3 +node = onnx.helper.make_node( + "LRN", + inputs=["x"], + outputs=["y"], + alpha=alpha, + beta=beta, + bias=bias, + size=nsize, +) +x = np.random.randn(5, 5, 5, 5).astype(np.float32) +square_sum = np.zeros((5, 5, 5, 5)).astype(np.float32) +for n, c, h, w in np.ndindex(x.shape): + square_sum[n, c, h, w] = sum( + x[ + n, + max(0, c - math.floor((nsize - 1) / 2)) : min( + 5, c + math.ceil((nsize - 1) / 2) + 1 + ), + h, + w, + ] + ** 2 + ) +y = x / ((bias + (alpha / nsize) * square_sum) ** beta) +expect(node, inputs=[x], outputs=[y], name="test_lrn") +``` + +
+ + +### LSTM +There are 4 test cases, listed as following: +
+batchwise + +```python +input = np.array([[[1.0, 2.0]], [[3.0, 4.0]], [[5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 7 +weight_scale = 0.3 +number_of_gates = 4 +layout = 1 + +node = onnx.helper.make_node( + "LSTM", + inputs=["X", "W", "R"], + outputs=["Y", "Y_h"], + hidden_size=hidden_size, + layout=layout, +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +lstm = LSTMHelper(X=input, W=W, R=R, layout=layout) +Y, Y_h = lstm.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y.astype(np.float32), Y_h.astype(np.float32)], + name="test_lstm_batchwise", +) +``` + +
+
+defaults + +```python +input = np.array([[[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 3 +weight_scale = 0.1 +number_of_gates = 4 + +node = onnx.helper.make_node( + "LSTM", inputs=["X", "W", "R"], outputs=["", "Y_h"], hidden_size=hidden_size +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +lstm = LSTMHelper(X=input, W=W, R=R) +_, Y_h = lstm.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y_h.astype(np.float32)], + name="test_lstm_defaults", +) +``` + +
+
+initial_bias + +```python +input = np.array([[[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]]]).astype( + np.float32 +) + +input_size = 3 +hidden_size = 4 +weight_scale = 0.1 +custom_bias = 0.1 +number_of_gates = 4 + +node = onnx.helper.make_node( + "LSTM", + inputs=["X", "W", "R", "B"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) + +# Adding custom bias +W_B = custom_bias * np.ones((1, number_of_gates * hidden_size)).astype( + np.float32 +) +R_B = np.zeros((1, number_of_gates * hidden_size)).astype(np.float32) +B = np.concatenate((W_B, R_B), 1) + +lstm = LSTMHelper(X=input, W=W, R=R, B=B) +_, Y_h = lstm.step() +expect( + node, + inputs=[input, W, R, B], + outputs=[Y_h.astype(np.float32)], + name="test_lstm_with_initial_bias", +) +``` + +
+
+peepholes + +```python +input = np.array([[[1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0]]]).astype( + np.float32 +) + +input_size = 4 +hidden_size = 3 +weight_scale = 0.1 +number_of_gates = 4 +number_of_peepholes = 3 + +node = onnx.helper.make_node( + "LSTM", + inputs=["X", "W", "R", "B", "sequence_lens", "initial_h", "initial_c", "P"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +# Initializing Inputs +W = weight_scale * np.ones( + (1, number_of_gates * hidden_size, input_size) +).astype(np.float32) +R = weight_scale * np.ones( + (1, number_of_gates * hidden_size, hidden_size) +).astype(np.float32) +B = np.zeros((1, 2 * number_of_gates * hidden_size)).astype(np.float32) +seq_lens = np.repeat(input.shape[0], input.shape[1]).astype(np.int32) +init_h = np.zeros((1, input.shape[1], hidden_size)).astype(np.float32) +init_c = np.zeros((1, input.shape[1], hidden_size)).astype(np.float32) +P = weight_scale * np.ones((1, number_of_peepholes * hidden_size)).astype( + np.float32 +) + +lstm = LSTMHelper( + X=input, W=W, R=R, B=B, P=P, initial_c=init_c, initial_h=init_h +) +_, Y_h = lstm.step() +expect( + node, + inputs=[input, W, R, B, seq_lens, init_h, init_c, P], + outputs=[Y_h.astype(np.float32)], + name="test_lstm_with_peepholes", +) +``` + +
+ + +### LayerNormalization +There are 4 test cases, listed as following: +
+d + +```python +X = np.random.randn(3, 4).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + B = np.random.randn(*normalized_shape).astype(np.float32) + Y, mean, inv_std_dev = _layer_normalization(X, W, B, axis=axis) + + node = onnx.helper.make_node( + "LayerNormalization", + inputs=["X", "W", "B"], + outputs=["Y", "Mean", "InvStdDev"], + axis=axis, + ) + + if axis < 0: + name = f"test_layer_normalization_2d_axis_negative_{-axis}" + else: + name = f"test_layer_normalization_2d_axis{axis}" + + expect(node, inputs=[X, W, B], outputs=[Y, mean, inv_std_dev], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+
+d_epsilon + +```python +epsilon = 1e-1 +X = np.random.randn(2, 3, 5).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + B = np.random.randn(*normalized_shape).astype(np.float32) + Y, mean, inv_std_dev = _layer_normalization(X, W, B, axis, epsilon) + node = onnx.helper.make_node( + "LayerNormalization", + inputs=["X", "W", "B"], + outputs=["Y", "Mean", "InvStdDev"], + axis=axis, + epsilon=epsilon, + ) + + if axis < 0: + name = f"test_layer_normalization_3d_axis_negative_{-axis}_epsilon" + else: + name = f"test_layer_normalization_3d_axis{axis}_epsilon" + + expect(node, inputs=[X, W, B], outputs=[Y, mean, inv_std_dev], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+
+default_axis + +```python +X = np.random.randn(2, 3, 4, 5).astype(np.float32) + +# Default axis in LayerNormalization is -1. +normalized_shape = calculate_normalized_shape(X.shape, -1) +W = np.random.randn(*normalized_shape).astype(np.float32) +B = np.random.randn(*normalized_shape).astype(np.float32) +# Axis is default to -1 in the reference implementation. +Y, mean, inv_std_dev = _layer_normalization(X, W, B) + +# Not specifying axis attribute means -1. +node = onnx.helper.make_node( + "LayerNormalization", + inputs=["X", "W", "B"], + outputs=["Y", "Mean", "InvStdDev"], +) + +expect( + node, + inputs=[X, W, B], + outputs=[Y, mean, inv_std_dev], + name="test_layer_normalization_default_axis", +) +``` + +
+
+layernormalization + +```python +X = np.random.randn(2, 3, 4, 5).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + B = np.random.randn(*normalized_shape).astype(np.float32) + Y, mean, inv_std_dev = _layer_normalization(X, W, B, axis) + + node = onnx.helper.make_node( + "LayerNormalization", + inputs=["X", "W", "B"], + outputs=["Y", "Mean", "InvStdDev"], + axis=axis, + ) + + if axis < 0: + name = f"test_layer_normalization_4d_axis_negative_{-axis}" + else: + name = f"test_layer_normalization_4d_axis{axis}" + + expect(node, inputs=[X, W, B], outputs=[Y, mean, inv_std_dev], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+ + +### LeakyRelu +There are 2 test cases, listed as following: +
+leakyrelu + +```python +node = onnx.helper.make_node( + "LeakyRelu", inputs=["x"], outputs=["y"], alpha=0.1 +) + +x = np.array([-1, 0, 1]).astype(np.float32) +# expected output [-0.1, 0., 1.] +y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * 0.1 +expect(node, inputs=[x], outputs=[y], name="test_leakyrelu_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * 0.1 +expect(node, inputs=[x], outputs=[y], name="test_leakyrelu") +``` + +
+
+leakyrelu_default + +```python +default_alpha = 0.01 +node = onnx.helper.make_node( + "LeakyRelu", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * default_alpha +expect(node, inputs=[x], outputs=[y], name="test_leakyrelu_default") +``` + +
+ + +### Less +There are 4 test cases, listed as following: +
+less + +```python +node = onnx.helper.make_node( + "Less", + inputs=["x", "y"], + outputs=["less"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less") + +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.random.randn(3, 4, 5).astype(np.int8) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_int8") + +x = np.random.randn(3, 4, 5).astype(np.int16) +y = np.random.randn(3, 4, 5).astype(np.int16) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_uint64") +``` + +
+
+less + +```python +node = onnx.helper.make_node( + "LessOrEqual", + inputs=["x", "y"], + outputs=["less_equal"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal") + +x = np.random.randn(3, 4, 5).astype(np.int8) +y = np.random.randn(3, 4, 5).astype(np.int8) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_int8") + +x = np.random.randn(3, 4, 5).astype(np.int16) +y = np.random.randn(3, 4, 5).astype(np.int16) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_int16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_uint8") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_uint16") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_uint32") + +x = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_uint64") +``` + +
+
+less_broadcast + +```python +node = onnx.helper.make_node( + "Less", + inputs=["x", "y"], + outputs=["less"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = np.less(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_bcast") +``` + +
+
+less_broadcast + +```python +node = onnx.helper.make_node( + "LessOrEqual", + inputs=["x", "y"], + outputs=["less_equal"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = np.less_equal(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_less_equal_bcast") +``` + +
+ + +### LinearAttention +There are 14 test cases, listed as following: +
+decode_step + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "past_state", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 1, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +past_state = np.random.randn(b, h_kv, d_k, d_v).astype(np.float32) * 0.1 +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + past_state=past_state, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, past_state, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_decode_step", + opset_imports=_OPSET, +) +``` + +
+
+delta + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "", "beta"], + outputs=["output", "present_state"], + update_rule="delta", + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, + update_rule="delta", +) +expect( + node, + inputs=[query, key, value, beta], + outputs=[output, present_state], + name="test_linear_attention_delta", + opset_imports=_OPSET, +) +``` + +
+
+explicit_scale + +```python +scale = 0.25 +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, + scale=scale, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, + scale=scale, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_explicit_scale", + opset_imports=_OPSET, +) +``` + +
+
+fp16 + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=8, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 8, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float16) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float16), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float16) +decay = (-np.abs(np.random.randn(b, t, h_kv * d_k)) * 0.1).astype(np.float16) +beta = np.random.rand(b, t, h_kv).astype(np.float16) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_fp16", + opset_imports=_OPSET, +) +``` + +
+
+gated + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay"], + outputs=["output", "present_state"], + update_rule="gated", + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = np.random.randn(b, t, h_kv * d_k).astype(np.float32) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +# Per-key-dim decay in log-space (negative -> decay). +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 + +output, present_state = _compute( + query, + key, + value, + decay=decay, + q_num_heads=h_q, + kv_num_heads=h_kv, + update_rule="gated", +) +expect( + node, + inputs=[query, key, value, decay], + outputs=[output, present_state], + name="test_linear_attention_gated", + opset_imports=_OPSET, +) +``` + +
+
+gated_delta + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_gated_delta", + opset_imports=_OPSET, +) +``` + +
+
+gated_delta_beta_scalar + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, 1).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_gated_delta_beta_scalar", + opset_imports=_OPSET, +) +``` + +
+
+gated_delta_gqa + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=8, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 8, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_gated_delta_gqa", + opset_imports=_OPSET, +) +``` + +
+
+gated_delta_mqa + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=8, + kv_num_heads=1, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 8, 1, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_gated_delta_mqa", + opset_imports=_OPSET, +) +``` + +
+
+gated_per_head_decay + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "", "decay"], + outputs=["output", "present_state"], + update_rule="gated", + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = np.random.randn(b, t, h_kv * d_k).astype(np.float32) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +# Per-head scalar decay. +decay = -np.abs(np.random.randn(b, t, h_kv)).astype(np.float32) * 0.1 + +output, present_state = _compute( + query, + key, + value, + decay=decay, + q_num_heads=h_q, + kv_num_heads=h_kv, + update_rule="gated", +) +expect( + node, + inputs=[query, key, value, decay], + outputs=[output, present_state], + name="test_linear_attention_gated_per_head_decay", + opset_imports=_OPSET, +) +``` + +
+
+linear + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value"], + outputs=["output", "present_state"], + update_rule="linear", + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = np.random.randn(b, t, h_kv * d_k).astype(np.float32) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + q_num_heads=h_q, + kv_num_heads=h_kv, + update_rule="linear", +) +expect( + node, + inputs=[query, key, value], + outputs=[output, present_state], + name="test_linear_attention_linear", + opset_imports=_OPSET, +) +``` + +
+
+linear_t1_no_past + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value"], + outputs=["output", "present_state"], + update_rule="linear", + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 1, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = np.random.randn(b, t, h_kv * d_k).astype(np.float32) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + q_num_heads=h_q, + kv_num_heads=h_kv, + update_rule="linear", +) +expect( + node, + inputs=[query, key, value], + outputs=[output, present_state], + name="test_linear_attention_linear_t1_no_past", + opset_imports=_OPSET, +) +``` + +
+
+no_past_explicit_zeros + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "past_state", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +past_state = np.zeros((b, h_kv, d_k, d_v), dtype=np.float32) +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + past_state=past_state, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, past_state, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_no_past_explicit_zeros", + opset_imports=_OPSET, +) +``` + +
+
+prefill_with_past + +```python +node = onnx.helper.make_node( + "LinearAttention", + inputs=["query", "key", "value", "past_state", "decay", "beta"], + outputs=["output", "present_state"], + q_num_heads=4, + kv_num_heads=4, +) +b, t, h_q, h_kv, d_k, d_v = 2, 4, 4, 4, 8, 8 +query = np.random.randn(b, t, h_q * d_k).astype(np.float32) +key = _l2_normalize(np.random.randn(b, t, h_kv * d_k).astype(np.float32), h_kv) +value = np.random.randn(b, t, h_kv * d_v).astype(np.float32) +past_state = np.random.randn(b, h_kv, d_k, d_v).astype(np.float32) * 0.1 +decay = -np.abs(np.random.randn(b, t, h_kv * d_k)).astype(np.float32) * 0.1 +beta = np.random.rand(b, t, h_kv).astype(np.float32) + +output, present_state = _compute( + query, + key, + value, + past_state=past_state, + decay=decay, + beta=beta, + q_num_heads=h_q, + kv_num_heads=h_kv, +) +expect( + node, + inputs=[query, key, value, past_state, decay, beta], + outputs=[output, present_state], + name="test_linear_attention_prefill_with_past", + opset_imports=_OPSET, +) +``` + +
+ + +### Log +There are 1 test cases, listed as following: +
+log + +```python +node = onnx.helper.make_node( + "Log", + inputs=["x"], + outputs=["y"], +) + +x = np.array([1, 10]).astype(np.float32) +y = np.log(x) # expected output [0., 2.30258512] +expect(node, inputs=[x], outputs=[y], name="test_log_example") + +x = np.exp(np.random.randn(3, 4, 5).astype(np.float32)) +y = np.log(x) +expect(node, inputs=[x], outputs=[y], name="test_log") +``` + +
+ + +### LogSoftmax +There are 2 test cases, listed as following: +
+logsoftmax + +```python +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], +) +x = np.array([[-1, 0, 1]]).astype(np.float32) +# expected output +# [[-2.4076061 -1.407606 -0.407606 ]] +y = logsoftmax(x) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_example_1") +``` + +
+
+logsoftmax_axis + +```python +x = np.array([[0, 1, 2, 3], [10000, 10001, 10002, 10003]]).astype(np.float32) +# expected output +# [[-3.4401896 -2.4401896 -1.4401896 -0.44018966] +# [-3.4401896 -2.4401896 -1.4401896 -0.44018966]] +y = logsoftmax(x) + +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], +) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_large_number") + +x = np.abs(np.random.randn(3, 4, 5).astype(np.float32)) +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], + axis=0, +) +y = logsoftmax(x, axis=0) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_axis_0") + +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], + axis=1, +) +y = logsoftmax(x, axis=1) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_axis_1") + +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], + axis=2, +) +y = logsoftmax(x, axis=2) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_axis_2") + +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], + axis=-1, +) +y = logsoftmax(x, axis=-1) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_negative_axis") + +# default axis is -1 +node = onnx.helper.make_node( + "LogSoftmax", + inputs=["x"], + outputs=["y"], +) +expect(node, inputs=[x], outputs=[y], name="test_logsoftmax_default_axis") +``` + +
+ + +### Loop +There are 3 test cases, listed as following: +
+loop_11 + +```python +# Given a tensor x of values [x1, ..., xN], and initial tensor y +# sum up its elements using a scan +# returning the final state (y+x1+x2+...+xN) as well the scan_output +# [y+x1, y+x1+x2, ..., y+x1+x2+...+xN] + +y_in = onnx.helper.make_tensor_value_info("y_in", onnx.TensorProto.FLOAT, [1]) +y_out = onnx.helper.make_tensor_value_info("y_out", onnx.TensorProto.FLOAT, [1]) +scan_out = onnx.helper.make_tensor_value_info( + "scan_out", onnx.TensorProto.FLOAT, [1] +) +cond_in = onnx.helper.make_tensor_value_info( + "cond_in", onnx.TensorProto.BOOL, [] +) +cond_out = onnx.helper.make_tensor_value_info( + "cond_out", onnx.TensorProto.BOOL, [] +) +iter_count = onnx.helper.make_tensor_value_info( + "iter_count", onnx.TensorProto.INT64, [] +) + +x = np.array([1, 2, 3, 4, 5]).astype(np.float32) +y = np.array([-2]).astype(np.float32) + +x_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["x"], + value=onnx.helper.make_tensor( + name="const_tensor_x", + data_type=onnx.TensorProto.FLOAT, + dims=x.shape, + vals=x.flatten().astype(float), + ), +) + +one_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["one"], + value=onnx.helper.make_tensor( + name="const_tensor_one", + data_type=onnx.TensorProto.INT64, + dims=(), + vals=[1], + ), +) + +i_add_node = onnx.helper.make_node( + "Add", inputs=["iter_count", "one"], outputs=["end"] +) + +start_unsqueeze_node = onnx.helper.make_node( + "Unsqueeze", inputs=["iter_count"], outputs=["slice_start"], axes=[0] +) + +end_unsqueeze_node = onnx.helper.make_node( + "Unsqueeze", inputs=["end"], outputs=["slice_end"], axes=[0] +) + +slice_node = onnx.helper.make_node( + "Slice", inputs=["x", "slice_start", "slice_end"], outputs=["slice_out"] +) + +y_add_node = onnx.helper.make_node( + "Add", inputs=["y_in", "slice_out"], outputs=["y_out"] +) + +identity_node = onnx.helper.make_node( + "Identity", inputs=["cond_in"], outputs=["cond_out"] +) + +scan_identity_node = onnx.helper.make_node( + "Identity", inputs=["y_out"], outputs=["scan_out"] +) + +loop_body = onnx.helper.make_graph( + [ + identity_node, + x_const_node, + one_const_node, + i_add_node, + start_unsqueeze_node, + end_unsqueeze_node, + slice_node, + y_add_node, + scan_identity_node, + ], + "loop_body", + [iter_count, cond_in, y_in], + [cond_out, y_out, scan_out], +) + +node = onnx.helper.make_node( + "Loop", + inputs=["trip_count", "cond", "y"], + outputs=["res_y", "res_scan"], + body=loop_body, +) + +trip_count = np.array(5).astype(np.int64) +res_y = np.array([13]).astype(np.float32) +cond = np.array(1).astype(bool) +res_scan = np.array([-1, 1, 4, 8, 13]).astype(np.float32).reshape((5, 1)) +expect( + node, + inputs=[trip_count, cond, y], + outputs=[res_y, res_scan], + name="test_loop11", + opset_imports=[onnx.helper.make_opsetid("", 11)], +) +``` + +
+
+loop_13 + +```python +# Given a tensor x of values [x1, ..., xN], +# Return a sequence of tensors of +# [[x1], [x1, x2], ..., [x1, ..., xN]] + +seq_in = onnx.helper.make_tensor_sequence_value_info( + "seq_in", onnx.TensorProto.FLOAT, None +) +seq_out = onnx.helper.make_tensor_sequence_value_info( + "seq_out", onnx.TensorProto.FLOAT, None +) +cond_in = onnx.helper.make_tensor_value_info( + "cond_in", onnx.TensorProto.BOOL, [] +) +cond_out = onnx.helper.make_tensor_value_info( + "cond_out", onnx.TensorProto.BOOL, [] +) +iter_count = onnx.helper.make_tensor_value_info( + "iter_count", onnx.TensorProto.INT64, [] +) + +x = np.array([1, 2, 3, 4, 5]).astype(np.float32) + +x_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["x"], + value=onnx.helper.make_tensor( + name="const_tensor_x", + data_type=onnx.TensorProto.FLOAT, + dims=x.shape, + vals=x.flatten().astype(float), + ), +) + +one_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["one"], + value=onnx.helper.make_tensor( + name="const_tensor_one", + data_type=onnx.TensorProto.INT64, + dims=(), + vals=[1], + ), +) + +zero_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["slice_start"], + value=onnx.helper.make_tensor( + name="const_tensor_zero", + data_type=onnx.TensorProto.INT64, + dims=(1,), + vals=[0], + ), +) + +axes_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["axes"], + value=onnx.helper.make_tensor( + name="const_tensor_axes", + data_type=onnx.TensorProto.INT64, + dims=(), + vals=[0], + ), +) + +add_node = onnx.helper.make_node( + "Add", inputs=["iter_count", "one"], outputs=["end"] +) + +end_unsqueeze_node = onnx.helper.make_node( + "Unsqueeze", inputs=["end", "axes"], outputs=["slice_end"] +) + +slice_node = onnx.helper.make_node( + "Slice", inputs=["x", "slice_start", "slice_end"], outputs=["slice_out"] +) + +insert_node = onnx.helper.make_node( + "SequenceInsert", inputs=["seq_in", "slice_out"], outputs=["seq_out"] +) + +identity_node = onnx.helper.make_node( + "Identity", inputs=["cond_in"], outputs=["cond_out"] +) + +loop_body = onnx.helper.make_graph( + [ + identity_node, + x_const_node, + one_const_node, + zero_const_node, + add_node, + axes_node, + end_unsqueeze_node, + slice_node, + insert_node, + ], + "loop_body", + [iter_count, cond_in, seq_in], + [cond_out, seq_out], +) + +node = onnx.helper.make_node( + "Loop", + inputs=["trip_count", "cond", "seq_empty"], + outputs=["seq_res"], + body=loop_body, +) + +trip_count = np.array(5).astype(np.int64) +seq_empty: list[Any] = [] +seq_res = [x[: int(i)] for i in x] +cond = np.array(1).astype(bool) +expect( + node, + inputs=[trip_count, cond, seq_empty], + outputs=[seq_res], + name="test_loop13_seq", + opset_imports=[onnx.helper.make_opsetid("", 13)], + input_type_protos=[ + onnx.helper.make_tensor_type_proto( + onnx.TensorProto.INT64, trip_count.shape + ), + onnx.helper.make_tensor_type_proto(onnx.TensorProto.BOOL, cond.shape), + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, []) + ), + ], +) +``` + +
+
+loop_16_none + +```python +# Given a tensor sequence of values [x1, ..., xN], and an initial optional sequence of tensors [x0], +# Return a concatenated sequence of tensors of +# [x0, [x1], [x1, x2], ..., [x1, ..., xN]] + +ten_in_tp = onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, []) +seq_in_tp = onnx.helper.make_sequence_type_proto(ten_in_tp) +opt_in_tp = onnx.helper.make_optional_type_proto(seq_in_tp) +opt_in = onnx.helper.make_value_info("opt_seq_in", opt_in_tp) +seq_out = onnx.helper.make_tensor_sequence_value_info( + "seq_out", onnx.TensorProto.FLOAT, [] +) +cond_in = onnx.helper.make_tensor_value_info( + "cond_in", onnx.TensorProto.BOOL, [] +) +cond_out = onnx.helper.make_tensor_value_info( + "cond_out", onnx.TensorProto.BOOL, [] +) +iter_count = onnx.helper.make_tensor_value_info( + "iter_count", onnx.TensorProto.INT64, [] +) + +x0 = np.array(0).astype(np.float32) +x = np.array([1, 2, 3, 4, 5]).astype(np.float32) + +optional_has_elem_node = onnx.helper.make_node( + "OptionalHasElement", inputs=["opt_seq_in"], outputs=["optional_has_elem"] +) + +optional_is_none = onnx.helper.make_node( + "Not", inputs=["optional_has_elem"], outputs=["optional_is_none"] +) + +optional_get_elem = onnx.helper.make_node( + "OptionalGetElement", inputs=["opt_seq_in"], outputs=["seq_in"] +) + +constant_in = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["constant_in"], + value=onnx.helper.make_tensor( + name="const_tensor", data_type=onnx.TensorProto.FLOAT, dims=(), vals=[0] + ), +) + +seq_const_in = onnx.helper.make_node( + "SequenceConstruct", inputs=["constant_in"], outputs=["init_seq_in"] +) + +then_seq_out = onnx.helper.make_tensor_sequence_value_info( + "init_seq_in", onnx.TensorProto.FLOAT, [] +) +then_body = onnx.helper.make_graph( + [constant_in, seq_const_in], "then_body", [], [then_seq_out] +) + +else_seq_out = onnx.helper.make_tensor_sequence_value_info( + "seq_in", onnx.TensorProto.FLOAT, [] +) +else_body = onnx.helper.make_graph( + [optional_get_elem], "else_body", [], [else_seq_out] +) + +if_node = onnx.helper.make_node( + "If", + inputs=["optional_is_none"], + outputs=["sequence"], + then_branch=then_body, + else_branch=else_body, +) + +x_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["x"], + value=onnx.helper.make_tensor( + name="const_tensor_x", + data_type=onnx.TensorProto.FLOAT, + dims=x.shape, + vals=x.flatten().astype(float), + ), +) + +one_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["one"], + value=onnx.helper.make_tensor( + name="const_tensor_one", + data_type=onnx.TensorProto.INT64, + dims=(), + vals=[1], + ), +) + +zero_const_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["slice_start"], + value=onnx.helper.make_tensor( + name="const_tensor_zero", + data_type=onnx.TensorProto.INT64, + dims=(1,), + vals=[0], + ), +) + +axes_node = onnx.helper.make_node( + "Constant", + inputs=[], + outputs=["axes"], + value=onnx.helper.make_tensor( + name="const_tensor_axes", + data_type=onnx.TensorProto.INT64, + dims=(), + vals=[0], + ), +) + +add_node = onnx.helper.make_node( + "Add", inputs=["iter_count", "one"], outputs=["end"] +) + +end_unsqueeze_node = onnx.helper.make_node( + "Unsqueeze", inputs=["end", "axes"], outputs=["slice_end"] +) + +slice_node = onnx.helper.make_node( + "Slice", inputs=["x", "slice_start", "slice_end"], outputs=["slice_out"] +) + +insert_node = onnx.helper.make_node( + "SequenceInsert", inputs=["sequence", "slice_out"], outputs=["seq_out"] +) + +identity_node = onnx.helper.make_node( + "Identity", inputs=["cond_in"], outputs=["cond_out"] +) + +loop_body = onnx.helper.make_graph( + [ + identity_node, + optional_has_elem_node, + optional_is_none, + if_node, + x_const_node, + one_const_node, + zero_const_node, + add_node, + axes_node, + end_unsqueeze_node, + slice_node, + insert_node, + ], + "loop_body", + [iter_count, cond_in, opt_in], + [cond_out, seq_out], +) + +node = onnx.helper.make_node( + "Loop", + inputs=["trip_count", "cond", "opt_seq"], + outputs=["seq_res"], + body=loop_body, +) + +trip_count = np.array(5).astype(np.int64) +cond = np.array(1).astype(bool) +seq_res = compute_loop_outputs(x, [x0], trip_count) +opt_seq_in: list[Any] = [x0] +expect( + node, + inputs=[trip_count, cond, opt_seq_in], + outputs=[seq_res], + name="test_loop16_seq_none", + opset_imports=[onnx.helper.make_opsetid("", 16)], + input_type_protos=[ + onnx.helper.make_tensor_type_proto( + onnx.TensorProto.INT64, trip_count.shape + ), + onnx.helper.make_tensor_type_proto(onnx.TensorProto.BOOL, cond.shape), + opt_in_tp, + ], +) +``` + +
+ + +### LpNormalization +There are 6 test cases, listed as following: +
+default + +```python +node = onnx.helper.make_node("LpNormalization", inputs=["x"], outputs=["y"]) +x = np.array( + [[[1.0, 2.0, 2.0], [3.0, 4.0, 0.0]], [[0.0, 5.0, 5.0], [6.0, 8.0, 0.0]]], + dtype=np.float32, +) +lp_norm_default = np.sqrt(np.sum(x**2, axis=-1, keepdims=True)) +y = x / lp_norm_default +expect(node, inputs=[x], outputs=[y], name="test_lpnormalization_default") +``` + +
+
+l1normalization_axis_0 + +```python +node = onnx.helper.make_node( + "LpNormalization", inputs=["x"], outputs=["y"], axis=0, p=1 +) +x = np.array([3.0, 4.0], dtype=np.float32) +l1_norm_axis_0 = np.sum(abs(x), axis=0, keepdims=True) +y = x / l1_norm_axis_0 +expect(node, inputs=[x], outputs=[y], name="test_l1normalization_axis_0") +``` + +
+
+l1normalization_axis_1 + +```python +node = onnx.helper.make_node( + "LpNormalization", inputs=["x"], outputs=["y"], axis=1, p=1 +) +x = np.array([[3.0, 4.0], [6.0, 8.0]], dtype=np.float32) +l1_norm_axis_1 = np.sum(abs(x), axis=1, keepdims=True) +y = x / l1_norm_axis_1 +expect(node, inputs=[x], outputs=[y], name="test_l1normalization_axis_1") +``` + +
+
+l1normalization_axis_last + +```python +node = onnx.helper.make_node( + "LpNormalization", inputs=["x"], outputs=["y"], axis=-1, p=1 +) +x = np.array( + [[[1.0, 2.0, 2.0], [3.0, 4.0, 0.0]], [[0.0, 5.0, 5.0], [6.0, 8.0, 0.0]]], + dtype=np.float32, +) +l1_norm_axis_last = np.sum(abs(x), axis=-1, keepdims=True) +y = x / l1_norm_axis_last +expect(node, inputs=[x], outputs=[y], name="test_l1normalization_axis_last") +``` + +
+
+l2normalization_axis_0 + +```python +node = onnx.helper.make_node( + "LpNormalization", inputs=["x"], outputs=["y"], axis=0, p=2 +) +x = np.array( + [[[1.0, 2.0, 2.0], [3.0, 4.0, 0.0]], [[0.0, 5.0, 5.0], [6.0, 8.0, 0.0]]], + dtype=np.float32, +) +l2_norm_axis_0 = np.sqrt(np.sum(x**2, axis=0, keepdims=True)) +# When norm is 0, output is 0 (0/0 = 0) +y = np.where(l2_norm_axis_0 == 0, 0, x / l2_norm_axis_0) +expect(node, inputs=[x], outputs=[y], name="test_l2normalization_axis_0") +``` + +
+
+l2normalization_axis_1 + +```python +node = onnx.helper.make_node( + "LpNormalization", inputs=["x"], outputs=["y"], axis=1, p=2 +) +x = np.array([[3.0, 4.0], [6.0, 8.0]], dtype=np.float32) +l2_norm_axis_1 = np.sqrt(np.sum(x**2, axis=1, keepdims=True)) +y = x / l2_norm_axis_1 +expect(node, inputs=[x], outputs=[y], name="test_l2normalization_axis_1") +``` + +
+ + +### LpPool +There are 8 test cases, listed as following: +
+lppool_1d_default + +```python +"""input_shape: [1, 3, 32] +output_shape: [1, 3, 31] +""" +p = 3 +kernel_shape = [2] +strides = [1] +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=kernel_shape, + strides=strides, + p=p, +) +x = np.random.randn(1, 3, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", p=p) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_1d_default") +``` + +
+
+lppool_2d_default + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 31, 31] +""" +p = 4 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + p=p, +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = (2, 2) +strides = (1, 1) +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", p=p) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_default") +``` + +
+
+lppool_2d_dilations + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +p = 2 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[1, 1], + dilations=[2, 2], + p=p, +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ] +).astype(np.float32) + +y = np.array( + [ + [ + [ + [14.560219778561036, 16.24807680927192], + [21.633307652783937, 23.49468024894146], + ] + ] + ] +).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_dilations") +``` + +
+
+lppool_2d_pads + +```python +"""input_shape: [1, 3, 28, 28] +output_shape: [1, 3, 30, 30] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +p = 3 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[2, 2, 2, 2], + p=p, +) +x = np.random.randn(1, 3, 28, 28).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (3, 3) +strides = (1, 1) +pad_bottom = pad_top = pad_right = pad_left = 2 +pads = [pad_top, pad_left, pad_bottom, pad_right] +out_shape, extra_pads = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = np.pad( + x, + ( + (0, 0), + (0, 0), + (extra_pads[0], extra_pads[2]), + (extra_pads[1], extra_pads[3]), + ), + mode="constant", + constant_values=0, +) +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "LPPOOL", + pads_required=extra_pads, + pads=pads, + p=p, +) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_pads") +``` + +
+
+lppool_2d_same_lower + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [1, 0, 1, 0] by axis +""" +p = 4 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_LOWER", + p=p, +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_LOWER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_LOWER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_bottom = pad_shape[0] // 2 +pad_top = pad_shape[0] - pad_bottom +pad_right = pad_shape[1] // 2 +pad_left = pad_shape[1] - pad_right +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=0, +) +pads = [pad_top, pad_left, pad_bottom, pad_right] +y = pool( + padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", pads, pads, p=p +) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_same_lower") +``` + +
+
+lppool_2d_same_upper + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [0, 1, 0, 1] by axis +""" +p = 2 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_UPPER", + p=p, +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_UPPER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_UPPER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_top = pad_shape[0] // 2 +pad_bottom = pad_shape[0] - pad_top +pad_left = pad_shape[1] // 2 +pad_right = pad_shape[1] - pad_left +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=0, +) +pads = [pad_top, pad_left, pad_bottom, pad_right] +y = pool( + padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", pads, pads, p=p +) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_same_upper") +``` + +
+
+lppool_2d_strides + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 10, 10] +""" +p = 2 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + strides=[3, 3], + p=p, +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = (5, 5) +strides = (3, 3) +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", p=p) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_2d_strides") +``` + +
+
+lppool_3d_default + +```python +"""input_shape: [1, 3, 32, 32, 32] +output_shape: [1, 3, 31, 31, 31] +""" +p = 3 +node = onnx.helper.make_node( + "LpPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], + p=p, +) +x = np.random.randn(1, 3, 32, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = [2, 2, 2] +strides = [1, 1, 1] +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "LPPOOL", p=p) + +expect(node, inputs=[x], outputs=[y], name="test_lppool_3d_default") +``` + +
+ + +### MatMul +There are 1 test cases, listed as following: +
+matmul + +```python +node = onnx.helper.make_node( + "MatMul", + inputs=["a", "b"], + outputs=["c"], +) + +# 2d +a = np.random.randn(3, 4).astype(np.float32) +b = np.random.randn(4, 3).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_2d") + +# 3d +a = np.random.randn(2, 3, 4).astype(np.float32) +b = np.random.randn(2, 4, 3).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_3d") + +# 4d +a = np.random.randn(1, 2, 3, 4).astype(np.float32) +b = np.random.randn(1, 2, 4, 3).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_4d") + +# broadcasting +a = np.random.randn(3, 1, 3, 4).astype(np.float32) +b = np.random.randn(1, 2, 4, 2).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_bcast") + +# 1d + 3d +a = np.random.randn(4).astype(np.float32) +b = np.random.randn(2, 4, 1).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_1d_3d") + +# 3d + 1d +a = np.random.randn(1, 2, 4, 3).astype(np.float32) +b = np.random.randn(3).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_4d_1d") + +# 1d + 1d +a = np.random.randn(3).astype(np.float32) +b = np.random.randn(3).astype(np.float32) +c = np.matmul(a, b) +expect(node, inputs=[a, b], outputs=[c], name="test_matmul_1d_1d") +``` + +
+ + +### MatMulInteger +There are 1 test cases, listed as following: +
+matmulinteger + +```python +node = onnx.helper.make_node( + "MatMulInteger", + inputs=["A", "B", "a_zero_point", "b_zero_point"], + outputs=["Y"], +) + +A = np.array( + [ + [11, 7, 3], + [10, 6, 2], + [9, 5, 1], + [8, 4, 0], + ], + dtype=np.uint8, +) + +a_zero_point = np.array([12], dtype=np.uint8) + +B = np.array( + [ + [1, 4], + [2, 5], + [3, 6], + ], + dtype=np.uint8, +) + +b_zero_point = np.array([0], dtype=np.uint8) + +output = np.array( + [ + [-38, -83], + [-44, -98], + [-50, -113], + [-56, -128], + ], + dtype=np.int32, +) + +expect( + node, + inputs=[A, B, a_zero_point, b_zero_point], + outputs=[output], + name="test_matmulinteger", +) +``` + +
+ + +### Max +There are 2 test cases, listed as following: +
+max + +```python +data_0 = np.array([3, 2, 1]).astype(np.float32) +data_1 = np.array([1, 4, 4]).astype(np.float32) +data_2 = np.array([2, 5, 3]).astype(np.float32) +result = np.array([3, 5, 4]).astype(np.float32) +node = onnx.helper.make_node( + "Max", + inputs=["data_0", "data_1", "data_2"], + outputs=["result"], +) +expect( + node, + inputs=[data_0, data_1, data_2], + outputs=[result], + name="test_max_example", +) + +node = onnx.helper.make_node( + "Max", + inputs=["data_0"], + outputs=["result"], +) +expect(node, inputs=[data_0], outputs=[data_0], name="test_max_one_input") + +result = np.maximum(data_0, data_1) +node = onnx.helper.make_node( + "Max", + inputs=["data_0", "data_1"], + outputs=["result"], +) +expect( + node, inputs=[data_0, data_1], outputs=[result], name="test_max_two_inputs" +) +``` + +
+
+max_all_numeric_types + +```python +for op_dtype in all_numeric_dtypes: + data_0 = np.array([3, 2, 1]).astype(op_dtype) + data_1 = np.array([1, 4, 4]).astype(op_dtype) + result = np.array([3, 4, 4]).astype(op_dtype) + node = onnx.helper.make_node( + "Max", + inputs=["data_0", "data_1"], + outputs=["result"], + ) + expect( + node, + inputs=[data_0, data_1], + outputs=[result], + name=f"test_max_{np.dtype(op_dtype).name}", + ) +``` + +
+ + +### MaxPool +There are 19 test cases, listed as following: +
+maxpool_1d_default + +```python +"""input_shape: [1, 3, 32] +output_shape: [1, 3, 31] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2], +) +x = np.random.randn(1, 3, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = [2] +strides = [1] +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX") + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_1d_default") +``` + +
+
+maxpool_2d_ceil + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + strides=[2, 2], + ceil_mode=True, +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[11, 12], [15, 16]]]]).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_ceil") +``` + +
+
+maxpool_2d_ceil_output_size_reduce_by_one + +```python +"""input_shape: [1, 1, 2, 2] +output_shape: [1, 1, 1, 1] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[1, 1], + strides=[2, 2], + ceil_mode=True, +) +x = np.array([[[[1, 2], [3, 4]]]]).astype(np.float32) +y = np.array([[[[1]]]]).astype(np.float32) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_maxpool_2d_ceil_output_size_reduce_by_one", +) +``` + +
+
+maxpool_2d_default + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 31, 31] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = (2, 2) +strides = (1, 1) +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX") + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_default") +``` + +
+
+maxpool_2d_dilations + +```python +"""input_shape: [1, 1, 4, 4] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[1, 1], + dilations=[2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[11, 12], [15, 16]]]]).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_dilations") +``` + +
+
+maxpool_2d_pads + +```python +"""input_shape: [1, 3, 28, 28] +output_shape: [1, 3, 30, 30] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + pads=[2, 2, 2, 2], +) +x = np.random.randn(1, 3, 28, 28).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (3, 3) +strides = (1, 1) +pad_bottom = pad_top = pad_right = pad_left = 2 +pads = [pad_top, pad_left, pad_bottom, pad_right] +out_shape, extra_pads = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=np.nan, +) + +y = pool( + padded, + x_shape, + kernel_shape, + strides, + out_shape, + "MAX", + pads_required=extra_pads, + pads=pads, +) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_pads") +``` + +
+
+maxpool_2d_precomputed_pads + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 5, 5] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + pads=[2, 2, 2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array( + [ + [ + [ + [13, 14, 15, 15, 15], + [18, 19, 20, 20, 20], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + ] + ] + ] +).astype(np.float32) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_precomputed_pads") +``` + +
+
+maxpool_2d_precomputed_same_upper + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 3, 3] +pad_shape: [2, 2] -> [1, 1, 1, 1] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[3, 3], + strides=[2, 2], + auto_pad="SAME_UPPER", +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[7, 9, 10], [17, 19, 20], [22, 24, 25]]]]).astype(np.float32) + +expect( + node, inputs=[x], outputs=[y], name="test_maxpool_2d_precomputed_same_upper" +) +``` + +
+
+maxpool_2d_precomputed_strides + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "MaxPool", inputs=["x"], outputs=["y"], kernel_shape=[2, 2], strides=[2, 2] +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[7, 9], [17, 19]]]]).astype(np.float32) + +expect( + node, inputs=[x], outputs=[y], name="test_maxpool_2d_precomputed_strides" +) +``` + +
+
+maxpool_2d_same_lower + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [1, 0, 1, 0] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_LOWER", +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_LOWER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_LOWER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_bottom = pad_shape[0] // 2 +pad_top = pad_shape[0] - pad_bottom +pad_right = pad_shape[1] // 2 +pad_left = pad_shape[1] - pad_right +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=np.nan, +) +pads = [pad_top, pad_left, pad_bottom, pad_right] +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX", pads, pads) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_same_lower") +``` + +
+
+maxpool_2d_same_upper + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 32, 32] +pad_shape: [1, 1] -> [0, 1, 0, 1] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2], + auto_pad="SAME_UPPER", +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +kernel_shape = (2, 2) +strides = (1, 1) +out_shape = get_output_shape_auto_pad( + "SAME_UPPER", x_shape[2:], kernel_shape, strides +) +pad_shape = get_pad_shape( + "SAME_UPPER", x_shape[2:], kernel_shape, strides, out_shape +) +pad_top = pad_shape[0] // 2 +pad_bottom = pad_shape[0] - pad_top +pad_left = pad_shape[1] // 2 +pad_right = pad_shape[1] - pad_left +padded = np.pad( + x, + ((0, 0), (0, 0), (pad_top, pad_bottom), (pad_left, pad_right)), + mode="constant", + constant_values=np.nan, +) +pads = [pad_top, pad_left, pad_bottom, pad_right] +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX", pads, pads) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_same_upper") +``` + +
+
+maxpool_2d_strides + +```python +"""input_shape: [1, 3, 32, 32] +output_shape: [1, 3, 10, 10] +""" +node = onnx.helper.make_node( + "MaxPool", inputs=["x"], outputs=["y"], kernel_shape=[5, 5], strides=[3, 3] +) +x = np.random.randn(1, 3, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = (5, 5) +strides = (3, 3) +out_shape, pads = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX") + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_strides") +``` + +
+
+maxpool_2d_uint8 + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 5, 5] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[5, 5], + pads=[2, 2, 2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.uint8) +y = np.array( + [ + [ + [ + [13, 14, 15, 15, 15], + [18, 19, 20, 20, 20], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + ] + ] + ] +).astype(np.uint8) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_2d_uint8") +``` + +
+
+maxpool_3d_default + +```python +"""input_shape: [1, 3, 32, 32, 32] +output_shape: [1, 3, 31, 31, 31] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], +) +x = np.random.randn(1, 3, 32, 32, 32).astype(np.float32) +x_shape = np.shape(x) +pads = None +kernel_shape = [2, 2, 2] +strides = [1, 1, 1] +out_shape, _ = get_output_shape_explicit_padding( + pads, x_shape[2:], kernel_shape, strides +) +padded = x +y = pool(padded, x_shape, kernel_shape, strides, out_shape, "MAX") + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_3d_default") +``` + +
+
+maxpool_3d_dilations + +```python +"""input_shape: [1, 1, 4, 4, 4] +output_shape: [1, 1, 2, 2, 2] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], + strides=[1, 1, 1], + dilations=[2, 2, 2], +) +x = np.array( + [ + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[[11, 12], [15, 16]], [[11, 12], [15, 16]]]]]).astype( + np.float32 +) + +expect(node, inputs=[x], outputs=[y], name="test_maxpool_3d_dilations") +``` + +
+
+maxpool_3d_dilations_use_ref_impl + +```python +"""input_shape: [1, 1, 4, 4, 4] +output_shape: [1, 1, 2, 2, 2] +""" +dilations = [2, 2, 2] +kernel_shape = [2, 2, 2] +strides = [1, 1, 1] +ceil_mode = False +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=[2, 2, 2], + strides=[1, 1, 1], + dilations=dilations, +) +x = np.array( + [ + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ], + ] + ] + ] +).astype(np.float32) + +x_shape = x.shape[2:] +out_shape, pads = get_output_shape_explicit_padding( + None, x_shape, kernel_shape, strides, dilations, ceil_mode=ceil_mode +) +padded = x +y = pool( + padded, + (1, 1, *x_shape), + kernel_shape, + strides, + out_shape, + "MAX", + pads_required=pads, + pads=None, + dilations=dilations, +) + +expect( + node, inputs=[x], outputs=[y], name="test_maxpool_3d_dilations_use_ref_impl" +) +``` + +
+
+maxpool_3d_dilations_use_ref_impl_large + +```python +x_shape = (32, 32, 32) +dilations = (2, 2, 2) +kernel_shape = (5, 5, 5) +strides = (3, 3, 3) +ceil_mode = True + +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y"], + kernel_shape=kernel_shape, + strides=strides, + dilations=dilations, + ceil_mode=ceil_mode, +) + +x = np.random.randn(1, 1, *x_shape).astype(np.float32) +out_shape, pads = get_output_shape_explicit_padding( + None, x_shape, kernel_shape, strides, dilations, ceil_mode=ceil_mode +) +padded = np.pad( + x, + ( + (0, 0), + (0, 0), + (pads[0], pads[3]), + (pads[1], pads[4]), + (pads[2], pads[5]), + ), + mode="constant", + constant_values=0, +) +y = pool( + padded, + (1, 1, *x_shape), + kernel_shape, + strides, + out_shape, + "MAX", + pads_required=pads, + pads=None, + dilations=dilations, +) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_maxpool_3d_dilations_use_ref_impl_large", +) +``` + +
+
+maxpool_with_argmax_2d_precomputed_pads + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 5, 5] +pad_shape: [4, 4] -> [2, 2, 2, 2] by axis +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y", "z"], + kernel_shape=[5, 5], + pads=[2, 2, 2, 2], +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array( + [ + [ + [ + [13, 14, 15, 15, 15], + [18, 19, 20, 20, 20], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + [23, 24, 25, 25, 25], + ] + ] + ] +).astype(np.float32) +z = np.array( + [ + [ + [ + [12, 13, 14, 14, 14], + [17, 18, 19, 19, 19], + [22, 23, 24, 24, 24], + [22, 23, 24, 24, 24], + [22, 23, 24, 24, 24], + ] + ] + ] +).astype(np.int64) + +expect( + node, + inputs=[x], + outputs=[y, z], + name="test_maxpool_with_argmax_2d_precomputed_pads", +) +``` + +
+
+maxpool_with_argmax_2d_precomputed_strides + +```python +"""input_shape: [1, 1, 5, 5] +output_shape: [1, 1, 2, 2] +""" +node = onnx.helper.make_node( + "MaxPool", + inputs=["x"], + outputs=["y", "z"], + kernel_shape=[2, 2], + strides=[2, 2], + storage_order=1, +) +x = np.array( + [ + [ + [ + [1, 2, 3, 4, 5], + [6, 7, 8, 9, 10], + [11, 12, 13, 14, 15], + [16, 17, 18, 19, 20], + [21, 22, 23, 24, 25], + ] + ] + ] +).astype(np.float32) +y = np.array([[[[7, 9], [17, 19]]]]).astype(np.float32) +z = np.array([[[[6, 16], [8, 18]]]]).astype(np.int64) + +expect( + node, + inputs=[x], + outputs=[y, z], + name="test_maxpool_with_argmax_2d_precomputed_strides", +) +``` + +
+ + +### MaxUnpool +There are 2 test cases, listed as following: +
+with_output_shape + +```python +node = onnx.helper.make_node( + "MaxUnpool", + inputs=["xT", "xI", "output_shape"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[2, 2], +) +xT = np.array([[[[5, 6], [7, 8]]]], dtype=np.float32) +xI = np.array([[[[5, 7], [13, 15]]]], dtype=np.int64) +output_shape = np.array((1, 1, 5, 5), dtype=np.int64) +y = np.array( + [ + [ + [ + [0, 0, 0, 0, 0], + [0, 5, 0, 6, 0], + [0, 0, 0, 0, 0], + [0, 7, 0, 8, 0], + [0, 0, 0, 0, 0], + ] + ] + ], + dtype=np.float32, +) +expect( + node, + inputs=[xT, xI, output_shape], + outputs=[y], + name="test_maxunpool_export_with_output_shape", +) +``` + +
+
+without_output_shape + +```python +node = onnx.helper.make_node( + "MaxUnpool", + inputs=["xT", "xI"], + outputs=["y"], + kernel_shape=[2, 2], + strides=[2, 2], +) +xT = np.array([[[[1, 2], [3, 4]]]], dtype=np.float32) +xI = np.array([[[[5, 7], [13, 15]]]], dtype=np.int64) +y = np.array( + [[[[0, 0, 0, 0], [0, 1, 0, 2], [0, 0, 0, 0], [0, 3, 0, 4]]]], + dtype=np.float32, +) +expect( + node, + inputs=[xT, xI], + outputs=[y], + name="test_maxunpool_export_without_output_shape", +) +``` + +
+ + +### Mean +There are 1 test cases, listed as following: +
+mean + +```python +data_0 = np.array([3, 0, 2]).astype(np.float32) +data_1 = np.array([1, 3, 4]).astype(np.float32) +data_2 = np.array([2, 6, 6]).astype(np.float32) +result = np.array([2, 3, 4]).astype(np.float32) +node = onnx.helper.make_node( + "Mean", + inputs=["data_0", "data_1", "data_2"], + outputs=["result"], +) +expect( + node, + inputs=[data_0, data_1, data_2], + outputs=[result], + name="test_mean_example", +) + +node = onnx.helper.make_node( + "Mean", + inputs=["data_0"], + outputs=["result"], +) +expect(node, inputs=[data_0], outputs=[data_0], name="test_mean_one_input") + +result = np.divide(np.add(data_0, data_1), 2.0) +node = onnx.helper.make_node( + "Mean", + inputs=["data_0", "data_1"], + outputs=["result"], +) +expect( + node, inputs=[data_0, data_1], outputs=[result], name="test_mean_two_inputs" +) +``` + +
+ + +### MeanVarianceNormalization +There are 1 test cases, listed as following: +
+meanvariancenormalization + +```python +node = onnx.helper.make_node( + "MeanVarianceNormalization", inputs=["X"], outputs=["Y"] +) + +input_data = np.array( + [ + [ + [[0.8439683], [0.5665144], [0.05836735]], + [[0.02916367], [0.12964272], [0.5060197]], + [[0.79538304], [0.9411346], [0.9546573]], + ], + [ + [[0.17730942], [0.46192095], [0.26480448]], + [[0.6746842], [0.01665257], [0.62473077]], + [[0.9240844], [0.9722341], [0.11965699]], + ], + [ + [[0.41356155], [0.9129373], [0.59330076]], + [[0.81929934], [0.7862604], [0.11799799]], + [[0.69248444], [0.54119414], [0.07513223]], + ], + ], + dtype=np.float32, +) + +# Calculate expected output data +data_mean = np.mean(input_data, axis=(0, 2, 3), keepdims=1) +data_mean_squared = np.power(data_mean, 2) +data_squared = np.power(input_data, 2) +data_squared_mean = np.mean(data_squared, axis=(0, 2, 3), keepdims=1) +std = np.sqrt(data_squared_mean - data_mean_squared) +expected_output = (input_data - data_mean) / (std + 1e-9) + +expect(node, inputs=[input_data], outputs=[expected_output], name="test_mvn") +``` + +
+ + +### MelWeightMatrix +There are 1 test cases, listed as following: +
+melweightmatrix + +```python +node = onnx.helper.make_node( + "MelWeightMatrix", + inputs=[ + "num_mel_bins", + "dft_length", + "sample_rate", + "lower_edge_hertz", + "upper_edge_hertz", + ], + outputs=["output"], +) + +num_mel_bins = np.int32(8) +dft_length = np.int32(16) +sample_rate = np.int32(8192) +lower_edge_hertz = np.float32(0) +upper_edge_hertz = np.float32(8192 / 2) + +num_spectrogram_bins = dft_length // 2 + 1 +frequency_bins = np.arange(0, num_mel_bins + 2) + +low_frequency_mel = 2595 * np.log10(1 + lower_edge_hertz / 700) +high_frequency_mel = 2595 * np.log10(1 + upper_edge_hertz / 700) +mel_step = (high_frequency_mel - low_frequency_mel) / frequency_bins.shape[0] + +frequency_bins = frequency_bins * mel_step + low_frequency_mel +frequency_bins = 700 * (np.power(10, (frequency_bins / 2595)) - 1) +frequency_bins = ((dft_length + 1) * frequency_bins) // sample_rate +frequency_bins = frequency_bins.astype(int) + +output = np.zeros((num_spectrogram_bins, num_mel_bins)) +output.flags.writeable = True + +for i in range(num_mel_bins): + lower_frequency_value = frequency_bins[i] # left + center_frequency_point = frequency_bins[i + 1] # center + higher_frequency_point = frequency_bins[i + 2] # right + low_to_center = center_frequency_point - lower_frequency_value + if low_to_center == 0: + output[center_frequency_point, i] = 1 + else: + for j in range(lower_frequency_value, center_frequency_point + 1): + output[j, i] = float(j - lower_frequency_value) / float( + low_to_center + ) + center_to_high = higher_frequency_point - center_frequency_point + if center_to_high > 0: + for j in range(center_frequency_point, higher_frequency_point): + output[j, i] = float(higher_frequency_point - j) / float( + center_to_high + ) + +# Expected output +# 1.000000, 1.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 1.000000, 1.000000, 0.000000, 0.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 1.000000, 0.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 1.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 1.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 1.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, +# 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, +output = output.astype(np.float32) +expect( + node, + inputs=[ + num_mel_bins, + dft_length, + sample_rate, + lower_edge_hertz, + upper_edge_hertz, + ], + outputs=[output], + name="test_melweightmatrix", +) +``` + +
+ + +### Min +There are 2 test cases, listed as following: +
+min + +```python +data_0 = np.array([3, 2, 1]).astype(np.float32) +data_1 = np.array([1, 4, 4]).astype(np.float32) +data_2 = np.array([2, 5, 0]).astype(np.float32) +result = np.array([1, 2, 0]).astype(np.float32) +node = onnx.helper.make_node( + "Min", + inputs=["data_0", "data_1", "data_2"], + outputs=["result"], +) +expect( + node, + inputs=[data_0, data_1, data_2], + outputs=[result], + name="test_min_example", +) + +node = onnx.helper.make_node( + "Min", + inputs=["data_0"], + outputs=["result"], +) +expect(node, inputs=[data_0], outputs=[data_0], name="test_min_one_input") + +result = np.minimum(data_0, data_1) +node = onnx.helper.make_node( + "Min", + inputs=["data_0", "data_1"], + outputs=["result"], +) +expect( + node, inputs=[data_0, data_1], outputs=[result], name="test_min_two_inputs" +) +``` + +
+
+min_all_numeric_types + +```python +for op_dtype in all_numeric_dtypes: + data_0 = np.array([3, 2, 1]).astype(op_dtype) + data_1 = np.array([1, 4, 4]).astype(op_dtype) + result = np.array([1, 2, 1]).astype(op_dtype) + node = onnx.helper.make_node( + "Min", + inputs=["data_0", "data_1"], + outputs=["result"], + ) + expect( + node, + inputs=[data_0, data_1], + outputs=[result], + name=f"test_min_{np.dtype(op_dtype).name}", + ) +``` + +
+ + +### Mish +There are 1 test cases, listed as following: +
+mish + +```python +node = onnx.helper.make_node("Mish", inputs=["X"], outputs=["Y"]) + +input_data = np.linspace(-10, 10, 10000, dtype=np.float32) + +# Calculate expected output data +expected_output = input_data * np.tanh(np.log1p(np.exp(input_data))) + +expect(node, inputs=[input_data], outputs=[expected_output], name="test_mish") +``` + +
+ + +### Mod +There are 13 test cases, listed as following: +
+mod_broadcast + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.arange(0, 30).reshape([3, 2, 5]).astype(np.int32) +y = np.array([7]).astype(np.int32) +z = np.mod(x, y) +# array([[[0, 1, 2, 3, 4], +# [5, 6, 0, 1, 2]], + +# [[3, 4, 5, 6, 0], +# [1, 2, 3, 4, 5]], + +# [[6, 0, 1, 2, 3], +# [4, 5, 6, 0, 1]]], dtype=int32) +expect(node, inputs=[x, y], outputs=[z], name="test_mod_broadcast") +``` + +
+
+mod_int64_fmod + +```python +node = onnx.helper.make_node("Mod", inputs=["x", "y"], outputs=["z"], fmod=1) + +x = np.array([-4, 7, 5, 4, -7, 8]).astype(np.int64) +y = np.array([2, -3, 8, -2, 3, 5]).astype(np.int64) +z = np.fmod(x, y) # expected output [ 0, 1, 5, 0, -1, 3] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_int64_fmod") +``` + +
+
+mod_mixed_sign_float16 + +```python +node = onnx.helper.make_node("Mod", inputs=["x", "y"], outputs=["z"], fmod=1) + +x = np.array([-4.3, 7.2, 5.0, 4.3, -7.2, 8.0]).astype(np.float16) +y = np.array([2.1, -3.4, 8.0, -2.1, 3.4, 5.0]).astype(np.float16) +z = np.fmod( + x, y +) # expected output [-0.10156, 0.3984 , 5. , 0.10156, -0.3984 , 3.] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_float16") +``` + +
+
+mod_mixed_sign_float32 + +```python +node = onnx.helper.make_node("Mod", inputs=["x", "y"], outputs=["z"], fmod=1) + +x = np.array([-4.3, 7.2, 5.0, 4.3, -7.2, 8.0]).astype(np.float32) +y = np.array([2.1, -3.4, 8.0, -2.1, 3.4, 5.0]).astype(np.float32) +z = np.fmod( + x, y +) # expected output [-0.10000038, 0.39999962, 5. , 0.10000038, -0.39999962, 3.] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_float32") +``` + +
+
+mod_mixed_sign_float64 + +```python +node = onnx.helper.make_node("Mod", inputs=["x", "y"], outputs=["z"], fmod=1) + +x = np.array([-4.3, 7.2, 5.0, 4.3, -7.2, 8.0]).astype(np.float64) +y = np.array([2.1, -3.4, 8.0, -2.1, 3.4, 5.0]).astype(np.float64) +z = np.fmod(x, y) # expected output [-0.1, 0.4, 5. , 0.1, -0.4, 3.] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_float64") +``` + +
+
+mod_mixed_sign_int16 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([-4, 7, 5, 4, -7, 8]).astype(np.int16) +y = np.array([2, -3, 8, -2, 3, 5]).astype(np.int16) +z = np.mod(x, y) # expected output [ 0, -2, 5, 0, 2, 3] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_int16") +``` + +
+
+mod_mixed_sign_int32 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([-4, 7, 5, 4, -7, 8]).astype(np.int32) +y = np.array([2, -3, 8, -2, 3, 5]).astype(np.int32) +z = np.mod(x, y) # expected output [ 0, -2, 5, 0, 2, 3] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_int32") +``` + +
+
+mod_mixed_sign_int64 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([-4, 7, 5, 4, -7, 8]).astype(np.int64) +y = np.array([2, -3, 8, -2, 3, 5]).astype(np.int64) +z = np.mod(x, y) # expected output [ 0, -2, 5, 0, 2, 3] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_int64") +``` + +
+
+mod_mixed_sign_int8 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([-4, 7, 5, 4, -7, 8]).astype(np.int8) +y = np.array([2, -3, 8, -2, 3, 5]).astype(np.int8) +z = np.mod(x, y) # expected output [ 0, -2, 5, 0, 2, 3] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_mixed_sign_int8") +``` + +
+
+mod_uint16 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([4, 7, 5]).astype(np.uint16) +y = np.array([2, 3, 8]).astype(np.uint16) +z = np.mod(x, y) # expected output [0, 1, 5] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_uint16") +``` + +
+
+mod_uint32 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([4, 7, 5]).astype(np.uint32) +y = np.array([2, 3, 8]).astype(np.uint32) +z = np.mod(x, y) # expected output [0, 1, 5] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_uint32") +``` + +
+
+mod_uint64 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([4, 7, 5]).astype(np.uint64) +y = np.array([2, 3, 8]).astype(np.uint64) +z = np.mod(x, y) # expected output [0, 1, 5] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_uint64") +``` + +
+
+mod_uint8 + +```python +node = onnx.helper.make_node( + "Mod", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([4, 7, 5]).astype(np.uint8) +y = np.array([2, 3, 8]).astype(np.uint8) +z = np.mod(x, y) # expected output [0, 1, 5] +expect(node, inputs=[x, y], outputs=[z], name="test_mod_uint8") +``` + +
+ + +### Momentum +There are 3 test cases, listed as following: +
+momentum + +```python +# Define operator attributes. +norm_coefficient = 0.001 +alpha = 0.95 +beta = 0.1 + +# Create operator. +node = onnx.helper.make_node( + "Momentum", + inputs=["R", "T", "X", "G", "V"], + outputs=["X_new", "V_new"], + norm_coefficient=norm_coefficient, + alpha=alpha, + beta=beta, + mode="standard", + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar +x = np.array([1.2, 2.8], dtype=np.float32) +g = np.array([-0.94, -2.5], dtype=np.float32) +v = np.array([1.7, 3.6], dtype=np.float32) + +# Compute expected outputs of Momentum. +x_new, v_new = apply_momentum(r, t, x, g, v, norm_coefficient, alpha, beta) + +# Check results. +expect( + node, + inputs=[r, t, x, g, v], + outputs=[x_new, v_new], + name="test_momentum", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+
+momentum_multiple + +```python +# Define operator attributes. +norm_coefficient = 0.001 +alpha = 0.95 +beta = 0.85 + +node = onnx.helper.make_node( + "Momentum", + inputs=["R", "T", "X1", "X2", "G1", "G2", "H1", "H2"], + outputs=["X1_new", "X2_new", "V1_new", "V2_new"], + norm_coefficient=norm_coefficient, + alpha=alpha, + beta=beta, + mode="standard", + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar + +x1 = np.array([1.0], dtype=np.float32) +g1 = np.array([-1.0], dtype=np.float32) +v1 = np.array([2.0], dtype=np.float32) + +x2 = np.array([1.0, 2.0], dtype=np.float32) +g2 = np.array([-1.0, -3.0], dtype=np.float32) +v2 = np.array([4.0, 1.0], dtype=np.float32) + +# Compute expected outputs of Momentum. +x1_new, v1_new = apply_momentum(r, t, x1, g1, v1, norm_coefficient, alpha, beta) +x2_new, v2_new = apply_momentum(r, t, x2, g2, v2, norm_coefficient, alpha, beta) + +# Check results. +expect( + node, + inputs=[r, t, x1, x2, g1, g2, v1, v2], + outputs=[x1_new, x2_new, v1_new, v2_new], + name="test_momentum_multiple", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+
+nesterov_momentum + +```python +# Define operator attributes. +norm_coefficient = 0.01 +alpha = 0.95 +beta = 1.0 + +# Create operator. +node = onnx.helper.make_node( + "Momentum", + inputs=["R", "T", "X", "G", "V"], + outputs=["X_new", "V_new"], + norm_coefficient=norm_coefficient, + alpha=alpha, + beta=beta, + mode="nesterov", + domain=AI_ONNX_PREVIEW_TRAINING_DOMAIN, +) + +# Define operator inputs. +r = np.array(0.1, dtype=np.float32) # scalar +t = np.array(0, dtype=np.int64) # scalar +x = np.array([1.2, 2.8], dtype=np.float32) +g = np.array([-0.94, -2.5], dtype=np.float32) +v = np.array([1.7, 3.6], dtype=np.float32) + +# Compute expected outputs of Momentum. +x_new, v_new = apply_nesterov(r, t, x, g, v, norm_coefficient, alpha, beta) + +# Check results. +expect( + node, + inputs=[r, t, x, g, v], + outputs=[x_new, v_new], + name="test_nesterov_momentum", + opset_imports=[ + onnx.helper.make_opsetid(AI_ONNX_PREVIEW_TRAINING_DOMAIN, 1) + ], +) +``` + +
+ + +### Mul +There are 2 test cases, listed as following: +
+mul + +```python +node = onnx.helper.make_node( + "Mul", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.float32) +z = x * y # expected output [4., 10., 18.] +expect(node, inputs=[x, y], outputs=[z], name="test_mul_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.int8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int8) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_int8") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.int16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.int16) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_int16") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint8) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_uint8") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint16) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_uint16") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint32) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_uint32") + +x = np.random.randint(4, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(24, size=(3, 4, 5), dtype=np.uint64) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_uint64") +``` + +
+
+mul_broadcast + +```python +node = onnx.helper.make_node( + "Mul", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = x * y +expect(node, inputs=[x, y], outputs=[z], name="test_mul_bcast") +``` + +
+ + +### Neg +There are 1 test cases, listed as following: +
+neg + +```python +node = onnx.helper.make_node( + "Neg", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-4, 2]).astype(np.float32) +y = np.negative(x) # expected output [4., -2.], +expect(node, inputs=[x], outputs=[y], name="test_neg_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.negative(x) +expect(node, inputs=[x], outputs=[y], name="test_neg") +``` + +
+ + +### NegativeLogLikelihoodLoss +There are 18 test cases, listed as following: +
+input_shape_is_NC + +```python +reduction = "none" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C = 3, 5 +np.random.seed(0) +input = np.random.rand(N, C).astype(np.float32) +target = np.random.randint(0, high=C, size=(N,)).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NC", +) +``` + +
+
+input_shape_is_NCd1 + +```python +reduction = "mean" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C, d1 = 3, 5, 2 +np.random.seed(0) +input = np.random.rand(N, C, d1).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, d1)).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1", +) +``` + +
+
+input_shape_is_NCd1_ii + +```python +reduction = "mean" +ignore_index = np.int64(1) +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, d1 = 3, 5, 2 +np.random.seed(0) +input = np.random.rand(N, C, d1).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, d1)).astype(np.int64) +target[0][0] = np.int64(1) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1_ii", +) +``` + +
+
+input_shape_is_NCd1_mean_weight_negative_ii + +```python +reduction = "mean" +ignore_index = np.int64(-1) + +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1 = 3, 5, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1)).astype(np.int64) +target[0][0] = -1 +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1_mean_weight_negative_ii", +) +``` + +
+
+input_shape_is_NCd1_weight + +```python +reduction = "mean" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, +) + +N, C, d1 = 3, 5, 2 +np.random.seed(0) +input = np.random.rand(N, C, d1).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, d1)).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1_weight", +) +``` + +
+
+input_shape_is_NCd1_weight_ii + +```python +reduction = "mean" +ignore_index = np.int64(1) +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, d1 = 3, 5, 2 +np.random.seed(0) +input = np.random.rand(N, C, d1).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, d1)).astype(np.int64) +target[0][0] = np.int64(1) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1_weight_ii", +) +``` + +
+
+input_shape_is_NCd1d2 + +```python +reduction = "none" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2", +) +``` + +
+
+input_shape_is_NCd1d2_no_weight_reduction_mean_ii + +```python +reduction = "mean" +ignore_index = np.int64(1) +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) +target[0][0][0] = np.int64(1) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_no_weight_reduction_mean_ii", +) +``` + +
+
+input_shape_is_NCd1d2_reduction_mean + +```python +reduction = "mean" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_reduction_mean", +) +``` + +
+
+input_shape_is_NCd1d2_reduction_sum + +```python +reduction = "sum" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=None, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_reduction_sum", +) +``` + +
+
+input_shape_is_NCd1d2_with_weight + +```python +reduction = "none" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_with_weight", +) +``` + +
+
+input_shape_is_NCd1d2_with_weight_reduction_mean + +```python +reduction = "mean" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_with_weight_reduction_mean", +) +``` + +
+
+input_shape_is_NCd1d2_with_weight_reduction_sum + +```python +reduction = "sum" +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_with_weight_reduction_sum", +) +``` + +
+
+input_shape_is_NCd1d2_with_weight_reduction_sum_ii + +```python +reduction = "sum" +ignore_index = np.int64(0) +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1, dim2 = 3, 5, 6, 6 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2)).astype(np.int64) +target[0][0][0] = np.int64(0) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2_with_weight_reduction_sum_ii", +) +``` + +
+
+input_shape_is_NCd1d2d3_none_no_weight_negative_ii + +```python +reduction = "none" +ignore_index = np.int64(-5) + +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1, dim2, dim3 = 3, 5, 6, 6, 5 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2, dim3).astype(np.float32) +target = np.random.randint(0, high=C, size=(N, dim1, dim2, dim3)).astype( + np.int64 +) +target[0][0][0][0] = -5 + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2d3_none_no_weight_negative_ii", +) +``` + +
+
+input_shape_is_NCd1d2d3_sum_weight_high_ii + +```python +reduction = "sum" +ignore_index = np.int64(10) + +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C = 3, 5 +np.random.seed(0) +input = np.random.rand(N, C).astype(np.float32) +target = np.random.randint(0, high=C, size=(N)).astype(np.int64) +target[0] = 10 +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2d3_sum_weight_high_ii", +) +``` + +
+
+input_shape_is_NCd1d2d3d4d5_mean_weight + +```python +reduction = "mean" + +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target", "weight"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +target = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, weight=weight, reduction=reduction +) + +expect( + node, + inputs=[input, target, weight], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2d3d4d5_mean_weight", +) +``` + +
+
+input_shape_is_NCd1d2d3d4d5_none_no_weight + +```python +reduction = "none" + +node = onnx.helper.make_node( + "NegativeLogLikelihoodLoss", + inputs=["input", "target"], + outputs=["loss"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +input = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +target = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) + +negative_log_likelihood_loss = compute_negative_log_likelihood_loss( + input, target, reduction=reduction +) + +expect( + node, + inputs=[input, target], + outputs=[negative_log_likelihood_loss], + name="test_nllloss_NCd1d2d3d4d5_none_no_weight", +) +``` + +
+ + +### NonMaxSuppression +There are 10 test cases, listed as following: +
+nonmaxsuppression_center_point_box_format + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], + center_point_box=1, +) +boxes = np.array( + [ + [ + [0.5, 0.5, 1.0, 1.0], + [0.5, 0.6, 1.0, 1.0], + [0.5, 0.4, 1.0, 1.0], + [0.5, 10.5, 1.0, 1.0], + [0.5, 10.6, 1.0, 1.0], + [0.5, 100.5, 1.0, 1.0], + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 3], [0, 0, 0], [0, 0, 5]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_center_point_box_format", +) +``` + +
+
+nonmaxsuppression_flipped_coordinates + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [1.0, 1.0, 0.0, 0.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, 0.9, 1.0, -0.1], + [0.0, 10.0, 1.0, 11.0], + [1.0, 10.1, 0.0, 11.1], + [1.0, 101.0, 0.0, 100.0], + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 3], [0, 0, 0], [0, 0, 5]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_flipped_coordinates", +) +``` + +
+
+nonmaxsuppression_identical_boxes + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.0, 1.0, 1.0], + ] + ] +).astype(np.float32) +scores = np.array( + [[[0.9, 0.9, 0.9, 0.9, 0.9, 0.9, 0.9, 0.9, 0.9, 0.9]]] +).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 0]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_identical_boxes", +) +``` + +
+
+nonmaxsuppression_iou_threshold_boundary + +```python +"""Test boundary condition where IoU exactly equals threshold. + +This test verifies that the comparison is strict (>), not inclusive (>=). +When IoU exactly equals the threshold, boxes should be KEPT, not suppressed. +This follows PyTorch's NMS implementation. +""" +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +# Two boxes with 50% overlap in each dimension +# box1=[0,0,1,1], box2=[0.5,0.5,1.5,1.5] +# Intersection area = 0.5 * 0.5 = 0.25 +# Union area = 1.0 + 1.0 - 0.25 = 1.75 +# IoU = 0.25 / 1.75 (exact value computed below as float32) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], # box 0 + [0.5, 0.5, 1.5, 1.5], # box 1 - overlaps box 0 + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.8]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +# Compute the exact IoU value and use it as threshold +# This ensures the threshold exactly equals the IoU +exact_iou = np.float32(0.25 / 1.75) +iou_threshold = np.array([exact_iou]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +# Both boxes should be selected because IoU == threshold (not > threshold) +selected_indices = np.array([[0, 0, 0], [0, 0, 1]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_iou_threshold_boundary", +) +``` + +
+
+nonmaxsuppression_limit_output_size + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]]).astype(np.float32) +max_output_boxes_per_class = np.array([2]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 3], [0, 0, 0]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_limit_output_size", +) +``` + +
+
+nonmaxsuppression_single_box + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array([[[0.0, 0.0, 1.0, 1.0]]]).astype(np.float32) +scores = np.array([[[0.9]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 0]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_single_box", +) +``` + +
+
+nonmaxsuppression_suppress_by_IOU + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array([[0, 0, 3], [0, 0, 0], [0, 0, 5]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_suppress_by_IOU", +) +``` + +
+
+nonmaxsuppression_suppress_by_IOU_and_scores + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ] + ] +).astype(np.float32) +scores = np.array([[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]]).astype(np.float32) +max_output_boxes_per_class = np.array([3]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.4]).astype(np.float32) +selected_indices = np.array([[0, 0, 3], [0, 0, 0]]).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_suppress_by_IOU_and_scores", +) +``` + +
+
+nonmaxsuppression_two_batches + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ], + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ], + ] +).astype(np.float32) +scores = np.array( + [[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]], [[0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]] +).astype(np.float32) +max_output_boxes_per_class = np.array([2]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array( + [[0, 0, 3], [0, 0, 0], [1, 0, 3], [1, 0, 0]] +).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_two_batches", +) +``` + +
+
+nonmaxsuppression_two_classes + +```python +node = onnx.helper.make_node( + "NonMaxSuppression", + inputs=[ + "boxes", + "scores", + "max_output_boxes_per_class", + "iou_threshold", + "score_threshold", + ], + outputs=["selected_indices"], +) +boxes = np.array( + [ + [ + [0.0, 0.0, 1.0, 1.0], + [0.0, 0.1, 1.0, 1.1], + [0.0, -0.1, 1.0, 0.9], + [0.0, 10.0, 1.0, 11.0], + [0.0, 10.1, 1.0, 11.1], + [0.0, 100.0, 1.0, 101.0], + ] + ] +).astype(np.float32) +scores = np.array( + [[[0.9, 0.75, 0.6, 0.95, 0.5, 0.3], [0.9, 0.75, 0.6, 0.95, 0.5, 0.3]]] +).astype(np.float32) +max_output_boxes_per_class = np.array([2]).astype(np.int64) +iou_threshold = np.array([0.5]).astype(np.float32) +score_threshold = np.array([0.0]).astype(np.float32) +selected_indices = np.array( + [[0, 0, 3], [0, 0, 0], [0, 1, 3], [0, 1, 0]] +).astype(np.int64) + +expect( + node, + inputs=[ + boxes, + scores, + max_output_boxes_per_class, + iou_threshold, + score_threshold, + ], + outputs=[selected_indices], + name="test_nonmaxsuppression_two_classes", +) +``` + +
+ + +### NonZero +There are 1 test cases, listed as following: +
+nonzero + +```python +node = onnx.helper.make_node( + "NonZero", + inputs=["condition"], + outputs=["result"], +) + +condition = np.array([[1, 0], [1, 1]], dtype=bool) +result = np.array( + np.nonzero(condition), dtype=np.int64 +) # expected output [[0, 1, 1], [0, 0, 1]] +expect(node, inputs=[condition], outputs=[result], name="test_nonzero_example") +``` + +
+ + +### Not +There are 1 test cases, listed as following: +
+not + +```python +node = onnx.helper.make_node( + "Not", + inputs=["x"], + outputs=["not"], +) + +# 2d +x = (np.random.randn(3, 4) > 0).astype(bool) +expect(node, inputs=[x], outputs=[np.logical_not(x)], name="test_not_2d") + +# 3d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +expect(node, inputs=[x], outputs=[np.logical_not(x)], name="test_not_3d") + +# 4d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +expect(node, inputs=[x], outputs=[np.logical_not(x)], name="test_not_4d") +``` + +
+ + +### OneHot +There are 5 test cases, listed as following: +
+with_axis + +```python +axisValue = 1 +on_value = 3 +off_value = 1 +output_type = np.float32 +node = onnx.helper.make_node( + "OneHot", + inputs=["indices", "depth", "values"], + outputs=["y"], + axis=axisValue, +) +indices = np.array([[1, 9], [2, 4]], dtype=np.float32) +depth = np.float32(10) +values = np.array([off_value, on_value], dtype=output_type) +y = one_hot(indices, depth, axis=axisValue, dtype=output_type) +y = y * (on_value - off_value) + off_value +expect( + node, + inputs=[indices, depth, values], + outputs=[y], + name="test_onehot_with_axis", +) +``` + +
+
+with_negative_axis + +```python +axisValue = -2 +on_value = 3 +off_value = 1 +output_type = np.float32 +node = onnx.helper.make_node( + "OneHot", + inputs=["indices", "depth", "values"], + outputs=["y"], + axis=axisValue, +) +indices = np.array([[1, 9], [2, 4]], dtype=np.float32) +depth = np.float32(10) +values = np.array([off_value, on_value], dtype=output_type) +y = one_hot(indices, depth, axis=axisValue, dtype=output_type) +y = y * (on_value - off_value) + off_value +expect( + node, + inputs=[indices, depth, values], + outputs=[y], + name="test_onehot_with_negative_axis", +) +``` + +
+
+with_negative_indices + +```python +axisValue = 1 +on_value = 3 +off_value = 1 +output_type = np.float32 +node = onnx.helper.make_node( + "OneHot", + inputs=["indices", "depth", "values"], + outputs=["y"], + axis=axisValue, +) +indices = np.array([0, -7, -8], dtype=np.int64) + +# print(y) +# [[3. 1. 1. 1. 1. 1. 1. 1. 1. 1.] +# [1. 1. 1. 3. 1. 1. 1. 1. 1. 1.] +# [1. 1. 3. 1. 1. 1. 1. 1. 1. 1.]] + +depth = np.float32(10) +values = np.array([off_value, on_value], dtype=output_type) +y = one_hot(indices, depth, axis=axisValue, dtype=output_type) +y = y * (on_value - off_value) + off_value +expect( + node, + inputs=[indices, depth, values], + outputs=[y], + name="test_onehot_negative_indices", +) +``` + +
+
+with_out_of_range_indices + +```python +axisValue = 1 +on_value = 3 +off_value = 1 +output_type = np.float32 +node = onnx.helper.make_node( + "OneHot", + inputs=["indices", "depth", "values"], + outputs=["y"], + axis=axisValue, +) +# Indices outside [-depth, depth-1] map to an all-off_value row. +indices = np.array([5, -6, -1], dtype=np.int64) + +# print(y) +# [[1. 1. 1. 1. 1.] +# [1. 1. 1. 1. 1.] +# [1. 1. 1. 1. 3.]] + +depth = np.float32(5) +values = np.array([off_value, on_value], dtype=output_type) +y = one_hot(indices, depth, axis=axisValue, dtype=output_type) +y = y * (on_value - off_value) + off_value +expect( + node, + inputs=[indices, depth, values], + outputs=[y], + name="test_onehot_out_of_range_indices", +) +``` + +
+
+without_axis + +```python +on_value = 5 +off_value = 2 +output_type = np.int32 +node = onnx.helper.make_node( + "OneHot", inputs=["indices", "depth", "values"], outputs=["y"] +) +indices = np.array([0, 7, 8], dtype=np.int64) +depth = np.float32(12) +values = np.array([off_value, on_value], dtype=output_type) +y = one_hot(indices, depth, dtype=output_type) +y = y * (on_value - off_value) + off_value +expect( + node, + inputs=[indices, depth, values], + outputs=[y], + name="test_onehot_without_axis", +) +``` + +
+ + +### OptionalHasElement +There are 4 test cases, listed as following: +
+empty + +```python +optional = None + +tensor_type_proto = onnx.helper.make_tensor_type_proto( + elem_type=onnx.TensorProto.INT32, shape=[] +) +optional_type_proto = onnx.helper.make_optional_type_proto(tensor_type_proto) + +# OptionalHasElement takes a tensor or optional as input +for input_type_proto in [tensor_type_proto, optional_type_proto]: + input_name_options = { + "empty": "optional_input", + "empty_no_input_name": "", + "empty_no_input": None, + } + for test_name_surfix, input_name in input_name_options.items(): + if input_type_proto == tensor_type_proto and input_name: + # the input tensor cannot be empty if input name is provided. + continue + node = onnx.helper.make_node( + "OptionalHasElement", + inputs=[] if input_name is None else [input_name], + outputs=["output"], + ) + output = optional_has_element_reference_implementation(optional) + test_name = ( + "test_optional_has_element_" + + test_name_surfix + + ( + "_optional_input" + if input_type_proto == optional_type_proto + else "_tensor_input" + ) + ) + expect( + node, + inputs=[optional] if input_name else [], + outputs=[output], + input_type_protos=[input_type_proto] if input_name else [], + name=test_name, + ) +``` + +
+
+get_element_sequence + +```python +optional = [np.array([1, 2, 3, 4]).astype(np.int32)] +tensor_type_proto = onnx.helper.make_tensor_type_proto( + elem_type=onnx.TensorProto.INT32, + shape=[ + 4, + ], +) +seq_type_proto = onnx.helper.make_sequence_type_proto(tensor_type_proto) +optional_type_proto = onnx.helper.make_optional_type_proto(seq_type_proto) + +node = onnx.helper.make_node( + "OptionalGetElement", inputs=["optional_input"], outputs=["output"] +) +output = optional_get_element_reference_implementation(optional) +expect( + node, + inputs=[optional], + outputs=[output], + input_type_protos=[optional_type_proto], + name="test_optional_get_element_optional_sequence", +) +expect( + node, + inputs=[optional], + outputs=[output], + input_type_protos=[seq_type_proto], + name="test_optional_get_element_sequence", +) +``` + +
+
+get_element_tensor + +```python +optional = np.array([1, 2, 3, 4]).astype(np.float32) +tensor_type_proto = onnx.helper.make_tensor_type_proto( + elem_type=onnx.TensorProto.FLOAT, + shape=[ + 4, + ], +) +optional_type_proto = onnx.helper.make_optional_type_proto(tensor_type_proto) + +node = onnx.helper.make_node( + "OptionalGetElement", inputs=["optional_input"], outputs=["output"] +) +output = optional_get_element_reference_implementation(optional) +expect( + node, + inputs=[optional], + outputs=[output], + input_type_protos=[optional_type_proto], + name="test_optional_get_element_optional_tensor", +) +expect( + node, + inputs=[optional], + outputs=[output], + input_type_protos=[tensor_type_proto], + name="test_optional_get_element_tensor", +) +``` + +
+
+optionalhaselement + +```python +optional = np.array([1, 2, 3, 4]).astype(np.float32) +tensor_type_proto = onnx.helper.make_tensor_type_proto( + elem_type=onnx.TensorProto.FLOAT, + shape=[ + 4, + ], +) +optional_type_proto = onnx.helper.make_optional_type_proto(tensor_type_proto) + +# OptionalHasElement takes a tensor or optional as input +for input_type_protos in [tensor_type_proto, optional_type_proto]: + node = onnx.helper.make_node( + "OptionalHasElement", inputs=["optional_input"], outputs=["output"] + ) + output = optional_has_element_reference_implementation(optional) + test_name = "test_optional_has_element_" + ( + "optional_input" + if input_type_protos == optional_type_proto + else "tensor_input" + ) + expect( + node, + inputs=[optional], + outputs=[output], + input_type_protos=[optional_type_proto], + name=test_name, + ) +``` + +
+ + +### Or +There are 2 test cases, listed as following: +
+or + +```python +node = onnx.helper.make_node( + "Or", + inputs=["x", "y"], + outputs=["or"], +) + +# 2d +x = (np.random.randn(3, 4) > 0).astype(bool) +y = (np.random.randn(3, 4) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or2d") + +# 3d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(3, 4, 5) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or3d") + +# 4d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or4d") +``` + +
+
+or_broadcast + +```python +node = onnx.helper.make_node( + "Or", + inputs=["x", "y"], + outputs=["or"], +) + +# 3d vs 1d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(5) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or_bcast3v1d") + +# 3d vs 2d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(4, 5) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or_bcast3v2d") + +# 4d vs 2d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(5, 6) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or_bcast4v2d") + +# 4d vs 3d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(4, 5, 6) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or_bcast4v3d") + +# 4d vs 4d +x = (np.random.randn(1, 4, 1, 6) > 0).astype(bool) +y = (np.random.randn(3, 1, 5, 6) > 0).astype(bool) +z = np.logical_or(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_or_bcast4v4d") +``` + +
+ + +### PRelu +There are 2 test cases, listed as following: +
+prelu + +```python +node = onnx.helper.make_node( + "PRelu", + inputs=["x", "slope"], + outputs=["y"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +slope = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * slope + +expect(node, inputs=[x, slope], outputs=[y], name="test_prelu_example") +``` + +
+
+prelu_broadcast + +```python +node = onnx.helper.make_node( + "PRelu", + inputs=["x", "slope"], + outputs=["y"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +slope = np.random.randn(5).astype(np.float32) +y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * slope + +expect(node, inputs=[x, slope], outputs=[y], name="test_prelu_broadcast") +``` + +
+ + +### Pad +There are 4 test cases, listed as following: +
+constant_pad + +```python +node = onnx.helper.make_node( + "Pad", inputs=["x", "pads", "value"], outputs=["y"], mode="constant" +) +x = np.random.randn(1, 3, 4, 5).astype(np.float32) +pads = np.array([0, 0, 1, 3, 0, 0, 2, 4]).astype( + np.int64 +) # pad order [x1_begin, x2_begin, ..., x1_end, x2_end, ...] +value = np.float32(1.2) +y = pad_impl(x, pads, "constant", 1.2) + +expect(node, inputs=[x, pads, value], outputs=[y], name="test_constant_pad") +``` + +
+
+constant_pad_axes + +```python +node = onnx.helper.make_node( + "Pad", inputs=["x", "pads", "value", "axes"], outputs=["y"], mode="constant" +) +x = np.random.randn(1, 3, 4, 5).astype(np.float32) +pads = np.array([0, 3, 0, 4]).astype( + np.int64 +) # pad order [x1_begin, x2_begin, ..., x1_end, x2_end, ...] +value = np.float32(1.2) +axes = np.array([1, 3], dtype=np.int64) +y = pad_impl( + x, + pads, + "constant", + 1.2, + [1, 3], +) + +expect( + node, + inputs=[x, pads, value, axes], + outputs=[y], + name="test_constant_pad_axes", +) +``` + +
+
+constant_pad_negative_axes + +```python +node = onnx.helper.make_node( + "Pad", inputs=["x", "pads", "value", "axes"], outputs=["y"], mode="constant" +) +x = np.random.randn(1, 3, 4, 5).astype(np.float32) +pads = np.array([0, 3, 0, 4]).astype( + np.int64 +) # pad order [x1_begin, x2_begin, ..., x1_end, x2_end, ...] +value = np.float32(1.2) +axes = np.array([-3, -1], dtype=np.int64) +y = pad_impl( + x, + pads, + "constant", + 1.2, + [-3, -1], +) + +expect( + node, + inputs=[x, pads, value, axes], + outputs=[y], + name="test_constant_pad_negative_axes", +) +``` + +
+
+reflection_edge_and_wrap_pad + +```python +for mode in ("edge", "reflect", "wrap"): + node = onnx.helper.make_node( + "Pad", inputs=["x", "pads"], outputs=["y"], mode=mode + ) + x = np.random.randn(1, 3, 4, 5).astype(np.int32) + pads = np.array([0, 0, 1, 1, 0, 0, 1, 1]).astype( + np.int64 + ) # pad order [x1_begin, x2_begin, ..., x1_end, x2_end, ...] + y = pad_impl(x, pads, mode) + + expect(node, inputs=[x, pads], outputs=[y], name=f"test_{mode}_pad") +``` + +
+ + +### Pow +There are 3 test cases, listed as following: +
+pow + +```python +node = onnx.helper.make_node( + "Pow", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.float32) +z = pow(x, y) # expected output [1., 32., 729.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_example") + +x = np.arange(60).reshape(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = pow(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_pow") +``` + +
+
+pow_broadcast + +```python +node = onnx.helper.make_node( + "Pow", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array(2).astype(np.float32) +z = pow(x, y) # expected output [1., 4., 9.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_bcast_scalar") + +node = onnx.helper.make_node( + "Pow", + inputs=["x", "y"], + outputs=["z"], +) +x = np.array([[1, 2, 3], [4, 5, 6]]).astype(np.float32) +y = np.array([1, 2, 3]).astype(np.float32) +# expected output [[1, 4, 27], [4, 25, 216]] +z = pow(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_pow_bcast_array") +``` + +
+
+types + +```python +node = onnx.helper.make_node( + "Pow", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.int64) +z = pow(x, y) # expected output [1., 32., 729.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_float32_int64") + +x = np.array([1, 2, 3]).astype(np.int64) +y = np.array([4, 5, 6]).astype(np.float32) +z = pow(x, y) # expected output [1, 32, 729] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_int64_float32") + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.int32) +z = pow(x, y) # expected output [1., 32., 729.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_float32_int32") + +x = np.array([1, 2, 3]).astype(np.int32) +y = np.array([4, 5, 6]).astype(np.float32) +z = pow(x, y) # expected output [1, 32, 729] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_int32_float32") + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.uint64) +z = pow(x, y) # expected output [1., 32., 729.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_float32_uint64") + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([4, 5, 6]).astype(np.uint32) +z = pow(x, y) # expected output [1., 32., 729.] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_float32_uint32") + +x = np.array([1, 2, 3]).astype(np.int64) +y = np.array([4, 5, 6]).astype(np.int64) +z = pow(x, y) # expected output [1, 32, 729] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_int64_int64") + +x = np.array([1, 2, 3]).astype(np.int32) +y = np.array([4, 5, 6]).astype(np.int32) +z = pow(x, y) # expected output [1, 32, 729] +expect(node, inputs=[x, y], outputs=[z], name="test_pow_types_int32_int32") +``` + +
+ + +### QLinearConv +There are 1 test cases, listed as following: +
+qlinearconv + +```python +node = onnx.helper.make_node( + "QLinearConv", + inputs=[ + "x", + "x_scale", + "x_zero_point", + "w", + "w_scale", + "w_zero_point", + "y_scale", + "y_zero_point", + ], + outputs=["y"], +) + +x = np.array( + [ + [255, 174, 162, 25, 203, 168, 58], + [15, 59, 237, 95, 129, 0, 64], + [56, 242, 153, 221, 168, 12, 166], + [232, 178, 186, 195, 237, 162, 237], + [188, 39, 124, 77, 80, 102, 43], + [127, 230, 21, 83, 41, 40, 134], + [255, 154, 92, 141, 42, 148, 247], + ], + dtype=np.uint8, +).reshape((1, 1, 7, 7)) + +x_scale = np.float32(0.00369204697) +x_zero_point = np.uint8(132) + +w = np.array([0], dtype=np.uint8).reshape((1, 1, 1, 1)) + +w_scale = np.array([0.00172794575], dtype=np.float32) +w_zero_point = np.array([255], dtype=np.uint8) + +y_scale = np.float32(0.00162681262) +y_zero_point = np.uint8(123) + +output = np.array( + [ + [0, 81, 93, 230, 52, 87, 197], + [240, 196, 18, 160, 126, 255, 191], + [199, 13, 102, 34, 87, 243, 89], + [23, 77, 69, 60, 18, 93, 18], + [67, 216, 131, 178, 175, 153, 212], + [128, 25, 234, 172, 214, 215, 121], + [0, 101, 163, 114, 213, 107, 8], + ], + dtype=np.uint8, +).reshape((1, 1, 7, 7)) + +expect( + node, + inputs=[ + x, + x_scale, + x_zero_point, + w, + w_scale, + w_zero_point, + y_scale, + y_zero_point, + ], + outputs=[output], + name="test_qlinearconv", +) +``` + +
+ + +### QLinearMatMul +There are 1 test cases, listed as following: +
+int + +```python +for quant_type_name in ["uint8", "int8"]: + quant_type = getattr(np, quant_type_name) + for dtype_name in ["float32", "float16"]: + dtype = getattr(np, dtype_name) + node = onnx.helper.make_node( + "QLinearMatMul", + inputs=[ + "a", + "a_scale", + "a_zero_point", + "b", + "b_scale", + "b_zero_point", + "y_scale", + "y_zero_point", + ], + outputs=["y"], + ) + + # 2D + a = np.array([[208, 236, 0, 238], [3, 214, 255, 29]]) + if quant_type == np.int8: + a -= 127 + a = a.astype(quant_type) + + a_scale = np.array([0.0066], dtype=dtype) + a_zero_point = np.array( + [113 - 127] if quant_type == np.int8 else [113], dtype=quant_type + ) + + b = np.array( + [[152, 51, 244], [60, 26, 255], [0, 127, 246], [127, 254, 247]] + ) + if quant_type == np.int8: + b -= 127 + b = b.astype(quant_type) + + b_scale = np.array([0.00705], dtype=dtype) + b_zero_point = np.array( + [114 - 127] if quant_type == np.int8 else [114], dtype=quant_type + ) + + y_scale = np.array([0.0107], dtype=dtype) + y_zero_point = np.array( + [118 - 127] if quant_type == np.int8 else [118], dtype=quant_type + ) + + if quant_type == np.int8: + output = np.array([[41, -12, -9], [1, -75, -128]]) + else: + output = np.array([[168, 115, 255], [1, 66, 151]]) + output = output.astype(quant_type) + + expect( + node, + inputs=[ + a, + a_scale, + a_zero_point, + b, + b_scale, + b_zero_point, + y_scale, + y_zero_point, + ], + outputs=[output], + name=f"test_qlinearmatmul_2D_{quant_type_name}_{dtype_name}", + ) + + # 3D + a = np.array( + [ + [[208, 236, 0, 238], [3, 214, 255, 29]], + [[208, 236, 0, 238], [3, 214, 255, 29]], + ], + ) + if quant_type == np.int8: + a -= 127 + a = a.astype(quant_type) + + a_scale = np.array([0.0066], dtype=dtype) + a_zero_point = np.array( + [113 - 127] if quant_type == np.int8 else [113], dtype=quant_type + ) + + b = np.array( + [ + [[152, 51, 244], [60, 26, 255], [0, 127, 246], [127, 254, 247]], + [[152, 51, 244], [60, 26, 255], [0, 127, 246], [127, 254, 247]], + ], + ) + if quant_type == np.int8: + b -= 127 + b = b.astype(quant_type) + + b_scale = np.array([0.00705], dtype=dtype) + b_zero_point = np.array([114], dtype=quant_type) + + y_scale = np.array([0.0107], dtype=dtype) + y_zero_point = np.array( + [118 - 127] if quant_type == np.int8 else [118], dtype=quant_type + ) + + if quant_type == np.int8: + output = np.array( + [ + [[-86, -128, -128], [115, 39, -121]], + [[-86, -128, -128], [115, 39, -121]], + ] + ) + else: + output = np.array( + [ + [[168, 115, 255], [1, 66, 151]], + [[168, 115, 255], [1, 66, 151]], + ] + ) + output = output.astype(quant_type) + + expect( + node, + inputs=[ + a, + a_scale, + a_zero_point, + b, + b_scale, + b_zero_point, + y_scale, + y_zero_point, + ], + outputs=[output], + name=f"test_qlinearmatmul_3D_{quant_type_name}_{dtype_name}", + ) +``` + +
+ + +### QuantizeLinear +There are 13 test cases, listed as following: +
+axis + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array( + [ + [ + [[-162, 10], [-100, 232], [-20, -50]], + [[-76, 0], [0, 252], [32, -44]], + [[245, -485], [-960, -270], [-375, -470]], + ], + ], + dtype=np.float32, +) +y_scale = np.array([2, 4, 5], dtype=np.float32) +y_zero_point = np.array([84, 24, 196], dtype=np.uint8) +y = (x / y_scale.reshape(1, 3, 1, 1) + y_zero_point.reshape(1, 3, 1, 1)).astype( + np.uint8 +) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_axis", +) +``` + +
+
+blocked_asymmetric + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=1, + block_size=2, +) + +x = np.array( + [ + [6.0, 12.0, 50.0, 5.0], + [1.0, 8.0, 4.0, 5.0], + [0.0, 20.0, 10.0, 4.0], + ], + dtype=np.float32, +) +y_scale = np.array( + [ + [1.5, 2.5], + [3.0, 4.9], + [5.1, 6.9], + ], + dtype=np.float32, +) +y_zero_point = np.array( + [ + [0, 1], + [1, 0], + [2, 3], + ], + dtype=np.uint8, +) +# x.shape = (3, 4) +# y_scale.shape = (3, 2) +assert y_scale.shape == y_zero_point.shape +block_axis = 1 +# The block shape is [x.shape[i] // y_scale.shape[i] for i in range(len(x.shape))] = (1, 2) +assert all( + x.shape[i] == y_scale.shape[i] + for i in range(len(x.shape)) + if i != block_axis +) +assert x.shape[block_axis] % y_scale.shape[block_axis] == 0 +repeats = x.shape[block_axis] // y_scale.shape[block_axis] + +# Create element-wise scale and zero point +y_scale_elementwise = np.repeat(y_scale, repeats=repeats, axis=block_axis) +y_zero_point_elementwise = np.repeat( + y_zero_point, repeats=repeats, axis=block_axis +) + +y = np.rint(x / y_scale_elementwise + y_zero_point_elementwise).astype(np.uint8) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_blocked_asymmetric", +) +``` + +
+
+blocked_symmetric + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale"], + outputs=["y"], + axis=1, + block_size=2, + output_dtype=TensorProto.INT16, +) + +x = np.array( + [ + [6.0, -8, -10, 5.0], + [1.0, 8.0, 4.0, 5.0], + [0.0, 20.0, 10.0, 4.0], + ], + dtype=np.float32, +) + +y_scale = np.array( + [ + [1.5, 2.5], + [3.0, 4.9], + [5.1, 6.9], + ], + dtype=np.float32, +) + +# x.shape = (3, 4) +# y_scale.shape = (3, 2) + +block_axis = 1 +# The block shape is [x.shape[i] // y_scale.shape[i] for i in range(len(x.shape))] = (1, 2) +assert all( + x.shape[i] == y_scale.shape[i] + for i in range(len(x.shape)) + if i != block_axis +) +assert x.shape[block_axis] % y_scale.shape[block_axis] == 0 +repeats = x.shape[block_axis] // y_scale.shape[block_axis] + +# Create element-wise scale and zero point +y_scale_elementwise = np.repeat(y_scale, repeats=repeats, axis=block_axis) + +y_val = np.clip( + np.rint(x / y_scale_elementwise), a_min=-32768, a_max=32767 +).astype(np.int16) +y = make_tensor( + "y", + TensorProto.INT16, + x.shape, + y_val, +) +expect( + node, + inputs=[x, y_scale], + outputs=[y], + name="test_quantizelinear_blocked_symmetric", +) +``` + +
+
+e4m3fn + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array([0.0, 1.0, 2.0, 100000.0, 200.0]).astype(np.float32) +y_scale = np.float32(2) +y_zero_point = make_tensor("y_zero_point", TensorProto.FLOAT8E4M3FN, [1], [0]) +y = make_tensor("y", TensorProto.FLOAT8E4M3FN, [5], [0, 0.5, 1, 448, 96]) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_e4m3fn", +) +``` + +
+
+e5m2 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array([0.0, 1.0, 2.0, 100000.0, 200.0]).astype(np.float32) +y_scale = np.float32(2) +y_zero_point = make_tensor("y_zero_point", TensorProto.FLOAT8E5M2, [1], [0.0]) +y = make_tensor("y", TensorProto.FLOAT8E5M2, [5], [0, 0.5, 1, 49152, 96]) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_e5m2", +) +``` + +
+
+float4e2m1 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=0, +) + +x = np.array( + [ + [0.0, 2.5, 4.8, 8.6], + [-30, -20, 6, 9], + [-0.0, -2.5, -4.8, -8.6], + ] +).astype(np.float32) + +y_scale = np.asarray([2.0, 3.0, 4.0], dtype=np.float32) +y_zero_point = make_tensor( + "y_zero_point", + TensorProto.FLOAT4E2M1, + y_scale.shape, + np.zeros_like(y_scale), +) +y = make_tensor( + "y", + TensorProto.FLOAT4E2M1, + x.shape, + [0, 1, 2, 4, -6, -6, 2, 3, 0, -0.5, -1, -2], +) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_float4e2m1", +) +``` + +
+
+int16 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array( + [ + 0.0, + -514.0, + 3.0, + -3.0, + 2.9, + -2.9, + 3.1, + -3.1, + 65022.0, + -66046.0, + 65023.0, + -66047.0, + 65024.0, + -66048.0, + 70000.0, + -70000.0, + ] +).astype(np.float32) +y_scale = np.float32(2.0) +y_zero_point = np.int16(256) +y = np.array( + [ + 256, + -1, + 258, + 254, + 257, + 255, + 258, + 254, + 32767, + -32767, + 32767, + -32768, + 32767, + -32768, + 32767, + -32768, + ] +).astype(np.int16) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_int16", +) +``` + +
+
+int2 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=0, +) +x = np.array( + [ + [0.0, 2.5, 4.8, 8.6], + [-4.0, -3.0, 1.0, 2.0], + [-0.0, -2.5, -4.8, -8.6], + ], + dtype=np.float32, +) +y_scale = np.asarray([2.0, 3.0, 4.0], dtype=np.float32) +y_zero_point = make_tensor( + "y_zero_point", TensorProto.INT2, y_scale.shape, np.zeros_like(y_scale) +) +y = make_tensor( + "y", TensorProto.INT2, x.shape, [0, 1, 1, 1, -1, -1, 0, 1, 0, -1, -1, -2] +) +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_int2", +) +``` + +
+
+int4 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=0, +) + +x = np.array( + [ + [0.0, 2.5, 4.8, 8.6], + [-30, -20, 6, 9], + [12, 15, 16, 40], + ] +).astype(np.float32) + +y_scale = np.asarray([2.0, 3.0, 4.0], dtype=np.float32) +y_zero_point = make_tensor( + "y_zero_point", TensorProto.INT4, y_scale.shape, np.ones_like(y_scale) +) +y = make_tensor( + "y", TensorProto.INT4, x.shape, [1, 2, 3, 5, -8, -6, 3, 4, 4, 5, 5, 7] +) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_int4", +) +``` + +
+
+quantizelinear + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array([0, 2, 3, 1000, -254, -1000]).astype(np.float32) +y_scale = np.float32(2) +y_zero_point = np.uint8(128) +y = np.array([128, 129, 130, 255, 1, 0]).astype(np.uint8) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear", +) +``` + +
+
+uint16 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], +) + +x = np.array( + [ + 0.0, + -128.0, + 3.0, + -3.0, + 2.9, + -2.9, + 3.1, + -3.1, + 65536.0, + -65534.0, + 70000.0, + -70000.0, + ] +).astype(np.float32) +y_scale = np.float32(2.0) +y_zero_point = np.uint16(32767) +y = np.array( + [ + 32767, + 32703, + 32769, + 32765, + 32768, + 32766, + 32769, + 32765, + 65535, + 0, + 65535, + 0, + ] +).astype(np.uint16) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_uint16", +) +``` + +
+
+uint2 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=0, +) + +x = np.array( + [ + [0.0, 2.5, 4.8, 8.6], + [-2.0, -1.0, 1.0, 3.0], + [4.0, 5.0, 6.0, 7.0], + ], + dtype=np.float32, +) +y_scale = np.asarray([2.0, 3.0, 4.0], dtype=np.float32) +y_zero_point = make_tensor( + "y_zero_point", TensorProto.UINT2, y_scale.shape, np.zeros_like(y_scale) +) +y = make_tensor( + "y", TensorProto.UINT2, x.shape, [0, 1, 2, 3, 0, 0, 0, 1, 1, 1, 2, 2] +) +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_uint2", +) +``` + +
+
+uint4 + +```python +node = onnx.helper.make_node( + "QuantizeLinear", + inputs=["x", "y_scale", "y_zero_point"], + outputs=["y"], + axis=0, +) + +x = np.array( + [ + [0.0, 2.5, 4.8, 8.6], + [-30, -20, 6, 9], + [12, 15, 16, 40], + ] +).astype(np.float32) + +y_scale = np.asarray([2.0, 3.0, 4.0], dtype=np.float32) +y_zero_point = make_tensor( + "y_zero_point", TensorProto.UINT4, y_scale.shape, np.ones_like(y_scale) +) +y = make_tensor( + "y", TensorProto.UINT4, x.shape, [1, 2, 3, 5, 0, 0, 3, 4, 4, 5, 5, 11] +) + +expect( + node, + inputs=[x, y_scale, y_zero_point], + outputs=[y], + name="test_quantizelinear_uint4", +) +``` + +
+ + +### RMSNormalization +There are 4 test cases, listed as following: +
+d + +```python +X = np.random.randn(3, 4).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + Y = _rms_normalization(X, W, axis=axis) + + node = onnx.helper.make_node( + "RMSNormalization", + inputs=["X", "W"], + outputs=["Y"], + axis=axis, + ) + + if axis < 0: + name = f"test_rms_normalization_2d_axis_negative_{-axis}" + else: + name = f"test_rms_normalization_2d_axis{axis}" + + expect(node, inputs=[X, W], outputs=[Y], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+
+d_epsilon + +```python +epsilon = 1e-1 +X = np.random.randn(2, 3, 5).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + Y = _rms_normalization(X, W, axis=axis, epsilon=epsilon) + node = onnx.helper.make_node( + "RMSNormalization", + inputs=["X", "W"], + outputs=["Y"], + axis=axis, + epsilon=epsilon, + ) + + if axis < 0: + name = f"test_rms_normalization_3d_axis_negative_{-axis}_epsilon" + else: + name = f"test_rms_normalization_3d_axis{axis}_epsilon" + + expect(node, inputs=[X, W], outputs=[Y], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+
+default_axis + +```python +X = np.random.randn(2, 3, 4, 5).astype(np.float32) + +# Default axis in RMSNormalization is -1. +normalized_shape = calculate_normalized_shape(X.shape, -1) +W = np.random.randn(*normalized_shape).astype(np.float32) +# Axis is default to -1 in the reference implementation. +Y = _rms_normalization(X, W) + +# Not specifying axis attribute means -1. +node = onnx.helper.make_node( + "RMSNormalization", + inputs=["X", "W"], + outputs=["Y"], +) + +expect( + node, + inputs=[X, W], + outputs=[Y], + name="test_rms_normalization_default_axis", +) +``` + +
+
+rmsnormalization + +```python +X = np.random.randn(2, 3, 4, 5).astype(np.float32) + +def case(axis: int) -> None: + normalized_shape = calculate_normalized_shape(X.shape, axis) + W = np.random.randn(*normalized_shape).astype(np.float32) + Y = _rms_normalization(X, W, axis=axis) + + node = onnx.helper.make_node( + "RMSNormalization", + inputs=["X", "W"], + outputs=["Y"], + axis=axis, + ) + + if axis < 0: + name = f"test_rms_normalization_4d_axis_negative_{-axis}" + else: + name = f"test_rms_normalization_4d_axis{axis}" + + expect(node, inputs=[X, W], outputs=[Y], name=name) + +for i in range(len(X.shape)): + case(i) + case(i - len(X.shape)) +``` + +
+ + +### RNN +There are 4 test cases, listed as following: +
+batchwise + +```python +input = np.array([[[1.0, 2.0]], [[3.0, 4.0]], [[5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 4 +weight_scale = 0.5 +layout = 1 + +node = onnx.helper.make_node( + "RNN", + inputs=["X", "W", "R"], + outputs=["Y", "Y_h"], + hidden_size=hidden_size, + layout=layout, +) + +W = weight_scale * np.ones((1, hidden_size, input_size)).astype(np.float32) +R = weight_scale * np.ones((1, hidden_size, hidden_size)).astype(np.float32) + +rnn = RNNHelper(X=input, W=W, R=R, layout=layout) +Y, Y_h = rnn.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y.astype(np.float32), Y_h.astype(np.float32)], + name="test_simple_rnn_batchwise", +) +``` + +
+
+defaults + +```python +input = np.array([[[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]]]).astype(np.float32) + +input_size = 2 +hidden_size = 4 +weight_scale = 0.1 + +node = onnx.helper.make_node( + "RNN", inputs=["X", "W", "R"], outputs=["", "Y_h"], hidden_size=hidden_size +) + +W = weight_scale * np.ones((1, hidden_size, input_size)).astype(np.float32) +R = weight_scale * np.ones((1, hidden_size, hidden_size)).astype(np.float32) + +rnn = RNNHelper(X=input, W=W, R=R) +_, Y_h = rnn.step() +expect( + node, + inputs=[input, W, R], + outputs=[Y_h.astype(np.float32)], + name="test_simple_rnn_defaults", +) +``` + +
+
+initial_bias + +```python +input = np.array([[[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]]]).astype( + np.float32 +) + +input_size = 3 +hidden_size = 5 +custom_bias = 0.1 +weight_scale = 0.1 + +node = onnx.helper.make_node( + "RNN", + inputs=["X", "W", "R", "B"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +W = weight_scale * np.ones((1, hidden_size, input_size)).astype(np.float32) +R = weight_scale * np.ones((1, hidden_size, hidden_size)).astype(np.float32) + +# Adding custom bias +W_B = custom_bias * np.ones((1, hidden_size)).astype(np.float32) +R_B = np.zeros((1, hidden_size)).astype(np.float32) +B = np.concatenate((W_B, R_B), axis=1) + +rnn = RNNHelper(X=input, W=W, R=R, B=B) +_, Y_h = rnn.step() +expect( + node, + inputs=[input, W, R, B], + outputs=[Y_h.astype(np.float32)], + name="test_simple_rnn_with_initial_bias", +) +``` + +
+
+seq_length + +```python +input = np.array( + [ + [[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]], + [[10.0, 11.0, 12.0], [13.0, 14.0, 15.0], [16.0, 17.0, 18.0]], + ] +).astype(np.float32) + +input_size = 3 +hidden_size = 5 + +node = onnx.helper.make_node( + "RNN", + inputs=["X", "W", "R", "B"], + outputs=["", "Y_h"], + hidden_size=hidden_size, +) + +W = np.random.randn(1, hidden_size, input_size).astype(np.float32) +R = np.random.randn(1, hidden_size, hidden_size).astype(np.float32) + +# Adding custom bias +W_B = np.random.randn(1, hidden_size).astype(np.float32) +R_B = np.random.randn(1, hidden_size).astype(np.float32) +B = np.concatenate((W_B, R_B), axis=1) + +rnn = RNNHelper(X=input, W=W, R=R, B=B) +_, Y_h = rnn.step() +expect( + node, + inputs=[input, W, R, B], + outputs=[Y_h.astype(np.float32)], + name="test_rnn_seq_length", +) +``` + +
+ + +### Range +There are 4 test cases, listed as following: +
+range_bfloat16_type_positive_delta + +```python +node = onnx.helper.make_node( + "Range", + inputs=["start", "limit", "delta"], + outputs=["output"], +) + +start = np.array(1.0, dtype=ml_dtypes.bfloat16) +limit = np.array(5.0, dtype=ml_dtypes.bfloat16) +delta = np.array(2.0, dtype=ml_dtypes.bfloat16) + +output = np.arange(1.0, 5.0, 2.0, dtype=np.float32).astype( + ml_dtypes.bfloat16 +) # expected output [1.0, 3.0] as bfloat16 + +expect( + node, + inputs=[start, limit, delta], + outputs=[output], + name="test_range_bfloat16_type_positive_delta", +) +``` + +
+
+range_float16_type_positive_delta + +```python +node = onnx.helper.make_node( + "Range", + inputs=["start", "limit", "delta"], + outputs=["output"], +) + +start = np.float16(1) +limit = np.float16(5) +delta = np.float16(2) + +output = np.arange( + start, limit, delta, dtype=np.float16 +) # expected output [1.0, 3.0] +expect( + node, + inputs=[start, limit, delta], + outputs=[output], + name="test_range_float16_type_positive_delta", +) +``` + +
+
+range_float_type_positive_delta + +```python +node = onnx.helper.make_node( + "Range", + inputs=["start", "limit", "delta"], + outputs=["output"], +) + +start = np.float32(1) +limit = np.float32(5) +delta = np.float32(2) + +output = np.arange( + start, limit, delta, dtype=np.float32 +) # expected output [1.0, 3.0] +expect( + node, + inputs=[start, limit, delta], + outputs=[output], + name="test_range_float_type_positive_delta", +) +``` + +
+
+range_int32_type_negative_delta + +```python +node = onnx.helper.make_node( + "Range", + inputs=["start", "limit", "delta"], + outputs=["output"], +) + +start = np.int32(10) +limit = np.int32(6) +delta = np.int32(-3) + +output = np.arange( + start, limit, delta, dtype=np.int32 +) # expected output [10, 7] +expect( + node, + inputs=[start, limit, delta], + outputs=[output], + name="test_range_int32_type_negative_delta", +) +``` + +
+ + +### Reciprocal +There are 1 test cases, listed as following: +
+reciprocal + +```python +node = onnx.helper.make_node( + "Reciprocal", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-4, 2]).astype(np.float32) +y = np.reciprocal(x) # expected output [-0.25, 0.5], +expect(node, inputs=[x], outputs=[y], name="test_reciprocal_example") + +x = np.random.rand(3, 4, 5).astype(np.float32) + 0.5 +y = np.reciprocal(x) +expect(node, inputs=[x], outputs=[y], name="test_reciprocal") +``` + +
+ + +### ReduceL1 +There are 5 test cases, listed as following: +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL1", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sum(a=np.abs(data), axis=None, keepdims=keepdims == 1) +# print(reduced) +# [[[78.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(a=np.abs(data), axis=None, keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_default_axes_keepdims_random", +) +``` + +
+
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([2], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceL1", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[3., 7.], [11., 15.], [19., 23.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_do_not_keepdims_random", +) +``` + +
+
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceL1", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +reduced = np.array(np.zeros(reduced_shape, dtype=np.float32)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_empty_set", +) +``` + +
+
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL1", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[3.], [7.]], [[11.], [15.]], [[19.], [23.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_keep_dims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_keep_dims_random", +) +``` + +
+
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL1", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[3.], [7.]], [[11.], [15.]], [[19.], [23.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_negative_axes_keep_dims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(a=np.abs(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l1_negative_axes_keep_dims_random", +) +``` + +
+ + +### ReduceL2 +There are 5 test cases, listed as following: +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL2", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sqrt(np.sum(a=np.square(data), axis=None, keepdims=keepdims == 1)) +# print(reduced) +# [[[25.49509757]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sqrt(np.sum(a=np.square(data), axis=None, keepdims=keepdims == 1)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_default_axes_keepdims_random", +) +``` + +
+
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([2], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceL2", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) +# print(reduced) +# [[2.23606798, 5.], +# [7.81024968, 10.63014581], +# [13.45362405, 16.2788206]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_do_not_keepdims_random", +) +``` + +
+
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceL2", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +reduced = np.array(np.zeros(reduced_shape, dtype=np.float32)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_empty_set", +) +``` + +
+
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL2", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) +# print(reduced) +# [[[2.23606798], [5.]] +# [[7.81024968], [10.63014581]] +# [[13.45362405], [16.2788206 ]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_keep_dims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_keep_dims_random", +) +``` + +
+
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceL2", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape) +# print(data) +# [[[1., 2.], [3., 4.]], [[5., 6.], [7., 8.]], [[9., 10.], [11., 12.]]] + +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) +# print(reduced) +# [[[2.23606798], [5.]] +# [[7.81024968], [10.63014581]] +# [[13.45362405], [16.2788206 ]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_negative_axes_keep_dims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sqrt( + np.sum(a=np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_l2_negative_axes_keep_dims_random", +) +``` + +
+ + +### ReduceLogSum +There are 4 test cases, listed as following: +
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceLogSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +zero = np.array(np.zeros(reduced_shape, dtype=np.float32)) +reduced = np.log(zero) # -inf + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_empty_set", +) +``` + +
+
+keepdims + +```python +node = onnx.helper.make_node( + "ReduceLogSum", inputs=["data", "axes"], outputs=["reduced"] +) +data = np.random.ranf([3, 4, 5]).astype(np.float32) +reduced = np.log(np.sum(data, keepdims=True)) +axes = np.array([], dtype=np.int64) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_default", +) +``` + +
+
+negative_axes_keepdims + +```python +axes = np.array([-2], dtype=np.int64) +node = onnx.helper.make_node( + "ReduceLogSum", inputs=["data", "axes"], outputs=["reduced"] +) +data = np.random.ranf([3, 4, 5]).astype(np.float32) +reduced = np.log(np.sum(data, axis=tuple(axes), keepdims=True)) +# print(reduced) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_negative_axes", +) +``` + +
+
+nokeepdims + +```python +shape = [3, 4, 5] +axes = np.array([2, 1], dtype=np.int64) + +node = onnx.helper.make_node( + "ReduceLogSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=0, +) +data = np.random.ranf(shape).astype(np.float32) +reduced = np.log(np.sum(data, axis=tuple(axes), keepdims=False)) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_desc_axes", +) + +axes = np.array([0, 1], dtype=np.int64) +node = onnx.helper.make_node( + "ReduceLogSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=0, +) +data = np.random.ranf(shape).astype(np.float32) +reduced = np.log(np.sum(data, axis=tuple(axes), keepdims=False)) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_asc_axes", +) +``` + +
+ + +### ReduceLogSumExp +There are 5 test cases, listed as following: +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceLogSumExp", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], dtype=np.double +) +reduced = np.log(np.sum(np.exp(data), axis=None, keepdims=keepdims == 1)) +# print(reduced) +# [[[60.00671387]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.double) +reduced = np.log(np.sum(np.exp(data), axis=None, keepdims=keepdims == 1)) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_default_axes_keepdims_random", +) +``` + +
+
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 +node = onnx.helper.make_node( + "ReduceLogSumExp", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], dtype=np.double +) +reduced = np.log(np.sum(np.exp(data), axis=tuple(axes), keepdims=keepdims == 1)) +# print(reduced) +# [[20., 2.31326175] +# [40.00004578, 2.31326175] +# [60.00671387, 2.31326175]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.double) +reduced = np.log(np.sum(np.exp(data), axis=tuple(axes), keepdims=keepdims == 1)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_do_not_keepdims_random", +) +``` + +
+
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceLogSumExp", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +zero = np.array(np.zeros(reduced_shape, dtype=np.float32)) +reduced = np.log(zero) # -inf + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_empty_set", +) +``` + +
+
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 +node = onnx.helper.make_node( + "ReduceLogSumExp", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], dtype=np.double +) +reduced = np.log(np.sum(np.exp(data), axis=tuple(axes), keepdims=keepdims == 1)) +# print(reduced) +# [[[20., 2.31326175]] +# [[40.00004578, 2.31326175]] +# [[60.00671387, 2.31326175]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.double) +reduced = np.log(np.sum(np.exp(data), axis=tuple(axes), keepdims=keepdims == 1)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_keepdims_random", +) +``` + +
+
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 +node = onnx.helper.make_node( + "ReduceLogSumExp", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], dtype=np.double +) +reduced = np.log(np.sum(np.exp(data), axis=tuple(axes), keepdims=keepdims == 1)) +# print(reduced) +# [[[20., 2.31326175]] +# [[40.00004578, 2.31326175]] +# [[60.00671387, 2.31326175]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_negative_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.double) +reduced = np.log( + np.sum(np.exp(data), axis=tuple(axes.tolist()), keepdims=keepdims == 1) +) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_log_sum_exp_negative_axes_keepdims_random", +) +``` + +
+ + +### ReduceMax +There are 6 test cases, listed as following: +
+bool_inputs + +```python +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMax", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[True, True], [True, False], [False, True], [False, False]], +) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=bool(keepdims)) +# print(reduced) +# [[True], +# [True], +# [True], +# [False]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_bool_inputs", +) +``` + +
+
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = None +keepdims = 1 +node = onnx.helper.make_node( + "ReduceMax", inputs=["data"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.maximum.reduce(data, axis=axes, keepdims=keepdims == 1) + +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_max_default_axes_keepdim_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.maximum.reduce(data, axis=axes, keepdims=keepdims == 1) + +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_max_default_axes_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceMax", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[20., 2.] +# [40., 2.] +# [60., 2.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_do_not_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_do_not_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceMax", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +one = np.array(np.ones(reduced_shape, dtype=np.float32)) +zero = np.array(np.zeros(reduced_shape, dtype=np.float32)) +reduced = -(one / zero) # -inf + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_empty_set", +) +``` + +
+
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMax", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[20., 2.]] +# [[40., 2.]] +# [[60., 2.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMax", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[20., 2.]] +# [[40., 2.]] +# [[60., 2.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_negative_axes_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.maximum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_max_negative_axes_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+ + +### ReduceMean +There are 4 test cases, listed as following: +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMean", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.mean(data, axis=None, keepdims=keepdims == 1) +# print(reduced) +# [[[18.25]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.mean(data, axis=None, keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_default_axes_keepdims_random", +) +``` + +
+
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceMean", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[12.5, 1.5] +# [35., 1.5] +# [57.5, 1.5]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_do_not_keepdims_random", +) +``` + +
+
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMean", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[12.5, 1.5]] +# [[35., 1.5]] +# [[57.5, 1.5]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_keepdims_random", +) +``` + +
+
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMean", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[12.5, 1.5]] +# [[35., 1.5]] +# [[57.5, 1.5]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_negative_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.mean(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_mean_negative_axes_keepdims_random", +) +``` + +
+ + +### ReduceMin +There are 6 test cases, listed as following: +
+bool_inputs + +```python +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMin", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[True, True], [True, False], [False, True], [False, False]], +) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=bool(keepdims)) +# print(reduced) +# [[ True], +# [False], +# [False], +# [False]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_bool_inputs", +) +``` + +
+
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = None +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMin", inputs=["data"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.minimum.reduce(data, axis=axes, keepdims=keepdims == 1) +# print(reduced) +# [[[1.]]] + +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_min_default_axes_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.minimum.reduce(data, axis=axes, keepdims=keepdims == 1) + +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_min_default_axes_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceMin", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[5., 1.] +# [30., 1.] +# [55., 1.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_do_not_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_do_not_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceMin", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +one = np.array(np.ones(reduced_shape, dtype=np.float32)) +zero = np.array(np.zeros(reduced_shape, dtype=np.float32)) +reduced = one / zero # inf + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_empty_set", +) +``` + +
+
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMin", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[5., 1.]] +# [[30., 1.]] +# [[55., 1.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceMin", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[5, 1], [20, 2]], [[30, 1], [40, 2]], [[55, 1], [60, 2]]], + dtype=np.float32, +) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[5., 1.]] +# [[30., 1.]] +# [[55., 1.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_negative_axes_keepdims_example", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.minimum.reduce(data, axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_min_negative_axes_keepdims_random", + opset_imports=[onnx.helper.make_opsetid("", 18)], +) +``` + +
+ + +### ReduceProd +There are 5 test cases, listed as following: +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = None +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceProd", inputs=["data"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.prod(data, axis=axes, keepdims=keepdims == 1) +# print(reduced) +# [[[4.790016e+08]]] + +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_prod_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.prod(data, axis=axes, keepdims=keepdims == 1) +expect( + node, + inputs=[data], + outputs=[reduced], + name="test_reduce_prod_default_axes_keepdims_random", +) +``` + +
+
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceProd", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[3., 8.] +# [35., 48.] +# [99., 120.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_do_not_keepdims_random", +) +``` + +
+
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceProd", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +reduced = np.array(np.ones(reduced_shape, dtype=np.float32)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_empty_set", +) +``` + +
+
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceProd", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[3., 8.]] +# [[35., 48.]] +# [[99., 120.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_keepdims_random", +) +``` + +
+
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceProd", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[3., 8.]] +# [[35., 48.]] +# [[99., 120.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_negative_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.prod(data, axis=tuple(axes), keepdims=keepdims == 1) +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_prod_negative_axes_keepdims_random", +) +``` + +
+ + +### ReduceSum +There are 7 test cases, listed as following: +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSum", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(data, axis=None, keepdims=keepdims == 1) +# print(reduced) +# [[[78.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(data, axis=None, keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_default_axes_keepdims_random", +) +``` + +
+
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceSum", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) +# print(reduced) +# [[4., 6.] +# [12., 14.] +# [20., 22.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_do_not_keepdims_random", +) +``` + +
+
+empty_axes_input_noop + +```python +shape = [3, 2, 2] +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, + noop_with_empty_axes=True, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +axes = np.array([], dtype=np.int64) +reduced = np.array(data) +# print(reduced) +# [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_empty_axes_input_noop_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.array(data) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_empty_axes_input_noop", +) +``` + +
+
+empty_set + +```python +"""Test case with the reduced-axis of size zero.""" +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +reduced = np.array(np.zeros(reduced_shape, dtype=np.float32)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_empty_set", +) +``` + +
+
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSum", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) +# print(reduced) +# [[[4., 6.]] +# [[12., 14.]] +# [[20., 22.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_keepdims_random", +) +``` + +
+
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSum", inputs=["data", "axes"], outputs=["reduced"], keepdims=keepdims +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) +# print(reduced) +# [[[4., 6.]] +# [[12., 14.]] +# [[20., 22.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_negative_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(data, axis=tuple(axes.tolist()), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_negative_axes_keepdims_random", +) +``` + +
+
+non_reduced_axis_zero + +```python +"""Test case with the non-reduced-axis of size zero.""" +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 0, 1] + +node = onnx.helper.make_node( + "ReduceSum", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([2], dtype=np.int64) +reduced = np.array([], dtype=np.float32).reshape(reduced_shape) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_empty_set_non_reduced_axis_zero", +) +``` + +
+ + +### ReduceSumSquare +There are 5 test cases, listed as following: +
+default_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSumSquare", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(np.square(data), axis=None, keepdims=keepdims == 1) +# print(reduced) +# [[[650.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_default_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(np.square(data), axis=None, keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_default_axes_keepdims_random", +) +``` + +
+
+do_not_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 0 + +node = onnx.helper.make_node( + "ReduceSumSquare", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[10., 20.] +# [74., 100.] +# [202., 244.]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_do_not_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_do_not_keepdims_random", +) +``` + +
+
+empty_set + +```python +shape = [2, 0, 4] +keepdims = 1 +reduced_shape = [2, 1, 4] + +node = onnx.helper.make_node( + "ReduceSumSquare", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array([], dtype=np.float32).reshape(shape) +axes = np.array([1], dtype=np.int64) +reduced = np.array(np.zeros(reduced_shape, dtype=np.float32)) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_empty_set", +) +``` + +
+
+keepdims + +```python +shape = [3, 2, 2] +axes = np.array([1], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSumSquare", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[10., 20.]] +# [[74., 100.]] +# [[202., 244.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_keepdims_random", +) +``` + +
+
+negative_axes_keepdims + +```python +shape = [3, 2, 2] +axes = np.array([-2], dtype=np.int64) +keepdims = 1 + +node = onnx.helper.make_node( + "ReduceSumSquare", + inputs=["data", "axes"], + outputs=["reduced"], + keepdims=keepdims, +) + +data = np.array( + [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]], dtype=np.float32 +) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) +# print(reduced) +# [[[10., 20.s]] +# [[74., 100.]] +# [[202., 244.]]] + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_negative_axes_keepdims_example", +) + +np.random.seed(0) +data = np.random.uniform(-10, 10, shape).astype(np.float32) +reduced = np.sum(np.square(data), axis=tuple(axes), keepdims=keepdims == 1) + +expect( + node, + inputs=[data, axes], + outputs=[reduced], + name="test_reduce_sum_square_negative_axes_keepdims_random", +) +``` + +
+ + +### RegexFullMatch +There are 3 test cases, listed as following: +
+basic + +```python +node = onnx.helper.make_node( + "RegexFullMatch", + inputs=["X"], + outputs=["Y"], + pattern=r"www\.[\w.-]+\.\bcom\b", +) + +x = np.array(["www.google.com", "www.facebook.com", "www.bbc.co.uk"]).astype( + object +) +result = np.array([True, True, False]) +expect(node, inputs=[x], outputs=[result], name="test_regex_full_match_basic") +``` + +
+
+match_email_domain + +```python +node = onnx.helper.make_node( + "RegexFullMatch", + inputs=["X"], + outputs=["Y"], + pattern=r"(\W|^)[\w.\-]{0,25}@(yahoo|gmail)\.com(\W|$)", +) + +x = np.array( + [ + ["account@gmail.com", "account@hotmail.com"], + ["not email", "account2@yahoo.com"], + ] +).astype(object) +result = np.array([[True, False], [False, True]]) +expect( + node, + inputs=[x], + outputs=[result], + name="test_regex_full_match_email_domain", +) +``` + +
+
+match_empty + +```python +node = onnx.helper.make_node( + "RegexFullMatch", + inputs=["X"], + outputs=["Y"], + pattern=r"(\W|^)[\w.\-]{0,25}@(yahoo|gmail)\.com(\W|$)", +) + +x = np.array([[], []]).astype(object) +result = np.array([[], []]).astype(bool) +expect( + node, + inputs=[x], + outputs=[result], + name="test_regex_full_match_empty", +) +``` + +
+ + +### Relu +There are 1 test cases, listed as following: +
+relu + +```python +node = onnx.helper.make_node( + "Relu", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, 0, np.inf) + +expect(node, inputs=[x], outputs=[y], name="test_relu") +``` + +
+ + +### Reshape +There are 2 test cases, listed as following: +
+allowzero + +```python +original_shape = [0, 3, 4] +test_cases = { + "allowzero_reordered": np.array([3, 4, 0], dtype=np.int64), +} +data = np.random.random_sample(original_shape).astype(np.float32) + +for test_name, shape in test_cases.items(): + node = onnx.helper.make_node( + "Reshape", + inputs=["data", "shape"], + outputs=["reshaped"], + allowzero=1, # if allowzero=1, final shape = (3, 4, 0) + # if allowzero=0, final shape = (3, 4, 4) + ) + + reshaped = reshape_reference_implementation(data, shape, allowzero=1) + + expect( + node, + inputs=[data, shape], + outputs=[reshaped], + name="test_reshape_" + test_name, + ) +``` + +
+
+reshape + +```python +original_shape = [2, 3, 4] +test_cases = { + "reordered_all_dims": np.array([4, 2, 3], dtype=np.int64), + "reordered_last_dims": np.array([2, 4, 3], dtype=np.int64), + "reduced_dims": np.array([2, 12], dtype=np.int64), + "extended_dims": np.array([2, 3, 2, 2], dtype=np.int64), + "one_dim": np.array([24], dtype=np.int64), + "negative_dim": np.array([2, -1, 2], dtype=np.int64), + "negative_extended_dims": np.array([-1, 2, 3, 4], dtype=np.int64), + "zero_dim": np.array([2, 0, 4, 1], dtype=np.int64), + "zero_and_negative_dim": np.array([2, 0, 1, -1], dtype=np.int64), +} +data = np.random.random_sample(original_shape).astype(np.float32) + +for test_name, shape in test_cases.items(): + node = onnx.helper.make_node( + "Reshape", + inputs=["data", "shape"], + outputs=["reshaped"], + ) + + reshaped = reshape_reference_implementation(data, shape) + + expect( + node, + inputs=[data, shape], + outputs=[reshaped], + name="test_reshape_" + test_name, + ) +``` + +
+ + +### Resize +There are 39 test cases, listed as following: +
+resize_downsample_scales_cubic + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.8, 0.8], dtype=np.float32) + +# [[[[ 1.47119141 2.78125 4.08251953] +# [ 6.71142578 8.02148438 9.32275391] +# [11.91650391 13.2265625 14.52783203]]]] +output = interpolate_nd( + data, lambda x, _: cubic_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_cubic", +) +``` + +
+
+resize_downsample_scales_cubic_A_n0p5_exclude_outside + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + cubic_coeff_a=-0.5, + exclude_outside=True, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.8, 0.8], dtype=np.float32) + +# [[[[ 1.36812675 2.6695014 4.0133367 ] +# [ 6.57362535 7.875 9.2188353 ] +# [11.94896657 13.25034122 14.59417652]]]] +output = interpolate_nd( + data, + lambda x, _: cubic_coeffs(x, A=-0.5), + scale_factors=scales, + exclude_outside=True, +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_cubic_A_n0p5_exclude_outside", +) +``` + +
+
+resize_downsample_scales_cubic_align_corners + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + coordinate_transformation_mode="align_corners", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.8, 0.8], dtype=np.float32) + +# [[[[ 1. 2.39519159 3.79038317] +# [ 6.58076634 7.97595793 9.37114951] +# [12.16153268 13.55672427 14.95191585]]]] +output = interpolate_nd( + data, + lambda x, _: cubic_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="align_corners", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_cubic_align_corners", +) +``` + +
+
+resize_downsample_scales_cubic_antialias + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + antialias=1, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.6, 0.6], dtype=np.float32) + +# [[[[ 2.5180721 4.2858863] +# [ 9.589329 11.357142 ]]]] +output = interpolate_nd( + data, cubic_coeffs_antialias, scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_cubic_antialias", +) +``` + +
+
+resize_downsample_scales_linear + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.6, 0.6], dtype=np.float32) + +# [[[[2.6666665 4.3333331]]]] +output = interpolate_nd( + data, lambda x, _: linear_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_linear", +) +``` + +
+
+resize_downsample_scales_linear_align_corners + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="align_corners", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.6, 0.6], dtype=np.float32) + +# [[[[1. 3.142857]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="align_corners", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_linear_align_corners", +) +``` + +
+
+resize_downsample_scales_linear_antialias + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", + antialias=1, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.6, 0.6], dtype=np.float32) + +# [[[[ 2.875 4.5 ] +# [ 9.375 11. ]]]] +output = interpolate_nd( + data, linear_coeffs_antialias, scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_linear_antialias", +) +``` + +
+
+resize_downsample_scales_linear_half_pixel_symmetric + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="half_pixel_symmetric", +) + +data = np.array([[[[1, 2, 3, 4]]]], dtype=np.float32) +scales = np.array([1.0, 1.0, 1.0, 0.6], dtype=np.float32) + +# [[[[1.6666667, 3.3333333]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="half_pixel_symmetric", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_linear_half_pixel_symmetric", +) +``` + +
+
+resize_downsample_scales_nearest + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="nearest", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 0.6, 0.6], dtype=np.float32) + +# [[[[1. 3.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_downsample_scales_nearest", +) +``` + +
+
+resize_downsample_sizes_cubic + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="cubic", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 3, 3], dtype=np.int64) + +# [[[[ 1.63078704 3.00462963 4.37847222] +# [ 7.12615741 8.5 9.87384259] +# [12.62152778 13.99537037 15.36921296]]]] +output = interpolate_nd( + data, lambda x, _: cubic_coeffs(x), output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_cubic", +) +``` + +
+
+resize_downsample_sizes_cubic_antialias + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="cubic", + antialias=1, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 3, 3], dtype=np.int64) + +# [[[[ 1.7750092 3.1200073 4.4650054] +# [ 7.1550016 8.5 9.844998 ] +# [12.534994 13.8799925 15.224991 ]]]] +output = interpolate_nd(data, cubic_coeffs_antialias, output_size=sizes).astype( + np.float32 +) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_cubic_antialias", +) +``` + +
+
+resize_downsample_sizes_linear_antialias + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="linear", + antialias=1, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 3, 3], dtype=np.int64) + +# [[[[ 2.3636363 3.590909 4.818182 ] +# [ 7.2727275 8.5 9.727273 ] +# [12.181818 13.409091 14.636364 ]]]] +output = interpolate_nd( + data, linear_coeffs_antialias, output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_linear_antialias", +) +``` + +
+
+resize_downsample_sizes_linear_pytorch_half_pixel + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="pytorch_half_pixel", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 3, 1], dtype=np.int64) + +# [[[[ 1.6666666] +# [ 7. ] +# [12.333333 ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + output_size=sizes, + coordinate_transformation_mode="pytorch_half_pixel", +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_linear_pytorch_half_pixel", +) +``` + +
+
+resize_downsample_sizes_nearest + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 1, 3], dtype=np.int64) + +# [[[[1. 2. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_nearest", +) +``` + +
+
+resize_downsample_sizes_nearest_not_larger + +```python +keep_aspect_ratio_policy = "not_larger" +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 3], dtype=np.int64) # Results in 1x2 + +# [[[[1. 3.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x), + output_size=sizes, + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_nearest_not_larger", +) +``` + +
+
+resize_downsample_sizes_nearest_not_smaller + +```python +keep_aspect_ratio_policy = "not_smaller" +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 3], dtype=np.int64) # Results in 2x3 + +# [[[[1. 2. 4.] +# [5. 6. 8.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x), + output_size=sizes, + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_downsample_sizes_nearest_not_smaller", +) +``` + +
+
+resize_tf_crop_and_resize + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "roi", "", "sizes"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="tf_crop_and_resize", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +# Note: for some rois, the result may be different with that of TF for inaccurate floating point +roi = np.array([0, 0, 0.4, 0.6, 1, 1, 0.6, 0.8], dtype=np.float32) +sizes = np.array([1, 1, 3, 3], dtype=np.int64) + +# [[[[ 7.6000004 7.9 8.2 ] +# [ 8.8 9.1 9.400001 ] +# [10. 10.3 10.6 ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + output_size=sizes, + roi=roi, + coordinate_transformation_mode="tf_crop_and_resize", +).astype(np.float32) + +expect( + node, + inputs=[data, roi, sizes], + outputs=[output], + name="test_resize_tf_crop_and_resize", +) +``` + +
+
+resize_tf_crop_and_resize_axes_2_3 + +```python +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "roi", "", "sizes"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="tf_crop_and_resize", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +# Note: for some rois, the result may be different with that of TF for inaccurate floating point +roi = np.array([0.4, 0.6, 0.6, 0.8], dtype=np.float32) +sizes = np.array([3, 3], dtype=np.int64) + +# [[[[ 7.6000004 7.9 8.2 ] +# [ 8.8 9.1 9.400001 ] +# [10. 10.3 10.6 ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + output_size=sizes, + roi=roi, + axes=axes, + coordinate_transformation_mode="tf_crop_and_resize", +).astype(np.float32) + +expect( + node, + inputs=[data, roi, sizes], + outputs=[output], + name="test_resize_tf_crop_and_resize_axes_2_3", +) +``` + +
+
+resize_tf_crop_and_resize_axes_3_2 + +```python +axes = [3, 2] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "roi", "", "sizes"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="tf_crop_and_resize", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +# Note: for some rois, the result may be different with that of TF for inaccurate floating point +roi = np.array([0.6, 0.4, 0.8, 0.6], dtype=np.float32) +sizes = np.array([3, 3], dtype=np.int64) + +# [[[[ 7.6000004 7.9 8.2 ] +# [ 8.8 9.1 9.400001 ] +# [10. 10.3 10.6 ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + output_size=sizes, + roi=roi, + axes=axes, + coordinate_transformation_mode="tf_crop_and_resize", +).astype(np.float32) + +expect( + node, + inputs=[data, roi, sizes], + outputs=[output], + name="test_resize_tf_crop_and_resize_axes_3_2", +) +``` + +
+
+resize_tf_crop_and_resize_extrapolation_value + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "roi", "", "sizes"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="tf_crop_and_resize", + extrapolation_value=10.0, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +# Note: for some rois, the result may be different with that of TF for inaccurate floating point +roi = np.array([0, 0, 0.4, 0.6, 1, 1, 1.2, 1.7], dtype=np.float32) +sizes = np.array([1, 1, 3, 3], dtype=np.int64) + +# [[[[ 7.6000004 10. 10. ] +# [12.400001 10. 10. ] +# [10. 10. 10. ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + output_size=sizes, + roi=roi, + coordinate_transformation_mode="tf_crop_and_resize", + extrapolation_value=10.0, +).astype(np.float32) + +expect( + node, + inputs=[data, roi, sizes], + outputs=[output], + name="test_resize_tf_crop_and_resize_extrapolation_value", +) +``` + +
+
+resize_upsample_scales_cubic + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[ 0.47265625 0.76953125 1.24609375 1.875 2.28125 +# 2.91015625 3.38671875 3.68359375] +# [ 1.66015625 1.95703125 2.43359375 3.0625 3.46875 +# 4.09765625 4.57421875 4.87109375] +# [ 3.56640625 3.86328125 4.33984375 4.96875 5.375 +# 6.00390625 6.48046875 6.77734375] +# [ 6.08203125 6.37890625 6.85546875 7.484375 7.890625 +# 8.51953125 8.99609375 9.29296875] +# [ 7.70703125 8.00390625 8.48046875 9.109375 9.515625 +# 10.14453125 10.62109375 10.91796875] +# [10.22265625 10.51953125 10.99609375 11.625 12.03125 +# 12.66015625 13.13671875 13.43359375] +# [12.12890625 12.42578125 12.90234375 13.53125 13.9375 +# 14.56640625 15.04296875 15.33984375] +# [13.31640625 13.61328125 14.08984375 14.71875 15.125 +# 15.75390625 16.23046875 16.52734375]]]] +output = interpolate_nd( + data, lambda x, _: cubic_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_cubic", +) +``` + +
+
+resize_upsample_scales_cubic_A_n0p5_exclude_outside + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + cubic_coeff_a=-0.5, + exclude_outside=True, +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[ 0.55882353 0.81494204 1.35698249 1.89705882 2.39705882 +# 2.93713516 3.47917561 3.73529412] +# [ 1.58329755 1.83941606 2.38145651 2.92153285 3.42153285 +# 3.96160918 4.50364964 4.75976814] +# [ 3.75145936 4.00757787 4.54961832 5.08969466 5.58969466 +# 6.12977099 6.67181144 6.92792995] +# [ 5.91176471 6.16788321 6.70992366 7.25 7.75 +# 8.29007634 8.83211679 9.08823529] +# [ 7.91176471 8.16788321 8.70992366 9.25 9.75 +# 10.29007634 10.83211679 11.08823529] +# [10.07207005 10.32818856 10.87022901 11.41030534 11.91030534 +# 12.45038168 12.99242213 13.24854064] +# [12.24023186 12.49635036 13.03839082 13.57846715 14.07846715 +# 14.61854349 15.16058394 15.41670245] +# [13.26470588 13.52082439 14.06286484 14.60294118 15.10294118 +# 15.64301751 16.18505796 16.44117647]]]] +output = interpolate_nd( + data, + lambda x, _: cubic_coeffs(x, A=-0.5), + scale_factors=scales, + exclude_outside=True, +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_cubic_A_n0p5_exclude_outside", +) +``` + +
+
+resize_upsample_scales_cubic_align_corners + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + coordinate_transformation_mode="align_corners", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[ 1. 1.34110787 1.80029155 2.32944606 2.67055394 +# 3.19970845 3.65889213 4. ] +# [ 2.36443149 2.70553936 3.16472303 3.69387755 4.03498542 +# 4.56413994 5.02332362 5.36443149] +# [ 4.20116618 4.54227405 5.00145773 5.53061224 5.87172012 +# 6.40087464 6.86005831 7.20116618] +# [ 6.31778426 6.65889213 7.1180758 7.64723032 7.98833819 +# 8.51749271 8.97667638 9.31778426] +# [ 7.68221574 8.02332362 8.48250729 9.01166181 9.35276968 +# 9.8819242 10.34110787 10.68221574] +# [ 9.79883382 10.13994169 10.59912536 11.12827988 11.46938776 +# 11.99854227 12.45772595 12.79883382] +# [11.63556851 11.97667638 12.43586006 12.96501458 13.30612245 +# 13.83527697 14.29446064 14.63556851] +# [13. 13.34110787 13.80029155 14.32944606 14.67055394 +# 15.19970845 15.65889213 16. ]]]] +output = interpolate_nd( + data, + lambda x, _: cubic_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="align_corners", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_cubic_align_corners", +) +``` + +
+
+resize_upsample_scales_cubic_asymmetric + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="cubic", + coordinate_transformation_mode="asymmetric", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[ 1. 1.40625 2. 2.5 3. 3.59375 4. +# 4.09375] +# [ 2.625 3.03125 3.625 4.125 4.625 5.21875 5.625 +# 5.71875] +# [ 5. 5.40625 6. 6.5 7. 7.59375 8. +# 8.09375] +# [ 7. 7.40625 8. 8.5 9. 9.59375 10. +# 10.09375] +# [ 9. 9.40625 10. 10.5 11. 11.59375 12. +# 12.09375] +# [11.375 11.78125 12.375 12.875 13.375 13.96875 14.375 +# 14.46875] +# [13. 13.40625 14. 14.5 15. 15.59375 16. +# 16.09375] +# [13.375 13.78125 14.375 14.875 15.375 15.96875 16.375 +# 16.46875]]]] +output = interpolate_nd( + data, + lambda x, _: cubic_coeffs(x, A=-0.75), + scale_factors=scales, + coordinate_transformation_mode="asymmetric", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_cubic_asymmetric", +) +``` + +
+
+resize_upsample_scales_linear + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[1. 1.25 1.75 2. ] +# [1.5 1.75 2.25 2.5 ] +# [2.5 2.75 3.25 3.5 ] +# [3. 3.25 3.75 4. ]]]] +output = interpolate_nd( + data, lambda x, _: linear_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_linear", +) +``` + +
+
+resize_upsample_scales_linear_align_corners + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="align_corners", +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 2.0], dtype=np.float32) + +# [[[[1. 1.33333333 1.66666667 2. ] +# [1.66666667 2. 2.33333333 2.66666667] +# [2.33333333 2.66666667 3. 3.33333333] +# [3. 3.33333333 3.66666667 4. ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="align_corners", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_linear_align_corners", +) +``` + +
+
+resize_upsample_scales_linear_half_pixel_symmetric + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="linear", + coordinate_transformation_mode="half_pixel_symmetric", +) + +data = np.array([[[[1, 2], [3, 4]]]], dtype=np.float32) +scales = np.array([1.0, 1.0, 2.3, 2.94], dtype=np.float32) + +# [[[[1. , 1.15986395, 1.5 , 1.84013605, 2. ], +# [1.56521738, 1.72508133, 2.06521738, 2.40535343, 2.56521738], +# [2.43478262, 2.59464657, 2.93478262, 3.27491867, 3.43478262], +# [3. , 3.15986395, 3.5 , 3.84013605, 4. ]]]] +output = interpolate_nd( + data, + lambda x, _: linear_coeffs(x), + scale_factors=scales, + coordinate_transformation_mode="half_pixel_symmetric", +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_linear_half_pixel_symmetric", +) +``` + +
+
+resize_upsample_scales_nearest + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="nearest", +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 3.0], dtype=np.float32) + +# [[[[1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 2. 2. 2.] +# [3. 3. 3. 4. 4. 4.] +# [3. 3. 3. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), scale_factors=scales +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_nearest", +) +``` + +
+
+resize_upsample_scales_nearest_axes_2_3 + +```python +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="nearest", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([2.0, 3.0], dtype=np.float32) + +# [[[[1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 2. 2. 2.] +# [3. 3. 3. 4. 4. 4.] +# [3. 3. 3. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), scale_factors=scales, axes=axes +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_nearest_axes_2_3", +) +``` + +
+
+resize_upsample_scales_nearest_axes_3_2 + +```python +axes = [3, 2] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "scales"], + outputs=["Y"], + mode="nearest", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([3.0, 2.0], dtype=np.float32) + +# [[[[1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 2. 2. 2.] +# [3. 3. 3. 4. 4. 4.] +# [3. 3. 3. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), scale_factors=scales, axes=axes +).astype(np.float32) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_resize_upsample_scales_nearest_axes_3_2", +) +``` + +
+
+resize_upsample_sizes_cubic + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="cubic", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 9, 10], dtype=np.int64) + +# [[[[ 0.45507922 0.64057922 0.97157922 1.42257922 1.90732922 +# 2.22332922 2.70807922 3.15907922 3.49007922 3.67557922] +# [ 1.39437963 1.57987963 1.91087963 2.36187963 2.84662963 +# 3.16262963 3.64737963 4.09837963 4.42937963 4.61487963] +# [ 2.95130693 3.13680693 3.46780693 3.91880693 4.40355693 +# 4.71955693 5.20430693 5.65530693 5.98630693 6.17180693] +# [ 5.20525069 5.39075069 5.72175069 6.17275069 6.65750069 +# 6.97350069 7.45825069 7.90925069 8.24025069 8.42575069] +# [ 6.88975 7.07525 7.40625 7.85725 8.342 +# 8.658 9.14275 9.59375 9.92475 10.11025 ] +# [ 8.57424931 8.75974931 9.09074931 9.54174931 10.02649931 +# 10.34249931 10.82724931 11.27824931 11.60924931 11.79474931] +# [10.82819307 11.01369307 11.34469307 11.79569307 12.28044307 +# 12.59644307 13.08119307 13.53219307 13.86319307 14.04869307] +# [12.38512037 12.57062037 12.90162037 13.35262037 13.83737037 +# 14.15337037 14.63812037 15.08912037 15.42012037 15.60562037] +# [13.32442078 13.50992078 13.84092078 14.29192078 14.77667078 +# 15.09267078 15.57742078 16.02842078 16.35942078 16.54492078]]]] +output = interpolate_nd( + data, lambda x, _: cubic_coeffs(x), output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_cubic", +) +``` + +
+
+resize_upsample_sizes_nearest + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 7, 8], dtype=np.int64) + +# [[[[1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest", +) +``` + +
+
+resize_upsample_sizes_nearest_axes_2_3 + +```python +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([7, 8], dtype=np.int64) + +# [[[[1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), output_size=sizes, axes=axes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_axes_2_3", +) +``` + +
+
+resize_upsample_sizes_nearest_axes_3_2 + +```python +axes = [3, 2] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([8, 7], dtype=np.int64) + +# [[[[1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x), output_size=sizes, axes=axes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_axes_3_2", +) +``` + +
+
+resize_upsample_sizes_nearest_ceil_half_pixel + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + coordinate_transformation_mode="half_pixel", + nearest_mode="ceil", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 8, 8], dtype=np.int64) + +# [[[[ 1. 2. 2. 3. 3. 4. 4. 4.] +# [ 5. 6. 6. 7. 7. 8. 8. 8.] +# [ 5. 6. 6. 7. 7. 8. 8. 8.] +# [ 9. 10. 10. 11. 11. 12. 12. 12.] +# [ 9. 10. 10. 11. 11. 12. 12. 12.] +# [13. 14. 14. 15. 15. 16. 16. 16.] +# [13. 14. 14. 15. 15. 16. 16. 16.] +# [13. 14. 14. 15. 15. 16. 16. 16.]]]] +output = interpolate_nd( + data, lambda x, _: nearest_coeffs(x, mode="ceil"), output_size=sizes +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_ceil_half_pixel", +) +``` + +
+
+resize_upsample_sizes_nearest_floor_align_corners + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + coordinate_transformation_mode="align_corners", + nearest_mode="floor", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 8, 8], dtype=np.int64) + +# [[[[ 1. 1. 1. 2. 2. 3. 3. 4.] +# [ 1. 1. 1. 2. 2. 3. 3. 4.] +# [ 1. 1. 1. 2. 2. 3. 3. 4.] +# [ 5. 5. 5. 6. 6. 7. 7. 8.] +# [ 5. 5. 5. 6. 6. 7. 7. 8.] +# [ 9. 9. 9. 10. 10. 11. 11. 12.] +# [ 9. 9. 9. 10. 10. 11. 11. 12.] +# [13. 13. 13. 14. 14. 15. 15. 16.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x, mode="floor"), + output_size=sizes, + coordinate_transformation_mode="align_corners", +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_floor_align_corners", +) +``` + +
+
+resize_upsample_sizes_nearest_not_larger + +```python +keep_aspect_ratio_policy = "not_larger" +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([7, 8], dtype=np.int64) # Results in 7x7 + +# [[[[1. 1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2.] +# [3. 3. 3. 3. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x), + output_size=sizes, + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_not_larger", +) +``` + +
+
+resize_upsample_sizes_nearest_not_smaller + +```python +keep_aspect_ratio_policy = "not_smaller" +axes = [2, 3] +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([7, 8], dtype=np.int64) # Results in 8x8 + +# [[[[1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [1. 1. 1. 1. 2. 2. 2. 2.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.] +# [3. 3. 3. 3. 4. 4. 4. 4.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x), + output_size=sizes, + axes=axes, + keep_aspect_ratio_policy=keep_aspect_ratio_policy, +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_not_smaller", +) +``` + +
+
+resize_upsample_sizes_nearest_round_prefer_ceil_asymmetric + +```python +node = onnx.helper.make_node( + "Resize", + inputs=["X", "", "", "sizes"], + outputs=["Y"], + mode="nearest", + coordinate_transformation_mode="asymmetric", + nearest_mode="round_prefer_ceil", +) + +data = np.array( + [ + [ + [ + [1, 2, 3, 4], + [5, 6, 7, 8], + [9, 10, 11, 12], + [13, 14, 15, 16], + ] + ] + ], + dtype=np.float32, +) + +sizes = np.array([1, 1, 8, 8], dtype=np.int64) + +# [[[[ 1. 2. 2. 3. 3. 4. 4. 4.] +# [ 5. 6. 6. 7. 7. 8. 8. 8.] +# [ 5. 6. 6. 7. 7. 8. 8. 8.] +# [ 9. 10. 10. 11. 11. 12. 12. 12.] +# [ 9. 10. 10. 11. 11. 12. 12. 12.] +# [13. 14. 14. 15. 15. 16. 16. 16.] +# [13. 14. 14. 15. 15. 16. 16. 16.] +# [13. 14. 14. 15. 15. 16. 16. 16.]]]] +output = interpolate_nd( + data, + lambda x, _: nearest_coeffs(x, mode="round_prefer_ceil"), + output_size=sizes, + coordinate_transformation_mode="asymmetric", +).astype(np.float32) + +expect( + node, + inputs=[data, sizes], + outputs=[output], + name="test_resize_upsample_sizes_nearest_round_prefer_ceil_asymmetric", +) +``` + +
+ + +### ReverseSequence +There are 2 test cases, listed as following: +
+reversesequence_batch + +```python +node = onnx.helper.make_node( + "ReverseSequence", + inputs=["x", "sequence_lens"], + outputs=["y"], + time_axis=1, + batch_axis=0, +) +x = np.array( + [ + [0.0, 1.0, 2.0, 3.0], + [4.0, 5.0, 6.0, 7.0], + [8.0, 9.0, 10.0, 11.0], + [12.0, 13.0, 14.0, 15.0], + ], + dtype=np.float32, +) +sequence_lens = np.array([1, 2, 3, 4], dtype=np.int64) + +y = np.array( + [ + [0.0, 1.0, 2.0, 3.0], + [5.0, 4.0, 6.0, 7.0], + [10.0, 9.0, 8.0, 11.0], + [15.0, 14.0, 13.0, 12.0], + ], + dtype=np.float32, +) + +expect( + node, + inputs=[x, sequence_lens], + outputs=[y], + name="test_reversesequence_batch", +) +``` + +
+
+reversesequence_time + +```python +node = onnx.helper.make_node( + "ReverseSequence", + inputs=["x", "sequence_lens"], + outputs=["y"], + time_axis=0, + batch_axis=1, +) +x = np.array( + [ + [0.0, 4.0, 8.0, 12.0], + [1.0, 5.0, 9.0, 13.0], + [2.0, 6.0, 10.0, 14.0], + [3.0, 7.0, 11.0, 15.0], + ], + dtype=np.float32, +) +sequence_lens = np.array([4, 3, 2, 1], dtype=np.int64) + +y = np.array( + [ + [3.0, 6.0, 9.0, 12.0], + [2.0, 5.0, 8.0, 13.0], + [1.0, 4.0, 10.0, 14.0], + [0.0, 7.0, 11.0, 15.0], + ], + dtype=np.float32, +) + +expect( + node, + inputs=[x, sequence_lens], + outputs=[y], + name="test_reversesequence_time", +) +``` + +
+ + +### RoiAlign +There are 3 test cases, listed as following: +
+roialign_aligned_false + +```python +node = onnx.helper.make_node( + "RoiAlign", + inputs=["X", "rois", "batch_indices"], + outputs=["Y"], + spatial_scale=1.0, + output_height=5, + output_width=5, + sampling_ratio=2, + coordinate_transformation_mode="output_half_pixel", +) + +X, batch_indices, rois = get_roi_align_input_values() +# (num_rois, C, output_height, output_width) +Y = np.array( + [ + [ + [ + [0.4664, 0.4466, 0.3405, 0.5688, 0.6068], + [0.3714, 0.4296, 0.3835, 0.5562, 0.3510], + [0.2768, 0.4883, 0.5222, 0.5528, 0.4171], + [0.4713, 0.4844, 0.6904, 0.4920, 0.8774], + [0.6239, 0.7125, 0.6289, 0.3355, 0.3495], + ] + ], + [ + [ + [0.3022, 0.4305, 0.4696, 0.3978, 0.5423], + [0.3656, 0.7050, 0.5165, 0.3172, 0.7015], + [0.2912, 0.5059, 0.6476, 0.6235, 0.8299], + [0.5916, 0.7389, 0.7048, 0.8372, 0.8893], + [0.6227, 0.6153, 0.7097, 0.6154, 0.4585], + ] + ], + [ + [ + [0.2384, 0.3379, 0.3717, 0.6100, 0.7601], + [0.3767, 0.3785, 0.7147, 0.9243, 0.9727], + [0.5749, 0.5826, 0.5709, 0.7619, 0.8770], + [0.5355, 0.2566, 0.2141, 0.2796, 0.3600], + [0.4365, 0.3504, 0.2887, 0.3661, 0.2349], + ] + ], + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, rois, batch_indices], + outputs=[Y], + name="test_roialign_aligned_false", +) +``` + +
+
+roialign_aligned_true + +```python +node = onnx.helper.make_node( + "RoiAlign", + inputs=["X", "rois", "batch_indices"], + outputs=["Y"], + spatial_scale=1.0, + output_height=5, + output_width=5, + sampling_ratio=2, + coordinate_transformation_mode="half_pixel", +) + +X, batch_indices, rois = get_roi_align_input_values() +# (num_rois, C, output_height, output_width) +Y = np.array( + [ + [ + [ + [0.5178, 0.3434, 0.3229, 0.4474, 0.6344], + [0.4031, 0.5366, 0.4428, 0.4861, 0.4023], + [0.2512, 0.4002, 0.5155, 0.6954, 0.3465], + [0.3350, 0.4601, 0.5881, 0.3439, 0.6849], + [0.4932, 0.7141, 0.8217, 0.4719, 0.4039], + ] + ], + [ + [ + [0.3070, 0.2187, 0.3337, 0.4880, 0.4870], + [0.1871, 0.4914, 0.5561, 0.4192, 0.3686], + [0.1433, 0.4608, 0.5971, 0.5310, 0.4982], + [0.2788, 0.4386, 0.6022, 0.7000, 0.7524], + [0.5774, 0.7024, 0.7251, 0.7338, 0.8163], + ] + ], + [ + [ + [0.2393, 0.4075, 0.3379, 0.2525, 0.4743], + [0.3671, 0.2702, 0.4105, 0.6419, 0.8308], + [0.5556, 0.4543, 0.5564, 0.7502, 0.9300], + [0.6626, 0.5617, 0.4813, 0.4954, 0.6663], + [0.6636, 0.3721, 0.2056, 0.1928, 0.2478], + ] + ], + ], + dtype=np.float32, +) + +expect( + node, + inputs=[X, rois, batch_indices], + outputs=[Y], + name="test_roialign_aligned_true", +) +``` + +
+
+roialign_mode_max + +```python +X = np.array( + [ + [ + [ + [ + 0.2764, + 0.715, + 0.1958, + 0.3416, + 0.4638, + 0.0259, + 0.2963, + 0.6518, + 0.4856, + 0.725, + ], + [ + 0.9637, + 0.0895, + 0.2919, + 0.6753, + 0.0234, + 0.6132, + 0.8085, + 0.5324, + 0.8992, + 0.4467, + ], + [ + 0.3265, + 0.8479, + 0.9698, + 0.2471, + 0.9336, + 0.1878, + 0.4766, + 0.4308, + 0.34, + 0.2162, + ], + [ + 0.0206, + 0.172, + 0.2155, + 0.4394, + 0.0653, + 0.3406, + 0.7724, + 0.3921, + 0.2541, + 0.5799, + ], + [ + 0.4062, + 0.2194, + 0.4473, + 0.4687, + 0.7109, + 0.9327, + 0.9815, + 0.632, + 0.1728, + 0.6119, + ], + [ + 0.3097, + 0.1283, + 0.4984, + 0.5068, + 0.4279, + 0.0173, + 0.4388, + 0.043, + 0.4671, + 0.7119, + ], + [ + 0.1011, + 0.8477, + 0.4726, + 0.1777, + 0.9923, + 0.4042, + 0.1869, + 0.7795, + 0.9946, + 0.9689, + ], + [ + 0.1366, + 0.3671, + 0.7011, + 0.6234, + 0.9867, + 0.5585, + 0.6985, + 0.5609, + 0.8788, + 0.9928, + ], + [ + 0.5697, + 0.8511, + 0.6711, + 0.9406, + 0.8751, + 0.7496, + 0.165, + 0.1049, + 0.1559, + 0.2514, + ], + [ + 0.7012, + 0.4056, + 0.7879, + 0.3461, + 0.0415, + 0.2998, + 0.5094, + 0.3727, + 0.5482, + 0.0502, + ], + ] + ] + ], + dtype=np.float32, +) +rois = np.array( + [[0.0, 0.0, 9.0, 9.0], [0.0, 5.0, 4.0, 9.0], [5.0, 5.0, 9.0, 9.0]], + dtype=np.float32, +) +batch_indices = np.array([0, 0, 0], dtype=np.int64) + +Y = np.array( + [ + [ + [ + [0.3445228, 0.37310338, 0.37865096, 0.446696, 0.37991184], + [0.4133513, 0.5455125, 0.6651902, 0.55805874, 0.27110294], + [0.21223956, 0.40924096, 0.8417618, 0.792561, 0.37196714], + [0.46835402, 0.39741728, 0.8012819, 0.4969306, 0.5495158], + [0.3595896, 0.5196813, 0.5403741, 0.23814403, 0.19992709], + ] + ], + [ + [ + [0.30517197, 0.5086199, 0.3189761, 0.4054401, 0.47630402], + [0.50862, 0.8477, 0.37808004, 0.24936005, 0.79384017], + [0.17620805, 0.29368007, 0.44870415, 0.4987201, 0.63148826], + [0.51066005, 0.8511, 0.5368801, 0.9406, 0.70008016], + [0.4487681, 0.51066035, 0.5042561, 0.5643603, 0.42004836], + ] + ], + [ + [ + [0.21062402, 0.3510401, 0.37416005, 0.5967599, 0.46507207], + [0.32336006, 0.31180006, 0.6236001, 0.9946, 0.7751202], + [0.35744014, 0.5588001, 0.35897616, 0.7030401, 0.6353923], + [0.5996801, 0.27940005, 0.17948808, 0.35152006, 0.31769615], + [0.3598083, 0.40752012, 0.2385281, 0.43856013, 0.26313624], + ] + ], + ], + dtype=np.float32, +) + +node = onnx.helper.make_node( + "RoiAlign", + inputs=["X", "rois", "batch_indices"], + mode="max", + outputs=["Y"], + spatial_scale=1.0, + output_height=5, + output_width=5, + sampling_ratio=2, + coordinate_transformation_mode="output_half_pixel", +) + +expect( + node, + inputs=[X, rois, batch_indices], + outputs=[Y], + name="test_roialign_mode_max", +) +``` + +
+ + +### RotaryEmbedding +There are 8 test cases, listed as following: +
+rotary_embedding + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache", "position_ids"], + outputs=["output"], +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +position_ids_data = np.random.uniform(0, 50, (2, 3)).astype(np.int64) +sin_cache_data = np.random.rand(50, 4).astype(np.float32) +cos_cache_data = np.random.rand(50, 4).astype(np.float32) + +expected_output = rotary_embedding( + input_data, cos_cache_data, sin_cache_data, position_ids=position_ids_data +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data, position_ids_data], + outputs=[expected_output], + name="test_rotary_embedding", +) +``` + +
+
+rotary_embedding_3d_input + +```python +num_heads = 4 +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache", "position_ids"], + outputs=["output"], + num_heads=num_heads, +) + +input_data = np.random.rand(2, 3, 32).astype(np.float32) +position_ids_data = np.random.uniform(0, 50, (2, 3)).astype(np.int64) +sin_cache_data = np.random.rand(50, 4).astype(np.float32) +cos_cache_data = np.random.rand(50, 4).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + position_ids=position_ids_data, + num_heads=num_heads, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data, position_ids_data], + outputs=[expected_output], + name="test_rotary_embedding_3d_input", +) +``` + +
+
+rotary_embedding_interleaved + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache", "position_ids"], + outputs=["output"], + interleaved=1, +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +position_ids_data = np.random.uniform(0, 50, (2, 3)).astype(np.int64) +sin_cache_data = np.random.rand(50, 4).astype(np.float32) +cos_cache_data = np.random.rand(50, 4).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + position_ids=position_ids_data, + interleaved=1, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data, position_ids_data], + outputs=[expected_output], + name="test_rotary_embedding_interleaved", +) +``` + +
+
+rotary_embedding_no_position_ids + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache"], + outputs=["output"], +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +sin_cache_data = np.random.rand(2, 3, 4).astype(np.float32) +cos_cache_data = np.random.rand(2, 3, 4).astype(np.float32) + +expected_output = rotary_embedding(input_data, cos_cache_data, sin_cache_data) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data], + outputs=[expected_output], + name="test_rotary_embedding_no_position_ids", +) +``` + +
+
+rotary_embedding_no_position_ids_interleaved + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache"], + outputs=["output"], + interleaved=1, +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +sin_cache_data = np.random.rand(2, 3, 4).astype(np.float32) +cos_cache_data = np.random.rand(2, 3, 4).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + interleaved=1, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data], + outputs=[expected_output], + name="test_rotary_embedding_no_position_ids_interleaved", +) +``` + +
+
+rotary_embedding_no_position_ids_rotary_dim + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache"], + outputs=["output"], + rotary_embedding_dim=4, +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +sin_cache_data = np.random.rand(2, 3, 2).astype(np.float32) +cos_cache_data = np.random.rand(2, 3, 2).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + rotary_embedding_dim=4, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data], + outputs=[expected_output], + name="test_rotary_embedding_no_position_ids_rotary_dim", +) +``` + +
+
+rotary_embedding_with_interleaved_rotary_dim + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache", "position_ids"], + outputs=["output"], + rotary_embedding_dim=4, + interleaved=1, +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +position_ids_data = np.random.uniform(0, 50, (2, 3)).astype(np.int64) +sin_cache_data = np.random.rand(50, 2).astype(np.float32) +cos_cache_data = np.random.rand(50, 2).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + position_ids=position_ids_data, + interleaved=1, + rotary_embedding_dim=4, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data, position_ids_data], + outputs=[expected_output], + name="test_rotary_embedding_with_interleaved_rotary_dim", +) +``` + +
+
+rotary_embedding_with_rotary_dim + +```python +node = onnx.helper.make_node( + "RotaryEmbedding", + inputs=["input", "cos_cache", "sin_cache", "position_ids"], + outputs=["output"], + rotary_embedding_dim=4, +) + +input_data = np.random.rand(2, 4, 3, 8).astype(np.float32) +position_ids_data = np.random.uniform(0, 50, (2, 3)).astype(np.int64) +sin_cache_data = np.random.rand(50, 2).astype(np.float32) +cos_cache_data = np.random.rand(50, 2).astype(np.float32) + +expected_output = rotary_embedding( + input_data, + cos_cache_data, + sin_cache_data, + position_ids=position_ids_data, + rotary_embedding_dim=4, +) + +expect( + node, + inputs=[input_data, cos_cache_data, sin_cache_data, position_ids_data], + outputs=[expected_output], + name="test_rotary_embedding_with_rotary_dim", +) +``` + +
+ + +### Round +There are 1 test cases, listed as following: +
+round + +```python +node = onnx.helper.make_node( + "Round", + inputs=["x"], + outputs=["y"], +) + +x = np.array( + [ + 0.1, + 0.5, + 0.9, + 1.2, + 1.5, + 1.8, + 2.3, + 2.5, + 2.7, + -1.1, + -1.5, + -1.9, + -2.2, + -2.5, + -2.8, + ] +).astype(np.float32) + +# expected output +y = np.array( + [ + 0.0, + 0.0, + 1.0, + 1.0, + 2.0, + 2.0, + 2.0, + 2.0, + 3.0, + -1.0, + -2.0, + -2.0, + -2.0, + -2.0, + -3.0, + ] +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_round") +``` + +
+ + +### STFT +There are 1 test cases, listed as following: +
+stft + +```python +signal = np.arange(0, 128, dtype=np.float32).reshape(1, 128, 1) +length = np.array(16).astype(np.int64) +onesided_length = (length >> 1) + 1 +step = np.array(8).astype(np.int64) + +no_window = "" # optional input, not supplied +node = onnx.helper.make_node( + "STFT", + inputs=["signal", "frame_step", no_window, "frame_length"], + outputs=["output"], +) + +nstfts = ((signal.shape[1] - length) // step) + 1 +# [batch_size][frames][frame_length][2] +output = np.empty([1, nstfts, onesided_length, 2], dtype=np.float32) +for i in range(nstfts): + start = i * step + stop = i * step + length + complex_out = np.fft.fft(signal[0, start:stop, 0])[0:onesided_length] + output[0, i] = np.stack((complex_out.real, complex_out.imag), axis=1) + +output = output.astype(signal.dtype) +expect(node, inputs=[signal, step, length], outputs=[output], name="test_stft") + +node = onnx.helper.make_node( + "STFT", + inputs=["signal", "frame_step", "window"], + outputs=["output"], +) + +# Test with window +a0 = 0.5 +a1 = 0.5 +window = a0 + a1 * np.cos( + 2 * np.pi * np.arange(0, length, 1, dtype=np.float32) / length +) +nstfts = 1 + (signal.shape[1] - window.shape[0]) // step + +# [batch_size][frames][frame_length][2] +output = np.empty([1, nstfts, onesided_length, 2], dtype=np.float32) +for i in range(nstfts): + start = i * step + stop = i * step + length + complex_out = np.fft.fft(signal[0, start:stop, 0] * window)[ + 0:onesided_length + ] + output[0, i] = np.stack((complex_out.real, complex_out.imag), axis=1) +window = window.astype(signal.dtype) +output = output.astype(signal.dtype) +expect( + node, + inputs=[signal, step, window], + outputs=[output], + name="test_stft_with_window", +) +``` + +
+ + +### Scan +There are 4 test cases, listed as following: +
+scan_8 + +```python +# Given an input sequence [x1, ..., xN], sum up its elements using a scan +# returning the final state (x1+x2+...+xN) as well the scan_output +# [x1, x1+x2, ..., x1+x2+...+xN] +# Note: the first input (sequence_lens) is optional and omitted via "". +node = onnx.parser.parse_node( + """ + y, z = Scan ("", initial, x) < + num_scan_inputs = 1, + body = scan_body (float[2] sum_in, float[2] next) + => (float[2] sum_out, float[2] scan_out) + { + sum_out = Add(sum_in, next) + scan_out = Identity(sum_out) + } + > + """ +) +# create inputs for batch-size 1, sequence-length 3, inner dimension 2 +initial = np.array([0, 0]).astype(np.float32).reshape((1, 2)) +x = np.array([1, 2, 3, 4, 5, 6]).astype(np.float32).reshape((1, 3, 2)) +# final state computed = [1 + 3 + 5, 2 + 4 + 6] +y = np.array([9, 12]).astype(np.float32).reshape((1, 2)) +# scan-output computed +z = np.array([1, 2, 4, 6, 9, 12]).astype(np.float32).reshape((1, 3, 2)) + +expect( + node, + inputs=[initial, x], + outputs=[y, z], + name="test_scan_sum", + opset_imports=[onnx.helper.make_opsetid("", 8)], +) +``` + +
+
+scan_9 + +```python +# Given an input sequence [x1, ..., xN], sum up its elements using a scan +# returning the final state (x1+x2+...+xN) as well the scan_output +# [x1, x1+x2, ..., x1+x2+...+xN] +node = onnx.parser.parse_node( + """ + y, z = Scan (initial, x) < + num_scan_inputs = 1, + body = scan_body (float[2] sum_in, float[2] next) + => (float[2] sum_out, float[2] scan_out) + { + sum_out = Add(sum_in, next) + scan_out = Identity(sum_out) + } + > + """ +) +# create inputs for sequence-length 3, inner dimension 2 +initial = np.array([0, 0]).astype(np.float32).reshape((2,)) +x = np.array([1, 2, 3, 4, 5, 6]).astype(np.float32).reshape((3, 2)) +# final state computed = [1 + 3 + 5, 2 + 4 + 6] +y = np.array([9, 12]).astype(np.float32).reshape((2,)) +# scan-output computed +z = np.array([1, 2, 4, 6, 9, 12]).astype(np.float32).reshape((3, 2)) + +expect( + node, + inputs=[initial, x], + outputs=[y, z], + name="test_scan9_sum", + opset_imports=[onnx.helper.make_opsetid("", 9)], +) +``` + +
+
+scan_9_multi_state + +```python +# Scan with two state variables: running sum and running product. +# This exercises the case where num_loop_state_vars (2) differs from +# num_scan_inputs (1). +# +# Body inputs: sum_in (state), prod_in (state), next (scan) +# Body outputs: sum_out (state), prod_out (state), scan_out (scan) +node = onnx.parser.parse_node( + """ + y_sum, y_prod, z = Scan (initial_sum, initial_prod, x) < + num_scan_inputs = 1, + body = scan_body (float[2] sum_in, float[2] prod_in, float[2] next) + => (float[2] sum_out, float[2] prod_out, float[2] scan_out) + { + sum_out = Add(sum_in, next) + prod_out = Mul(prod_in, next) + scan_out = Identity(sum_out) + } + > + """ +) +# x = [[1, 2], [3, 4], [5, 6]] +initial_sum = np.array([0, 0]).astype(np.float32) +initial_prod = np.array([1, 1]).astype(np.float32) +x = np.array([1, 2, 3, 4, 5, 6]).astype(np.float32).reshape((3, 2)) +# final sum = [1+3+5, 2+4+6] = [9, 12] +y_sum = np.array([9, 12]).astype(np.float32) +# final product = [1*3*5, 2*4*6] = [15, 48] +y_prod = np.array([15, 48]).astype(np.float32) +# scan output (running sum) = [[1,2], [4,6], [9,12]] +z = np.array([1, 2, 4, 6, 9, 12]).astype(np.float32).reshape((3, 2)) + +expect( + node, + inputs=[initial_sum, initial_prod, x], + outputs=[y_sum, y_prod, z], + name="test_scan9_multi_state", + opset_imports=[onnx.helper.make_opsetid("", 9)], +) +``` + +
+
+scan_9_scalar + +```python +# Scan with scalar state and scan output to verify that output +# shapes are not distorted (e.g. (T,) not (T, 1)). +node = onnx.parser.parse_node( + """ + y, z = Scan (initial, x) < + num_scan_inputs = 1, + body = scan_body (float sum_in, float next) + => (float sum_out, float scan_out) + { + sum_out = Add(sum_in, next) + scan_out = Identity(sum_out) + } + > + """ +) +initial = np.float32(0.0) +x = np.array([1, 2, 3, 4, 5]).astype(np.float32) +# final state = 1+2+3+4+5 = 15 +y = np.float32(15.0) +# scan output = [1, 3, 6, 10, 15], shape (5,) +z = np.array([1, 3, 6, 10, 15]).astype(np.float32) + +expect( + node, + inputs=[initial, x], + outputs=[y, z], + name="test_scan9_scalar", + opset_imports=[onnx.helper.make_opsetid("", 9)], +) +``` + +
+ + +### Scatter +There are 2 test cases, listed as following: +
+scatter_with_axis + +```python +axis = 1 +node = onnx.helper.make_node( + "Scatter", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 3]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter(data, indices, updates, axis=axis) +# print(y) produces +# [[1.0, 1.1, 3.0, 2.1, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_with_axis", + opset_imports=[helper.make_opsetid("", 10)], +) +``` + +
+
+scatter_without_axis + +```python +node = onnx.helper.make_node( + "Scatter", + inputs=["data", "indices", "updates"], + outputs=["y"], +) +data = np.zeros((3, 3), dtype=np.float32) +indices = np.array([[1, 0, 2], [0, 2, 1]], dtype=np.int64) +updates = np.array([[1.0, 1.1, 1.2], [2.0, 2.1, 2.2]], dtype=np.float32) + +y = scatter(data, indices, updates) +# print(y) produces +# [[2.0, 1.1, 0.0], +# [1.0, 0.0, 2.2], +# [0.0, 2.1, 1.2]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_without_axis", + opset_imports=[helper.make_opsetid("", 10)], +) +``` + +
+ + +### ScatterElements +There are 7 test cases, listed as following: +
+scatter_elements_with_axis + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 3]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis) +# print(y) produces +# [[1.0, 1.1, 3.0, 2.1, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_axis", +) +``` + +
+
+scatter_elements_with_duplicate_indices + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, + reduction="add", +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 1]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis, reduction="add") +# print(y) produces +# [[1.0, 5.2, 3.0, 4.0, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_duplicate_indices", +) +``` + +
+
+scatter_elements_with_negative_indices + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, -3]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis) +# print(y) produces +# [[1.0, 1.1, 2.1, 4.0, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_negative_indices", +) +``` + +
+
+scatter_elements_with_reduction_max + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, + reduction="max", +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 1]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis, reduction="max") +# print(y) produces +# [[1.0, 2.1, 3.0, 4.0, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_reduction_max", +) +``` + +
+
+scatter_elements_with_reduction_min + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, + reduction="min", +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 1]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis, reduction="min") +# print(y) produces +# [[1.0, 1.1, 3.0, 4.0, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_reduction_min", +) +``` + +
+
+scatter_elements_with_reduction_mul + +```python +axis = 1 +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], + axis=axis, + reduction="mul", +) +data = np.array([[1.0, 2.0, 3.0, 4.0, 5.0]], dtype=np.float32) +indices = np.array([[1, 1]], dtype=np.int64) +updates = np.array([[1.1, 2.1]], dtype=np.float32) + +y = scatter_elements(data, indices, updates, axis, reduction="mul") +# print(y) produces +# [[1.0, 4.62, 3.0, 4.0, 5.0]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_with_reduction_mul", +) +``` + +
+
+scatter_elements_without_axis + +```python +node = onnx.helper.make_node( + "ScatterElements", + inputs=["data", "indices", "updates"], + outputs=["y"], +) +data = np.zeros((3, 3), dtype=np.float32) +indices = np.array([[1, 0, 2], [0, 2, 1]], dtype=np.int64) +updates = np.array([[1.0, 1.1, 1.2], [2.0, 2.1, 2.2]], dtype=np.float32) + +y = scatter_elements(data, indices, updates) +# print(y) produces +# [[2.0, 1.1, 0.0], +# [1.0, 0.0, 2.2], +# [0.0, 2.1, 1.2]] + +expect( + node, + inputs=[data, indices, updates], + outputs=[y], + name="test_scatter_elements_without_axis", +) +``` + +
+ + +### ScatterND +There are 7 test cases, listed as following: +
+scatternd + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], +) +data = np.array( + [ + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + ], + dtype=np.float32, +) +indices = np.array([[0], [2]], dtype=np.int64) +updates = np.array( + [ + [[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + ], + dtype=np.float32, +) +# Expecting output as np.array( +# [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], +# [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], +# [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], +# [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates) +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd", +) +``` + +
+
+scatternd_add + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="add", +) +data = np.array( + [ + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + ], + dtype=np.float32, +) +indices = np.array([[0], [0]], dtype=np.int64) +updates = np.array( + [ + [[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + ], + dtype=np.float32, +) +# Expecting output as np.array( +# [[[7, 8, 9, 10], [13, 14, 15, 16], [18, 17, 16, 15], [16, 15, 14, 13]], +# [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], +# [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], +# [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="add") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_add", +) +``` + +
+
+scatternd_max + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="max", +) +data = np.array( + [ + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + ], + dtype=np.float32, +) +indices = np.array([[0], [0]], dtype=np.int64) +updates = np.array( + [ + [[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + ], + dtype=np.float32, +) +# Expecting output as np.array( +# [[[5, 5, 5, 5], [6, 6, 7, 8], [8, 7, 7, 7], [8, 8 ,8, 8]], +# [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], +# [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], +# [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="max") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_max", +) +``` + +
+
+scatternd_max_with_element_indices + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="max", +) +data = np.array([[1, 2], [3, 4]], dtype=np.float32) +# Indices address individual elements (index rank == data rank), +# which exercises the reduction at the element level. +indices = np.array([[0, 0], [1, 1]], dtype=np.int64) +updates = np.array([5, 1], dtype=np.float32) +# Expecting output as np.array([[5, 2], [3, 4]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="max") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_max_with_element_indices", +) +``` + +
+
+scatternd_min + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="min", +) +data = np.array( + [ + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + ], + dtype=np.float32, +) +indices = np.array([[0], [0]], dtype=np.int64) +updates = np.array( + [ + [[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + ], + dtype=np.float32, +) +# Expecting output as np.array( +# [[[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 3, 2, 1]], +# [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], +# [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], +# [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="min") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_min", +) +``` + +
+
+scatternd_min_with_element_indices + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="min", +) +data = np.array([[1, 2], [3, 4]], dtype=np.float32) +indices = np.array([[0, 0], [1, 1]], dtype=np.int64) +updates = np.array([5, 1], dtype=np.float32) +# Expecting output as np.array([[1, 2], [3, 1]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="min") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_min_with_element_indices", +) +``` + +
+
+scatternd_multiply + +```python +node = onnx.helper.make_node( + "ScatterND", + inputs=["data", "indices", "updates"], + outputs=["y"], + reduction="mul", +) +data = np.array( + [ + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], + ], + dtype=np.float32, +) +indices = np.array([[0], [0]], dtype=np.int64) +updates = np.array( + [ + [[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], + [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], + ], + dtype=np.float32, +) +# Expecting output as np.array( +# [[[5, 10, 15, 20], [60, 72, 84, 96], [168, 147, 126, 105], [128, 96, 64, 32]], +# [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], +# [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], +# [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]], dtype=np.float32) +output = scatter_nd_impl(data, indices, updates, reduction="mul") +expect( + node, + inputs=[data, indices, updates], + outputs=[output], + name="test_scatternd_multiply", +) +``` + +
+ + +### Selu +There are 2 test cases, listed as following: +
+selu + +```python +node = onnx.helper.make_node( + "Selu", inputs=["x"], outputs=["y"], alpha=2.0, gamma=3.0 +) + +x = np.array([-1, 0, 1]).astype(np.float32) +# expected output [-3.79272318, 0., 3.] +y = ( + np.clip(x, 0, np.inf) * 3.0 + + (np.exp(np.clip(x, -np.inf, 0)) - 1) * 2.0 * 3.0 +) +expect(node, inputs=[x], outputs=[y], name="test_selu_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = ( + np.clip(x, 0, np.inf) * 3.0 + + (np.exp(np.clip(x, -np.inf, 0)) - 1) * 2.0 * 3.0 +) +expect(node, inputs=[x], outputs=[y], name="test_selu") +``` + +
+
+selu_default + +```python +default_alpha = 1.67326319217681884765625 +default_gamma = 1.05070102214813232421875 +node = onnx.helper.make_node( + "Selu", + inputs=["x"], + outputs=["y"], +) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = ( + np.clip(x, 0, np.inf) * default_gamma + + (np.exp(np.clip(x, -np.inf, 0)) - 1) * default_alpha * default_gamma +) +expect(node, inputs=[x], outputs=[y], name="test_selu_default") +``` + +
+ + +### SequenceInsert +There are 1 test cases, listed as following: +
+sequenceinsert + +```python +test_cases = { + "at_back": [np.array([10, 11, 12]).astype(np.int64)], + "at_front": [np.array([-2, -1, 0]), np.array([0]).astype(np.int64)], +} +sequence = [ + np.array([1, 2, 3, 4]).astype(np.int64), + np.array([5, 6, 7]).astype(np.int64), + np.array([8, 9]).astype(np.int64), +] + +for test_name, test_inputs in test_cases.items(): + tensor = test_inputs[0].astype(np.int64) + + if len(test_inputs) > 1: + node = onnx.helper.make_node( + "SequenceInsert", + inputs=["sequence", "tensor", "position"], + outputs=["output_sequence"], + ) + position = test_inputs[1] + inserted = sequence_insert_reference_implementation( + sequence, tensor, position + ) + expect( + node, + inputs=[sequence, tensor, position], + outputs=[inserted], + name="test_sequence_insert_" + test_name, + ) + else: + node = onnx.helper.make_node( + "SequenceInsert", + inputs=["sequence", "tensor"], + outputs=["output_sequence"], + ) + inserted = sequence_insert_reference_implementation(sequence, tensor) + expect( + node, + inputs=[sequence, tensor], + outputs=[inserted], + name="test_sequence_insert_" + test_name, + ) +``` + +
+ + +### SequenceMap +There are 6 test cases, listed as following: +
+sequence_map_add_1_sequence_1_tensor + +```python +body = onnx.helper.make_graph( + [onnx.helper.make_node("Add", ["in0", "in1"], ["out0"])], + "seq_map_body", + [ + onnx.helper.make_tensor_value_info( + "in0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "in1", onnx.TensorProto.FLOAT, ["N"] + ), + ], + [onnx.helper.make_tensor_value_info("out0", onnx.TensorProto.FLOAT, ["N"])], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["x0", "x1"], outputs=["y0"], body=body +) + +x0 = [np.random.uniform(0.0, 1.0, 10).astype(np.float32) for k in range(3)] +x1 = np.random.uniform(0.0, 1.0, 10).astype(np.float32) +y0 = [x0[i] + x1 for i in range(3)] +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), +] +expect( + node, + inputs=[x0, x1], + outputs=[y0], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_add_1_sequence_1_tensor", +) +``` + +
+
+sequence_map_add_2_sequences + +```python +body = onnx.helper.make_graph( + [onnx.helper.make_node("Add", ["in0", "in1"], ["out0"])], + "seq_map_body", + [ + onnx.helper.make_tensor_value_info( + "in0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "in1", onnx.TensorProto.FLOAT, ["N"] + ), + ], + [onnx.helper.make_tensor_value_info("out0", onnx.TensorProto.FLOAT, ["N"])], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["x0", "x1"], outputs=["y0"], body=body +) + +N = [np.random.randint(1, 10) for _ in range(3)] +x0 = [np.random.uniform(0.0, 1.0, N[k]).astype(np.float32) for k in range(3)] +x1 = [np.random.uniform(0.0, 1.0, N[k]).astype(np.float32) for k in range(3)] +y0 = [x0[k] + x1[k] for k in range(3)] +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), +] +expect( + node, + inputs=[x0, x1], + outputs=[y0], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_add_2_sequences", +) +``` + +
+
+sequence_map_extract_shapes + +```python +body = onnx.helper.make_graph( + [onnx.helper.make_node("Shape", ["x"], ["shape"])], + "seq_map_body", + [ + onnx.helper.make_tensor_value_info( + "x", onnx.TensorProto.FLOAT, ["H", "W", "C"] + ) + ], + [onnx.helper.make_tensor_value_info("shape", onnx.TensorProto.INT64, [3])], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["in_seq"], outputs=["shapes"], body=body +) + +shapes = [ + np.array([40, 30, 3], dtype=np.int64), + np.array([20, 10, 3], dtype=np.int64), + np.array([10, 5, 3], dtype=np.int64), +] +x0 = [np.zeros(shape, dtype=np.float32) for shape in shapes] +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto( + onnx.TensorProto.FLOAT, ["H", "W", "C"] + ) + ), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.INT64, [3]) + ), +] +expect( + node, + inputs=[x0], + outputs=[shapes], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_extract_shapes", +) +``` + +
+
+sequence_map_identity_1_sequence + +```python +body = onnx.helper.make_graph( + [onnx.helper.make_node("Identity", ["in0"], ["out0"])], + "seq_map_body", + [onnx.helper.make_tensor_value_info("in0", onnx.TensorProto.FLOAT, ["N"])], + [onnx.helper.make_tensor_value_info("out0", onnx.TensorProto.FLOAT, ["M"])], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["x"], outputs=["y"], body=body +) + +x = [np.random.uniform(0.0, 1.0, 10).astype(np.float32) for _ in range(3)] +y = x +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), +] +expect( + node, + inputs=[x], + outputs=[y], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_identity_1_sequence", +) +``` + +
+
+sequence_map_identity_1_sequence_1_tensor + +```python +body = onnx.helper.make_graph( + [ + onnx.helper.make_node("Identity", ["in0"], ["out0"]), + onnx.helper.make_node("Identity", ["in1"], ["out1"]), + ], + "seq_map_body", + [ + onnx.helper.make_tensor_value_info( + "in0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "in1", onnx.TensorProto.FLOAT, ["M"] + ), + ], + [ + onnx.helper.make_tensor_value_info( + "out0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "out1", onnx.TensorProto.FLOAT, ["M"] + ), + ], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["x0", "x1"], outputs=["y0", "y1"], body=body +) + +x0 = [ + np.random.uniform(0.0, 1.0, np.random.randint(1, 10)).astype(np.float32) + for _ in range(3) +] +x1 = np.random.uniform(0.0, 1.0, np.random.randint(1, 10)).astype(np.float32) +y0 = x0 +y1 = [x1 for _ in range(3)] +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["M"]), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["M"]) + ), +] +expect( + node, + inputs=[x0, x1], + outputs=[y0, y1], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_identity_1_sequence_1_tensor", +) +``` + +
+
+sequence_map_identity_2_sequences + +```python +body = onnx.helper.make_graph( + [ + onnx.helper.make_node("Identity", ["in0"], ["out0"]), + onnx.helper.make_node("Identity", ["in1"], ["out1"]), + ], + "seq_map_body", + [ + onnx.helper.make_tensor_value_info( + "in0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "in1", onnx.TensorProto.FLOAT, ["M"] + ), + ], + [ + onnx.helper.make_tensor_value_info( + "out0", onnx.TensorProto.FLOAT, ["N"] + ), + onnx.helper.make_tensor_value_info( + "out1", onnx.TensorProto.FLOAT, ["M"] + ), + ], +) + +node = onnx.helper.make_node( + "SequenceMap", inputs=["x0", "x1"], outputs=["y0", "y1"], body=body +) + +x0 = [ + np.random.uniform(0.0, 1.0, np.random.randint(1, 10)).astype(np.float32) + for _ in range(3) +] +x1 = [ + np.random.uniform(0.0, 1.0, np.random.randint(1, 10)).astype(np.float32) + for _ in range(3) +] +y0 = x0 +y1 = x1 +input_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["M"]) + ), +] +output_type_protos = [ + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["N"]) + ), + onnx.helper.make_sequence_type_proto( + onnx.helper.make_tensor_type_proto(onnx.TensorProto.FLOAT, ["M"]) + ), +] +expect( + node, + inputs=[x0, x1], + outputs=[y0, y1], + input_type_protos=input_type_protos, + output_type_protos=output_type_protos, + name="test_sequence_map_identity_2_sequences", +) +``` + +
+ + +### Shape +There are 1 test cases, listed as following: +
+shape + +```python +x = np.array( + [ + [1, 2, 3], + [4, 5, 6], + ] +).astype(np.float32) +test_shape("_example", x) # preserve names of original test cases + +x = np.random.randn(3, 4, 5).astype(np.float32) + +test_shape("", x) # preserve names of original test cases + +test_shape("_start_1", x, start=1) + +test_shape("_end_1", x, end=1) + +test_shape("_start_negative_1", x, start=-1) + +test_shape("_end_negative_1", x, end=-1) + +test_shape("_start_1_end_negative_1", x, start=1, end=-1) + +test_shape("_start_1_end_2", x, start=1, end=2) + +test_shape("_clip_start", x, start=-10) + +test_shape("_clip_end", x, end=10) + +test_shape("_start_greater_than_end", x, start=2, end=1) +``` + +
+ + +### Shrink +There are 2 test cases, listed as following: +
+hard_shrink + +```python +node = onnx.helper.make_node( + "Shrink", + inputs=["x"], + outputs=["y"], + lambd=1.5, +) +X = np.arange(-2.0, 2.1, dtype=np.float32) +Y = np.array([-2, 0, 0, 0, 2], dtype=np.float32) +expect(node, inputs=[X], outputs=[Y], name="test_shrink_hard") +``` + +
+
+soft_shrink + +```python +node = onnx.helper.make_node( + "Shrink", + inputs=["x"], + outputs=["y"], + lambd=1.5, + bias=1.5, +) +X = np.arange(-2.0, 2.1, dtype=np.float32) +Y = np.array([-0.5, 0, 0, 0, 0.5], dtype=np.float32) +expect(node, inputs=[X], outputs=[Y], name="test_shrink_soft") +``` + +
+ + +### Sigmoid +There are 1 test cases, listed as following: +
+sigmoid + +```python +node = onnx.helper.make_node( + "Sigmoid", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = 1.0 / ( + 1.0 + np.exp(np.negative(x)) +) # expected output [0.26894143, 0.5, 0.7310586] +expect(node, inputs=[x], outputs=[y], name="test_sigmoid_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = 1.0 / (1.0 + np.exp(np.negative(x))) +expect(node, inputs=[x], outputs=[y], name="test_sigmoid") +``` + +
+ + +### Sign +There are 1 test cases, listed as following: +
+sign + +```python +node = onnx.helper.make_node( + "Sign", + inputs=["x"], + outputs=["y"], +) + +x = np.array(range(-5, 6)).astype(np.float32) +y = np.sign(x) +expect(node, inputs=[x], outputs=[y], name="test_sign") +``` + +
+ + +### Sin +There are 1 test cases, listed as following: +
+sin + +```python +node = onnx.helper.make_node( + "Sin", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.sin(x) +expect(node, inputs=[x], outputs=[y], name="test_sin_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.sin(x) +expect(node, inputs=[x], outputs=[y], name="test_sin") +``` + +
+ + +### Sinh +There are 1 test cases, listed as following: +
+sinh + +```python +node = onnx.helper.make_node( + "Sinh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.sinh(x) # expected output [-1.17520118, 0., 1.17520118] +expect(node, inputs=[x], outputs=[y], name="test_sinh_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.sinh(x) +expect(node, inputs=[x], outputs=[y], name="test_sinh") +``` + +
+ + +### Size +There are 1 test cases, listed as following: +
+size + +```python +node = onnx.helper.make_node( + "Size", + inputs=["x"], + outputs=["y"], +) + +x = np.array( + [ + [1, 2, 3], + [4, 5, 6], + ] +).astype(np.float32) +y = np.array(6).astype(np.int64) + +expect(node, inputs=[x], outputs=[y], name="test_size_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.array(x.size).astype(np.int64) + +expect(node, inputs=[x], outputs=[y], name="test_size") +``` + +
+ + +### Slice +There are 8 test cases, listed as following: +
+slice + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes", "steps"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +y = x[0:3, 0:10] +starts = np.array([0, 0], dtype=np.int64) +ends = np.array([3, 10], dtype=np.int64) +axes = np.array([0, 1], dtype=np.int64) +steps = np.array([1, 1], dtype=np.int64) + +expect( + node, inputs=[x, starts, ends, axes, steps], outputs=[y], name="test_slice" +) +``` + +
+
+slice_default_axes + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([0, 0, 3], dtype=np.int64) +ends = np.array([20, 10, 4], dtype=np.int64) +y = x[:, :, 3:4] + +expect( + node, inputs=[x, starts, ends], outputs=[y], name="test_slice_default_axes" +) +``` + +
+
+slice_default_steps + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([0, 0, 3], dtype=np.int64) +ends = np.array([20, 10, 4], dtype=np.int64) +axes = np.array([0, 1, 2], dtype=np.int64) +y = x[:, :, 3:4] + +expect( + node, + inputs=[x, starts, ends, axes], + outputs=[y], + name="test_slice_default_steps", +) +``` + +
+
+slice_end_out_of_bounds + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes", "steps"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([1], dtype=np.int64) +ends = np.array([1000], dtype=np.int64) +axes = np.array([1], dtype=np.int64) +steps = np.array([1], dtype=np.int64) +y = x[:, 1:1000] + +expect( + node, + inputs=[x, starts, ends, axes, steps], + outputs=[y], + name="test_slice_end_out_of_bounds", +) +``` + +
+
+slice_neg + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes", "steps"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([0], dtype=np.int64) +ends = np.array([-1], dtype=np.int64) +axes = np.array([1], dtype=np.int64) +steps = np.array([1], dtype=np.int64) +y = x[:, 0:-1] + +expect( + node, + inputs=[x, starts, ends, axes, steps], + outputs=[y], + name="test_slice_neg", +) +``` + +
+
+slice_neg_steps + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes", "steps"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([20, 10, 4], dtype=np.int64) +ends = np.array([0, 0, 1], dtype=np.int64) +axes = np.array([0, 1, 2], dtype=np.int64) +steps = np.array([-1, -3, -2]).astype(np.int64) +y = x[20:0:-1, 10:0:-3, 4:1:-2] + +expect( + node, + inputs=[x, starts, ends, axes, steps], + outputs=[y], + name="test_slice_neg_steps", +) +``` + +
+
+slice_negative_axes + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([0, 0, 3], dtype=np.int64) +ends = np.array([20, 10, 4], dtype=np.int64) +axes = np.array([0, -2, -1], dtype=np.int64) +y = x[:, :, 3:4] + +expect( + node, + inputs=[x, starts, ends, axes], + outputs=[y], + name="test_slice_negative_axes", +) +``` + +
+
+slice_start_out_of_bounds + +```python +node = onnx.helper.make_node( + "Slice", + inputs=["x", "starts", "ends", "axes", "steps"], + outputs=["y"], +) + +x = np.random.randn(20, 10, 5).astype(np.float32) +starts = np.array([1000], dtype=np.int64) +ends = np.array([1000], dtype=np.int64) +axes = np.array([1], dtype=np.int64) +steps = np.array([1], dtype=np.int64) +y = x[:, 1000:1000] + +expect( + node, + inputs=[x, starts, ends, axes, steps], + outputs=[y], + name="test_slice_start_out_of_bounds", +) +``` + +
+ + +### Softmax +There are 2 test cases, listed as following: +
+softmax + +```python +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], +) +x = np.array([[-1, 0, 1]]).astype(np.float32) +# expected output [[0.09003058, 0.24472848, 0.66524094]] +y = softmax(x, axis=1) +expect(node, inputs=[x], outputs=[y], name="test_softmax_example") +``` + +
+
+softmax_axis + +```python +x = np.array([[0, 1, 2, 3], [10000, 10001, 10002, 10003]]).astype(np.float32) +# expected output +# [[0.032058604 0.08714432 0.23688284 0.6439143 ] +# [0.032058604 0.08714432 0.23688284 0.6439143 ]] +y = softmax(x) + +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], +) +expect(node, inputs=[x], outputs=[y], name="test_softmax_large_number") + +x = np.abs(np.random.randn(3, 4, 5).astype(np.float32)) +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], + axis=0, +) +y = softmax(x, axis=0) +expect(node, inputs=[x], outputs=[y], name="test_softmax_axis_0") + +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], + axis=1, +) +y = softmax(x, axis=1) +expect(node, inputs=[x], outputs=[y], name="test_softmax_axis_1") + +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], + axis=2, +) +y = softmax(x, axis=2) +expect(node, inputs=[x], outputs=[y], name="test_softmax_axis_2") + +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], + axis=-1, +) +y = softmax(x, axis=-1) +expect(node, inputs=[x], outputs=[y], name="test_softmax_negative_axis") + +# default axis is -1 +node = onnx.helper.make_node( + "Softmax", + inputs=["x"], + outputs=["y"], +) +expect(node, inputs=[x], outputs=[y], name="test_softmax_default_axis") +``` + +
+ + +### SoftmaxCrossEntropyLoss +There are 34 test cases, listed as following: +
+input_shape_is_NCd1_mean_weight_negative_ii + +```python +reduction = "mean" +ignore_index = np.int64(-1) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1 = 3, 5, 6 +np.random.seed(0) +x = np.random.rand(N, C, dim1).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N, dim1)).astype(np.int64) +labels[0][0] = -1 +weight = np.random.rand(C).astype(np.float32) + +sce = softmaxcrossentropy( + x, labels, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[x, labels, weight], + outputs=[sce], + name="test_sce_NCd1_mean_weight_negative_ii", +) +``` + +
+
+input_shape_is_NCd1_mean_weight_negative_ii_log_prob + +```python +reduction = "mean" +ignore_index = np.int64(-1) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1 = 3, 5, 6 +np.random.seed(0) +x = np.random.rand(N, C, dim1).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N, dim1)).astype(np.int64) +labels[0][0] = -1 +weight = np.random.rand(C).astype(np.float32) + +loss, log_prob = softmaxcrossentropy( + x, + labels, + weight=weight, + reduction=reduction, + ignore_index=ignore_index, + get_log_prob=True, +) + +expect( + node, + inputs=[x, labels, weight], + outputs=[loss, log_prob], + name="test_sce_NCd1_mean_weight_negative_ii_log_prob", +) +``` + +
+
+input_shape_is_NCd1d2d3_none_no_weight_negative_ii + +```python +reduction = "none" +ignore_index = np.int64(-5) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1, dim2, dim3 = 3, 5, 6, 6, 5 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N, dim1, dim2, dim3)).astype( + np.int64 +) +labels[0][0][0][0] = -5 + +sce = softmaxcrossentropy( + x, labels, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[x, labels], + outputs=[sce], + name="test_sce_NCd1d2d3_none_no_weight_negative_ii", +) +``` + +
+
+input_shape_is_NCd1d2d3_none_no_weight_negative_ii_log_prob + +```python +reduction = "none" +ignore_index = np.int64(-5) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C, dim1, dim2, dim3 = 3, 5, 6, 6, 5 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N, dim1, dim2, dim3)).astype( + np.int64 +) +labels[0][0][0][0] = -5 + +loss, log_prob = softmaxcrossentropy( + x, labels, reduction=reduction, ignore_index=ignore_index, get_log_prob=True +) + +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_NCd1d2d3_none_no_weight_negative_ii_log_prob", +) +``` + +
+
+input_shape_is_NCd1d2d3_sum_weight_high_ii + +```python +reduction = "sum" +ignore_index = np.int64(10) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C = 3, 5 +np.random.seed(0) +x = np.random.rand(N, C).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N)).astype(np.int64) +labels[0] = 10 +weight = np.random.rand(C).astype(np.float32) + +sce = softmaxcrossentropy( + x, labels, weight=weight, reduction=reduction, ignore_index=ignore_index +) + +expect( + node, + inputs=[x, labels, weight], + outputs=[sce], + name="test_sce_NCd1d2d3_sum_weight_high_ii", +) +``` + +
+
+input_shape_is_NCd1d2d3_sum_weight_high_ii_log_prob + +```python +reduction = "sum" +ignore_index = np.int64(10) + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +N, C = 3, 5 +np.random.seed(0) +x = np.random.rand(N, C).astype(np.float32) +labels = np.random.randint(0, high=C, size=(N)).astype(np.int64) +labels[0] = 10 +weight = np.random.rand(C).astype(np.float32) + +loss, log_prob = softmaxcrossentropy( + x, + labels, + weight=weight, + reduction=reduction, + ignore_index=ignore_index, + get_log_prob=True, +) + +expect( + node, + inputs=[x, labels, weight], + outputs=[loss, log_prob], + name="test_sce_NCd1d2d3_sum_weight_high_ii_log_prob", +) +``` + +
+
+input_shape_is_NCd1d2d3d4d5_mean_weight + +```python +reduction = "mean" + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +labels = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +sce = softmaxcrossentropy(x, labels, weight=weight, reduction=reduction) + +expect( + node, + inputs=[x, labels, weight], + outputs=[sce], + name="test_sce_NCd1d2d3d4d5_mean_weight", +) +``` + +
+
+input_shape_is_NCd1d2d3d4d5_mean_weight_log_prob + +```python +reduction = "mean" + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +labels = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) +weight = np.random.rand(C).astype(np.float32) + +loss, log_prob = softmaxcrossentropy( + x, labels, weight=weight, reduction=reduction, get_log_prob=True +) + +expect( + node, + inputs=[x, labels, weight], + outputs=[loss, log_prob], + name="test_sce_NCd1d2d3d4d5_mean_weight_log_prob", +) +``` + +
+
+input_shape_is_NCd1d2d3d4d5_none_no_weight + +```python +reduction = "none" + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +labels = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) + +sce = softmaxcrossentropy(x, labels, reduction=reduction) + +expect( + node, + inputs=[x, labels], + outputs=[sce], + name="test_sce_NCd1d2d3d4d5_none_no_weight", +) +``` + +
+
+input_shape_is_NCd1d2d3d4d5_none_no_weight_log_prob + +```python +reduction = "none" + +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +N, C, dim1, dim2, dim3, dim4, dim5 = 3, 5, 6, 6, 5, 3, 4 +np.random.seed(0) +x = np.random.rand(N, C, dim1, dim2, dim3, dim4, dim5).astype(np.float32) +labels = np.random.randint( + 0, high=C, size=(N, dim1, dim2, dim3, dim4, dim5) +).astype(np.int64) + +loss, log_prob = softmaxcrossentropy( + x, labels, reduction=reduction, get_log_prob=True +) + +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_NCd1d2d3d4d5_none_no_weight_log_prob", +) +``` + +
+
+softmaxcrossentropy_mean + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels) + +# Check results +expect(node, inputs=[x, labels], outputs=[sce], name="test_sce_mean") +``` + +
+
+softmaxcrossentropy_mean_3d + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +y = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, y) + +# Check results +expect(node, inputs=[x, y], outputs=[sce], name="test_sce_mean_3d") +``` + +
+
+softmaxcrossentropy_mean_3d_log_prob + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +y = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy(x, y, get_log_prob=True) + +# Check results +expect( + node, + inputs=[x, y], + outputs=[loss, log_prob], + name="test_sce_mean_3d_log_prob", +) +``` + +
+
+softmaxcrossentropy_mean_log_prob + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy(x, labels, get_log_prob=True) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_mean_log_prob", +) +``` + +
+
+softmaxcrossentropy_mean_no_weights_ii + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +labels[0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, ignore_index=ignore_index) + +# Check results +expect( + node, inputs=[x, labels], outputs=[sce], name="test_sce_mean_no_weight_ii" +) +``` + +
+
+softmaxcrossentropy_mean_no_weights_ii_3d + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) +labels[0][0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, ignore_index=ignore_index) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[sce], + name="test_sce_mean_no_weight_ii_3d", +) +``` + +
+
+softmaxcrossentropy_mean_no_weights_ii_3d_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) +labels[0][0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, ignore_index=ignore_index, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_mean_no_weight_ii_3d_log_prob", +) +``` + +
+
+softmaxcrossentropy_mean_no_weights_ii_4d + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2, 7).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2, 7)).astype(np.int64) +labels[0][0][0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy( + x, labels, reduction=reduction, ignore_index=ignore_index +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[sce], + name="test_sce_mean_no_weight_ii_4d", +) +``` + +
+
+softmaxcrossentropy_mean_no_weights_ii_4d_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2, 7).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2, 7)).astype(np.int64) +labels[0][0][0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, reduction=reduction, ignore_index=ignore_index, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_mean_no_weight_ii_4d_log_prob", +) +``` + +
+
+softmaxcrossentropy_mean_no_weights_ii_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +labels[0] = np.int64(2) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, ignore_index=ignore_index, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_mean_no_weight_ii_log_prob", +) +``` + +
+
+softmaxcrossentropy_mean_weights + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, weight=weights) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[sce], + name="test_sce_mean_weight", +) +``` + +
+
+softmaxcrossentropy_mean_weights_ii + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(0) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +labels[0] = np.int64(0) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, weight=weights, ignore_index=ignore_index) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[sce], + name="test_sce_mean_weight_ii", +) +``` + +
+
+softmaxcrossentropy_mean_weights_ii_3d + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(1) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) +labels[0][0] = np.int64(1) +weights = np.array([0.2, 0.3, 0.6, 0.1, 0.5], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, weight=weights, ignore_index=ignore_index) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[sce], + name="test_sce_mean_weight_ii_3d", +) +``` + +
+
+softmaxcrossentropy_mean_weights_ii_3d_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(1) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2)).astype(np.int64) +labels[0][0] = np.int64(1) +weights = np.array([0.2, 0.3, 0.6, 0.1, 0.5], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, weight=weights, ignore_index=ignore_index, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[loss, log_prob], + name="test_sce_mean_weight_ii_3d_log_prob", +) +``` + +
+
+softmaxcrossentropy_mean_weights_ii_4d + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2, 7).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2, 7)).astype(np.int64) +labels[0][0][0] = np.int64(2) +weights = np.array([0.2, 0.3, 0.6, 0.1, 0.5], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy( + x, labels, reduction=reduction, weight=weights, ignore_index=ignore_index +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[sce], + name="test_sce_mean_weight_ii_4d", +) +``` + +
+
+softmaxcrossentropy_mean_weights_ii_4d_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(2) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5, 2, 7).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3, 2, 7)).astype(np.int64) +labels[0][0][0] = np.int64(2) +weights = np.array([0.2, 0.3, 0.6, 0.1, 0.5], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, + labels, + reduction=reduction, + weight=weights, + ignore_index=ignore_index, + get_log_prob=True, +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[loss, log_prob], + name="test_sce_mean_weight_ii_4d_log_prob", +) +``` + +
+
+softmaxcrossentropy_mean_weights_ii_log_prob + +```python +# Define operator attributes. +reduction = "mean" +ignore_index = np.int64(0) + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, + ignore_index=ignore_index, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +labels[0] = np.int64(0) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, weight=weights, ignore_index=ignore_index, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[loss, log_prob], + name="test_sce_mean_weight_ii_log_prob", +) +``` + +
+
+softmaxcrossentropy_mean_weights_log_prob + +```python +# Define operator attributes. +reduction = "mean" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, weight=weights, get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[loss, log_prob], + name="test_sce_mean_weight_log_prob", +) +``` + +
+
+softmaxcrossentropy_none + +```python +# Define operator attributes. +reduction = "none" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, reduction="none") + +# Check results +expect(node, inputs=[x, labels], outputs=[sce], name="test_sce_none") +``` + +
+
+softmaxcrossentropy_none_log_prob + +```python +# Define operator attributes. +reduction = "none" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, reduction="none", get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_none_log_prob", +) +``` + +
+
+softmaxcrossentropy_none_weights + +```python +# Define operator attributes. +reduction = "none" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, weight=weights, reduction="none") + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[sce], + name="test_sce_none_weights", +) +``` + +
+
+softmaxcrossentropy_none_weights_log_prob + +```python +# Define operator attributes. +reduction = "none" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y", "w"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) +weights = np.array([0.9, 0.7, 0.8, 0.9, 0.9], dtype=np.float32) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, weight=weights, reduction="none", get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels, weights], + outputs=[loss, log_prob], + name="test_sce_none_weights_log_prob", +) +``` + +
+
+softmaxcrossentropy_sum + +```python +# Define operator attributes. +reduction = "sum" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +sce = softmaxcrossentropy(x, labels, reduction="sum") + +# Check results +expect(node, inputs=[x, labels], outputs=[sce], name="test_sce_sum") +``` + +
+
+softmaxcrossentropy_sum_log_prob + +```python +# Define operator attributes. +reduction = "sum" + +# Create operator. +node = onnx.helper.make_node( + "SoftmaxCrossEntropyLoss", + inputs=["x", "y"], + outputs=["z", "log_prob"], + reduction=reduction, +) + +# Define operator inputs. +np.random.seed(0) +x = np.random.rand(3, 5).astype(np.float32) +labels = np.random.randint(0, high=5, size=(3,)).astype(np.int64) + +# Compute SoftmaxCrossEntropyLoss +loss, log_prob = softmaxcrossentropy( + x, labels, reduction="sum", get_log_prob=True +) + +# Check results +expect( + node, + inputs=[x, labels], + outputs=[loss, log_prob], + name="test_sce_sum_log_prob", +) +``` + +
+ + +### Softplus +There are 1 test cases, listed as following: +
+softplus + +```python +node = onnx.helper.make_node( + "Softplus", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.log( + np.exp(x) + 1 +) # expected output [0.31326166, 0.69314718, 1.31326163] +expect(node, inputs=[x], outputs=[y], name="test_softplus_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.log(np.exp(x) + 1) +expect(node, inputs=[x], outputs=[y], name="test_softplus") +``` + +
+ + +### Softsign +There are 1 test cases, listed as following: +
+softsign + +```python +node = onnx.helper.make_node( + "Softsign", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.array([-0.5, 0, 0.5]).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_softsign_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = x / (1 + np.abs(x)) +expect(node, inputs=[x], outputs=[y], name="test_softsign") +``` + +
+ + +### SpaceToDepth +There are 2 test cases, listed as following: +
+example + +```python +node = onnx.helper.make_node( + "SpaceToDepth", + inputs=["x"], + outputs=["y"], + blocksize=2, +) + +# (1, 1, 4, 6) input tensor +x = np.array( + [ + [ + [ + [0, 6, 1, 7, 2, 8], + [12, 18, 13, 19, 14, 20], + [3, 9, 4, 10, 5, 11], + [15, 21, 16, 22, 17, 23], + ] + ] + ] +).astype(np.float32) + +# (1, 4, 2, 3) output tensor +y = np.array( + [ + [ + [[0, 1, 2], [3, 4, 5]], + [[6, 7, 8], [9, 10, 11]], + [[12, 13, 14], [15, 16, 17]], + [[18, 19, 20], [21, 22, 23]], + ] + ] +).astype(np.float32) +expect(node, inputs=[x], outputs=[y], name="test_spacetodepth_example") +``` + +
+
+spacetodepth + +```python +b, c, h, w = shape = (2, 2, 6, 6) +blocksize = 2 +node = onnx.helper.make_node( + "SpaceToDepth", + inputs=["x"], + outputs=["y"], + blocksize=blocksize, +) +x = np.random.random_sample(shape).astype(np.float32) +tmp = np.reshape( + x, [b, c, h // blocksize, blocksize, w // blocksize, blocksize] +) +tmp = np.transpose(tmp, [0, 3, 5, 1, 2, 4]) +y = np.reshape(tmp, [b, c * (blocksize**2), h // blocksize, w // blocksize]) +expect(node, inputs=[x], outputs=[y], name="test_spacetodepth") +``` + +
+ + +### Split +There are 10 test cases, listed as following: +
+1d_opset13 + +```python +node_input = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float32) + +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2", "output_3"], + axis=0, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0]).astype(np.float32), + np.array([5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_1d_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2"], + axis=0, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0, 5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_1d_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) +``` + +
+
+1d_opset18 + +```python +node_input = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float32) + +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2", "output_3"], + axis=0, + num_outputs=3, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0]).astype(np.float32), + np.array([5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_1d_opset18", +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2"], + axis=0, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0, 5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_1d_opset18", +) +``` + +
+
+1d_uneven_split_opset18 + +```python +node_input = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0]).astype(np.float32) + +# If axis is not specified, split is applied on default axis 0 +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2", "output_3", "output_4"], + num_outputs=4, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0]).astype(np.float32), + np.array([5.0, 6.0]).astype(np.float32), + np.array([7.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_1d_uneven_split_opset18", +) +``` + +
+
+2d_opset13 + +```python +node_input = np.array( + [[1.0, 2.0, 3.0, 4.0, 5.0, 6.0], [7.0, 8.0, 9.0, 10.0, 11.0, 12.0]] +).astype(np.float32) + +node = onnx.helper.make_node( + "Split", inputs=["input"], outputs=["output_1", "output_2"], axis=1 +) + +expected_outputs = [ + np.array([[1.0, 2.0, 3.0], [7.0, 8.0, 9.0]]).astype(np.float32), + np.array([[4.0, 5.0, 6.0], [10.0, 11.0, 12.0]]).astype(np.float32), +] + +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_2d_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2"], + axis=1, +) + +expected_outputs = [ + np.array([[1.0, 2.0], [7.0, 8.0]]).astype(np.float32), + np.array([[3.0, 4.0, 5.0, 6.0], [9.0, 10.0, 11.0, 12.0]]).astype( + np.float32 + ), +] + +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_2d_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) +``` + +
+
+2d_opset18 + +```python +node_input = np.array( + [[1.0, 2.0, 3.0, 4.0, 5.0, 6.0], [7.0, 8.0, 9.0, 10.0, 11.0, 12.0]] +).astype(np.float32) + +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2"], + axis=1, + num_outputs=2, +) + +expected_outputs = [ + np.array([[1.0, 2.0, 3.0], [7.0, 8.0, 9.0]]).astype(np.float32), + np.array([[4.0, 5.0, 6.0], [10.0, 11.0, 12.0]]).astype(np.float32), +] + +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_2d", +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2"], + axis=1, +) + +expected_outputs = [ + np.array([[1.0, 2.0], [7.0, 8.0]]).astype(np.float32), + np.array([[3.0, 4.0, 5.0, 6.0], [9.0, 10.0, 11.0, 12.0]]).astype( + np.float32 + ), +] + +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_2d_opset18", +) +``` + +
+
+2d_uneven_split_opset18 + +```python +node_input = np.array( + [ + [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0], + [9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0], + ] +).astype(np.float32) + +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2", "output_3"], + axis=1, + num_outputs=3, +) + +expected_outputs = [ + np.array([[1.0, 2.0, 3.0], [9.0, 10.0, 11.0]]).astype(np.float32), + np.array([[4.0, 5.0, 6.0], [12.0, 13.0, 14.0]]).astype(np.float32), + np.array([[7.0, 8.0], [15.0, 16.0]]).astype(np.float32), +] + +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_2d_uneven_split_opset18", +) +``` + +
+
+default_values_opset13 + +```python +node_input = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float32) + +# If axis is not specified, split is applied on default axis 0 +node = onnx.helper.make_node( + "Split", inputs=["input"], outputs=["output_1", "output_2", "output_3"] +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0]).astype(np.float32), + np.array([5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_default_axis_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", inputs=["input", "split"], outputs=["output_1", "output_2"] +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0, 5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_default_axis_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) +``` + +
+
+default_values_opset18 + +```python +node_input = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).astype(np.float32) + +# If axis is not specified, split is applied on default axis 0 +node = onnx.helper.make_node( + "Split", + inputs=["input"], + outputs=["output_1", "output_2", "output_3"], + num_outputs=3, +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0]).astype(np.float32), + np.array([5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input], + outputs=expected_outputs, + name="test_split_equal_parts_default_axis_opset18", +) + +split = np.array([2, 4]).astype(np.int64) +node = onnx.helper.make_node( + "Split", inputs=["input", "split"], outputs=["output_1", "output_2"] +) + +expected_outputs = [ + np.array([1.0, 2.0]).astype(np.float32), + np.array([3.0, 4.0, 5.0, 6.0]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_variable_parts_default_axis_opset18", +) +``` + +
+
+zero_size_splits_opset13 + +```python +# 1-dimensional tensor with dimension_size=0 +node_input = np.array([]).astype(np.float32) + +# Split empty tensor to tensors of size zero +split = np.array([0, 0, 0]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2", "output_3"], +) + +expected_outputs = [ + np.array([]).astype(np.float32), + np.array([]).astype(np.float32), + np.array([]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_zero_size_splits_opset13", + opset_imports=[onnx.helper.make_opsetid("", 13)], +) +``` + +
+
+zero_size_splits_opset18 + +```python +# 1-dimensional tensor with dimension_size=0 +node_input = np.array([]).astype(np.float32) + +# Split empty tensor to tensors of size zero +split = np.array([0, 0, 0]).astype(np.int64) +node = onnx.helper.make_node( + "Split", + inputs=["input", "split"], + outputs=["output_1", "output_2", "output_3"], +) + +expected_outputs = [ + np.array([]).astype(np.float32), + np.array([]).astype(np.float32), + np.array([]).astype(np.float32), +] +expect( + node, + inputs=[node_input, split], + outputs=expected_outputs, + name="test_split_zero_size_splits_opset18", +) +``` + +
+ + +### SplitToSequence +There are 3 test cases, listed as following: +
+nokeepdims + +```python +data = np.arange(18).reshape((3, 6)).astype(np.float32) + +node = onnx.helper.make_node( + "SplitToSequence", + ["data"], + ["seq"], + axis=1, + keepdims=0, +) + +expected_outputs = [[data[:, i] for i in range(data.shape[1])]] + +expect( + node, + inputs=[data], + outputs=expected_outputs, + name="test_split_to_sequence_nokeepdims", +) +``` + +
+
+with_split_1 + +```python +data = np.arange(18).reshape((3, 6)).astype(np.float32) +split = np.array(2, dtype=np.int64) + +node = onnx.helper.make_node( + "SplitToSequence", ["data", "split"], ["seq"], axis=1 +) + +expected_outputs = [ + [ + np.array([[0.0, 1.0], [6.0, 7.0], [12.0, 13.0]], dtype=np.float32), + np.array([[2.0, 3.0], [8.0, 9.0], [14.0, 15.0]], dtype=np.float32), + np.array([[4.0, 5.0], [10.0, 11.0], [16.0, 17.0]], dtype=np.float32), + ] +] + +expect( + node, + inputs=[data, split], + outputs=expected_outputs, + name="test_split_to_sequence_1", +) +``` + +
+
+with_split_2 + +```python +data = np.arange(18).reshape((3, 6)).astype(np.float32) +split = np.array([1, 2], dtype=np.int64) + +node = onnx.helper.make_node( + "SplitToSequence", ["data", "split"], ["seq"], axis=0 +) + +expected_outputs = [ + [ + data[:1], + data[1:], + ] +] + +expect( + node, + inputs=[data, split], + outputs=expected_outputs, + name="test_split_to_sequence_2", +) +``` + +
+ + +### Sqrt +There are 1 test cases, listed as following: +
+sqrt + +```python +node = onnx.helper.make_node( + "Sqrt", + inputs=["x"], + outputs=["y"], +) + +x = np.array([1, 4, 9]).astype(np.float32) +y = np.sqrt(x) # expected output [1., 2., 3.] +expect(node, inputs=[x], outputs=[y], name="test_sqrt_example") + +x = np.abs(np.random.randn(3, 4, 5).astype(np.float32)) +y = np.sqrt(x) +expect(node, inputs=[x], outputs=[y], name="test_sqrt") +``` + +
+ + +### Squeeze +There are 2 test cases, listed as following: +
+squeeze + +```python +node = onnx.helper.make_node( + "Squeeze", + inputs=["x", "axes"], + outputs=["y"], +) +x = np.random.randn(1, 3, 4, 5).astype(np.float32) +axes = np.array([0], dtype=np.int64) +y = np.squeeze(x, axis=0) + +expect(node, inputs=[x, axes], outputs=[y], name="test_squeeze") +``` + +
+
+squeeze_negative_axes + +```python +node = onnx.helper.make_node( + "Squeeze", + inputs=["x", "axes"], + outputs=["y"], +) +x = np.random.randn(1, 3, 1, 5).astype(np.float32) +axes = np.array([-2], dtype=np.int64) +y = np.squeeze(x, axis=-2) +expect(node, inputs=[x, axes], outputs=[y], name="test_squeeze_negative_axes") +``` + +
+ + +### StringConcat +There are 1 test cases, listed as following: +
+stringconcat + +```python +node = onnx.helper.make_node( + "StringConcat", + inputs=["x", "y"], + outputs=["result"], +) +x = np.array(["abc", "def"]).astype("object") +y = np.array([".com", ".net"]).astype("object") +result = np.array(["abc.com", "def.net"]).astype("object") + +expect(node, inputs=[x, y], outputs=[result], name="test_string_concat") + +x = np.array(["cat", "dog", "snake"]).astype("object") +y = np.array(["s"]).astype("object") +result = np.array(["cats", "dogs", "snakes"]).astype("object") + +expect( + node, + inputs=[x, y], + outputs=[result], + name="test_string_concat_broadcasting", +) + +x = np.array("cat").astype("object") +y = np.array("s").astype("object") +result = np.array("cats").astype("object") + +expect( + node, + inputs=[x, y], + outputs=[result], + name="test_string_concat_zero_dimensional", +) + +x = np.array(["abc", ""]).astype("object") +y = np.array(["", "abc"]).astype("object") +result = np.array(["abc", "abc"]).astype("object") + +expect( + node, + inputs=[x, y], + outputs=[result], + name="test_string_concat_empty_string", +) + +x = np.array(["的", "中"]).astype("object") +y = np.array(["的", "中"]).astype("object") +result = np.array(["的的", "中中"]).astype("object") + +expect( + node, + inputs=[x, y], + outputs=[result], + name="test_string_concat_utf8", +) +``` + +
+ + +### StringNormalizer +There are 6 test cases, listed as following: +
+monday_casesensintive_lower + +```python +input = np.array(["monday", "tuesday", "wednesday", "thursday"]).astype(object) +output = np.array(["tuesday", "wednesday", "thursday"]).astype(object) +stopwords = ["monday"] + +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + case_change_action="LOWER", + is_case_sensitive=1, + stopwords=stopwords, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_export_monday_casesensintive_lower", +) +``` + +
+
+monday_casesensintive_nochangecase + +```python +input = np.array(["monday", "tuesday", "wednesday", "thursday"]).astype(object) +output = np.array(["tuesday", "wednesday", "thursday"]).astype(object) +stopwords = ["monday"] + +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + is_case_sensitive=1, + stopwords=stopwords, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_export_monday_casesensintive_nochangecase", +) +``` + +
+
+monday_casesensintive_upper + +```python +input = np.array(["monday", "tuesday", "wednesday", "thursday"]).astype(object) +output = np.array(["TUESDAY", "WEDNESDAY", "THURSDAY"]).astype(object) +stopwords = ["monday"] + +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + case_change_action="UPPER", + is_case_sensitive=1, + stopwords=stopwords, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_export_monday_casesensintive_upper", +) +``` + +
+
+monday_empty_output + +```python +input = np.array(["monday", "monday"]).astype(object) +output = np.array([""]).astype(object) +stopwords = ["monday"] + +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + case_change_action="UPPER", + is_case_sensitive=1, + stopwords=stopwords, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_export_monday_empty_output", +) +``` + +
+
+monday_insensintive_upper_twodim + +```python +input = ( + np.array( + ["Monday", "tuesday", "wednesday", "Monday", "tuesday", "wednesday"] + ) + .astype(object) + .reshape([1, 6]) +) + +# It does upper case cecedille, accented E +# and german umlaut but fails +# with german eszett +output = ( + np.array(["TUESDAY", "WEDNESDAY", "TUESDAY", "WEDNESDAY"]) + .astype(object) + .reshape([1, 4]) +) +stopwords = ["monday"] + +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + case_change_action="UPPER", + stopwords=stopwords, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_export_monday_insensintive_upper_twodim", +) +``` + +
+
+nostopwords_nochangecase + +```python +input = np.array(["monday", "tuesday"]).astype(object) +output = input + +# No stopwords. This is a NOOP +node = onnx.helper.make_node( + "StringNormalizer", + inputs=["x"], + outputs=["y"], + is_case_sensitive=1, +) +expect( + node, + inputs=[input], + outputs=[output], + name="test_strnormalizer_nostopwords_nochangecase", +) +``` + +
+ + +### StringSplit +There are 5 test cases, listed as following: +
+basic + +```python +node = onnx.helper.make_node( + "StringSplit", + inputs=["x"], + outputs=["substrings", "length"], + delimiter=".", + maxsplit=None, +) + +x = np.array(["abc.com", "def.net"]).astype(object) + +substrings = np.array([["abc", "com"], ["def", "net"]]).astype(object) + +length = np.array([2, 2], dtype=np.int64) + +expect( + node, + inputs=[x], + outputs=[substrings, length], + name="test_string_split_basic", +) +``` + +
+
+consecutive_delimiters + +```python +node = onnx.helper.make_node( + "StringSplit", + inputs=["x"], + outputs=["substrings", "length"], + delimiter="-", + maxsplit=None, +) + +x = np.array(["o-n-n--x-", "o-n----nx"]).astype(object) + +substrings = np.array( + [["o", "n", "n", "", "x", ""], ["o", "n", "", "", "", "nx"]] +).astype(object) + +length = np.array([6, 6], dtype=np.int64) + +expect( + node, + inputs=[x], + outputs=[substrings, length], + name="test_string_split_consecutive_delimiters", +) +``` + +
+
+empty_string_delimiter + +```python +for delimiter, test_name in ( + ("", "test_string_split_empty_string_delimiter"), + (None, "test_string_split_no_delimiter"), +): + node = onnx.helper.make_node( + "StringSplit", + inputs=["x"], + outputs=["substrings", "length"], + delimiter=delimiter, + maxsplit=None, + ) + + x = np.array( + ["hello world !", " hello world !", " hello world ! "] + ).astype(object) + + substrings = np.array( + [ + ["hello", "world", "!"], + ["hello", "world", "!"], + ["hello", "world", "!"], + ] + ).astype(object) + + length = np.array([3, 3, 3], dtype=np.int64) + + expect( + node, + inputs=[x], + outputs=[substrings, length], + name=test_name, + ) +``` + +
+
+empty_string_split + +```python +node = onnx.helper.make_node( + "StringSplit", + inputs=["x"], + outputs=["substrings", "length"], + delimiter=None, + maxsplit=None, +) + +x = np.array([]).astype(object) + +substrings = np.array([]).astype(object).reshape(0, 0) + +length = np.array([], dtype=np.int64) + +expect( + node, + inputs=[x], + outputs=[substrings, length], + name="test_string_split_empty_tensor", + output_type_protos=[ + onnx.helper.make_tensor_type_proto(onnx.TensorProto.STRING, (0, None)), + None, + ], +) +``` + +
+
+maxsplit + +```python +node = onnx.helper.make_node( + "StringSplit", + inputs=["x"], + outputs=["substrings", "length"], + maxsplit=2, +) + +x = np.array( + [["hello world", "def.net"], ["o n n x", "the quick brown fox"]] +).astype(object) + +substrings = np.array( + [ + [["hello", "world", ""], ["def.net", "", ""]], + [["o", "n", "n x"], ["the", "quick", "brown fox"]], + ] +).astype(object) + +length = np.array([[2, 1], [3, 3]], np.int64) + +expect( + node, + inputs=[x], + outputs=[substrings, length], + name="test_string_split_maxsplit", +) +``` + +
+ + +### Sub +There are 2 test cases, listed as following: +
+sub + +```python +node = onnx.helper.make_node( + "Sub", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.array([1, 2, 3]).astype(np.float32) +y = np.array([3, 2, 1]).astype(np.float32) +z = x - y # expected output [-2., 0., 2.] +expect(node, inputs=[x, y], outputs=[z], name="test_sub_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(3, 4, 5).astype(np.float32) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.int8) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.int8) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_int8") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.int16) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.int16) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_int16") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.uint8) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.uint8) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_uint8") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.uint16) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.uint16) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_uint16") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.uint32) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.uint32) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_uint32") + +x = np.random.randint(12, 24, size=(3, 4, 5), dtype=np.uint64) +y = np.random.randint(12, size=(3, 4, 5), dtype=np.uint64) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_uint64") +``` + +
+
+sub_broadcast + +```python +node = onnx.helper.make_node( + "Sub", + inputs=["x", "y"], + outputs=["z"], +) + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.random.randn(5).astype(np.float32) +z = x - y +expect(node, inputs=[x, y], outputs=[z], name="test_sub_bcast") +``` + +
+ + +### Sum +There are 1 test cases, listed as following: +
+sum + +```python +data_0 = np.array([3, 0, 2]).astype(np.float32) +data_1 = np.array([1, 3, 4]).astype(np.float32) +data_2 = np.array([2, 6, 6]).astype(np.float32) +result = np.array([6, 9, 12]).astype(np.float32) +node = onnx.helper.make_node( + "Sum", + inputs=["data_0", "data_1", "data_2"], + outputs=["result"], +) +expect( + node, + inputs=[data_0, data_1, data_2], + outputs=[result], + name="test_sum_example", +) + +node = onnx.helper.make_node( + "Sum", + inputs=["data_0"], + outputs=["result"], +) +expect(node, inputs=[data_0], outputs=[data_0], name="test_sum_one_input") + +result = np.add(data_0, data_1) +node = onnx.helper.make_node( + "Sum", + inputs=["data_0", "data_1"], + outputs=["result"], +) +expect( + node, inputs=[data_0, data_1], outputs=[result], name="test_sum_two_inputs" +) +``` + +
+ + +### Swish +There are 1 test cases, listed as following: +
+swish + +```python +node = onnx.helper.make_node( + "Swish", + inputs=["x"], + outputs=["y"], + alpha=1.0, # pass alpha as attribute +) + +x = np.array([3, 4, 5], dtype=np.float32) +y = swish(x, alpha=1.0) + +expect( + node, + inputs=[x], + outputs=[y], + name="test_swish", + opset_imports=[onnx.helper.make_opsetid("", 24)], +) +``` + +
+ + +### Tan +There are 1 test cases, listed as following: +
+tan + +```python +node = onnx.helper.make_node( + "Tan", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.tan(x) +expect(node, inputs=[x], outputs=[y], name="test_tan_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.tan(x) +expect(node, inputs=[x], outputs=[y], name="test_tan") +``` + +
+ + +### Tanh +There are 1 test cases, listed as following: +
+tanh + +```python +node = onnx.helper.make_node( + "Tanh", + inputs=["x"], + outputs=["y"], +) + +x = np.array([-1, 0, 1]).astype(np.float32) +y = np.tanh(x) # expected output [-0.76159418, 0., 0.76159418] +expect(node, inputs=[x], outputs=[y], name="test_tanh_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.tanh(x) +expect(node, inputs=[x], outputs=[y], name="test_tanh") +``` + +
+ + +### TensorScatter +There are 3 test cases, listed as following: +
+tensorscatter + +```python +node = onnx.helper.make_node( + "TensorScatter", + inputs=["past_cache", "update", "write_indices"], + outputs=["present_cache"], + mode="linear", +) +past_cache = np.array( + [ + [[[1, 2, 3, 4, 5], [5, 6, 7, 8, 9], [8, 7, 6, 5, 4], [4, 3, 2, 1, 0]]], + [[[1, 2, 3, 4, 5], [5, 6, 7, 8, 9], [8, 7, 6, 5, 4], [4, 3, 2, 1, 0]]], + ], + dtype=np.float32, +) +update = np.array( + [ + [[[5, 5, 5, 5, 5]]], + [[[1, 1, 1, 1, 1]]], + ], + dtype=np.float32, +) +write_indices = np.array([1, 2], dtype=np.int64) +present_cache = np.array( + [ + [[[1, 2, 3, 4, 5], [5, 5, 5, 5, 5], [8, 7, 6, 5, 4], [4, 3, 2, 1, 0]]], + [[[1, 2, 3, 4, 5], [5, 6, 7, 8, 9], [1, 1, 1, 1, 1], [4, 3, 2, 1, 0]]], + ], + dtype=np.float32, +) +expect( + node, + inputs=[past_cache, update, write_indices], + outputs=[present_cache], + name="test_tensorscatter", +) +``` + +
+
+tensorscatter_3d + +```python +node = onnx.helper.make_node( + "TensorScatter", + inputs=["past_cache", "update", "write_indices"], + outputs=["present_cache"], +) +past_cache = np.array( + [ + [ + [1, 2, 3, 4, 5], + [5, 6, 7, 8, 9], + [8, 7, 6, 5, 4], + [5, 4, 3, 2, 1], + ], + [ + [1, 2, 3, 4, 5], + [5, 6, 7, 8, 9], + [8, 7, 6, 5, 4], + [5, 4, 3, 2, 1], + ], + [ + [1, 2, 3, 4, 5], + [5, 6, 7, 8, 9], + [8, 7, 6, 5, 4], + [5, 4, 3, 2, 1], + ], + ], + dtype=np.float32, +) +update = np.array( + [ + [ + [4, 4, 4, 4, 4], + [5, 5, 5, 5, 5], + ], + [ + [6, 6, 6, 6, 6], + [7, 7, 7, 7, 7], + ], + [ + [2, 2, 2, 2, 2], + [3, 3, 3, 3, 3], + ], + ], + dtype=np.float32, +) +write_indices = np.array([1, 2, 0], dtype=np.int64) +present_cache = np.array( + [ + [ + [1, 2, 3, 4, 5], + [4, 4, 4, 4, 4], + [5, 5, 5, 5, 5], + [5, 4, 3, 2, 1], + ], + [ + [1, 2, 3, 4, 5], + [5, 6, 7, 8, 9], + [6, 6, 6, 6, 6], + [7, 7, 7, 7, 7], + ], + [ + [2, 2, 2, 2, 2], + [3, 3, 3, 3, 3], + [8, 7, 6, 5, 4], + [5, 4, 3, 2, 1], + ], + ], + dtype=np.float32, +) +expect( + node, + inputs=[past_cache, update, write_indices], + outputs=[present_cache], + name="test_tensorscatter_3d", +) +``` + +
+
+tensorscatter_circular + +```python +node = onnx.helper.make_node( + "TensorScatter", + inputs=["past_cache", "update", "write_indices"], + outputs=["present_cache"], + mode="circular", +) +past_cache = np.array( + [ + [[[1, 2, 3, 4, 5], [5, 6, 7, 8, 9], [8, 7, 6, 5, 4], [4, 3, 2, 1, 0]]], + [[[1, 2, 3, 4, 5], [5, 6, 7, 8, 9], [8, 7, 6, 5, 4], [4, 3, 2, 1, 0]]], + ], + dtype=np.float32, +) +update = np.array( + [ + [ + [ + [5, 5, 5, 5, 5], + [6, 6, 6, 6, 6], + ] + ], + [ + [ + [1, 1, 1, 1, 1], + [2, 2, 2, 2, 2], + ] + ], + ], + dtype=np.float32, +) +write_indices = np.array([1, 3], dtype=np.int64) +present_cache = np.array( + [ + [[[1, 2, 3, 4, 5], [5, 5, 5, 5, 5], [6, 6, 6, 6, 6], [4, 3, 2, 1, 0]]], + [[[2, 2, 2, 2, 2], [5, 6, 7, 8, 9], [8, 7, 6, 5, 4], [1, 1, 1, 1, 1]]], + ], + dtype=np.float32, +) +expect( + node, + inputs=[past_cache, update, write_indices], + outputs=[present_cache], + name="test_tensorscatter_circular", +) +``` + +
+ + +### TfIdfVectorizer +There are 7 test cases, listed as following: +
+tf_batch_onlybigrams_skip0 + +```python +input = np.array([[1, 1, 3, 3, 3, 7], [8, 6, 7, 5, 6, 8]]).astype(np.int32) +output = np.array( + [[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0]] +).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=2, + max_gram_length=2, + max_skip_count=0, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_batch_onlybigrams_skip0", +) +``` + +
+
+tf_batch_onlybigrams_skip5 + +```python +input = np.array([[1, 1, 3, 3, 3, 7], [8, 6, 7, 5, 6, 8]]).astype(np.int32) +output = np.array( + [[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0]] +).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=2, + max_gram_length=2, + max_skip_count=5, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_batch_onlybigrams_skip5", +) +``` + +
+
+tf_batch_uniandbigrams_skip5 + +```python +input = np.array([[1, 1, 3, 3, 3, 7], [8, 6, 7, 5, 6, 8]]).astype(np.int32) +output = np.array( + [[0.0, 3.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0]] +).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=1, + max_gram_length=2, + max_skip_count=5, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_batch_uniandbigrams_skip5", +) +``` + +
+
+tf_only_bigrams_skip0 + +```python +input = np.array([1, 1, 3, 3, 3, 7, 8, 6, 7, 5, 6, 8]).astype(np.int32) +output = np.array([0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0]).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=2, + max_gram_length=2, + max_skip_count=0, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_only_bigrams_skip0", +) +``` + +
+
+tf_onlybigrams_levelempty + +```python +input = np.array([1, 1, 3, 3, 3, 7, 8, 6, 7, 5, 6, 8]).astype(np.int32) +output = np.array([1.0, 1.0, 1.0]).astype(np.float32) + +ngram_counts = np.array([0, 0]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2]).astype(np.int64) +pool_int64s = np.array([5, 6, 7, 8, 6, 7]).astype( # unigrams none + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=2, + max_gram_length=2, + max_skip_count=0, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_onlybigrams_levelempty", +) +``` + +
+
+tf_onlybigrams_skip5 + +```python +input = np.array([1, 1, 3, 3, 3, 7, 8, 6, 7, 5, 6, 8]).astype(np.int32) +output = np.array([0.0, 0.0, 0.0, 0.0, 1.0, 3.0, 1.0]).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=2, + max_gram_length=2, + max_skip_count=5, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_onlybigrams_skip5", +) +``` + +
+
+tf_uniandbigrams_skip5 + +```python +input = np.array([1, 1, 3, 3, 3, 7, 8, 6, 7, 5, 6, 8]).astype(np.int32) +output = np.array([0.0, 3.0, 1.0, 0.0, 1.0, 3.0, 1.0]).astype(np.float32) + +ngram_counts = np.array([0, 4]).astype(np.int64) +ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64) +pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype( # unigrams + np.int64 +) # bigrams + +helper = TfIdfVectorizerHelper( + mode="TF", + min_gram_length=1, + max_gram_length=2, + max_skip_count=5, + ngram_counts=ngram_counts, + ngram_indexes=ngram_indexes, + pool_int64s=pool_int64s, +) +node = helper.make_node_noweights() +expect( + node, + inputs=[input], + outputs=[output], + name="test_tfidfvectorizer_tf_uniandbigrams_skip5", +) +``` + +
+ + +### ThresholdedRelu +There are 2 test cases, listed as following: +
+default + +```python +default_alpha = 1.0 +node = onnx.helper.make_node("ThresholdedRelu", inputs=["x"], outputs=["y"]) +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, default_alpha, np.inf) +y[y == default_alpha] = 0 + +expect(node, inputs=[x], outputs=[y], name="test_thresholdedrelu_default") +``` + +
+
+thresholdedrelu + +```python +alpha = 2.0 +node = onnx.helper.make_node( + "ThresholdedRelu", inputs=["x"], outputs=["y"], alpha=alpha +) + +x = np.array([-1.5, 0.0, 1.2, 2.0, 2.2]).astype(np.float32) +y = np.clip(x, alpha, np.inf) # expected output [0., 0., 0., 0., 2.2] +y[y == alpha] = 0 + +expect(node, inputs=[x], outputs=[y], name="test_thresholdedrelu_example") + +x = np.random.randn(3, 4, 5).astype(np.float32) +y = np.clip(x, alpha, np.inf) +y[y == alpha] = 0 + +expect(node, inputs=[x], outputs=[y], name="test_thresholdedrelu") +``` + +
+ + +### Tile +There are 2 test cases, listed as following: +
+tile + +```python +node = onnx.helper.make_node("Tile", inputs=["x", "y"], outputs=["z"]) + +x = np.random.rand(2, 3, 4, 5).astype(np.float32) + +repeats = np.random.randint(low=1, high=10, size=(np.ndim(x),)).astype(np.int64) + +z = np.tile(x, repeats) + +expect(node, inputs=[x, repeats], outputs=[z], name="test_tile") +``` + +
+
+tile_precomputed + +```python +node = onnx.helper.make_node("Tile", inputs=["x", "y"], outputs=["z"]) + +x = np.array([[0, 1], [2, 3]], dtype=np.float32) + +repeats = np.array([2, 2], dtype=np.int64) + +z = np.array( + [[0, 1, 0, 1], [2, 3, 2, 3], [0, 1, 0, 1], [2, 3, 2, 3]], dtype=np.float32 +) + +expect(node, inputs=[x, repeats], outputs=[z], name="test_tile_precomputed") +``` + +
+ + +### TopK +There are 7 test cases, listed as following: +
+top_k + +```python +axis = 1 +largest = 1 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [ + [0, 1, 2, 3], + [4, 5, 6, 7], + [8, 9, 10, 11], + ], + dtype=np.float32, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [[ 3. 2. 1.] +# [ 7. 6. 5.] +# [11. 10. 9.]] +# print(indices_ref) +# [[3 2 1] +# [3 2 1] +# [3 2 1]] + +expect( + node, inputs=[X, K], outputs=[values_ref, indices_ref], name="test_top_k" +) +``` + +
+
+top_k_negative_axis + +```python +axis = -1 +largest = 1 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [ + [0, 1, 2, 3], + [4, 5, 6, 7], + [8, 9, 10, 11], + ], + dtype=np.float32, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [[ 3. 2. 1.] +# [ 7. 6. 5.] +# [11. 10. 9.]] +# print(indices_ref) +# [[3 2 1] +# [3 2 1] +# [3 2 1]] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_negative_axis", +) +``` + +
+
+top_k_same_values + +```python +axis = 0 +largest = 0 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [0, 0, 0, 0], + dtype=np.int64, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# (Pdb) print(values_ref) +# [0 0 0] +# (Pdb) print(indices_ref) +# [0 1 2] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_same_values", +) +``` + +
+
+top_k_same_values_2d + +```python +axis = 1 +largest = 1 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [[0, 0, 0, 0], [1, 1, 1, 1], [2, 2, 1, 1]], + dtype=np.int64, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [[0 0 0] +# [1 1 1] +# [1 1 2]] +# print(indices_ref) +# [[0 1 2] +# [0 1 2] +# [2 3 0]] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_same_values_2d", +) +``` + +
+
+top_k_same_values_largest + +```python +axis = 0 +largest = 1 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [0, 0, 0, 0], + dtype=np.int64, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [0 0 0] +# print(indices_ref) +# [0 1 2] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_same_values_largest", +) +``` + +
+
+top_k_smallest + +```python +axis = 1 +largest = 0 +sorted_ = 1 +k = 3 + +node = onnx.helper.make_node( + "TopK", + inputs=["x", "k"], + outputs=["values", "indices"], + axis=axis, + largest=largest, + sorted=sorted_, +) + +X = np.array( + [ + [0, 1, 2, 3], + [4, 5, 6, 7], + [11, 10, 9, 8], + ], + dtype=np.float32, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [[ 0. 1. 2.] +# [ 4. 5. 6.] +# [ 8. 9. 10.]] +# print(indices_ref) +# [[0 1 2] +# [0 1 2] +# [3 2 1]] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_smallest", +) +``` + +
+
+top_k_uint64 + +```python +axis = 1 +largest = 1 + +k = 3 +node = onnx.helper.make_node( + "TopK", inputs=["x", "k"], outputs=["values", "indices"], axis=axis +) +X = np.array( + [ + [0, 1, 2, 3], + [4, 5, 6, 7], + [8, 9, 10, 11], + ], + dtype=np.uint64, +) +K = np.array([k], dtype=np.int64) +values_ref, indices_ref = topk_sorted_implementation(X, k, axis, largest) + +# print(values_ref) +# [[ 3 2 1] +# [ 7 6 5] +# [11 10 9]] +# print(indices_ref) +# [[3 2 1] +# [3 2 1] +# [3 2 1]] + +expect( + node, + inputs=[X, K], + outputs=[values_ref, indices_ref], + name="test_top_k_uint64", +) +``` + +
+ + +### Transpose +There are 2 test cases, listed as following: +
+all_permutations + +```python +shape = (2, 3, 4) +data = np.random.random_sample(shape).astype(np.float32) +permutations = list(itertools.permutations(np.arange(len(shape)))) + +for i, permutation in enumerate(permutations): + node = onnx.helper.make_node( + "Transpose", + inputs=["data"], + outputs=["transposed"], + perm=permutation, + ) + transposed = np.transpose(data, permutation) + expect( + node, + inputs=[data], + outputs=[transposed], + name=f"test_transpose_all_permutations_{i}", + ) +``` + +
+
+default + +```python +shape = (2, 3, 4) +data = np.random.random_sample(shape).astype(np.float32) + +node = onnx.helper.make_node( + "Transpose", inputs=["data"], outputs=["transposed"] +) + +transposed = np.transpose(data) +expect(node, inputs=[data], outputs=[transposed], name="test_transpose_default") +``` + +
+ + +### Trilu +There are 18 test cases, listed as following: +
+tril + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 0, 0, 0, 0], +# [1, 2, 0, 0, 0], +# [9, 4, 1, 0, 0], +# [4, 3, 4, 2, 0]] +y = tril_reference_implementation(x) +expect(node, inputs=[x], outputs=[y], name="test_tril") +``` + +
+
+tril_neg + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(-1).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[0, 0, 0, 0, 0], +# [1, 0, 0, 0, 0], +# [9, 4, 0, 0, 0], +# [4, 3, 4, 0, 0]] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_neg") +``` + +
+
+tril_one_row + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(3, 1, 5)).astype(np.int64) +# X: +# [[[6, 2, 4, 1, 6]], +# +# [[8, 3, 8, 7, 0]], +# +# [[2, 2, 9, 5, 9]]] +# expect result: +# [[[6, 0, 0, 0, 0]], +# +# [[8, 0, 0, 0, 0]], +# +# [[2, 0, 0, 0, 0]]] +y = tril_reference_implementation(x) +expect(node, inputs=[x], outputs=[y], name="test_tril_one_row_neg") +``` + +
+
+tril_out_neg + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(-7).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0]] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_out_neg") +``` + +
+
+tril_out_pos + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(6).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_out_pos") +``` + +
+
+tril_pos + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(2).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 7, 3, 0, 0], +# [1, 2, 8, 6, 0], +# [9, 4, 1, 8, 7], +# [4, 3, 4, 2, 4]] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_pos") +``` + +
+
+tril_square + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(2, 3, 3)).astype(np.int64) +# X: +# [[[0, 4, 3], +# [2, 0, 9], +# [8, 2, 5]], +# +# [[2, 7, 2], +# [2, 6, 0], +# [2, 6, 5]]] +# expect result: +# [[[0, 0, 0], +# [2, 0, 0], +# [8, 2, 5]], +# +# [[2, 0, 0], +# [2, 6, 0], +# [2, 6, 5]]] +y = tril_reference_implementation(x) +expect(node, inputs=[x], outputs=[y], name="test_tril_square") +``` + +
+
+tril_square_neg + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(2, 3, 3)).astype(np.int64) +k = np.array(-1).astype(np.int64) +# X: +# [[[0, 4, 3], +# [2, 0, 9], +# [8, 2, 5]], +# +# [[2, 7, 2], +# [2, 6, 0], +# [2, 6, 5]]] +# expect result: +# [[[0, 0, 0], +# [2, 0, 0], +# [8, 2, 0]], +# +# [[0, 0, 0], +# [2, 0, 0], +# [2, 6, 0]]] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_square_neg") +``` + +
+
+tril_zero + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], + upper=0, +) + +x = np.random.randint(10, size=(3, 0, 5)).astype(np.int64) +k = np.array(6).astype(np.int64) +# X: +# [] +# expect result: +# [] +y = tril_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_tril_zero") +``` + +
+
+triu + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x"], + outputs=["y"], +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 7, 3, 7, 9], +# [0, 2, 8, 6, 9], +# [0, 0, 0, 8, 7], +# [0, 0, 0, 2, 4]] +y = triu_reference_implementation(x) +expect(node, inputs=[x], outputs=[y], name="test_triu") +``` + +
+
+triu_neg + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(-1).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [0, 4, 0, 8, 7], +# [0, 0, 4, 2, 4]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_neg") +``` + +
+
+triu_one_row + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(3, 1, 5)).astype(np.int64) +k = np.array(1).astype(np.int64) +# X: +# [[[1, 4, 9, 7, 1]], +# +# [[9, 2, 8, 8, 4]], +# +# [[3, 9, 7, 4, 2]]] +# expect result: +# [[[0, 4, 9, 7, 1]], +# +# [[0, 2, 8, 8, 4]], +# +# [[0, 9, 7, 4, 2]]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_one_row") +``` + +
+
+triu_out_neg_out + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(-7).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_out_neg_out") +``` + +
+
+triu_out_pos + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(6).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0], +# [0, 0, 0, 0, 0]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_out_pos") +``` + +
+
+triu_pos + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(4, 5)).astype(np.int64) +k = np.array(2).astype(np.int64) +# X: +# [[4, 7, 3, 7, 9], +# [1, 2, 8, 6, 9], +# [9, 4, 0, 8, 7], +# [4, 3, 4, 2, 4]] +# expect result: +# [[0, 0, 3, 7, 9], +# [0, 0, 0, 6, 9], +# [0, 0, 0, 0, 7], +# [0, 0, 0, 0, 0]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_pos") +``` + +
+
+triu_square + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x"], + outputs=["y"], +) + +x = np.random.randint(10, size=(2, 3, 3)).astype(np.int64) +y = triu_reference_implementation(x) +# X: +# [[[4, 6, 9], +# [7, 5, 4], +# [8, 1, 2]], +# +# [[1, 4, 9], +# [9, 6, 3], +# [8, 9, 8]]] +# expect result: +# [[[4, 6, 9], +# [0, 5, 4], +# [0, 0, 2]], +# +# [[1, 4, 9], +# [0, 6, 3], +# [0, 0, 8]]] +expect(node, inputs=[x], outputs=[y], name="test_triu_square") +``` + +
+
+triu_square_neg + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(2, 3, 3)).astype(np.int64) +k = np.array(-1).astype(np.int64) +# X: +# [[[4, 6, 9], +# [7, 5, 4], +# [8, 1, 2]], +# +# [[1, 4, 9], +# [9, 6, 3], +# [8, 9, 8]]] +# expect result: +# [[[4, 6, 9], +# [7, 5, 4], +# [0, 1, 2]], +# +# [[1, 4, 9], +# [9, 6, 3], +# [0, 9, 8]]] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_square_neg") +``` + +
+
+triu_zero + +```python +node = onnx.helper.make_node( + "Trilu", + inputs=["x", "k"], + outputs=["y"], +) + +x = np.random.randint(10, size=(0, 5)).astype(np.int64) +k = np.array(6).astype(np.int64) +# X: +# [] +# expect result: +# [] +y = triu_reference_implementation(x, int(k)) +expect(node, inputs=[x, k], outputs=[y], name="test_triu_zero") +``` + +
+ + +### Unique +There are 6 test cases, listed as following: +
+length_1 + +```python +node_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], + sorted=1, +) + +x = np.array([0], dtype=np.int64) +y, indices, inverse_indices, counts = np.unique(x, True, True, True) +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +# behavior changed with numpy >= 2.0 +inverse_indices = inverse_indices.reshape(-1) +# print(y) +# [0] +# print(indices) +# [0] +# print(inverse_indices) +# [0] +# print(counts) +# [1] + +expect( + node_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_length_1", +) +``` + +
+
+not_sorted_without_axis + +```python +node_not_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], + sorted=0, +) +# numpy unique does not retain original order (it sorts the output unique values) +# https://github.com/numpy/numpy/issues/8621 +# we need to recover unsorted output and indices +x = np.array([2.0, 1.0, 1.0, 3.0, 4.0, 3.0], dtype=np.float32) +y, indices, inverse_indices, counts = np.unique(x, True, True, True) + +# prepare index mapping from sorted to unsorted +argsorted_indices = np.argsort(indices) +inverse_indices_map = dict( + zip(argsorted_indices, np.arange(len(argsorted_indices)), strict=True) +) + +indices = indices[argsorted_indices] +y = np.take(x, indices, axis=0) +inverse_indices = np.asarray( + [inverse_indices_map[i] for i in inverse_indices], dtype=np.int64 +) +counts = counts[argsorted_indices] +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +# print(y) +# [2.0, 1.0, 3.0, 4.0] +# print(indices) +# [0 1 3 4] +# print(inverse_indices) +# [0, 1, 1, 2, 3, 2] +# print(counts) +# [1, 2, 2, 1] + +expect( + node_not_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_not_sorted_without_axis", +) +``` + +
+
+sorted_with_axis + +```python +node_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], + sorted=1, + axis=0, +) + +x = np.array([[1, 0, 0], [1, 0, 0], [2, 3, 4]], dtype=np.float32) +y, indices, inverse_indices, counts = np.unique(x, True, True, True, axis=0) +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +# behavior changed with numpy >= 2.0 +inverse_indices = inverse_indices.reshape(-1) +# print(y) +# [[1. 0. 0.] +# [2. 3. 4.]] +# print(indices) +# [0 2] +# print(inverse_indices) +# [0 0 1] +# print(counts) +# [2 1] + +expect( + node_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_sorted_with_axis", +) +``` + +
+
+sorted_with_axis_3d + +```python +node_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], + sorted=1, + axis=1, +) + +x = np.array( + [ + [[1.0, 1.0], [0.0, 1.0], [2.0, 1.0], [0.0, 1.0]], + [[1.0, 1.0], [0.0, 1.0], [2.0, 1.0], [0.0, 1.0]], + ], + dtype=np.float32, +) +y, indices, inverse_indices, counts = np.unique(x, True, True, True, axis=1) +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +# behavior changed with numpy >= 2.0 +inverse_indices = inverse_indices.reshape(-1) +# print(y) +# [[[0. 1.] +# [1. 1.] +# [2. 1.]] +# [[0. 1.] +# [1. 1.] +# [2. 1.]]] +# print(indices) +# [1 0 2] +# print(inverse_indices) +# [1 0 2 0] +# print(counts) +# [2 1 1] +expect( + node_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_sorted_with_axis_3d", +) +``` + +
+
+sorted_with_negative_axis + +```python +node_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], + sorted=1, + axis=-1, +) + +x = np.array([[1, 0, 0], [1, 0, 0], [2, 3, 3]], dtype=np.float32) +y, indices, inverse_indices, counts = np.unique(x, True, True, True, axis=-1) +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +# behavior changed with numpy >= 2.0 +inverse_indices = inverse_indices.reshape(-1) +# print(y) +# [[0. 1.] +# [0. 1.] +# [3. 2.]] +# print(indices) +# [1 0] +# print(inverse_indices) +# [1 0 0] +# print(counts) +# [2 1] + +expect( + node_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_sorted_with_negative_axis", +) +``` + +
+
+sorted_without_axis + +```python +node_sorted = onnx.helper.make_node( + "Unique", + inputs=["X"], + outputs=["Y", "indices", "inverse_indices", "counts"], +) + +x = np.array([2.0, 1.0, 1.0, 3.0, 4.0, 3.0], dtype=np.float32) +y, indices, inverse_indices, counts = np.unique(x, True, True, True) +indices, inverse_indices, counts = specify_int64( + indices, inverse_indices, counts +) +expect( + node_sorted, + inputs=[x], + outputs=[y, indices, inverse_indices, counts], + name="test_unique_sorted_without_axis", +) +``` + +
+ + +### Unsqueeze +There are 5 test cases, listed as following: +
+unsqueeze_negative_axes + +```python +node = onnx.helper.make_node( + "Unsqueeze", + inputs=["x", "axes"], + outputs=["y"], +) +x = np.random.randn(1, 3, 1, 5).astype(np.float32) +axes = np.array([-2]).astype(np.int64) +y = np.expand_dims(x, axis=-2) +expect(node, inputs=[x, axes], outputs=[y], name="test_unsqueeze_negative_axes") +``` + +
+
+unsqueeze_one_axis + +```python +x = np.random.randn(3, 4, 5).astype(np.float32) + +for i in range(x.ndim): + axes = np.array([i]).astype(np.int64) + node = onnx.helper.make_node( + "Unsqueeze", + inputs=["x", "axes"], + outputs=["y"], + ) + y = np.expand_dims(x, axis=i) + + expect( + node, + inputs=[x, axes], + outputs=[y], + name="test_unsqueeze_axis_" + str(i), + ) +``` + +
+
+unsqueeze_three_axes + +```python +x = np.random.randn(3, 4, 5).astype(np.float32) +axes = np.array([2, 4, 5]).astype(np.int64) + +node = onnx.helper.make_node( + "Unsqueeze", + inputs=["x", "axes"], + outputs=["y"], +) +y = np.expand_dims(x, axis=2) +y = np.expand_dims(y, axis=4) +y = np.expand_dims(y, axis=5) + +expect(node, inputs=[x, axes], outputs=[y], name="test_unsqueeze_three_axes") +``` + +
+
+unsqueeze_two_axes + +```python +x = np.random.randn(3, 4, 5).astype(np.float32) +axes = np.array([1, 4]).astype(np.int64) + +node = onnx.helper.make_node( + "Unsqueeze", + inputs=["x", "axes"], + outputs=["y"], +) +y = np.expand_dims(x, axis=1) +y = np.expand_dims(y, axis=4) + +expect(node, inputs=[x, axes], outputs=[y], name="test_unsqueeze_two_axes") +``` + +
+
+unsqueeze_unsorted_axes + +```python +x = np.random.randn(3, 4, 5).astype(np.float32) +axes = np.array([5, 4, 2]).astype(np.int64) + +node = onnx.helper.make_node( + "Unsqueeze", + inputs=["x", "axes"], + outputs=["y"], +) +y = np.expand_dims(x, axis=2) +y = np.expand_dims(y, axis=4) +y = np.expand_dims(y, axis=5) + +expect(node, inputs=[x, axes], outputs=[y], name="test_unsqueeze_unsorted_axes") +``` + +
+ + +### Upsample +There are 1 test cases, listed as following: +
+nearest + +```python +node = onnx.helper.make_node( + "Upsample", + inputs=["X", "scales"], + outputs=["Y"], + mode="nearest", +) + +data = np.array( + [ + [ + [ + [1, 2], + [3, 4], + ] + ] + ], + dtype=np.float32, +) + +scales = np.array([1.0, 1.0, 2.0, 3.0], dtype=np.float32) + +output = np.array( + [ + [ + [ + [1, 1, 1, 2, 2, 2], + [1, 1, 1, 2, 2, 2], + [3, 3, 3, 4, 4, 4], + [3, 3, 3, 4, 4, 4], + ] + ] + ], + dtype=np.float32, +) + +expect( + node, + inputs=[data, scales], + outputs=[output], + name="test_upsample_nearest", + opset_imports=[helper.make_opsetid("", 9)], +) +``` + +
+ + +### Where +There are 2 test cases, listed as following: +
+long + +```python +node = onnx.helper.make_node( + "Where", + inputs=["condition", "x", "y"], + outputs=["z"], +) + +condition = np.array([[1, 0], [1, 1]], dtype=bool) +x = np.array([[1, 2], [3, 4]], dtype=np.int64) +y = np.array([[9, 8], [7, 6]], dtype=np.int64) +z = np.where(condition, x, y) # expected output [[1, 8], [3, 4]] +expect( + node, inputs=[condition, x, y], outputs=[z], name="test_where_long_example" +) +``` + +
+
+where + +```python +node = onnx.helper.make_node( + "Where", + inputs=["condition", "x", "y"], + outputs=["z"], +) + +condition = np.array([[1, 0], [1, 1]], dtype=bool) +x = np.array([[1, 2], [3, 4]], dtype=np.float32) +y = np.array([[9, 8], [7, 6]], dtype=np.float32) +z = np.where(condition, x, y) # expected output [[1, 8], [3, 4]] +expect(node, inputs=[condition, x, y], outputs=[z], name="test_where_example") +``` + +
+ + +### Xor +There are 2 test cases, listed as following: +
+xor + +```python +node = onnx.helper.make_node( + "Xor", + inputs=["x", "y"], + outputs=["xor"], +) + +# 2d +x = (np.random.randn(3, 4) > 0).astype(bool) +y = (np.random.randn(3, 4) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor2d") + +# 3d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(3, 4, 5) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor3d") + +# 4d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor4d") +``` + +
+
+xor_broadcast + +```python +node = onnx.helper.make_node( + "Xor", + inputs=["x", "y"], + outputs=["xor"], +) + +# 3d vs 1d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(5) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor_bcast3v1d") + +# 3d vs 2d +x = (np.random.randn(3, 4, 5) > 0).astype(bool) +y = (np.random.randn(4, 5) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor_bcast3v2d") + +# 4d vs 2d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(5, 6) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor_bcast4v2d") + +# 4d vs 3d +x = (np.random.randn(3, 4, 5, 6) > 0).astype(bool) +y = (np.random.randn(4, 5, 6) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor_bcast4v3d") + +# 4d vs 4d +x = (np.random.randn(1, 4, 1, 6) > 0).astype(bool) +y = (np.random.randn(3, 1, 5, 6) > 0).astype(bool) +z = np.logical_xor(x, y) +expect(node, inputs=[x, y], outputs=[z], name="test_xor_bcast4v4d") +``` + +
+ + +
+ +## 💔No Cover Common Operators +### ConcatFromSequence (call for test cases) + + +### GlobalLpPool (call for test cases) + + +### GreaterOrEqual (call for test cases) + + +### LessOrEqual (call for test cases) + + +### MaxRoiPool (call for test cases) + + +### Multinomial (random generator operator) + + +### Optional (call for test cases) + + +### OptionalGetElement (call for test cases) + + +### RandomNormal (random generator operator) + + +### RandomNormalLike (random generator operator) + + +### RandomUniform (random generator operator) + + +### RandomUniformLike (random generator operator) + + +### SequenceAt (call for test cases) + + +### SequenceConstruct (call for test cases) + + +### SequenceEmpty (call for test cases) + + +### SequenceErase (call for test cases) + + +### SequenceLength (call for test cases) + + +
+ +## 💚Covered Experimental Operators +### FlexAttention +There are 11 test cases, listed as following: +
+flexattention + +```python +"""Basic FlexAttention test with default settings.""" +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, L, E = 2, 4, 8, 16 +S, Ev = 6, 16 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +(Y,) = _compute_flex_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+
+flexattention_causal_mask + +```python +"""FlexAttention with causal masking score_mod (Qwen-3, Gemma-3, Llama-3 pattern).""" +score_mod_graph = _make_score_mod_causal_mask_graph(TensorProto.FLOAT) + +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) +score_mod_attr = helper.make_attribute("score_mod", score_mod_graph) +node.attribute.append(score_mod_attr) + +B, Hq, L, E = 1, 2, 4, 8 +S, Ev = 4, 8 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +# Manually compute expected output with causal masking +scale = 1.0 / np.sqrt(E) +scores = np.einsum("bhle,bhse->bhls", Q, K) * scale +# Apply causal mask: set future positions to -inf +q_idx = np.arange(L).reshape(1, 1, L, 1) +k_idx = np.arange(S).reshape(1, 1, 1, S) +mask = q_idx >= k_idx +scores = np.where(mask, scores, -np.inf) +probs = np.exp(scores - scores.max(axis=-1, keepdims=True)) +probs = probs / probs.sum(axis=-1, keepdims=True) +Y = np.einsum("bhls,bhsv->bhlv", probs, V).astype(np.float32) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_causal_mask", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+
+flexattention_diff_head_sizes + +```python +"""FlexAttention with different head sizes for Q/K vs V.""" +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, L, E = 2, 4, 8, 16 +S, Ev = 6, 32 # V has different head size + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +(Y,) = _compute_flex_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_diff_head_sizes", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+
+flexattention_double + +```python +"""FlexAttention with double precision inputs.""" +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, L, E = 2, 4, 8, 16 +S, Ev = 6, 16 + +Q = np.random.rand(B, Hq, L, E).astype(np.float64) +K = np.random.rand(B, Hq, S, E).astype(np.float64) +V = np.random.rand(B, Hq, S, Ev).astype(np.float64) + +(Y,) = _compute_flex_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_double", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+
+flexattention_fp16 + +```python +"""FlexAttention with float16 inputs.""" +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, L, E = 2, 4, 8, 16 +S, Ev = 6, 16 + +Q = np.random.rand(B, Hq, L, E).astype(np.float16) +K = np.random.rand(B, Hq, S, E).astype(np.float16) +V = np.random.rand(B, Hq, S, Ev).astype(np.float16) + +(Y,) = _compute_flex_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_fp16", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+
+flexattention_gqa + +```python +"""FlexAttention with Grouped Query Attention (GQA).""" +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, Hkv, L, S, E, Ev = 2, 8, 2, 4, 6, 16, 16 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hkv, S, E).astype(np.float32) +V = np.random.rand(B, Hkv, S, Ev).astype(np.float32) + +(Y,) = _compute_flex_attention(Q, K, V) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_gqa", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+
+flexattention_prob_mod + +```python +"""FlexAttention with prob_mod subgraph (scales probabilities).""" +scale_value = 0.5 +prob_mod_graph = _make_prob_mod_scale_graph(scale_value, TensorProto.FLOAT) + +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) +prob_mod_attr = helper.make_attribute("prob_mod", prob_mod_graph) +node.attribute.append(prob_mod_attr) + +B, Hq, L, E = 1, 2, 3, 4 +S, Ev = 3, 4 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +scale = 1.0 / np.sqrt(E) +scores = np.einsum("bhle,bhse->bhls", Q, K) * scale +probs = np.exp(scores - scores.max(axis=-1, keepdims=True)) +probs = probs / probs.sum(axis=-1, keepdims=True) +probs = probs * scale_value +Y = np.einsum("bhls,bhsv->bhlv", probs, V).astype(np.float32) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_prob_mod", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+
+flexattention_relative_positional + +```python +"""FlexAttention with relative positional bias score_mod.""" +score_mod_graph = _make_score_mod_relative_positional_graph(TensorProto.FLOAT) + +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) +score_mod_attr = helper.make_attribute("score_mod", score_mod_graph) +node.attribute.append(score_mod_attr) + +B, Hq, L, E = 1, 2, 4, 8 +S, Ev = 4, 8 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +# Manually compute expected output with relative positional bias +scale = 1.0 / np.sqrt(E) +scores = np.einsum("bhle,bhse->bhls", Q, K) * scale +q_idx = np.arange(L).reshape(-1, 1) +k_idx = np.arange(S).reshape(1, -1) +rel_pos = (q_idx - k_idx).astype(np.float32) +scores = scores + rel_pos +probs = np.exp(scores - scores.max(axis=-1, keepdims=True)) +probs = probs / probs.sum(axis=-1, keepdims=True) +Y = np.einsum("bhls,bhsv->bhlv", probs, V).astype(np.float32) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_relative_positional", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+
+flexattention_scaled + +```python +"""FlexAttention with explicit scale attribute.""" +scale = 0.1 +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + scale=scale, + domain=AI_ONNX_PREVIEW_DOMAIN, +) + +B, Hq, L, E = 2, 4, 8, 16 +S, Ev = 6, 16 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +(Y,) = _compute_flex_attention(Q, K, V, scale=scale) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_scaled", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+
+flexattention_score_mod + +```python +"""FlexAttention with score_mod subgraph (adds bias to scores).""" +bias_value = 0.5 +score_mod_graph = _make_score_mod_bias_graph(bias_value, TensorProto.FLOAT) + +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) +# Add score_mod as a graph attribute +score_mod_attr = helper.make_attribute("score_mod", score_mod_graph) +node.attribute.append(score_mod_attr) + +B, Hq, L, E = 1, 2, 3, 4 +S, Ev = 3, 4 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +scale = 1.0 / np.sqrt(E) +scores = np.einsum("bhle,bhse->bhls", Q, K) * scale +scores = scores + bias_value # score_mod: add bias +probs = np.exp(scores - scores.max(axis=-1, keepdims=True)) +probs = probs / probs.sum(axis=-1, keepdims=True) +Y = np.einsum("bhls,bhsv->bhlv", probs, V).astype(np.float32) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_score_mod", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+
+flexattention_soft_cap + +```python +"""FlexAttention with soft capping score_mod (Gemma-2 pattern).""" +cap_value = 20.0 +score_mod_graph = _make_score_mod_soft_cap_graph(cap_value, TensorProto.FLOAT) + +node = helper.make_node( + "FlexAttention", + inputs=["Q", "K", "V"], + outputs=["Y"], + domain=AI_ONNX_PREVIEW_DOMAIN, +) +score_mod_attr = helper.make_attribute("score_mod", score_mod_graph) +node.attribute.append(score_mod_attr) + +B, Hq, L, E = 1, 2, 4, 8 +S, Ev = 4, 8 + +Q = np.random.rand(B, Hq, L, E).astype(np.float32) +K = np.random.rand(B, Hq, S, E).astype(np.float32) +V = np.random.rand(B, Hq, S, Ev).astype(np.float32) + +# Manually compute expected output with soft capping +scale = 1.0 / np.sqrt(E) +scores = np.einsum("bhle,bhse->bhls", Q, K) * scale +scores = np.tanh(scores / cap_value) * cap_value +probs = np.exp(scores - scores.max(axis=-1, keepdims=True)) +probs = probs / probs.sum(axis=-1, keepdims=True) +Y = np.einsum("bhls,bhsv->bhlv", probs, V).astype(np.float32) + +expect( + node, + inputs=[Q, K, V], + outputs=[Y], + name="test_flexattention_soft_cap", + opset_imports=[ + helper.make_opsetid("", 26), + helper.make_opsetid(AI_ONNX_PREVIEW_DOMAIN, 1), + ], +) +``` + +
+ + +
+ +## 💔No Cover Experimental Operators +
+ +# Model Test Coverage +## bvlc_alexnet + +bvlc_alexnet has 40 nodes. Of these, 40 are covered by node tests (100.0%) + + +
+nodes + +
+ConstantOfShape: 1 out of 1 attributes covered + +value: 1 +
+
+Conv: 4 out of 6 attributes covered + +auto_pad: 0 +dilations: 0 +group: 1 +kernel_shape: 3 +pads: 3 +strides: 2 +
+
+Dropout: 1 out of 1 attributes covered + +seed: 0 +
+
+Gemm: 1 out of 4 attributes covered + +alpha: 0 +beta: 0 +transA: 0 +transB: 1 +
+
+LRN: 4 out of 4 attributes covered + +alpha: 1 +beta: 1 +bias: 1 +size: 1 +
+
+MaxPool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +dilations: 0 +kernel_shape: 1 +pads: 2 +storage_order: 0 +strides: 1 +
+
+ + +## densenet121 + +densenet121 has 1746 nodes. Of these, 1746 are covered by node tests (100.0%) + + +
+nodes + +
+AveragePool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +count_include_pad: 0 +dilations: 0 +kernel_shape: 1 +pads: 1 +strides: 1 +
+
+BatchNormalization: 1 out of 3 attributes covered + +epsilon: 1 +momentum: 0 +training_mode: 0 +
+
+Concat: 1 out of 1 attributes covered + +axis: 1 +
+
+ConstantOfShape: 1 out of 1 attributes covered + +value: 1 +
+
+Conv: 4 out of 6 attributes covered + +auto_pad: 0 +dilations: 0 +group: 1 +kernel_shape: 5 +pads: 4 +strides: 3 +
+
+Dropout: 1 out of 1 attributes covered + +seed: 0 +
+
+Gemm: 1 out of 4 attributes covered + +alpha: 0 +beta: 0 +transA: 0 +transB: 1 +
+
+LRN: 4 out of 4 attributes covered + +alpha: 1 +beta: 1 +bias: 1 +size: 1 +
+
+MaxPool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +dilations: 0 +kernel_shape: 1 +pads: 3 +storage_order: 0 +strides: 1 +
+
+Unsqueeze: 1 out of 0 attributes covered + +
+
+ + +## inception_v1 + +inception_v1 has 237 nodes. Of these, 237 are covered by node tests (100.0%) + + +
+nodes + +
+AveragePool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +count_include_pad: 0 +dilations: 0 +kernel_shape: 2 +pads: 2 +strides: 2 +
+
+BatchNormalization: 1 out of 3 attributes covered + +epsilon: 1 +momentum: 0 +training_mode: 0 +
+
+Concat: 1 out of 1 attributes covered + +axis: 1 +
+
+ConstantOfShape: 1 out of 1 attributes covered + +value: 1 +
+
+Conv: 4 out of 6 attributes covered + +auto_pad: 0 +dilations: 0 +group: 1 +kernel_shape: 5 +pads: 4 +strides: 3 +
+
+Dropout: 1 out of 1 attributes covered + +seed: 0 +
+
+Gemm: 1 out of 4 attributes covered + +alpha: 0 +beta: 0 +transA: 0 +transB: 1 +
+
+LRN: 4 out of 4 attributes covered + +alpha: 1 +beta: 1 +bias: 1 +size: 1 +
+
+MaxPool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +dilations: 0 +kernel_shape: 1 +pads: 3 +storage_order: 0 +strides: 2 +
+
+Unsqueeze: 1 out of 0 attributes covered + +
+
+ + +## inception_v2 + +inception_v2 has 916 nodes. Of these, 916 are covered by node tests (100.0%) + + +
+nodes + +
+AveragePool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +count_include_pad: 0 +dilations: 0 +kernel_shape: 3 +pads: 3 +strides: 2 +
+
+BatchNormalization: 1 out of 3 attributes covered + +epsilon: 1 +momentum: 0 +training_mode: 0 +
+
+Concat: 1 out of 1 attributes covered + +axis: 1 +
+
+ConstantOfShape: 1 out of 1 attributes covered + +value: 1 +
+
+Conv: 4 out of 6 attributes covered + +auto_pad: 0 +dilations: 0 +group: 1 +kernel_shape: 5 +pads: 4 +strides: 3 +
+
+Dropout: 1 out of 1 attributes covered + +seed: 0 +
+
+Gemm: 1 out of 4 attributes covered + +alpha: 0 +beta: 0 +transA: 0 +transB: 1 +
+
+LRN: 4 out of 4 attributes covered + +alpha: 1 +beta: 1 +bias: 1 +size: 1 +
+
+MaxPool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +dilations: 0 +kernel_shape: 1 +pads: 3 +storage_order: 0 +strides: 2 +
+
+Unsqueeze: 1 out of 0 attributes covered + +
+
+ + +## resnet50 + +resnet50 has 415 nodes. Of these, 415 are covered by node tests (100.0%) + + +
+nodes + +
+AveragePool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +count_include_pad: 0 +dilations: 0 +kernel_shape: 3 +pads: 3 +strides: 2 +
+
+BatchNormalization: 1 out of 3 attributes covered + +epsilon: 2 +momentum: 0 +training_mode: 0 +
+
+Concat: 1 out of 1 attributes covered + +axis: 1 +
+
+ConstantOfShape: 1 out of 1 attributes covered + +value: 1 +
+
+Conv: 4 out of 6 attributes covered + +auto_pad: 0 +dilations: 0 +group: 1 +kernel_shape: 5 +pads: 4 +strides: 3 +
+
+Dropout: 1 out of 1 attributes covered + +seed: 0 +
+
+Gemm: 1 out of 4 attributes covered + +alpha: 0 +beta: 0 +transA: 0 +transB: 1 +
+
+LRN: 4 out of 4 attributes covered + +alpha: 1 +beta: 1 +bias: 1 +size: 1 +
+
+MaxPool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +dilations: 0 +kernel_shape: 1 +pads: 3 +storage_order: 0 +strides: 2 +
+
+Unsqueeze: 1 out of 0 attributes covered + +
+
+ + +## shufflenet + +shufflenet has 446 nodes. Of these, 446 are covered by node tests (100.0%) + + +
+nodes + +
+AveragePool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +count_include_pad: 0 +dilations: 0 +kernel_shape: 3 +pads: 3 +strides: 2 +
+
+BatchNormalization: 1 out of 3 attributes covered + +epsilon: 2 +momentum: 0 +training_mode: 0 +
+
+Concat: 1 out of 1 attributes covered + +axis: 1 +
+
+ConstantOfShape: 1 out of 1 attributes covered + +value: 1 +
+
+Conv: 4 out of 6 attributes covered + +auto_pad: 0 +dilations: 0 +group: 6 +kernel_shape: 5 +pads: 4 +strides: 3 +
+
+Dropout: 1 out of 1 attributes covered + +seed: 0 +
+
+Gemm: 1 out of 4 attributes covered + +alpha: 0 +beta: 0 +transA: 0 +transB: 1 +
+
+LRN: 4 out of 4 attributes covered + +alpha: 1 +beta: 1 +bias: 1 +size: 1 +
+
+MaxPool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +dilations: 0 +kernel_shape: 1 +pads: 3 +storage_order: 0 +strides: 2 +
+
+Transpose: 1 out of 1 attributes covered + +perm: 1 +
+
+Unsqueeze: 1 out of 0 attributes covered + +
+
+ + +## squeezenet_old + +squeezenet_old has 105 nodes. Of these, 105 are covered by node tests (100.0%) + + +
+nodes + +
+AveragePool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +count_include_pad: 0 +dilations: 0 +kernel_shape: 3 +pads: 3 +strides: 2 +
+
+BatchNormalization: 1 out of 3 attributes covered + +epsilon: 2 +momentum: 0 +training_mode: 0 +
+
+Concat: 1 out of 1 attributes covered + +axis: 1 +
+
+ConstantOfShape: 1 out of 1 attributes covered + +value: 1 +
+
+Conv: 4 out of 6 attributes covered + +auto_pad: 0 +dilations: 0 +group: 6 +kernel_shape: 5 +pads: 4 +strides: 3 +
+
+Dropout: 1 out of 1 attributes covered + +seed: 0 +
+
+Gemm: 1 out of 4 attributes covered + +alpha: 0 +beta: 0 +transA: 0 +transB: 1 +
+
+LRN: 4 out of 4 attributes covered + +alpha: 1 +beta: 1 +bias: 1 +size: 1 +
+
+MaxPool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +dilations: 0 +kernel_shape: 1 +pads: 3 +storage_order: 0 +strides: 2 +
+
+Transpose: 1 out of 1 attributes covered + +perm: 1 +
+
+Unsqueeze: 1 out of 0 attributes covered + +
+
+ + +## vgg19 + +vgg19 has 82 nodes. Of these, 82 are covered by node tests (100.0%) + + +
+nodes + +
+AveragePool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +count_include_pad: 0 +dilations: 0 +kernel_shape: 3 +pads: 3 +strides: 2 +
+
+BatchNormalization: 1 out of 3 attributes covered + +epsilon: 2 +momentum: 0 +training_mode: 0 +
+
+Concat: 1 out of 1 attributes covered + +axis: 1 +
+
+ConstantOfShape: 1 out of 1 attributes covered + +value: 1 +
+
+Conv: 4 out of 6 attributes covered + +auto_pad: 0 +dilations: 0 +group: 6 +kernel_shape: 5 +pads: 4 +strides: 3 +
+
+Dropout: 1 out of 1 attributes covered + +seed: 0 +
+
+Gemm: 1 out of 4 attributes covered + +alpha: 0 +beta: 0 +transA: 0 +transB: 1 +
+
+LRN: 4 out of 4 attributes covered + +alpha: 1 +beta: 1 +bias: 1 +size: 1 +
+
+MaxPool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +dilations: 0 +kernel_shape: 2 +pads: 3 +storage_order: 0 +strides: 2 +
+
+Transpose: 1 out of 1 attributes covered + +perm: 1 +
+
+Unsqueeze: 1 out of 0 attributes covered + +
+
+ + +## zfnet512 + +zfnet512 has 38 nodes. Of these, 38 are covered by node tests (100.0%) + + +
+nodes + +
+AveragePool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +count_include_pad: 0 +dilations: 0 +kernel_shape: 3 +pads: 3 +strides: 2 +
+
+BatchNormalization: 1 out of 3 attributes covered + +epsilon: 2 +momentum: 0 +training_mode: 0 +
+
+Concat: 1 out of 1 attributes covered + +axis: 1 +
+
+ConstantOfShape: 1 out of 1 attributes covered + +value: 1 +
+
+Conv: 4 out of 6 attributes covered + +auto_pad: 0 +dilations: 0 +group: 6 +kernel_shape: 5 +pads: 4 +strides: 3 +
+
+Dropout: 1 out of 1 attributes covered + +seed: 0 +
+
+Gemm: 1 out of 4 attributes covered + +alpha: 0 +beta: 0 +transA: 0 +transB: 1 +
+
+LRN: 4 out of 4 attributes covered + +alpha: 2 +beta: 1 +bias: 2 +size: 1 +
+
+MaxPool: 3 out of 7 attributes covered + +auto_pad: 0 +ceil_mode: 0 +dilations: 0 +kernel_shape: 2 +pads: 3 +storage_order: 0 +strides: 2 +
+
+Transpose: 1 out of 1 attributes covered + +perm: 1 +
+
+Unsqueeze: 1 out of 0 attributes covered + +
+
+ + +# Overall Test Coverage +## To be filled. diff --git a/docs/TypeDenotation.md b/docs/TypeDenotation.md new file mode 100644 index 0000000..b3c8c3b --- /dev/null +++ b/docs/TypeDenotation.md @@ -0,0 +1,62 @@ + + +# Type Denotation + +Type Denotation is used to describe semantic information around what the inputs and outputs are. It is stored on the TypeProto message. + +## Motivation + +The motivation of such a mechanism can be illustrated via a simple example. In the neural network SqueezeNet, it takes in an NCHW image input float[1,3,244,244] and produces a output float[1,1000,1,1]: + +```text +input_in_NCHW -> data_0 -> SqueezeNet() -> output_softmaxout_1 +``` + +In order to run this model the user needs a lot of information. In this case the user needs to know: + +* the input is an image +* the image is in the format of NCHW +* the color channels are in the order of bgr +* the pixel data is 8 bit +* the pixel data is normalized as values 0-255 + +This proposal consists of three key components to provide all of this information: + +* Type Denotation, +* [Dimension Denotation](DimensionDenotation.md), +* [Model Metadata](MetadataProps.md). + +## Type Denotation Definition + +To begin with, we define a set of semantic types that define what models generally consume as inputs and produce as outputs. + +Specifically, in our first proposal we define the following set of standard denotations: + +0. `TENSOR` describes that a type holds a generic tensor using the standard TypeProto message. +1. `IMAGE` describes that a type holds an image. You can use dimension denotation to learn more about the layout of the image, and also the optional model metadata_props. +2. `AUDIO` describes that a type holds an audio clip. +3. `TEXT` describes that a type holds a block of text. + +Model authors SHOULD add type denotation to inputs and outputs for the model as appropriate. + +## An Example with input IMAGE + +Let's use the same SqueezeNet example from above and show everything to properly annotate the model: + +* First set the TypeProto.denotation =`IMAGE` for the ValueInfoProto `data_0` +* Because it's an image, the model consumer now knows to go look for image metadata on the model +* Then include 3 metadata strings on ModelProto.metadata_props + * `Image.BitmapPixelFormat` = `Bgr8` + * `Image.ColorSpaceGamma` = `SRGB` + * `Image.NominalPixelRange` = `NominalRange_0_255` +* For that same ValueInfoProto, make sure to also use Dimension Denotations to denote NCHW + * TensorShapeProto.Dimension[0].denotation = `DATA_BATCH` + * TensorShapeProto.Dimension[1].denotation = `DATA_CHANNEL` + * TensorShapeProto.Dimension[2].denotation = `DATA_FEATURE` + * TensorShapeProto.Dimension[3].denotation = `DATA_FEATURE` + +Now there is enough information in the model to know everything about how to pass a correct image into the model. diff --git a/docs/VersionConverter.md b/docs/VersionConverter.md new file mode 100644 index 0000000..2aa0701 --- /dev/null +++ b/docs/VersionConverter.md @@ -0,0 +1,55 @@ + + +# ONNX Version Converter + +ONNX provides a library for converting ONNX models between different +opset versions. The primary motivation is to improve backwards compatibility of ONNX +models without having to strengthen the spec for ONNX backends. This +allows backend developers to offer support for a particular opset version +and for users to write or export models to a particular opset version but +run in an environment with a different opset version. Implementation wise, the library leverages the in-memory representation that is much more convenient to manipulate than the raw protobuf structs, and converters to and from the protobuf format which were developed for the ONNX Optimizer. + +You may be interested in invoking the provided op-specific adapters, or in +implementing new ones (or both). Default adapters only work in the default +domain, but can be generalized to work cross-domain or utilizing new +conversion methods, dependent on the nature of relevant breaking changes. + +## Invoking The Version Converter + +The version converter may be invoked either via C++ or Python. + +The Python API +is described, with example, +[here](PythonAPIOverview.md#converting-version-of-an-onnx-model-within-default-domain-aionnx). + +The C++ API consists of a single function + +```cpp +ModelProto ConvertVersion( + const ModelProto& mp_in, + const OpSetID& initial_version, + const OpSetID& target_version); +``` + +which accepts an input `ModelProto`, the initial opset version of the model, +and the target opset version, and which returns a new `ModelProto` which +is the result of apply all relevant adapters between initial_version and +target_version. For a list of available passes, see +[convert.h](/onnx/version_converter/convert.h). + +## Implementing Adapters + +You can implement a new adapter by subclassing `Adapter`, and registering +your new adapter with `VersionConverter::registerAdapter()`. Adapters operate +on an in-memory graph representation defined in [ir.h](/onnx/common/ir.h). +There are a number of examples in the [adapters](/onnx/version_converter/adapters) +directory. Please ensure that all adapters convert from opset version i to i + 1 +or i - 1, i.e. from Version 6 to Version 5 or vice versa, even if the 2 versions +being converted between are Version 1 and Version 6. + +If your adapter applies in the default domain, please consider adding it +to the core ONNX repository diff --git a/docs/Versioning.md b/docs/Versioning.md new file mode 100644 index 0000000..5356a2e --- /dev/null +++ b/docs/Versioning.md @@ -0,0 +1,205 @@ + + +# ONNX Versioning + +This document describes the rules for versioning ONNX. MUST, SHOULD et al are used consistent with [RFC2119](https://tools.ietf.org/html/rfc2119). + +## Versioning Principles + +ONNX defines the versioning policy and mechanism for three classes of entities: + +* The [intermediate representation (IR) specification](IR.md), which is the abstract model for graphs and operators and the concrete format that represents them. These are always versioned atomically and are referred to as the *IR version*. +* Operator specifications that may be referenced by a given ONNX graph. We refer to this as the *operator version*. +* A defined/trained model that defines a specific graph in terms of specific operators. We refer to this as the *model version*. + +The versioning of all three of these entity types is distinct and largely independent. The IR specification evolves at a different (generally slower) rate than the operator specifications. Model versions are entirely independent of the other two versions. + +Specific policies for version management are mandated only for IR version and operator version. For model versioning, they are merely recommendations. For model versioning, ONNX users and systems MAY follow whichever local customs make sense; however, to facilitate easily managing shared collections of ONNX models, they SHOULD adhere to the policies described under model versioning. + +New IR and operator versions are released as part of ONNX _releases_, which have their own versioning scheme. The release versioning scheme is not described as part of the standard itself. It is discussed in the [ONNX release management document](../RELEASE-MANAGEMENT.md). + +### Semantic Versioning or Simple Numbers? + +The ONNX versioning system allows for simple monotonically increasing numbers or [semantic versioning (SemVer)](https://semver.org/). For IR and operator sets, versioning is based on simple numbers. For models, ONNX does not require any scheme, but recommends a set of shared conventions. + +Which versioning scheme is in use by a model is made clear by inspecting the most significant four bytes, which MUST be non-zero when using semantic versioning and MUST be zero when using simple numbers. In other words, when using SemVer, at least one of the MAJOR or MINOR numbers must be non-zero. + +### SemVer, Files and Consumers + +For model and release versioning, ONNX builds on the principles and syntax defined by [SemVer 2.0.0](http://semver.org/spec/v2.0.0.html). Throughout this document, we use the terms *breaking change*, *non-breaking change*, and *patch* consistent with SemVer 2.0.0. + +Because ONNX models are serialized files (not APIs), it's worth making clear the relationship between a serialized model and a piece of software that consumes that model. As a rough approximation, the serialized model plays the role of an API's *callee*, while the consumer of the serialized model plays the role of the API's *caller*. + +The ONNX versioning principles are based on the [robustness principle](https://en.wikipedia.org/wiki/Robustness_principle): "be conservative in what you do, be liberal in what you accept from others". + +1. A producer of a given ONNX model (and the ONNX specification itself) MUST strictly adhere to the rules for breaking vs. non-breaking changes defined in this specification. +2. A consumer of a given ONNX model SHOULD consume an updated ONNX file, provided there are no breaking changes in the new ONNX file's IR version, referenced operator versions, or model version (meaning the MAJOR version numbers have not changed between the two ONNX files). +3. A consumer of a given ONNX model MAY consume an updated ONNX file, provided there are one or more breaking changes in the new ONNX file's IR version, referenced operator versions, or model version. + +### Serializing SemVer version numbers in protobuf + +For efficiency, ONNX serializes the MAJOR, MINOR, and PATCH values as a bit-packed 64-bit integer; the two most significant bytes are the MAJOR component, the next two most significant bytes are the MINOR component, and the least significant four bytes are the PATCH component. + +For example, *1.2.345* is represented as *0x0001000200000159*. + +Pre-release and build metadata are not stored in the model. + +## IR versioning + +The IR format is versioned using simple numbers, which MUST be monotonically increasing. Breaking changes to the format or semantics of the ONNX specification require an increment of the version. Non-breaking changes to the IR format do not require changing the version number. + +NOTE: breaking changes include those that do not alter the serialized binary format, but still break software using libraries that write or read it. For example, changing the spelling of a message property will cause code accessing the property break. + +The IR format adheres to the versioning guidelines defined in the [Updating a Message Type](https://developers.google.com/protocol-buffers/docs/proto3#updating) section of the proto3 specification. + +As a general principle, implementations SHOULD be robust in the face of missing fields. However, to ensure basic interoperation, a subset of message fields will be marked as required for a given IR version and all producers MUST set these fields correctly. Required fields MUST always be marked with the comment: + + // This field MUST be present for this version of the IR. + +For example, the `ModelProto.ir_version` property MUST be present in every model. The ONNX checker (`onnx/checker.py`) will enforce these rules. + +Because the protocol buffer message definitions (.proto / .proto3 files) are expected to be consumed by multiple independent developers, changes to those definitions SHOULD NOT break code that depends on generated language bindings (e.g., changing the type of an existing field). + +## Operator versioning + +The IR can evolve independently from the set of operators. Operators represent both the signature and semantics of a given operation. Operators are abstract interfaces in that they do not imply a specific implementation; rather, they are simply the contract between a model author and the implementations that model may execute on. + +A given operator is identified by a three-tuple: `(domain, op_type, since_version)`, written as `domain.op_type:since_version` in prose (e.g., `com.acme.FastConv:3`). `since_version` is the version of the operator set that introduced the operator. Breaking operator changes include: + +* Adding/removing/renaming an attribute. This even includes the case of adding a new optional attribute, where omitting the attribute would imply a default value yielding semantics identical to the previous operator version. + +* Adding/removing/reordering inputs or outputs. + +* Adding/removing types supported by inputs and outputs, and changing types used by attributes. + +* Supporting new behavior even when the existing parameter signature is otherwise identical (e.g. implicitly supporting tensor broadcasting in the Mean operator). + +The following are not breaking: + +* Clarifications of specification ambiguities to match prevailing + implementation practice. + +Changes to the semantics of an operator or function MUST be introduced in a new operator, which MUST be introduced in a new [operator set](#operator-sets). + +> In practice, this means that BC-breaking changes in the ONNX +> repository require contributors to follow these steps: +> +> 1. Increment the maximum version in `DomainToVersionRange`. +> 2. Copy the old operator schema to an `old.cc` file. +> 3. Update the `SinceVersion` signifier to the new max version from +> step (1). +> 4. Register the new operator in the corresponding `operator_sets` +> header file. +> 5. Add a version adapter to `convert.h` so that the version +> converter can upgrade the old version of the operator to the new +> one. This can be a `CompatibleAdapter` in case operators following +> the old schema are still valid under the new one (which is usually +> true). +> 6. A version adapter to downgrade the new operator to the older version +> can also be added to `convert.h` but it's not mandatory. + +How nodes bind to operator declarations is strictly defined, and are designed to increase model compatibility across ONNX implementations, in the spirit of the conservative clause of the robustness principle. + +How ONNX implementations bind an operator declaration to a specific implementation is outside the scope of this specification. Implementations of ONNX MAY elect to introduce more sophisticated operator declaration/implementation binding modes, in the spirit of the liberal clause of the robustness principle. + +### Operator sets + +ONNX uses operator sets to group together immutable operator specifications. An operator set represents a specific version of a domain, indicated by a pair (domain, version). This represents the set of all operators belonging to the specified domain with the specified version (referred to as the `opset_version`). When the inventory of a given operator set changes either by addition, removal, or a change in semantics of a contained operator, its version MUST increase. + +Models declare which operator sets they require as a list of `(domain, opset_version)` pairs in `ModelProto.opset_import`. The empty string ("") domain indicates the operators defined as part of the ONNX specification; other domains correspond to operator sets of other vendors (meaning they can be used to provide vendor-specific extensions to ONNX). The union of the operator sets specified by a given model MUST have a compatible operator declaration for each node in the model's graph. + +### Example + +This section is not normative and informational only. + +Given the following operator sets: + +OpSet|Operators|Comments +-|-|- +1|{A} | A introduced +2|{A, B} | B introduced +3|{A', B, C} | A updated (to A'), C introduced +4|{B, C'} | A removed, C updated (to C') + +The operators for a given operator set will have the following `since_version` values: + +Operator|OpSet 1|OpSet 2|OpSet 3|OpSet 4 +-|-|-|-|- +A|**1** |1 |**3** |**-** +B|- |**2** |2 |2 +C|- |- |**3** |**4** + +Notes: + +- Values that are new or updated from a previous OpSet version are in **bold**. + +## Model versioning + +This section of the specification is not normative. It simply outlines a set of recommended practices. + +Model authors and applications/systems MAY elect to ignore the model versioning mechanism and policy rules. For models that will be shared across developers, teams, or organizations, model authors and applications/systems SHOULD adhere to the following version policies: + +### Signature Changes + +1. Breaking changes to the ModelProto.graph.GraphProto.input or .output MUST increment the MAJOR version of `ModelProto.model_version`. Breaking changes include: + + * Breaking changes to the semantics of an input or output (e.g., changing the required contents of an input tensor from a color image to a black and white image). + * Changing the declared type of an input or output to an incompatible type (e.g., `tensor(int)->tensor(string)`). + * Adding a new input for which there is no meaningful or specified default value. Recall that default values for inputs are specified in the initializer list. + * Removing an existing output for which there is no meaningful or specified default value. + +2. Non-breaking changes to the ModelProto.graph.GraphProto.input or .output MUST increment the MINOR version of `ModelProto.model_version`. Non-breaking changes include: + + * Changing the declared type of an input or output to a compatible/widening type (e.g., `tensor(int32)->tensor(int64)`, `tensor(float16)->tensor(float32)`). + * Adding a new input for which there is a meaningful or specified default value. + * Adding new behavior that is only triggered in the presence of inputs that were not + possible in prior versions of the graph (typically by the presence of a new input + or allowing a previously invalid input value). + +### Accuracy or performance changes + +Changes that impact accuracy or performance significantly but do not change the model's inputs or outputs SHOULD increment the PATCH version of `ModelProto.model_version`. + +## Released Versions + +ONNX version|IR version|Opset version ai.onnx|Opset version ai.onnx.ml|Opset version ai.onnx.training +------------|-------------------|---------------------|------------------------|------------------------------ +1.0|3|1|1|- +1.1|3|5|1|- +1.1.2|3|6|1|- +1.2|3|7|1|- +1.3|3|8|1|- +1.4.1|4|9|1|- +1.5.0|5|10|1|- +1.6.0|6|11|2|- +1.7.0|7|12|2|1 +1.8.0|7|13|2|1 +1.8.1|7|13|2|1 +1.9.0|7|14|2|1 +1.10.0|8|15|2|1 +1.10.1|8|15|2|1 +1.10.2|8|15|2|1 +1.11.0|8|16|3|1 +1.12.0|8|17|3|1 +1.13.0|8|18|3|1 +1.13.1|8|18|3|1 +1.14.0|9|19|3|1 +1.14.1|9|19|3|1 +1.15.0|9|20|4|1 +1.16.0|10|21|5|1 +1.16.1|10|21|5|1 +1.16.2|10|21|5|1 +1.17.0|10|22|5|1 +1.18.0|11|23|5|1 +1.19.0|12|24|5|1 +1.19.1|12|24|5|1 +1.20.0|13|25|5|1 +1.20.1|13|25|5|1 +1.21.0|13|26|5|1 + +The version number is centrally defined in [VERSION_NUMBER](../VERSION_NUMBER). +Programmatically accessible version information is available from [onnx/helper.py](../onnx/helper.py), `onnx/common/version.h` (auto-generated) and [onnx/defs/schema.h](../onnx/defs/schema.h). diff --git a/docs/docsgen/Makefile b/docs/docsgen/Makefile new file mode 100644 index 0000000..c7fb98f --- /dev/null +++ b/docs/docsgen/Makefile @@ -0,0 +1,24 @@ +# Copyright (c) Sphinx Project +# +# SPDX-License-Identifier: BSD-2-Clause + +# Minimal makefile for Sphinx documentation +# + +# You can set these variables from the command line, and also +# from the environment for the first two. +SPHINXOPTS ?= +SPHINXBUILD ?= sphinx-build +SOURCEDIR = source +BUILDDIR = build + +# Put it first so that "make" without argument is like "make help". +help: + @$(SPHINXBUILD) -M help "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O) --keep-going + +.PHONY: help Makefile + +# Catch-all target: route all unknown targets to Sphinx using the new +# "make mode" option. $(O) is meant as a shortcut for $(SPHINXOPTS). +%: Makefile + @$(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O) --keep-going diff --git a/docs/docsgen/make.bat b/docs/docsgen/make.bat new file mode 100644 index 0000000..ed82127 --- /dev/null +++ b/docs/docsgen/make.bat @@ -0,0 +1,39 @@ +REM Copyright (c) Sphinx Project +REM +REM SPDX-License-Identifier: BSD-2-Clause + +@ECHO OFF + +pushd %~dp0 + +REM Command file for Sphinx documentation + +if "%SPHINXBUILD%" == "" ( + set SPHINXBUILD=sphinx-build +) +set SOURCEDIR=source +set BUILDDIR=build + +%SPHINXBUILD% >NUL 2>NUL +if errorlevel 9009 ( + echo. + echo.The 'sphinx-build' command was not found. Make sure you have Sphinx + echo.installed, then set the SPHINXBUILD environment variable to point + echo.to the full path of the 'sphinx-build' executable. Alternatively you + echo.may add the Sphinx directory to PATH. + echo. + echo.If you don't have Sphinx installed, grab it from + echo.https://www.sphinx-doc.org/ + exit /b 1 +) + +if "%1" == "" goto help + +%SPHINXBUILD% -M %1 %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% %O% +goto end + +:help +%SPHINXBUILD% -M help %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% %O% + +:end +popd diff --git a/docs/docsgen/source/_static/ONNX-Logo.svg b/docs/docsgen/source/_static/ONNX-Logo.svg new file mode 100644 index 0000000..d2d3adf --- /dev/null +++ b/docs/docsgen/source/_static/ONNX-Logo.svg @@ -0,0 +1,83 @@ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + diff --git a/docs/docsgen/source/_static/css/custom.css b/docs/docsgen/source/_static/css/custom.css new file mode 100644 index 0000000..8db8be6 --- /dev/null +++ b/docs/docsgen/source/_static/css/custom.css @@ -0,0 +1,36 @@ +@import "../basic.css"; + +html[data-theme="dark"] { + --pst-color-text-base: #5c8fda; +} + +.d2h-del { + background-color: var(--pst-color-danger-highlight); + color: var(--pst-color-text-base); +} + +.d2h-ins { + background-color: var(--pst-color-success); + color: var(--pst-color-text-base); +} + +.d2h-change { + background-color: var(--yellow); + color: var(--pst-color-text-base); +} + +.sphinx-tabs-panel { + background-color: var(--pst-color-background); +} + +.sphinx-tabs-tab { + background-color: var(--pst-color-background); +} + +.sphinx-tabs { + background-color: var(--pst-color-background); +} + +.closeable{ + background-color: var(--pst-color-background); +} diff --git a/docs/docsgen/source/_static/diff2html-ui-slim.min.js b/docs/docsgen/source/_static/diff2html-ui-slim.min.js new file mode 100644 index 0000000..da87c3e --- /dev/null +++ b/docs/docsgen/source/_static/diff2html-ui-slim.min.js @@ -0,0 +1 @@ +!function(e,n){if("object"==typeof exports&&"object"==typeof module)module.exports=n();else if("function"==typeof define&&define.amd)define([],n);else{var t=n();for(var i in t)("object"==typeof exports?exports:e)[i]=t[i]}}(this,(()=>(()=>{var e={696:(e,n)=>{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.convertChangesToDMP=function(e){for(var n,t,i=[],a=0;a{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.convertChangesToXML=function(e){for(var n=[],t=0;t"):i.removed&&n.push(""),n.push((a=i.value,void 0,a.replace(/&/g,"&").replace(//g,">").replace(/"/g,"""))),i.added?n.push(""):i.removed&&n.push("")}var a;return n.join("")}},6976:(e,n,t)=>{"use strict";var i;Object.defineProperty(n,"__esModule",{value:!0}),n.diffArrays=function(e,n,t){return a.diff(e,n,t)},n.arrayDiff=void 0;var a=new(((i=t(5913))&&i.__esModule?i:{default:i}).default);n.arrayDiff=a,a.tokenize=function(e){return e.slice()},a.join=a.removeEmpty=function(e){return e}},5913:(e,n)=>{"use strict";function t(){}function i(e,n,t,i,a){for(var r=0,s=n.length,o=0,l=0;re.length?t:e})),c.value=e.join(u)}else c.value=e.join(t.slice(o,o+c.count));o+=c.count,c.added||(l+=c.count)}}var g=n[s-1];return s>1&&"string"==typeof g.value&&(g.added||g.removed)&&e.equals("",g.value)&&(n[s-2].value+=g.value,n.pop()),n}Object.defineProperty(n,"__esModule",{value:!0}),n.default=t,t.prototype={diff:function(e,n){var t=arguments.length>2&&void 0!==arguments[2]?arguments[2]:{},a=t.callback;"function"==typeof t&&(a=t,t={}),this.options=t;var r=this;function s(e){return a?(setTimeout((function(){a(void 0,e)}),0),!0):e}e=this.castInput(e),n=this.castInput(n),e=this.removeEmpty(this.tokenize(e));var o=(n=this.removeEmpty(this.tokenize(n))).length,l=e.length,c=1,d=o+l;t.maxEditLength&&(d=Math.min(d,t.maxEditLength));var u=[{newPos:-1,components:[]}],g=this.extractCommon(u[0],n,e,0);if(u[0].newPos+1>=o&&g+1>=l)return s([{value:this.join(n),count:n.length}]);function b(){for(var t=-1*c;t<=c;t+=2){var a=void 0,d=u[t-1],g=u[t+1],b=(g?g.newPos:0)-t;d&&(u[t-1]=void 0);var p=d&&d.newPos+1=o&&b+1>=l)return s(i(r,a.components,n,e,r.useLongestToken));u[t]=a}else u[t]=void 0}var m;c++}if(a)!function e(){setTimeout((function(){if(c>d)return a();b()||e()}),0)}();else for(;c<=d;){var p=b();if(p)return p}},pushComponent:function(e,n,t){var i=e[e.length-1];i&&i.added===n&&i.removed===t?e[e.length-1]={count:i.count+1,added:n,removed:t}:e.push({count:1,added:n,removed:t})},extractCommon:function(e,n,t,i){for(var a=n.length,r=t.length,s=e.newPos,o=s-i,l=0;s+1{"use strict";var i;Object.defineProperty(n,"__esModule",{value:!0}),n.diffChars=function(e,n,t){return a.diff(e,n,t)},n.characterDiff=void 0;var a=new(((i=t(5913))&&i.__esModule?i:{default:i}).default);n.characterDiff=a},4852:(e,n,t)=>{"use strict";var i;Object.defineProperty(n,"__esModule",{value:!0}),n.diffCss=function(e,n,t){return a.diff(e,n,t)},n.cssDiff=void 0;var a=new(((i=t(5913))&&i.__esModule?i:{default:i}).default);n.cssDiff=a,a.tokenize=function(e){return e.split(/([{}:;,]|\s+)/)}},4276:(e,n,t)=>{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.diffJson=function(e,n,t){return l.diff(e,n,t)},n.canonicalize=c,n.jsonDiff=void 0;var i,a=(i=t(5913))&&i.__esModule?i:{default:i},r=t(8187);function s(e){return s="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},s(e)}var o=Object.prototype.toString,l=new a.default;function c(e,n,t,i,a){var r,l;for(n=n||[],t=t||[],i&&(e=i(a,e)),r=0;r{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.diffLines=function(e,n,t){return s.diff(e,n,t)},n.diffTrimmedLines=function(e,n,t){var i=(0,r.generateOptions)(t,{ignoreWhitespace:!0});return s.diff(e,n,i)},n.lineDiff=void 0;var i,a=(i=t(5913))&&i.__esModule?i:{default:i},r=t(8009),s=new a.default;n.lineDiff=s,s.tokenize=function(e){var n=[],t=e.split(/(\n|\r\n)/);t[t.length-1]||t.pop();for(var i=0;i{"use strict";var i;Object.defineProperty(n,"__esModule",{value:!0}),n.diffSentences=function(e,n,t){return a.diff(e,n,t)},n.sentenceDiff=void 0;var a=new(((i=t(5913))&&i.__esModule?i:{default:i}).default);n.sentenceDiff=a,a.tokenize=function(e){return e.split(/(\S.+?[.!?])(?=\s+|$)/)}},5303:(e,n,t)=>{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.diffWords=function(e,n,t){return t=(0,r.generateOptions)(t,{ignoreWhitespace:!0}),l.diff(e,n,t)},n.diffWordsWithSpace=function(e,n,t){return l.diff(e,n,t)},n.wordDiff=void 0;var i,a=(i=t(5913))&&i.__esModule?i:{default:i},r=t(8009),s=/^[A-Za-z\xC0-\u02C6\u02C8-\u02D7\u02DE-\u02FF\u1E00-\u1EFF]+$/,o=/\S/,l=new a.default;n.wordDiff=l,l.equals=function(e,n){return this.options.ignoreCase&&(e=e.toLowerCase(),n=n.toLowerCase()),e===n||this.options.ignoreWhitespace&&!o.test(e)&&!o.test(n)},l.tokenize=function(e){for(var n=e.split(/([^\S\r\n]+|[()[\]{}'"\r\n]|\b)/),t=0;t{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),Object.defineProperty(n,"Diff",{enumerable:!0,get:function(){return a.default}}),Object.defineProperty(n,"diffChars",{enumerable:!0,get:function(){return r.diffChars}}),Object.defineProperty(n,"diffWords",{enumerable:!0,get:function(){return s.diffWords}}),Object.defineProperty(n,"diffWordsWithSpace",{enumerable:!0,get:function(){return s.diffWordsWithSpace}}),Object.defineProperty(n,"diffLines",{enumerable:!0,get:function(){return o.diffLines}}),Object.defineProperty(n,"diffTrimmedLines",{enumerable:!0,get:function(){return o.diffTrimmedLines}}),Object.defineProperty(n,"diffSentences",{enumerable:!0,get:function(){return l.diffSentences}}),Object.defineProperty(n,"diffCss",{enumerable:!0,get:function(){return c.diffCss}}),Object.defineProperty(n,"diffJson",{enumerable:!0,get:function(){return d.diffJson}}),Object.defineProperty(n,"canonicalize",{enumerable:!0,get:function(){return d.canonicalize}}),Object.defineProperty(n,"diffArrays",{enumerable:!0,get:function(){return u.diffArrays}}),Object.defineProperty(n,"applyPatch",{enumerable:!0,get:function(){return g.applyPatch}}),Object.defineProperty(n,"applyPatches",{enumerable:!0,get:function(){return g.applyPatches}}),Object.defineProperty(n,"parsePatch",{enumerable:!0,get:function(){return b.parsePatch}}),Object.defineProperty(n,"merge",{enumerable:!0,get:function(){return p.merge}}),Object.defineProperty(n,"structuredPatch",{enumerable:!0,get:function(){return f.structuredPatch}}),Object.defineProperty(n,"createTwoFilesPatch",{enumerable:!0,get:function(){return f.createTwoFilesPatch}}),Object.defineProperty(n,"createPatch",{enumerable:!0,get:function(){return f.createPatch}}),Object.defineProperty(n,"convertChangesToDMP",{enumerable:!0,get:function(){return m.convertChangesToDMP}}),Object.defineProperty(n,"convertChangesToXML",{enumerable:!0,get:function(){return E.convertChangesToXML}});var i,a=(i=t(5913))&&i.__esModule?i:{default:i},r=t(7630),s=t(5303),o=t(8187),l=t(4146),c=t(4852),d=t(4276),u=t(6976),g=t(3690),b=t(3719),p=t(3051),f=t(1286),m=t(696),E=t(5826)},3690:(e,n,t)=>{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.applyPatch=s,n.applyPatches=function(e,n){"string"==typeof e&&(e=(0,a.parsePatch)(e));var t=0;!function i(){var a=e[t++];if(!a)return n.complete();n.loadFile(a,(function(e,t){if(e)return n.complete(e);var r=s(t,a,n);n.patched(a,r,(function(e){if(e)return n.complete(e);i()}))}))}()};var i,a=t(3719),r=(i=t(8169))&&i.__esModule?i:{default:i};function s(e,n){var t=arguments.length>2&&void 0!==arguments[2]?arguments[2]:{};if("string"==typeof n&&(n=(0,a.parsePatch)(n)),Array.isArray(n)){if(n.length>1)throw new Error("applyPatch only works with a single input.");n=n[0]}var i,s,o=e.split(/\r\n|[\n\v\f\r\x85]/),l=e.match(/\r\n|[\n\v\f\r\x85]/g)||[],c=n.hunks,d=t.compareLine||function(e,n,t,i){return n===i},u=0,g=t.fuzzFactor||0,b=0,p=0;function f(e,n){for(var t=0;t0?i[0]:" ",r=i.length>0?i.substr(1):i;if(" "===a||"-"===a){if(!d(n+1,o[n],a,r)&&++u>g)return!1;n++}}return!0}for(var m=0;m0?R[0]:" ",w=R.length>0?R.substr(1):R,C=v.linedelimiters[S];if(" "===I)y++;else if("-"===I)o.splice(y,1),l.splice(y,1);else if("+"===I)o.splice(y,0,w),l.splice(y,0,C),y++;else if("\\"===I){var L=v.lines[S-1]?v.lines[S-1][0]:null;"+"===L?i=!0:"-"===L&&(s=!0)}}}if(i)for(;!o[o.length-1];)o.pop(),l.pop();else s&&(o.push(""),l.push("\n"));for(var M=0;M{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.structuredPatch=s,n.formatPatch=o,n.createTwoFilesPatch=l,n.createPatch=function(e,n,t,i,a,r){return l(e,e,n,t,i,a,r)};var i=t(8187);function a(e){return function(e){if(Array.isArray(e))return r(e)}(e)||function(e){if("undefined"!=typeof Symbol&&Symbol.iterator in Object(e))return Array.from(e)}(e)||function(e,n){if(e){if("string"==typeof e)return r(e,n);var t=Object.prototype.toString.call(e).slice(8,-1);return"Object"===t&&e.constructor&&(t=e.constructor.name),"Map"===t||"Set"===t?Array.from(e):"Arguments"===t||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(t)?r(e,n):void 0}}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function r(e,n){(null==n||n>e.length)&&(n=e.length);for(var t=0,i=new Array(n);t0?h(o.lines.slice(-l.context)):[],u-=b.length,g-=b.length)}(s=b).push.apply(s,a(i.map((function(e){return(n.added?"+":"-")+e})))),n.added?f+=i.length:p+=i.length}else{if(u)if(i.length<=2*l.context&&e=c.length-2&&i.length<=l.context){var T=/\n$/.test(t),O=/\n$/.test(r),A=0==i.length&&b.length>N.oldLines;!T&&A&&t.length>0&&b.splice(N.oldLines,0,"\\ No newline at end of file"),(T||A)&&O||b.push("\\ No newline at end of file")}d.push(N),u=0,g=0,b=[]}p+=i.length,f+=i.length}},E=0;E{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.calcLineCount=l,n.merge=function(e,n,t){e=c(e,t),n=c(n,t);var i={};(e.index||n.index)&&(i.index=e.index||n.index),(e.newFileName||n.newFileName)&&(d(e)?d(n)?(i.oldFileName=u(i,e.oldFileName,n.oldFileName),i.newFileName=u(i,e.newFileName,n.newFileName),i.oldHeader=u(i,e.oldHeader,n.oldHeader),i.newHeader=u(i,e.newHeader,n.newHeader)):(i.oldFileName=e.oldFileName,i.newFileName=e.newFileName,i.oldHeader=e.oldHeader,i.newHeader=e.newHeader):(i.oldFileName=n.oldFileName||e.oldFileName,i.newFileName=n.newFileName||e.newFileName,i.oldHeader=n.oldHeader||e.oldHeader,i.newHeader=n.newHeader||e.newHeader)),i.hunks=[];for(var a=0,r=0,s=0,o=0;ae.length)&&(n=e.length);for(var t=0,i=new Array(n);t{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.parsePatch=function(e){var n=arguments.length>1&&void 0!==arguments[1]?arguments[1]:{},t=e.split(/\r\n|[\n\v\f\r\x85]/),i=e.match(/\r\n|[\n\v\f\r\x85]/g)||[],a=[],r=0;function s(){var e={};for(a.push(e);r{"use strict";function t(e,n){if(n.length>e.length)return!1;for(var t=0;t{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.default=function(e,n,t){var i=!0,a=!1,r=!1,s=1;return function o(){if(i&&!r){if(a?s++:i=!1,e+s<=t)return s;r=!0}if(!a)return r||(i=!0),n<=e-s?-s++:(a=!0,o())}}},8009:(e,n)=>{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.generateOptions=function(e,n){if("function"==typeof e)n.callback=e;else if(e)for(var t in e)e.hasOwnProperty(t)&&(n[t]=e[t]);return n}},9397:(e,n)=>{!function(e){var n=/\S/,t=/\"/g,i=/\n/g,a=/\r/g,r=/\\/g,s=/\u2028/,o=/\u2029/;function l(e){return e.trim?e.trim():e.replace(/^\s*|\s*$/g,"")}function c(e,n,t){if(n.charAt(t)!=e.charAt(0))return!1;for(var i=1,a=e.length;i":7,"=":8,_v:9,"{":10,"&":11,_t:12},e.scan=function(t,i){var a,r=t.length,s=0,o=null,d=null,u="",g=[],b=!1,p=0,f=0,m="{{",E="}}";function h(){u.length>0&&(g.push({tag:"_t",text:new String(u)}),u="")}function _(t,i){if(h(),t&&function(){for(var t=!0,i=f;i"==a.tag&&(a.indent=g[r].text.toString()),g.splice(r,1));else i||g.push({tag:"\n"});b=!1,f=g.length}function N(e,n){var t="="+E,i=e.indexOf(t,n),a=l(e.substring(e.indexOf("=",n)+1,i)).split(" ");return m=a[0],E=a[a.length-1],i+t.length-1}for(i&&(i=i.split(" "),m=i[0],E=i[1]),p=0;p0;){if(l=n.shift(),r&&"<"==r.tag&&!(l.tag in d))throw new Error("Illegal content in < super tag.");if(e.tags[l.tag]<=e.tags.$||g(l,a))i.push(l),l.nodes=u(n,l.tag,i,a);else{if("/"==l.tag){if(0===i.length)throw new Error("Closing tag without opener: /"+l.n);if(o=i.pop(),l.n!=o.n&&!b(l.n,o.n,a))throw new Error("Nesting error: "+o.n+" vs. "+l.n);return o.end=l.i,s}"\n"==l.tag&&(l.last=0==n.length||"\n"==n[0].tag)}s.push(l)}if(i.length>0)throw new Error("missing closing tag: "+i.pop().n);return s}function g(e,n){for(var t=0,i=n.length;t":h,"<":function(n,t){var i={partials:{},code:"",subs:{},inPartial:!0};e.walk(n.nodes,i);var a=t.partials[h(n,t)];a.subs=i.subs,a.partials=i.partials},$:function(n,t){var i={subs:{},code:"",partials:t.partials,prefix:n.n};e.walk(n.nodes,i),t.subs[n.n]=i.code,t.inPartial||(t.code+='t.sub("'+m(n.n)+'",c,p,i);')},"\n":function(e,n){n.code+=N('"\\n"'+(e.last?"":" + i"))},_v:function(e,n){n.code+="t.b(t.v(t."+E(e.n)+'("'+m(e.n)+'",c,p,0)));'},_t:function(e,n){n.code+=N('"'+m(e.text)+'"')},"{":_,"&":_},e.walk=function(n,t){for(var i,a=0,r=n.length;a{var i=t(9397);i.Template=t(2882).Template,i.template=i.Template,e.exports=i},2882:(e,n)=>{!function(e){function n(e,n,t){var i;return n&&"object"==typeof n&&(void 0!==n[e]?i=n[e]:t&&n.get&&"function"==typeof n.get&&(i=n.get(e))),i}e.Template=function(e,n,t,i){e=e||{},this.r=e.code||this.r,this.c=t,this.options=i||{},this.text=n||"",this.partials=e.partials||{},this.subs=e.subs||{},this.buf=""},e.Template.prototype={r:function(e,n,t){return""},v:function(e){return e=l(e),o.test(e)?e.replace(t,"&").replace(i,"<").replace(a,">").replace(r,"'").replace(s,"""):e},t:l,render:function(e,n,t){return this.ri([e],n||{},t)},ri:function(e,n,t){return this.r(e,n,t)},ep:function(e,n){var t=this.partials[e],i=n[t.name];if(t.instance&&t.base==i)return t.instance;if("string"==typeof i){if(!this.c)throw new Error("No compiler available.");i=this.c.compile(i,this.options)}if(!i)return null;if(this.partials[e].base=i,t.subs){for(key in n.stackText||(n.stackText={}),t.subs)n.stackText[key]||(n.stackText[key]=void 0!==this.activeSub&&n.stackText[this.activeSub]?n.stackText[this.activeSub]:this.text);i=function(e,n,t,i,a,r){function s(){}function o(){}var l;s.prototype=e,o.prototype=e.subs;var c=new s;for(l in c.subs=new o,c.subsText={},c.buf="",i=i||{},c.stackSubs=i,c.subsText=r,n)i[l]||(i[l]=n[l]);for(l in i)c.subs[l]=i[l];for(l in a=a||{},c.stackPartials=a,t)a[l]||(a[l]=t[l]);for(l in a)c.partials[l]=a[l];return c}(i,t.subs,t.partials,this.stackSubs,this.stackPartials,n.stackText)}return this.partials[e].instance=i,i},rp:function(e,n,t,i){var a=this.ep(e,t);return a?a.ri(n,t,i):""},rs:function(e,n,t){var i=e[e.length-1];if(c(i))for(var a=0;a=0;l--)if(void 0!==(r=n(e,t[l],o))){s=!0;break}return s?(a||"function"!=typeof r||(r=this.mv(r,t,i)),r):!a&&""},ls:function(e,n,t,i,a){var r=this.options.delimiters;return this.options.delimiters=a,this.b(this.ct(l(e.call(n,i)),n,t)),this.options.delimiters=r,!1},ct:function(e,n,t){if(this.options.disableLambda)throw new Error("Lambda features disabled.");return this.c.compile(e,this.options).render(n,t)},b:function(e){this.buf+=e},fl:function(){var e=this.buf;return this.buf="",e},ms:function(e,n,t,i,a,r,s){var o,l=n[n.length-1],c=e.call(l);return"function"==typeof c?!!i||(o=this.activeSub&&this.subsText&&this.subsText[this.activeSub]?this.subsText[this.activeSub]:this.text,this.ls(c,l,t,o.substring(a,r),s)):c},mv:function(e,n,t){var i=n[n.length-1],a=e.call(i);return"function"==typeof a?this.ct(l(a.call(i)),i,t):a},sub:function(e,n,t,i){var a=this.subs[e];a&&(this.activeSub=e,a(n,t,this,i),this.activeSub=!1)}};var t=/&/g,i=//g,r=/\'/g,s=/\"/g,o=/[&<>\"\']/;function l(e){return String(null==e?"":e)}var c=Array.isArray||function(e){return"[object Array]"===Object.prototype.toString.call(e)}}(n)},8468:(e,n,t)=>{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.parse=void 0;const i=t(9699),a=t(8593);function r(e,n){const t=e.split(".");return t.length>1?t[t.length-1]:n}function s(e,n){return n.reduce(((n,t)=>n||e.startsWith(t)),!1)}const o=["a/","b/","i/","w/","c/","o/"];function l(e,n,t){const i=void 0!==t?[...o,t]:o,r=n?new RegExp(`^${(0,a.escapeForRegExp)(n)} "?(.+?)"?$`):new RegExp('^"?(.+?)"?$'),[,s=""]=r.exec(e)||[],l=i.find((e=>0===s.indexOf(e)));return(l?s.slice(l.length):s).replace(/\s+\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}(?:\.\d+)? [+-]\d{4}.*$/,"")}n.parse=function(e,n={}){const t=[];let a=null,o=null,c=null,d=null,u=null,g=null,b=null;const p="--- ",f="+++ ",m="@@",E=/^old mode (\d{6})/,h=/^new mode (\d{6})/,_=/^deleted file mode (\d{6})/,N=/^new file mode (\d{6})/,T=/^copy from "?(.+)"?/,O=/^copy to "?(.+)"?/,A=/^rename from "?(.+)"?/,v=/^rename to "?(.+)"?/,y=/^similarity index (\d+)%/,S=/^dissimilarity index (\d+)%/,R=/^index ([\da-z]+)\.\.([\da-z]+)\s*(\d{6})?/,I=/^Binary files (.*) and (.*) differ/,w=/^GIT binary patch/,C=/^index ([\da-z]+),([\da-z]+)\.\.([\da-z]+)/,L=/^mode (\d{6}),(\d{6})\.\.(\d{6})/,M=/^new file mode (\d{6})/,D=/^deleted file mode (\d{6}),(\d{6})/,x=e.replace(/\\ No newline at end of file/g,"").replace(/\r\n?/g,"\n").split("\n");function k(){null!==o&&null!==a&&(a.blocks.push(o),o=null)}function P(){null!==a&&(a.oldName||null===g||(a.oldName=g),a.newName||null===b||(a.newName=b),a.newName&&(t.push(a),a=null)),g=null,b=null}function U(){k(),P(),a={blocks:[],deletedLines:0,addedLines:0}}function B(e){let n;k(),null!==a&&((n=/^@@ -(\d+)(?:,\d+)? \+(\d+)(?:,\d+)? @@.*/.exec(e))?(a.isCombined=!1,c=parseInt(n[1],10),u=parseInt(n[2],10)):(n=/^@@@ -(\d+)(?:,\d+)? -(\d+)(?:,\d+)? \+(\d+)(?:,\d+)? @@@.*/.exec(e))?(a.isCombined=!0,c=parseInt(n[1],10),d=parseInt(n[2],10),u=parseInt(n[3],10)):(e.startsWith(m)&&console.error("Failed to parse lines, starting in 0!"),c=0,u=0,a.isCombined=!1)),o={lines:[],oldStartLine:c,oldStartLine2:d,newStartLine:u,header:e}}return x.forEach(((e,d)=>{if(!e||e.startsWith("*"))return;let k;const P=x[d-1],F=x[d+1],j=x[d+2];if(e.startsWith("diff --git")||e.startsWith("diff --combined")){if(U(),(k=/^diff --git "?([a-ciow]\/.+)"? "?([a-ciow]\/.+)"?/.exec(e))&&(g=l(k[1],void 0,n.dstPrefix),b=l(k[2],void 0,n.srcPrefix)),null===a)throw new Error("Where is my file !!!");return void(a.isGitDiff=!0)}if(e.startsWith("Binary files")&&!(null==a?void 0:a.isGitDiff)){if(U(),(k=/^Binary files "?([a-ciow]\/.+)"? and "?([a-ciow]\/.+)"? differ/.exec(e))&&(g=l(k[1],void 0,n.dstPrefix),b=l(k[2],void 0,n.srcPrefix)),null===a)throw new Error("Where is my file !!!");return void(a.isBinary=!0)}if((!a||!a.isGitDiff&&a&&e.startsWith(p)&&F.startsWith(f)&&j.startsWith(m))&&U(),null==a?void 0:a.isTooBig)return;if(a&&("number"==typeof n.diffMaxChanges&&a.addedLines+a.deletedLines>n.diffMaxChanges||"number"==typeof n.diffMaxLineLength&&e.length>n.diffMaxLineLength))return a.isTooBig=!0,a.addedLines=0,a.deletedLines=0,a.blocks=[],o=null,void B("function"==typeof n.diffTooBigMessage?n.diffTooBigMessage(t.length):"Diff too big to be displayed");if(e.startsWith(p)&&F.startsWith(f)||e.startsWith(f)&&P.startsWith(p)){if(a&&!a.oldName&&e.startsWith("--- ")&&(k=function(e,n){return l(e,"---",n)}(e,n.srcPrefix)))return a.oldName=k,void(a.language=r(a.oldName,a.language));if(a&&!a.newName&&e.startsWith("+++ ")&&(k=function(e,n){return l(e,"+++",n)}(e,n.dstPrefix)))return a.newName=k,void(a.language=r(a.newName,a.language))}if(a&&(e.startsWith(m)||a.isGitDiff&&a.oldName&&a.newName&&!o))return void B(e);if(o&&(e.startsWith("+")||e.startsWith("-")||e.startsWith(" ")))return void function(e){if(null===a||null===o||null===c||null===u)return;const n={content:e},t=a.isCombined?["+ "," +","++"]:["+"],r=a.isCombined?["- "," -","--"]:["-"];s(e,t)?(a.addedLines++,n.type=i.LineType.INSERT,n.oldNumber=void 0,n.newNumber=u++):s(e,r)?(a.deletedLines++,n.type=i.LineType.DELETE,n.oldNumber=c++,n.newNumber=void 0):(n.type=i.LineType.CONTEXT,n.oldNumber=c++,n.newNumber=u++),o.lines.push(n)}(e);const H=!function(e,n){let t=n;for(;t'),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b(i.rp("'),i.b(i.v(i.f("fileName",e,n,0))),i.b(""),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b(' '),i.b(i.v(i.f("addedLines",e,n,0))),i.b(""),i.b("\n"+t),i.b(' '),i.b(i.v(i.f("deletedLines",e,n,0))),i.b(""),i.b("\n"+t),i.b(" "),i.b("\n"+t),i.b(" "),i.b("\n"+t),i.b(""),i.fl()},partials:{"'),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b(' Files changed ('),i.b(i.v(i.f("filesNumber",e,n,0))),i.b(")"),i.b("\n"+t),i.b(' hide'),i.b("\n"+t),i.b(' show'),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b('
    '),i.b("\n"+t),i.b(" "),i.b(i.t(i.f("files",e,n,0))),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b(""),i.fl()},partials:{},subs:{}}),n.defaultTemplates["generic-block-header"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b(""),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b('
'),i.s(i.f("blockHeader",e,n,1),e,n,0,156,173,"{{ }}")&&(i.rs(e,n,(function(e,n,t){t.b(t.t(t.f("blockHeader",e,n,0)))})),e.pop()),i.s(i.f("blockHeader",e,n,1),e,n,1,0,0,"")||i.b(" "),i.b("
"),i.b("\n"+t),i.b(" "),i.b("\n"+t),i.b(""),i.fl()},partials:{},subs:{}}),n.defaultTemplates["generic-empty-diff"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b(""),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b(" File without changes"),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b(" "),i.b("\n"+t),i.b(""),i.fl()},partials:{},subs:{}}),n.defaultTemplates["generic-file-path"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b(''),i.b("\n"+t),i.b(i.rp("'),i.b(i.v(i.f("fileDiffName",e,n,0))),i.b(""),i.b("\n"+t),i.b(i.rp(""),i.b("\n"+t),i.b('"),i.fl()},partials:{""),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b(" "),i.b(i.t(i.f("lineNumber",e,n,0))),i.b("\n"+t),i.b(" "),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.s(i.f("prefix",e,n,1),e,n,0,162,238,"{{ }}")&&(i.rs(e,n,(function(e,n,i){i.b(' '),i.b(i.t(i.f("prefix",e,n,0))),i.b(""),i.b("\n"+t)})),e.pop()),i.s(i.f("prefix",e,n,1),e,n,1,0,0,"")||(i.b('  '),i.b("\n"+t)),i.s(i.f("content",e,n,1),e,n,0,371,445,"{{ }}")&&(i.rs(e,n,(function(e,n,i){i.b(' '),i.b(i.t(i.f("content",e,n,0))),i.b(""),i.b("\n"+t)})),e.pop()),i.s(i.f("content",e,n,1),e,n,1,0,0,"")||(i.b('
'),i.b("\n"+t)),i.b("
"),i.b("\n"+t),i.b(" "),i.b("\n"+t),i.b(""),i.fl()},partials:{},subs:{}}),n.defaultTemplates["generic-wrapper"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('
'),i.b("\n"+t),i.b(" "),i.b(i.t(i.f("content",e,n,0))),i.b("\n"+t),i.b("
"),i.fl()},partials:{},subs:{}}),n.defaultTemplates["icon-file-added"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('"),i.fl()},partials:{},subs:{}}),n.defaultTemplates["icon-file-changed"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('"),i.fl()},partials:{},subs:{}}),n.defaultTemplates["icon-file-deleted"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('"),i.fl()},partials:{},subs:{}}),n.defaultTemplates["icon-file-renamed"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('"),i.fl()},partials:{},subs:{}}),n.defaultTemplates["icon-file"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('"),i.fl()},partials:{},subs:{}}),n.defaultTemplates["line-by-line-file-diff"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('
'),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b(" "),i.b(i.t(i.f("filePath",e,n,0))),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b(" "),i.b(i.t(i.f("diffs",e,n,0))),i.b("\n"+t),i.b(" "),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b("
"),i.fl()},partials:{},subs:{}}),n.defaultTemplates["line-by-line-numbers"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('
'),i.b(i.v(i.f("oldNumber",e,n,0))),i.b("
"),i.b("\n"+t),i.b('
'),i.b(i.v(i.f("newNumber",e,n,0))),i.b("
"),i.fl()},partials:{},subs:{}}),n.defaultTemplates["side-by-side-file-diff"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('
'),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b(" "),i.b(i.t(i.f("filePath",e,n,0))),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b(" "),i.b(i.t(i.d("diffs.left",e,n,0))),i.b("\n"+t),i.b(" "),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b('
'),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b(' '),i.b("\n"+t),i.b(" "),i.b(i.t(i.d("diffs.right",e,n,0))),i.b("\n"+t),i.b(" "),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b("
"),i.b("\n"+t),i.b("
"),i.fl()},partials:{},subs:{}}),n.defaultTemplates["tag-file-added"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('ADDED'),i.fl()},partials:{},subs:{}}),n.defaultTemplates["tag-file-changed"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('CHANGED'),i.fl()},partials:{},subs:{}}),n.defaultTemplates["tag-file-deleted"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('DELETED'),i.fl()},partials:{},subs:{}}),n.defaultTemplates["tag-file-renamed"]=new s.Template({code:function(e,n,t){var i=this;return i.b(t=t||""),i.b('RENAMED'),i.fl()},partials:{},subs:{}})},6834:function(e,n,t){"use strict";var i=this&&this.__createBinding||(Object.create?function(e,n,t,i){void 0===i&&(i=t);var a=Object.getOwnPropertyDescriptor(n,t);a&&!("get"in a?!n.__esModule:a.writable||a.configurable)||(a={enumerable:!0,get:function(){return n[t]}}),Object.defineProperty(e,i,a)}:function(e,n,t,i){void 0===i&&(i=t),e[i]=n[t]}),a=this&&this.__setModuleDefault||(Object.create?function(e,n){Object.defineProperty(e,"default",{enumerable:!0,value:n})}:function(e,n){e.default=n}),r=this&&this.__importStar||function(e){if(e&&e.__esModule)return e;var n={};if(null!=e)for(var t in e)"default"!==t&&Object.prototype.hasOwnProperty.call(e,t)&&i(n,e,t);return a(n,e),n},s=this&&this.__importDefault||function(e){return e&&e.__esModule?e:{default:e}};Object.defineProperty(n,"__esModule",{value:!0}),n.html=n.parse=n.defaultDiff2HtmlConfig=void 0;const o=r(t(8468)),l=t(3479),c=r(t(2378)),d=r(t(170)),u=t(9699),g=s(t(4063));n.defaultDiff2HtmlConfig=Object.assign(Object.assign(Object.assign({},c.defaultLineByLineRendererConfig),d.defaultSideBySideRendererConfig),{outputFormat:u.OutputFormatType.LINE_BY_LINE,drawFileList:!0}),n.parse=function(e,t={}){return o.parse(e,Object.assign(Object.assign({},n.defaultDiff2HtmlConfig),t))},n.html=function(e,t={}){const i=Object.assign(Object.assign({},n.defaultDiff2HtmlConfig),t),a="string"==typeof e?o.parse(e,i):e,r=new g.default(i),{colorScheme:s}=i,u={colorScheme:s};return(i.drawFileList?new l.FileListRenderer(r,u).render(a):"")+("side-by-side"===i.outputFormat?new d.default(r,i).render(a):new c.default(r,i).render(a))}},3479:function(e,n,t){"use strict";var i=this&&this.__createBinding||(Object.create?function(e,n,t,i){void 0===i&&(i=t);var a=Object.getOwnPropertyDescriptor(n,t);a&&!("get"in a?!n.__esModule:a.writable||a.configurable)||(a={enumerable:!0,get:function(){return n[t]}}),Object.defineProperty(e,i,a)}:function(e,n,t,i){void 0===i&&(i=t),e[i]=n[t]}),a=this&&this.__setModuleDefault||(Object.create?function(e,n){Object.defineProperty(e,"default",{enumerable:!0,value:n})}:function(e,n){e.default=n}),r=this&&this.__importStar||function(e){if(e&&e.__esModule)return e;var n={};if(null!=e)for(var t in e)"default"!==t&&Object.prototype.hasOwnProperty.call(e,t)&&i(n,e,t);return a(n,e),n};Object.defineProperty(n,"__esModule",{value:!0}),n.FileListRenderer=n.defaultFileListRendererConfig=void 0;const s=r(t(5741)),o="file-summary";n.defaultFileListRendererConfig={colorScheme:s.defaultRenderConfig.colorScheme},n.FileListRenderer=class{constructor(e,t={}){this.hoganUtils=e,this.config=Object.assign(Object.assign({},n.defaultFileListRendererConfig),t)}render(e){const n=e.map((e=>this.hoganUtils.render(o,"line",{fileHtmlId:s.getHtmlId(e),oldName:e.oldName,newName:e.newName,fileName:s.filenameDiff(e),deletedLines:"-"+e.deletedLines,addedLines:"+"+e.addedLines},{fileIcon:this.hoganUtils.template("icon",s.getFileIcon(e))}))).join("\n");return this.hoganUtils.render(o,"wrapper",{colorScheme:s.colorSchemeToCss(this.config.colorScheme),filesNumber:e.length,files:n})}}},4063:function(e,n,t){"use strict";var i=this&&this.__createBinding||(Object.create?function(e,n,t,i){void 0===i&&(i=t);var a=Object.getOwnPropertyDescriptor(n,t);a&&!("get"in a?!n.__esModule:a.writable||a.configurable)||(a={enumerable:!0,get:function(){return n[t]}}),Object.defineProperty(e,i,a)}:function(e,n,t,i){void 0===i&&(i=t),e[i]=n[t]}),a=this&&this.__setModuleDefault||(Object.create?function(e,n){Object.defineProperty(e,"default",{enumerable:!0,value:n})}:function(e,n){e.default=n}),r=this&&this.__importStar||function(e){if(e&&e.__esModule)return e;var n={};if(null!=e)for(var t in e)"default"!==t&&Object.prototype.hasOwnProperty.call(e,t)&&i(n,e,t);return a(n,e),n};Object.defineProperty(n,"__esModule",{value:!0});const s=r(t(5485)),o=t(979);n.default=class{constructor({compiledTemplates:e={},rawTemplates:n={}}){const t=Object.entries(n).reduce(((e,[n,t])=>{const i=s.compile(t,{asString:!1});return Object.assign(Object.assign({},e),{[n]:i})}),{});this.preCompiledTemplates=Object.assign(Object.assign(Object.assign({},o.defaultTemplates),e),t)}static compile(e){return s.compile(e,{asString:!1})}render(e,n,t,i,a){const r=this.templateKey(e,n);try{return this.preCompiledTemplates[r].render(t,i,a)}catch(e){throw new Error(`Could not find template to render '${r}'`)}}template(e,n){return this.preCompiledTemplates[this.templateKey(e,n)]}templateKey(e,n){return`${e}-${n}`}}},2378:function(e,n,t){"use strict";var i=this&&this.__createBinding||(Object.create?function(e,n,t,i){void 0===i&&(i=t);var a=Object.getOwnPropertyDescriptor(n,t);a&&!("get"in a?!n.__esModule:a.writable||a.configurable)||(a={enumerable:!0,get:function(){return n[t]}}),Object.defineProperty(e,i,a)}:function(e,n,t,i){void 0===i&&(i=t),e[i]=n[t]}),a=this&&this.__setModuleDefault||(Object.create?function(e,n){Object.defineProperty(e,"default",{enumerable:!0,value:n})}:function(e,n){e.default=n}),r=this&&this.__importStar||function(e){if(e&&e.__esModule)return e;var n={};if(null!=e)for(var t in e)"default"!==t&&Object.prototype.hasOwnProperty.call(e,t)&&i(n,e,t);return a(n,e),n};Object.defineProperty(n,"__esModule",{value:!0}),n.defaultLineByLineRendererConfig=void 0;const s=r(t(1483)),o=r(t(5741)),l=t(9699);n.defaultLineByLineRendererConfig=Object.assign(Object.assign({},o.defaultRenderConfig),{renderNothingWhenEmpty:!1,matchingMaxComparisons:2500,maxLineSizeInBlockForComparison:200});const c="generic",d="line-by-line";n.default=class{constructor(e,t={}){this.hoganUtils=e,this.config=Object.assign(Object.assign({},n.defaultLineByLineRendererConfig),t)}render(e){const n=e.map((e=>{let n;return n=e.blocks.length?this.generateFileHtml(e):this.generateEmptyDiff(),this.makeFileDiffHtml(e,n)})).join("\n");return this.hoganUtils.render(c,"wrapper",{colorScheme:o.colorSchemeToCss(this.config.colorScheme),content:n})}makeFileDiffHtml(e,n){if(this.config.renderNothingWhenEmpty&&Array.isArray(e.blocks)&&0===e.blocks.length)return"";const t=this.hoganUtils.template(d,"file-diff"),i=this.hoganUtils.template(c,"file-path"),a=this.hoganUtils.template("icon","file"),r=this.hoganUtils.template("tag",o.getFileIcon(e));return t.render({file:e,fileHtmlId:o.getHtmlId(e),diffs:n,filePath:i.render({fileDiffName:o.filenameDiff(e)},{fileIcon:a,fileTag:r})})}generateEmptyDiff(){return this.hoganUtils.render(c,"empty-diff",{contentClass:"d2h-code-line",CSSLineClass:o.CSSLineClass})}generateFileHtml(e){const n=s.newMatcherFn(s.newDistanceFn((n=>o.deconstructLine(n.content,e.isCombined).content)));return e.blocks.map((t=>{let i=this.hoganUtils.render(c,"block-header",{CSSLineClass:o.CSSLineClass,blockHeader:e.isTooBig?t.header:o.escapeForHtml(t.header),lineClass:"d2h-code-linenumber",contentClass:"d2h-code-line"});return this.applyLineGroupping(t).forEach((([t,a,r])=>{if(a.length&&r.length&&!t.length)this.applyRematchMatching(a,r,n).map((([n,t])=>{const{left:a,right:r}=this.processChangedLines(e,e.isCombined,n,t);i+=a,i+=r}));else if(t.length)t.forEach((n=>{const{prefix:t,content:a}=o.deconstructLine(n.content,e.isCombined);i+=this.generateSingleLineHtml(e,{type:o.CSSLineClass.CONTEXT,prefix:t,content:a,oldNumber:n.oldNumber,newNumber:n.newNumber})}));else if(a.length||r.length){const{left:n,right:t}=this.processChangedLines(e,e.isCombined,a,r);i+=n,i+=t}else console.error("Unknown state reached while processing groups of lines",t,a,r)})),i})).join("\n")}applyLineGroupping(e){const n=[];let t=[],i=[];for(let a=0;a0)&&(n.push([[],t,i]),t=[],i=[]),r.type===l.LineType.CONTEXT?n.push([[r],[],[]]):r.type===l.LineType.INSERT&&0===t.length?n.push([[],[],[r]]):r.type===l.LineType.INSERT&&t.length>0?i.push(r):r.type===l.LineType.DELETE&&t.push(r)}return(t.length||i.length)&&(n.push([[],t,i]),t=[],i=[]),n}applyRematchMatching(e,n,t){const i=e.length*n.length,a=Math.max.apply(null,[0].concat(e.concat(n).map((e=>e.content.length))));return i{"use strict";function t(e,n){if(0===e.length)return n.length;if(0===n.length)return e.length;const t=[];let i,a;for(i=0;i<=n.length;i++)t[i]=[i];for(a=0;a<=e.length;a++)t[0][a]=a;for(i=1;i<=n.length;i++)for(a=1;a<=e.length;a++)n.charAt(i-1)===e.charAt(a-1)?t[i][a]=t[i-1][a-1]:t[i][a]=Math.min(t[i-1][a-1]+1,Math.min(t[i][a-1]+1,t[i-1][a]+1));return t[n.length][e.length]}Object.defineProperty(n,"__esModule",{value:!0}),n.newMatcherFn=n.newDistanceFn=n.levenshtein=void 0,n.levenshtein=t,n.newDistanceFn=function(e){return(n,i)=>{const a=e(n).trim(),r=e(i).trim();return t(a,r)/(a.length+r.length)}},n.newMatcherFn=function(e){return function n(t,i,a=0,r=new Map){const s=function(n,t,i=new Map){let a,r=1/0;for(let s=0;s0||s.indexB>0)&&(h=f.concat(h)),(t.length>u||i.length>g)&&(h=h.concat(E)),h}}},5741:function(e,n,t){"use strict";var i=this&&this.__createBinding||(Object.create?function(e,n,t,i){void 0===i&&(i=t);var a=Object.getOwnPropertyDescriptor(n,t);a&&!("get"in a?!n.__esModule:a.writable||a.configurable)||(a={enumerable:!0,get:function(){return n[t]}}),Object.defineProperty(e,i,a)}:function(e,n,t,i){void 0===i&&(i=t),e[i]=n[t]}),a=this&&this.__setModuleDefault||(Object.create?function(e,n){Object.defineProperty(e,"default",{enumerable:!0,value:n})}:function(e,n){e.default=n}),r=this&&this.__importStar||function(e){if(e&&e.__esModule)return e;var n={};if(null!=e)for(var t in e)"default"!==t&&Object.prototype.hasOwnProperty.call(e,t)&&i(n,e,t);return a(n,e),n};Object.defineProperty(n,"__esModule",{value:!0}),n.diffHighlight=n.getFileIcon=n.getHtmlId=n.filenameDiff=n.deconstructLine=n.escapeForHtml=n.colorSchemeToCss=n.toCSSClass=n.defaultRenderConfig=n.CSSLineClass=void 0;const s=r(t(9479)),o=t(8593),l=r(t(1483)),c=t(9699);n.CSSLineClass={INSERTS:"d2h-ins",DELETES:"d2h-del",CONTEXT:"d2h-cntx",INFO:"d2h-info",INSERT_CHANGES:"d2h-ins d2h-change",DELETE_CHANGES:"d2h-del d2h-change"},n.defaultRenderConfig={matching:c.LineMatchingType.NONE,matchWordsThreshold:.25,maxLineLengthHighlight:1e4,diffStyle:c.DiffStyleType.WORD,colorScheme:c.ColorSchemeType.LIGHT};const d="/",u=l.newDistanceFn((e=>e.value)),g=l.newMatcherFn(u);function b(e){return-1!==e.indexOf("dev/null")}function p(e){return e.replace(/(]*>((.|\n)*?)<\/del>)/g,"")}function f(e){return e.slice(0).replace(/&/g,"&").replace(//g,">").replace(/"/g,""").replace(/'/g,"'").replace(/\//g,"/")}function m(e,n,t=!0){const i=function(e){return e?2:1}(n);return{prefix:e.substring(0,i),content:t?f(e.substring(i)):e.substring(i)}}function E(e){const n=(0,o.unifyPath)(e.oldName),t=(0,o.unifyPath)(e.newName);if(n===t||b(n)||b(t))return b(t)?n:t;{const e=[],i=[],a=n.split(d),r=t.split(d);let s=0,o=a.length-1,l=r.length-1;for(;ss&&l>s&&a[o]===r[l];)i.unshift(r[l]),o-=1,l-=1;const c=e.join(d),u=i.join(d),g=a.slice(s,o+1).join(d),b=r.slice(s,l+1).join(d);return c.length&&u.length?c+d+"{"+g+" → "+b+"}"+d+u:c.length?c+d+"{"+g+" → "+b+"}":u.length?"{"+g+" → "+b+"}"+d+u:n+" → "+t}}n.toCSSClass=function(e){switch(e){case c.LineType.CONTEXT:return n.CSSLineClass.CONTEXT;case c.LineType.INSERT:return n.CSSLineClass.INSERTS;case c.LineType.DELETE:return n.CSSLineClass.DELETES}},n.colorSchemeToCss=function(e){switch(e){case c.ColorSchemeType.DARK:return"d2h-dark-color-scheme";case c.ColorSchemeType.AUTO:return"d2h-auto-color-scheme";case c.ColorSchemeType.LIGHT:default:return"d2h-light-color-scheme"}},n.escapeForHtml=f,n.deconstructLine=m,n.filenameDiff=E,n.getHtmlId=function(e){return`d2h-${(0,o.hashCode)(E(e)).toString().slice(-6)}`},n.getFileIcon=function(e){let n="file-changed";return e.isRename||e.isCopy?n="file-renamed":e.isNew?n="file-added":e.isDeleted?n="file-deleted":e.newName!==e.oldName&&(n="file-renamed"),n},n.diffHighlight=function(e,t,i,a={}){const{matching:r,maxLineLengthHighlight:o,matchWordsThreshold:l,diffStyle:c}=Object.assign(Object.assign({},n.defaultRenderConfig),a),d=m(e,i,!1),b=m(t,i,!1);if(d.content.length>o||b.content.length>o)return{oldLine:{prefix:d.prefix,content:f(d.content)},newLine:{prefix:b.prefix,content:f(b.content)}};const E="char"===c?s.diffChars(d.content,b.content):s.diffWordsWithSpace(d.content,b.content),h=[];if("word"===c&&"words"===r){const e=E.filter((e=>e.removed)),n=E.filter((e=>e.added));g(n,e).forEach((e=>{1===e[0].length&&1===e[1].length&&u(e[0][0],e[1][0]){const t=n.added?"ins":n.removed?"del":null,i=h.indexOf(n)>-1?' class="d2h-change"':"",a=f(n.value);return null!==t?`${e}<${t}${i}>${a}`:`${e}${a}`}),"");return{oldLine:{prefix:d.prefix,content:(N=_,N.replace(/(]*>((.|\n)*?)<\/ins>)/g,""))},newLine:{prefix:b.prefix,content:p(_)}};var N}},170:function(e,n,t){"use strict";var i=this&&this.__createBinding||(Object.create?function(e,n,t,i){void 0===i&&(i=t);var a=Object.getOwnPropertyDescriptor(n,t);a&&!("get"in a?!n.__esModule:a.writable||a.configurable)||(a={enumerable:!0,get:function(){return n[t]}}),Object.defineProperty(e,i,a)}:function(e,n,t,i){void 0===i&&(i=t),e[i]=n[t]}),a=this&&this.__setModuleDefault||(Object.create?function(e,n){Object.defineProperty(e,"default",{enumerable:!0,value:n})}:function(e,n){e.default=n}),r=this&&this.__importStar||function(e){if(e&&e.__esModule)return e;var n={};if(null!=e)for(var t in e)"default"!==t&&Object.prototype.hasOwnProperty.call(e,t)&&i(n,e,t);return a(n,e),n};Object.defineProperty(n,"__esModule",{value:!0}),n.defaultSideBySideRendererConfig=void 0;const s=r(t(1483)),o=r(t(5741)),l=t(9699);n.defaultSideBySideRendererConfig=Object.assign(Object.assign({},o.defaultRenderConfig),{renderNothingWhenEmpty:!1,matchingMaxComparisons:2500,maxLineSizeInBlockForComparison:200});const c="generic";n.default=class{constructor(e,t={}){this.hoganUtils=e,this.config=Object.assign(Object.assign({},n.defaultSideBySideRendererConfig),t)}render(e){const n=e.map((e=>{let n;return n=e.blocks.length?this.generateFileHtml(e):this.generateEmptyDiff(),this.makeFileDiffHtml(e,n)})).join("\n");return this.hoganUtils.render(c,"wrapper",{colorScheme:o.colorSchemeToCss(this.config.colorScheme),content:n})}makeFileDiffHtml(e,n){if(this.config.renderNothingWhenEmpty&&Array.isArray(e.blocks)&&0===e.blocks.length)return"";const t=this.hoganUtils.template("side-by-side","file-diff"),i=this.hoganUtils.template(c,"file-path"),a=this.hoganUtils.template("icon","file"),r=this.hoganUtils.template("tag",o.getFileIcon(e));return t.render({file:e,fileHtmlId:o.getHtmlId(e),diffs:n,filePath:i.render({fileDiffName:o.filenameDiff(e)},{fileIcon:a,fileTag:r})})}generateEmptyDiff(){return{right:"",left:this.hoganUtils.render(c,"empty-diff",{contentClass:"d2h-code-side-line",CSSLineClass:o.CSSLineClass})}}generateFileHtml(e){const n=s.newMatcherFn(s.newDistanceFn((n=>o.deconstructLine(n.content,e.isCombined).content)));return e.blocks.map((t=>{const i={left:this.makeHeaderHtml(t.header,e),right:this.makeHeaderHtml("")};return this.applyLineGroupping(t).forEach((([t,a,r])=>{if(a.length&&r.length&&!t.length)this.applyRematchMatching(a,r,n).map((([n,t])=>{const{left:a,right:r}=this.processChangedLines(e.isCombined,n,t);i.left+=a,i.right+=r}));else if(t.length)t.forEach((n=>{const{prefix:t,content:a}=o.deconstructLine(n.content,e.isCombined),{left:r,right:s}=this.generateLineHtml({type:o.CSSLineClass.CONTEXT,prefix:t,content:a,number:n.oldNumber},{type:o.CSSLineClass.CONTEXT,prefix:t,content:a,number:n.newNumber});i.left+=r,i.right+=s}));else if(a.length||r.length){const{left:n,right:t}=this.processChangedLines(e.isCombined,a,r);i.left+=n,i.right+=t}else console.error("Unknown state reached while processing groups of lines",t,a,r)})),i})).reduce(((e,n)=>({left:e.left+n.left,right:e.right+n.right})),{left:"",right:""})}applyLineGroupping(e){const n=[];let t=[],i=[];for(let a=0;a0)&&(n.push([[],t,i]),t=[],i=[]),r.type===l.LineType.CONTEXT?n.push([[r],[],[]]):r.type===l.LineType.INSERT&&0===t.length?n.push([[],[],[r]]):r.type===l.LineType.INSERT&&t.length>0?i.push(r):r.type===l.LineType.DELETE&&t.push(r)}return(t.length||i.length)&&(n.push([[],t,i]),t=[],i=[]),n}applyRematchMatching(e,n,t){const i=e.length*n.length,a=Math.max.apply(null,[0].concat(e.concat(n).map((e=>e.content.length))));return i{"use strict";var t,i;Object.defineProperty(n,"__esModule",{value:!0}),n.ColorSchemeType=n.DiffStyleType=n.LineMatchingType=n.OutputFormatType=n.LineType=void 0,function(e){e.INSERT="insert",e.DELETE="delete",e.CONTEXT="context"}(t||(n.LineType=t={})),n.OutputFormatType={LINE_BY_LINE:"line-by-line",SIDE_BY_SIDE:"side-by-side"},n.LineMatchingType={LINES:"lines",WORDS:"words",NONE:"none"},n.DiffStyleType={WORD:"word",CHAR:"char"},function(e){e.AUTO="auto",e.DARK="dark",e.LIGHT="light"}(i||(n.ColorSchemeType=i={}))},4169:(e,n,t)=>{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.Diff2HtmlUI=n.defaultDiff2HtmlUIConfig=void 0;const i=t(1529),a=t(6834);n.defaultDiff2HtmlUIConfig=Object.assign(Object.assign({},a.defaultDiff2HtmlConfig),{synchronisedScroll:!0,highlight:!0,fileListToggle:!0,fileListStartVisible:!1,highlightLanguages:new Map,smartSelection:!0,fileContentToggle:!0,stickyFileHeaders:!0}),n.Diff2HtmlUI=class{constructor(e,t,i={},r){this.hljs=null,this.currentSelectionColumnId=-1,this.config=Object.assign(Object.assign({},n.defaultDiff2HtmlUIConfig),i),this.diffHtml=void 0!==t?(0,a.html)(t,this.config):e.innerHTML,this.targetElement=e,void 0!==r&&(this.hljs=r)}draw(){this.targetElement.innerHTML=this.diffHtml,this.config.synchronisedScroll&&this.synchronisedScroll(),this.config.highlight&&this.highlightCode(),this.config.fileListToggle&&this.fileListToggle(this.config.fileListStartVisible),this.config.fileContentToggle&&this.fileContentToggle(),this.config.stickyFileHeaders&&this.stickyFileHeaders()}synchronisedScroll(){this.targetElement.querySelectorAll(".d2h-file-wrapper").forEach((e=>{const[n,t]=Array().slice.call(e.querySelectorAll(".d2h-file-side-diff"));if(void 0===n||void 0===t)return;const i=e=>{null!==e&&null!==e.target&&(e.target===n?(t.scrollTop=n.scrollTop,t.scrollLeft=n.scrollLeft):(n.scrollTop=t.scrollTop,n.scrollLeft=t.scrollLeft))};n.addEventListener("scroll",i),t.addEventListener("scroll",i)}))}fileListToggle(e){const n=this.targetElement.querySelector(".d2h-show"),t=this.targetElement.querySelector(".d2h-hide"),i=this.targetElement.querySelector(".d2h-file-list");if(null===n||null===t||null===i)return;const a=()=>{n.style.display="none",t.style.display="inline",i.style.display="block"},r=()=>{n.style.display="inline",t.style.display="none",i.style.display="none"};n.addEventListener("click",(()=>a())),t.addEventListener("click",(()=>r()));const s=this.getHashTag();"files-summary-show"===s?a():"files-summary-hide"===s?r():e?a():r()}fileContentToggle(){this.targetElement.querySelectorAll(".d2h-file-collapse").forEach((e=>{e.style.display="flex";const n=n=>{var t;const i=null===(t=e.closest(".d2h-file-wrapper"))||void 0===t?void 0:t.querySelector(n);null!=i&&(e.classList.toggle("d2h-selected"),i.classList.toggle("d2h-d-none"))};e.addEventListener("click",(t=>(t=>{e!==t.target&&(n(".d2h-file-diff"),n(".d2h-files-diff"))})(t)))}))}highlightCode(){const e=this.hljs;if(null===e)throw new Error("Missing a `highlight.js` implementation. Please provide one when instantiating Diff2HtmlUI.");this.targetElement.querySelectorAll(".d2h-file-wrapper").forEach((n=>{const t=n.getAttribute("data-lang");this.config.highlightLanguages instanceof Map||(this.config.highlightLanguages=new Map(Object.entries(this.config.highlightLanguages)));let a=t&&this.config.highlightLanguages.has(t)?this.config.highlightLanguages.get(t):t?(0,i.getLanguage)(t):"plaintext";void 0===e.getLanguage(a)&&(a="plaintext"),n.querySelectorAll(".d2h-code-line-ctn").forEach((n=>{const t=n.textContent,r=n.parentNode;if(null===t||null===r||!this.isElement(r))return;const s=(0,i.closeTags)(e.highlight(t,{language:a,ignoreIllegals:!0})),o=(0,i.nodeStream)(n);if(o.length){const e=document.createElementNS("http://www.w3.org/1999/xhtml","div");e.innerHTML=s.value,s.value=(0,i.mergeStreams)(o,(0,i.nodeStream)(e),t)}n.classList.add("hljs"),s.language&&n.classList.add(s.language),n.innerHTML=s.value}))}))}stickyFileHeaders(){this.targetElement.querySelectorAll(".d2h-file-header").forEach((e=>{e.classList.add("d2h-sticky-header")}))}smartSelection(){console.warn("Smart selection is now enabled by default with CSS. No need to call this method anymore.")}getHashTag(){const e=document.URL,n=e.indexOf("#");let t=null;return-1!==n&&(t=e.substr(n+1)),t}isElement(e){return null!==e&&void 0!==(null==e?void 0:e.classList)}}},1529:(e,n)=>{"use strict";function t(e){return e.replace(/&/gm,"&").replace(//gm,">")}function i(e){return e.nodeName.toLowerCase()}Object.defineProperty(n,"__esModule",{value:!0}),n.getLanguage=n.closeTags=n.mergeStreams=n.nodeStream=void 0,n.nodeStream=function(e){const n=[],t=(e,a)=>{for(let r=e.firstChild;r;r=r.nextSibling)3===r.nodeType&&null!==r.nodeValue?a+=r.nodeValue.length:1===r.nodeType&&(n.push({event:"start",offset:a,node:r}),a=t(r,a),i(r).match(/br|hr|img|input/)||n.push({event:"stop",offset:a,node:r}));return a};return t(e,0),n},n.mergeStreams=function(e,n,a){let r=0,s="";const o=[];function l(){return e.length&&n.length?e[0].offset!==n[0].offset?e[0].offset`${e.nodeName}="${t(e.value).replace(/"/g,""")}"`)).join(" ")}>`}function d(e){s+=""}function u(e){("start"===e.event?c:d)(e.node)}for(;e.length||n.length;){let n=l();if(s+=t(a.substring(r,n[0].offset)),r=n[0].offset,n===e){o.reverse().forEach(d);do{u(n.splice(0,1)[0]),n=l()}while(n===e&&n.length&&n[0].offset===r);o.reverse().forEach(c)}else"start"===n[0].event?o.push(n[0].node):o.pop(),u(n.splice(0,1)[0])}return s+t(a.substr(r))},n.closeTags=function(e){const n=new Array;return e.value=e.value.split("\n").map((e=>{const t=n.map((e=>``)).join(""),i=e.matchAll(/(|<\/span>)/g);return Array.from(i).forEach((e=>{""===e[0]?n.shift():n.unshift(e[2])})),t+e+"".repeat(n.length)})).join("\n"),e};const a={"1c":"1c",abnf:"abnf",accesslog:"accesslog",as:"actionscript",adb:"ada",ada:"ada",ads:"ada",angelscript:"angelscript",apache:"apache",applescript:"applescript",scpt:"applescript",arcade:"arcade",cpp:"cpp",hpp:"cpp",arduino:"arduino",ino:"arduino",armasm:"armasm",arm:"armasm",xml:"xml",html:"xml",xhtml:"xml",rss:"xml",atom:"xml",xjb:"xml",xsd:"xml",xsl:"xml",plist:"xml",svg:"xml",asciidoc:"asciidoc",adoc:"asciidoc",asc:"asciidoc",aspectj:"aspectj",ahk:"autohotkey",ahkl:"autohotkey",au3:"autoit",avrasm:"avrasm",awk:"awk",axapta:"axapta","x++":"axapta",bash:"bash",sh:"bash",zsh:"bash",b:"basic",bnf:"bnf",bf:"brainfuck",c:"c",h:"c",cats:"c",idc:"c",cal:"cal",capnproto:"capnproto",capnp:"capnproto",ceylon:"ceylon",clean:"clean",clj:"clojure",boot:"clojure",cl2:"clojure",cljc:"clojure",cljs:"clojure","cljs.hl":"clojure",cljscm:"clojure",cljx:"clojure",hic:"clojure","clojure-repl":"clojure-repl",cmake:"cmake","cmake.in":"cmake",coffee:"coffeescript",_coffee:"coffeescript",cake:"coffeescript",cjsx:"coffeescript",iced:"coffeescript",cson:"coffeescript",coq:"coq",cos:"cos",cls:"cos",crmsh:"crmsh",crm:"crmsh",pcmk:"crmsh",cr:"crystal",cs:"csharp",csx:"csharp",csp:"csp",css:"css",d:"d",di:"d",md:"markdown",markdown:"markdown",mdown:"markdown",mdwn:"markdown",mkd:"markdown",mkdn:"markdown",mkdown:"markdown",ronn:"markdown",workbook:"markdown",dart:"dart",dpr:"delphi",dfm:"delphi",pas:"delphi",pascal:"delphi",diff:"diff",patch:"diff",django:"django",jinja:"django",dns:"dns",zone:"dns",bind:"dns",dockerfile:"dockerfile",docker:"dockerfile",dos:"dos",bat:"dos",cmd:"dos",dsconfig:"dsconfig",dts:"dts",dust:"dust",dst:"dust",ebnf:"ebnf",ex:"elixir",exs:"elixir",elm:"elm",rb:"ruby",builder:"ruby",eye:"ruby",gemspec:"ruby",god:"ruby",jbuilder:"ruby",mspec:"ruby",pluginspec:"ruby",podspec:"ruby",rabl:"ruby",rake:"ruby",rbuild:"ruby",rbw:"ruby",rbx:"ruby",ru:"ruby",ruby:"ruby",spec:"ruby",thor:"ruby",watchr:"ruby",erb:"erb","erlang-repl":"erlang-repl",erl:"erlang","app.src":"erlang",escript:"erlang",hrl:"erlang",xrl:"erlang",yrl:"erlang",excel:"excel",xls:"excel",xlsx:"excel",fix:"fix",flix:"flix",f90:"fortran",f:"fortran",f03:"fortran",f08:"fortran",f77:"fortran",f95:"fortran",for:"fortran",fpp:"fortran",fs:"fsharp",fsx:"fsharp",gams:"gams",gms:"gams",gauss:"gauss",gss:"gauss",gcode:"gcode",nc:"gcode",gherkin:"gherkin",glsl:"glsl",fp:"glsl",frag:"glsl",frg:"glsl",fsh:"glsl",fshader:"glsl",geo:"glsl",geom:"glsl",glslv:"glsl",gshader:"glsl",shader:"glsl",tesc:"glsl",tese:"glsl",vert:"glsl",vrx:"glsl",vsh:"glsl",vshader:"glsl",gml:"gml",go:"go",bal:"go",golo:"golo",gololang:"golo",gradle:"gradle",groovy:"groovy",grt:"groovy",gtpl:"groovy",gvy:"groovy",haml:"haml","haml.deface":"haml",handlebars:"handlebars",hbs:"handlebars","html.hbs":"handlebars","html.handlebars":"handlebars",hs:"haskell",hsc:"haskell",idr:"haskell",purs:"haskell",hx:"haxe",hxsl:"haxe",hsp:"hsp",htmlbars:"htmlbars",http:"http",https:"http",hy:"hy",inform7:"inform7",i7:"inform7",ini:"ini",toml:"ini",cfg:"ini",prefs:"ini",irpf90:"irpf90",isbl:"isbl",java:"java",jsp:"java",js:"javascript",jsx:"javascript",_js:"javascript",bones:"javascript",es:"javascript",es6:"javascript",gs:"javascript",jake:"javascript",jsb:"javascript",jscad:"javascript",jsfl:"javascript",jsm:"javascript",jss:"javascript",mjs:"javascript",njs:"javascript",pac:"javascript",sjs:"javascript",ssjs:"javascript",xsjs:"javascript",xsjslib:"javascript",cfc:"javascript","jboss-cli":"jboss-cli",json:"json",avsc:"json",geojson:"json",gltf:"json","JSON-tmLanguage":"json",jsonl:"json",tfstate:"json","tfstate.backup":"json",topojson:"json",webapp:"json",webmanifest:"json",jl:"julia","julia-repl":"julia-repl",kt:"kotlin",ktm:"kotlin",kts:"kotlin",lasso:"lasso",lassoscript:"lasso",tex:"latex",ldif:"ldif",leaf:"leaf",less:"less",lisp:"lisp",factor:"lisp",livecodeserver:"livecodeserver",ls:"livescript",_ls:"livescript",llvm:"llvm",lsl:"lsl",lua:"lua",nse:"lua",p8:"lua",pd_lua:"lua",rbxs:"lua",wlua:"lua",mak:"makefile",make:"makefile",mk:"makefile",mkfile:"makefile",mathematica:"mathematica",mma:"mathematica",wl:"mathematica",matlab:"matlab",maxima:"maxima",mel:"mel",mercury:"mercury",mipsasm:"mipsasm",miz:"mizar",voc:"mizar",al:"perl",cgi:"perl",fcgi:"perl",perl:"perl",ph:"perl",plx:"perl",pl:"perl",pm:"perl",psgi:"perl",t:"perl",mojolicious:"mojolicious",monkey:"monkey",monkey2:"monkey",moonscript:"moonscript",moon:"moonscript",n1ql:"n1ql",nginxconf:"nginx",nim:"nim",nimrod:"nim",nix:"nix",nsi:"nsis",nsh:"nsis",m:"objectivec",objc:"objectivec",mm:"objectivec","obj-c":"objectivec","obj-c++":"objectivec","objective-c++":"objectivec",fun:"ocaml",sig:"ocaml",ml:"ocaml",mli:"ocaml",eliom:"ocaml",eliomi:"ocaml",ml4:"ocaml",mll:"ocaml",mly:"ocaml",openscad:"openscad",oxygene:"oxygene",parser3:"parser3",pf:"pf","pf.conf":"pf",pgsql:"pgsql",postgres:"pgsql",postgresql:"pgsql",php:"php",aw:"php",ctp:"php",inc:"php",php3:"php",php4:"php",php5:"php",phps:"php",phpt:"php","php-template":"php-template",plaintext:"plaintext",txt:"plaintext",text:"plaintext",pony:"pony",ps:"powershell",ps1:"powershell",psd1:"powershell",psm1:"powershell",pde:"processing",profile:"profile",pro:"prolog",prolog:"prolog",yap:"prolog",properties:"properties",proto:"protobuf",puppet:"puppet",pp:"puppet",purebasic:"purebasic",py:"python",bzl:"python",gyp:"python",gypi:"python",lmi:"python",py3:"python",pyde:"python",pyi:"python",pyp:"python",pyt:"python",pyw:"python",rpy:"python",tac:"python",wsgi:"python",xpy:"python","python-repl":"python-repl",pycon:"python-repl",q:"q",k:"q",kdb:"q",qml:"qml",r:"r",rd:"r",rsx:"r",reasonml:"reasonml",re:"reasonml",rib:"rib",roboconf:"roboconf",graph:"roboconf",instances:"roboconf",routeros:"routeros",rsl:"rsl",ruleslanguage:"ruleslanguage",rs:"rust","rs.in":"rust",sas:"sas",scala:"scala",kojo:"scala",sbt:"scala",sc:"scala",scm:"scheme",sch:"scheme",sld:"scheme",sls:"scheme",sps:"scheme",ss:"scheme",rkt:"scheme",scilab:"scilab",scss:"scss",shell:"shell",smali:"smali",st:"smalltalk",sml:"sml",sqf:"sqf",sql:"sql",cql:"sql",ddl:"sql",mysql:"sql",prc:"sql",tab:"sql",udf:"sql",viw:"sql",stan:"stan",stanfuncs:"stan",stata:"stata",step21:"step21",step:"step21",stp:"step21",styl:"stylus",subunit:"subunit",swift:"swift",taggerscript:"taggerscript",yml:"yaml",mir:"yaml",reek:"yaml",rviz:"yaml","sublime-syntax":"yaml",syntax:"yaml",yaml:"yaml","yaml-tmlanguage":"yaml","yml.mysql":"yaml",tap:"tap",tcl:"tcl",adp:"tcl",tm:"tcl",thrift:"thrift",tp:"tp",twig:"twig",craftcms:"twig",ts:"typescript",tsx:"typescript",vala:"vala",vbnet:"vbnet",vb:"vbnet",vbscript:"vbscript",vbs:"vbscript","vbscript-html":"vbscript-html",v:"verilog",veo:"verilog",vhdl:"vhdl",vhd:"vhdl",vhf:"vhdl",vhi:"vhdl",vho:"vhdl",vhs:"vhdl",vht:"vhdl",vhw:"vhdl",vim:"vim",x86asm:"x86asm",xl:"xl",xquery:"xquery",xpath:"xquery",xq:"xquery",zephir:"zephir",zep:"zephir"};n.getLanguage=function(e){var n;return null!==(n=a[e])&&void 0!==n?n:"plaintext"}},1761:function(e,n,t){"use strict";var i=this&&this.__importDefault||function(e){return e&&e.__esModule?e:{default:e}};Object.defineProperty(n,"__esModule",{value:!0}),n.hljs=void 0;const a=i(t(3390)),r=i(t(6248)),s=i(t(4610)),o=i(t(4868)),l=i(t(8780)),c=i(t(612)),d=i(t(2177)),u=i(t(3147)),g=i(t(3707)),b=i(t(9534)),p=i(t(5064)),f=i(t(2003)),m=i(t(6642)),E=i(t(7731)),h=i(t(7360)),_=i(t(2543)),N=i(t(5658)),T=i(t(7905)),O=i(t(7569)),A=i(t(6652)),v=i(t(2399)),y=i(t(9878)),S=i(t(4658)),R=i(t(1407)),I=i(t(7077)),w=i(t(4762)),C=i(t(8257)),L=i(t(978)),M=i(t(14)),D=i(t(5812)),x=i(t(4210)),k=i(t(1943)),P=i(t(4981)),U=i(t(7903)),B=i(t(2482)),F=i(t(4028)),j=i(t(2446)),H=i(t(9814)),G=i(t(2656)),$=i(t(2437)),z=i(t(5040)),V=i(t(7546)),K=i(t(5559)),W=i(t(8245)),q=i(t(9880)),X=i(t(729)),Y=i(t(1062)),Z=i(t(7874)),Q=i(t(8935)),J=i(t(7690)),ee=i(t(1392)),ne=i(t(8987));a.default.registerLanguage("cpp",r.default),a.default.registerLanguage("xml",s.default),a.default.registerLanguage("awk",o.default),a.default.registerLanguage("bash",l.default),a.default.registerLanguage("c",c.default),a.default.registerLanguage("clojure",d.default),a.default.registerLanguage("crystal",u.default),a.default.registerLanguage("csharp",g.default),a.default.registerLanguage("csp",b.default),a.default.registerLanguage("css",p.default),a.default.registerLanguage("markdown",f.default),a.default.registerLanguage("dart",m.default),a.default.registerLanguage("diff",E.default),a.default.registerLanguage("dockerfile",h.default),a.default.registerLanguage("elixir",_.default),a.default.registerLanguage("elm",N.default),a.default.registerLanguage("ruby",T.default),a.default.registerLanguage("erlang",O.default),a.default.registerLanguage("fsharp",A.default),a.default.registerLanguage("go",v.default),a.default.registerLanguage("gradle",y.default),a.default.registerLanguage("groovy",S.default),a.default.registerLanguage("handlebars",R.default),a.default.registerLanguage("haskell",I.default),a.default.registerLanguage("ini",w.default),a.default.registerLanguage("java",C.default),a.default.registerLanguage("javascript",L.default),a.default.registerLanguage("json",M.default),a.default.registerLanguage("kotlin",D.default),a.default.registerLanguage("less",x.default),a.default.registerLanguage("lisp",k.default),a.default.registerLanguage("lua",P.default),a.default.registerLanguage("makefile",U.default),a.default.registerLanguage("perl",B.default),a.default.registerLanguage("nginx",F.default),a.default.registerLanguage("objectivec",j.default),a.default.registerLanguage("pgsql",H.default),a.default.registerLanguage("php",G.default),a.default.registerLanguage("plaintext",$.default),a.default.registerLanguage("powershell",z.default),a.default.registerLanguage("properties",V.default),a.default.registerLanguage("protobuf",K.default),a.default.registerLanguage("python",W.default),a.default.registerLanguage("rust",q.default),a.default.registerLanguage("scala",X.default),a.default.registerLanguage("scss",Y.default),a.default.registerLanguage("shell",Z.default),a.default.registerLanguage("sql",Q.default),a.default.registerLanguage("swift",J.default),a.default.registerLanguage("yaml",ee.default),a.default.registerLanguage("typescript",ne.default),n.hljs=a.default},8593:(e,n)=>{"use strict";Object.defineProperty(n,"__esModule",{value:!0}),n.hashCode=n.unifyPath=n.escapeForRegExp=void 0;const t=RegExp("["+["-","[","]","/","{","}","(",")","*","+","?",".","\\","^","$","|"].join("\\")+"]","g");n.escapeForRegExp=function(e){return e.replace(t,"\\$&")},n.unifyPath=function(e){return e?e.replace(/\\/g,"/"):e},n.hashCode=function(e){let n,t,i,a=0;for(n=0,i=e.length;n{function n(e){return e instanceof Map?e.clear=e.delete=e.set=function(){throw new Error("map is read-only")}:e instanceof Set&&(e.add=e.clear=e.delete=function(){throw new Error("set is read-only")}),Object.freeze(e),Object.getOwnPropertyNames(e).forEach((t=>{const i=e[t],a=typeof i;"object"!==a&&"function"!==a||Object.isFrozen(i)||n(i)})),e}class t{constructor(e){void 0===e.data&&(e.data={}),this.data=e.data,this.isMatchIgnored=!1}ignoreMatch(){this.isMatchIgnored=!0}}function i(e){return e.replace(/&/g,"&").replace(//g,">").replace(/"/g,""").replace(/'/g,"'")}function a(e,...n){const t=Object.create(null);for(const n in e)t[n]=e[n];return n.forEach((function(e){for(const n in e)t[n]=e[n]})),t}const r=e=>!!e.scope;class s{constructor(e,n){this.buffer="",this.classPrefix=n.classPrefix,e.walk(this)}addText(e){this.buffer+=i(e)}openNode(e){if(!r(e))return;const n=((e,{prefix:n})=>{if(e.startsWith("language:"))return e.replace("language:","language-");if(e.includes(".")){const t=e.split(".");return[`${n}${t.shift()}`,...t.map(((e,n)=>`${e}${"_".repeat(n+1)}`))].join(" ")}return`${n}${e}`})(e.scope,{prefix:this.classPrefix});this.span(n)}closeNode(e){r(e)&&(this.buffer+="")}value(){return this.buffer}span(e){this.buffer+=``}}const o=(e={})=>{const n={children:[]};return Object.assign(n,e),n};class l{constructor(){this.rootNode=o(),this.stack=[this.rootNode]}get top(){return this.stack[this.stack.length-1]}get root(){return this.rootNode}add(e){this.top.children.push(e)}openNode(e){const n=o({scope:e});this.add(n),this.stack.push(n)}closeNode(){if(this.stack.length>1)return this.stack.pop()}closeAllNodes(){for(;this.closeNode(););}toJSON(){return JSON.stringify(this.rootNode,null,4)}walk(e){return this.constructor._walk(e,this.rootNode)}static _walk(e,n){return"string"==typeof n?e.addText(n):n.children&&(e.openNode(n),n.children.forEach((n=>this._walk(e,n))),e.closeNode(n)),e}static _collapse(e){"string"!=typeof e&&e.children&&(e.children.every((e=>"string"==typeof e))?e.children=[e.children.join("")]:e.children.forEach((e=>{l._collapse(e)})))}}class c extends l{constructor(e){super(),this.options=e}addText(e){""!==e&&this.add(e)}startScope(e){this.openNode(e)}endScope(){this.closeNode()}__addSublanguage(e,n){const t=e.root;n&&(t.scope=`language:${n}`),this.add(t)}toHTML(){return new s(this,this.options).value()}finalize(){return this.closeAllNodes(),!0}}function d(e){return e?"string"==typeof e?e:e.source:null}function u(e){return p("(?=",e,")")}function g(e){return p("(?:",e,")*")}function b(e){return p("(?:",e,")?")}function p(...e){return e.map((e=>d(e))).join("")}function f(...e){const n=function(e){const n=e[e.length-1];return"object"==typeof n&&n.constructor===Object?(e.splice(e.length-1,1),n):{}}(e);return"("+(n.capture?"":"?:")+e.map((e=>d(e))).join("|")+")"}function m(e){return new RegExp(e.toString()+"|").exec("").length-1}const E=/\[(?:[^\\\]]|\\.)*\]|\(\??|\\([1-9][0-9]*)|\\./;function h(e,{joinWith:n}){let t=0;return e.map((e=>{t+=1;const n=t;let i=d(e),a="";for(;i.length>0;){const e=E.exec(i);if(!e){a+=i;break}a+=i.substring(0,e.index),i=i.substring(e.index+e[0].length),"\\"===e[0][0]&&e[1]?a+="\\"+String(Number(e[1])+n):(a+=e[0],"("===e[0]&&t++)}return a})).map((e=>`(${e})`)).join(n)}const _="[a-zA-Z]\\w*",N="[a-zA-Z_]\\w*",T="\\b\\d+(\\.\\d+)?",O="(-?)(\\b0[xX][a-fA-F0-9]+|(\\b\\d+(\\.\\d*)?|\\.\\d+)([eE][-+]?\\d+)?)",A="\\b(0b[01]+)",v={begin:"\\\\[\\s\\S]",relevance:0},y={scope:"string",begin:"'",end:"'",illegal:"\\n",contains:[v]},S={scope:"string",begin:'"',end:'"',illegal:"\\n",contains:[v]},R=function(e,n,t={}){const i=a({scope:"comment",begin:e,end:n,contains:[]},t);i.contains.push({scope:"doctag",begin:"[ ]*(?=(TODO|FIXME|NOTE|BUG|OPTIMIZE|HACK|XXX):)",end:/(TODO|FIXME|NOTE|BUG|OPTIMIZE|HACK|XXX):/,excludeBegin:!0,relevance:0});const r=f("I","a","is","so","us","to","at","if","in","it","on",/[A-Za-z]+['](d|ve|re|ll|t|s|n)/,/[A-Za-z]+[-][a-z]+/,/[A-Za-z][a-z]{2,}/);return i.contains.push({begin:p(/[ ]+/,"(",r,/[.]?[:]?([.][ ]|[ ])/,"){3}")}),i},I=R("//","$"),w=R("/\\*","\\*/"),C=R("#","$"),L={scope:"number",begin:T,relevance:0},M={scope:"number",begin:O,relevance:0},D={scope:"number",begin:A,relevance:0},x={begin:/(?=\/[^/\n]*\/)/,contains:[{scope:"regexp",begin:/\//,end:/\/[gimuy]*/,illegal:/\n/,contains:[v,{begin:/\[/,end:/\]/,relevance:0,contains:[v]}]}]},k={scope:"title",begin:_,relevance:0},P={scope:"title",begin:N,relevance:0},U={begin:"\\.\\s*"+N,relevance:0};var B=Object.freeze({__proto__:null,MATCH_NOTHING_RE:/\b\B/,IDENT_RE:_,UNDERSCORE_IDENT_RE:N,NUMBER_RE:T,C_NUMBER_RE:O,BINARY_NUMBER_RE:A,RE_STARTERS_RE:"!|!=|!==|%|%=|&|&&|&=|\\*|\\*=|\\+|\\+=|,|-|-=|/=|/|:|;|<<|<<=|<=|<|===|==|=|>>>=|>>=|>=|>>>|>>|>|\\?|\\[|\\{|\\(|\\^|\\^=|\\||\\|=|\\|\\||~",SHEBANG:(e={})=>{const n=/^#![ ]*\//;return e.binary&&(e.begin=p(n,/.*\b/,e.binary,/\b.*/)),a({scope:"meta",begin:n,end:/$/,relevance:0,"on:begin":(e,n)=>{0!==e.index&&n.ignoreMatch()}},e)},BACKSLASH_ESCAPE:v,APOS_STRING_MODE:y,QUOTE_STRING_MODE:S,PHRASAL_WORDS_MODE:{begin:/\b(a|an|the|are|I'm|isn't|don't|doesn't|won't|but|just|should|pretty|simply|enough|gonna|going|wtf|so|such|will|you|your|they|like|more)\b/},COMMENT:R,C_LINE_COMMENT_MODE:I,C_BLOCK_COMMENT_MODE:w,HASH_COMMENT_MODE:C,NUMBER_MODE:L,C_NUMBER_MODE:M,BINARY_NUMBER_MODE:D,REGEXP_MODE:x,TITLE_MODE:k,UNDERSCORE_TITLE_MODE:P,METHOD_GUARD:U,END_SAME_AS_BEGIN:function(e){return Object.assign(e,{"on:begin":(e,n)=>{n.data._beginMatch=e[1]},"on:end":(e,n)=>{n.data._beginMatch!==e[1]&&n.ignoreMatch()}})}});function F(e,n){"."===e.input[e.index-1]&&n.ignoreMatch()}function j(e,n){void 0!==e.className&&(e.scope=e.className,delete e.className)}function H(e,n){n&&e.beginKeywords&&(e.begin="\\b("+e.beginKeywords.split(" ").join("|")+")(?!\\.)(?=\\b|\\s)",e.__beforeBegin=F,e.keywords=e.keywords||e.beginKeywords,delete e.beginKeywords,void 0===e.relevance&&(e.relevance=0))}function G(e,n){Array.isArray(e.illegal)&&(e.illegal=f(...e.illegal))}function $(e,n){if(e.match){if(e.begin||e.end)throw new Error("begin & end are not supported with match");e.begin=e.match,delete e.match}}function z(e,n){void 0===e.relevance&&(e.relevance=1)}const V=(e,n)=>{if(!e.beforeMatch)return;if(e.starts)throw new Error("beforeMatch cannot be used with starts");const t=Object.assign({},e);Object.keys(e).forEach((n=>{delete e[n]})),e.keywords=t.keywords,e.begin=p(t.beforeMatch,u(t.begin)),e.starts={relevance:0,contains:[Object.assign(t,{endsParent:!0})]},e.relevance=0,delete t.beforeMatch},K=["of","and","for","in","not","or","if","then","parent","list","value"],W="keyword";function q(e,n,t=W){const i=Object.create(null);return"string"==typeof e?a(t,e.split(" ")):Array.isArray(e)?a(t,e):Object.keys(e).forEach((function(t){Object.assign(i,q(e[t],n,t))})),i;function a(e,t){n&&(t=t.map((e=>e.toLowerCase()))),t.forEach((function(n){const t=n.split("|");i[t[0]]=[e,X(t[0],t[1])]}))}}function X(e,n){return n?Number(n):function(e){return K.includes(e.toLowerCase())}(e)?0:1}const Y={},Z=e=>{console.error(e)},Q=(e,...n)=>{console.log(`WARN: ${e}`,...n)},J=(e,n)=>{Y[`${e}/${n}`]||(console.log(`Deprecated as of ${e}. ${n}`),Y[`${e}/${n}`]=!0)},ee=new Error;function ne(e,n,{key:t}){let i=0;const a=e[t],r={},s={};for(let e=1;e<=n.length;e++)s[e+i]=a[e],r[e+i]=!0,i+=m(n[e-1]);e[t]=s,e[t]._emit=r,e[t]._multi=!0}function te(e){!function(e){e.scope&&"object"==typeof e.scope&&null!==e.scope&&(e.beginScope=e.scope,delete e.scope)}(e),"string"==typeof e.beginScope&&(e.beginScope={_wrap:e.beginScope}),"string"==typeof e.endScope&&(e.endScope={_wrap:e.endScope}),function(e){if(Array.isArray(e.begin)){if(e.skip||e.excludeBegin||e.returnBegin)throw Z("skip, excludeBegin, returnBegin not compatible with beginScope: {}"),ee;if("object"!=typeof e.beginScope||null===e.beginScope)throw Z("beginScope must be object"),ee;ne(e,e.begin,{key:"beginScope"}),e.begin=h(e.begin,{joinWith:""})}}(e),function(e){if(Array.isArray(e.end)){if(e.skip||e.excludeEnd||e.returnEnd)throw Z("skip, excludeEnd, returnEnd not compatible with endScope: {}"),ee;if("object"!=typeof e.endScope||null===e.endScope)throw Z("endScope must be object"),ee;ne(e,e.end,{key:"endScope"}),e.end=h(e.end,{joinWith:""})}}(e)}function ie(e){function n(n,t){return new RegExp(d(n),"m"+(e.case_insensitive?"i":"")+(e.unicodeRegex?"u":"")+(t?"g":""))}class t{constructor(){this.matchIndexes={},this.regexes=[],this.matchAt=1,this.position=0}addRule(e,n){n.position=this.position++,this.matchIndexes[this.matchAt]=n,this.regexes.push([n,e]),this.matchAt+=m(e)+1}compile(){0===this.regexes.length&&(this.exec=()=>null);const e=this.regexes.map((e=>e[1]));this.matcherRe=n(h(e,{joinWith:"|"}),!0),this.lastIndex=0}exec(e){this.matcherRe.lastIndex=this.lastIndex;const n=this.matcherRe.exec(e);if(!n)return null;const t=n.findIndex(((e,n)=>n>0&&void 0!==e)),i=this.matchIndexes[t];return n.splice(0,t),Object.assign(n,i)}}class i{constructor(){this.rules=[],this.multiRegexes=[],this.count=0,this.lastIndex=0,this.regexIndex=0}getMatcher(e){if(this.multiRegexes[e])return this.multiRegexes[e];const n=new t;return this.rules.slice(e).forEach((([e,t])=>n.addRule(e,t))),n.compile(),this.multiRegexes[e]=n,n}resumingScanAtSamePosition(){return 0!==this.regexIndex}considerAll(){this.regexIndex=0}addRule(e,n){this.rules.push([e,n]),"begin"===n.type&&this.count++}exec(e){const n=this.getMatcher(this.regexIndex);n.lastIndex=this.lastIndex;let t=n.exec(e);if(this.resumingScanAtSamePosition())if(t&&t.index===this.lastIndex);else{const n=this.getMatcher(0);n.lastIndex=this.lastIndex+1,t=n.exec(e)}return t&&(this.regexIndex+=t.position+1,this.regexIndex===this.count&&this.considerAll()),t}}if(e.compilerExtensions||(e.compilerExtensions=[]),e.contains&&e.contains.includes("self"))throw new Error("ERR: contains `self` is not supported at the top-level of a language. See documentation.");return e.classNameAliases=a(e.classNameAliases||{}),function t(r,s){const o=r;if(r.isCompiled)return o;[j,$,te,V].forEach((e=>e(r,s))),e.compilerExtensions.forEach((e=>e(r,s))),r.__beforeBegin=null,[H,G,z].forEach((e=>e(r,s))),r.isCompiled=!0;let l=null;return"object"==typeof r.keywords&&r.keywords.$pattern&&(r.keywords=Object.assign({},r.keywords),l=r.keywords.$pattern,delete r.keywords.$pattern),l=l||/\w+/,r.keywords&&(r.keywords=q(r.keywords,e.case_insensitive)),o.keywordPatternRe=n(l,!0),s&&(r.begin||(r.begin=/\B|\b/),o.beginRe=n(o.begin),r.end||r.endsWithParent||(r.end=/\B|\b/),r.end&&(o.endRe=n(o.end)),o.terminatorEnd=d(o.end)||"",r.endsWithParent&&s.terminatorEnd&&(o.terminatorEnd+=(r.end?"|":"")+s.terminatorEnd)),r.illegal&&(o.illegalRe=n(r.illegal)),r.contains||(r.contains=[]),r.contains=[].concat(...r.contains.map((function(e){return function(e){return e.variants&&!e.cachedVariants&&(e.cachedVariants=e.variants.map((function(n){return a(e,{variants:null},n)}))),e.cachedVariants?e.cachedVariants:ae(e)?a(e,{starts:e.starts?a(e.starts):null}):Object.isFrozen(e)?a(e):e}("self"===e?r:e)}))),r.contains.forEach((function(e){t(e,o)})),r.starts&&t(r.starts,s),o.matcher=function(e){const n=new i;return e.contains.forEach((e=>n.addRule(e.begin,{rule:e,type:"begin"}))),e.terminatorEnd&&n.addRule(e.terminatorEnd,{type:"end"}),e.illegal&&n.addRule(e.illegal,{type:"illegal"}),n}(o),o}(e)}function ae(e){return!!e&&(e.endsWithParent||ae(e.starts))}class re extends Error{constructor(e,n){super(e),this.name="HTMLInjectionError",this.html=n}}const se=i,oe=a,le=Symbol("nomatch"),ce=function(e){const i=Object.create(null),a=Object.create(null),r=[];let s=!0;const o="Could not find the language '{}', did you forget to load/include a language module?",l={disableAutodetect:!0,name:"Plain text",contains:[]};let d={ignoreUnescapedHTML:!1,throwUnescapedHTML:!1,noHighlightRe:/^(no-?highlight)$/i,languageDetectRe:/\blang(?:uage)?-([\w-]+)\b/i,classPrefix:"hljs-",cssSelector:"pre code",languages:null,__emitter:c};function m(e){return d.noHighlightRe.test(e)}function E(e,n,t){let i="",a="";"object"==typeof n?(i=e,t=n.ignoreIllegals,a=n.language):(J("10.7.0","highlight(lang, code, ...args) has been deprecated."),J("10.7.0","Please use highlight(code, options) instead.\nhttps://github.com/highlightjs/highlight.js/issues/2277"),a=e,i=n),void 0===t&&(t=!0);const r={code:i,language:a};S("before:highlight",r);const s=r.result?r.result:h(r.language,r.code,t);return s.code=r.code,S("after:highlight",s),s}function h(e,n,a,r){const l=Object.create(null);function c(){if(!S.keywords)return void I.addText(w);let e=0;S.keywordPatternRe.lastIndex=0;let n=S.keywordPatternRe.exec(w),t="";for(;n;){t+=w.substring(e,n.index);const a=O.case_insensitive?n[0].toLowerCase():n[0],r=(i=a,S.keywords[i]);if(r){const[e,i]=r;if(I.addText(t),t="",l[a]=(l[a]||0)+1,l[a]<=7&&(C+=i),e.startsWith("_"))t+=n[0];else{const t=O.classNameAliases[e]||e;g(n[0],t)}}else t+=n[0];e=S.keywordPatternRe.lastIndex,n=S.keywordPatternRe.exec(w)}var i;t+=w.substring(e),I.addText(t)}function u(){null!=S.subLanguage?function(){if(""===w)return;let e=null;if("string"==typeof S.subLanguage){if(!i[S.subLanguage])return void I.addText(w);e=h(S.subLanguage,w,!0,R[S.subLanguage]),R[S.subLanguage]=e._top}else e=_(w,S.subLanguage.length?S.subLanguage:null);S.relevance>0&&(C+=e.relevance),I.__addSublanguage(e._emitter,e.language)}():c(),w=""}function g(e,n){""!==e&&(I.startScope(n),I.addText(e),I.endScope())}function b(e,n){let t=1;const i=n.length-1;for(;t<=i;){if(!e._emit[t]){t++;continue}const i=O.classNameAliases[e[t]]||e[t],a=n[t];i?g(a,i):(w=a,c(),w=""),t++}}function p(e,n){return e.scope&&"string"==typeof e.scope&&I.openNode(O.classNameAliases[e.scope]||e.scope),e.beginScope&&(e.beginScope._wrap?(g(w,O.classNameAliases[e.beginScope._wrap]||e.beginScope._wrap),w=""):e.beginScope._multi&&(b(e.beginScope,n),w="")),S=Object.create(e,{parent:{value:S}}),S}function f(e,n,i){let a=function(e,n){const t=e&&e.exec(n);return t&&0===t.index}(e.endRe,i);if(a){if(e["on:end"]){const i=new t(e);e["on:end"](n,i),i.isMatchIgnored&&(a=!1)}if(a){for(;e.endsParent&&e.parent;)e=e.parent;return e}}if(e.endsWithParent)return f(e.parent,n,i)}function m(e){return 0===S.matcher.regexIndex?(w+=e[0],1):(D=!0,0)}function E(e){const t=e[0],i=n.substring(e.index),a=f(S,e,i);if(!a)return le;const r=S;S.endScope&&S.endScope._wrap?(u(),g(t,S.endScope._wrap)):S.endScope&&S.endScope._multi?(u(),b(S.endScope,e)):r.skip?w+=t:(r.returnEnd||r.excludeEnd||(w+=t),u(),r.excludeEnd&&(w=t));do{S.scope&&I.closeNode(),S.skip||S.subLanguage||(C+=S.relevance),S=S.parent}while(S!==a.parent);return a.starts&&p(a.starts,e),r.returnEnd?0:t.length}let N={};function T(i,r){const o=r&&r[0];if(w+=i,null==o)return u(),0;if("begin"===N.type&&"end"===r.type&&N.index===r.index&&""===o){if(w+=n.slice(r.index,r.index+1),!s){const n=new Error(`0 width match regex (${e})`);throw n.languageName=e,n.badRule=N.rule,n}return 1}if(N=r,"begin"===r.type)return function(e){const n=e[0],i=e.rule,a=new t(i),r=[i.__beforeBegin,i["on:begin"]];for(const t of r)if(t&&(t(e,a),a.isMatchIgnored))return m(n);return i.skip?w+=n:(i.excludeBegin&&(w+=n),u(),i.returnBegin||i.excludeBegin||(w=n)),p(i,e),i.returnBegin?0:n.length}(r);if("illegal"===r.type&&!a){const e=new Error('Illegal lexeme "'+o+'" for mode "'+(S.scope||"")+'"');throw e.mode=S,e}if("end"===r.type){const e=E(r);if(e!==le)return e}if("illegal"===r.type&&""===o)return 1;if(M>1e5&&M>3*r.index)throw new Error("potential infinite loop, way more iterations than matches");return w+=o,o.length}const O=A(e);if(!O)throw Z(o.replace("{}",e)),new Error('Unknown language: "'+e+'"');const v=ie(O);let y="",S=r||v;const R={},I=new d.__emitter(d);!function(){const e=[];for(let n=S;n!==O;n=n.parent)n.scope&&e.unshift(n.scope);e.forEach((e=>I.openNode(e)))}();let w="",C=0,L=0,M=0,D=!1;try{if(O.__emitTokens)O.__emitTokens(n,I);else{for(S.matcher.considerAll();;){M++,D?D=!1:S.matcher.considerAll(),S.matcher.lastIndex=L;const e=S.matcher.exec(n);if(!e)break;const t=T(n.substring(L,e.index),e);L=e.index+t}T(n.substring(L))}return I.finalize(),y=I.toHTML(),{language:e,value:y,relevance:C,illegal:!1,_emitter:I,_top:S}}catch(t){if(t.message&&t.message.includes("Illegal"))return{language:e,value:se(n),illegal:!0,relevance:0,_illegalBy:{message:t.message,index:L,context:n.slice(L-100,L+100),mode:t.mode,resultSoFar:y},_emitter:I};if(s)return{language:e,value:se(n),illegal:!1,relevance:0,errorRaised:t,_emitter:I,_top:S};throw t}}function _(e,n){n=n||d.languages||Object.keys(i);const t=function(e){const n={value:se(e),illegal:!1,relevance:0,_top:l,_emitter:new d.__emitter(d)};return n._emitter.addText(e),n}(e),a=n.filter(A).filter(y).map((n=>h(n,e,!1)));a.unshift(t);const r=a.sort(((e,n)=>{if(e.relevance!==n.relevance)return n.relevance-e.relevance;if(e.language&&n.language){if(A(e.language).supersetOf===n.language)return 1;if(A(n.language).supersetOf===e.language)return-1}return 0})),[s,o]=r,c=s;return c.secondBest=o,c}function N(e){let n=null;const t=function(e){let n=e.className+" ";n+=e.parentNode?e.parentNode.className:"";const t=d.languageDetectRe.exec(n);if(t){const n=A(t[1]);return n||(Q(o.replace("{}",t[1])),Q("Falling back to no-highlight mode for this block.",e)),n?t[1]:"no-highlight"}return n.split(/\s+/).find((e=>m(e)||A(e)))}(e);if(m(t))return;if(S("before:highlightElement",{el:e,language:t}),e.children.length>0&&(d.ignoreUnescapedHTML||(console.warn("One of your code blocks includes unescaped HTML. This is a potentially serious security risk."),console.warn("https://github.com/highlightjs/highlight.js/wiki/security"),console.warn("The element with unescaped HTML:"),console.warn(e)),d.throwUnescapedHTML))throw new re("One of your code blocks includes unescaped HTML.",e.innerHTML);n=e;const i=n.textContent,r=t?E(i,{language:t,ignoreIllegals:!0}):_(i);e.innerHTML=r.value,function(e,n,t){const i=n&&a[n]||t;e.classList.add("hljs"),e.classList.add(`language-${i}`)}(e,t,r.language),e.result={language:r.language,re:r.relevance,relevance:r.relevance},r.secondBest&&(e.secondBest={language:r.secondBest.language,relevance:r.secondBest.relevance}),S("after:highlightElement",{el:e,result:r,text:i})}let T=!1;function O(){"loading"!==document.readyState?document.querySelectorAll(d.cssSelector).forEach(N):T=!0}function A(e){return e=(e||"").toLowerCase(),i[e]||i[a[e]]}function v(e,{languageName:n}){"string"==typeof e&&(e=[e]),e.forEach((e=>{a[e.toLowerCase()]=n}))}function y(e){const n=A(e);return n&&!n.disableAutodetect}function S(e,n){const t=e;r.forEach((function(e){e[t]&&e[t](n)}))}"undefined"!=typeof window&&window.addEventListener&&window.addEventListener("DOMContentLoaded",(function(){T&&O()}),!1),Object.assign(e,{highlight:E,highlightAuto:_,highlightAll:O,highlightElement:N,highlightBlock:function(e){return J("10.7.0","highlightBlock will be removed entirely in v12.0"),J("10.7.0","Please use highlightElement now."),N(e)},configure:function(e){d=oe(d,e)},initHighlighting:()=>{O(),J("10.6.0","initHighlighting() deprecated. Use highlightAll() now.")},initHighlightingOnLoad:function(){O(),J("10.6.0","initHighlightingOnLoad() deprecated. Use highlightAll() now.")},registerLanguage:function(n,t){let a=null;try{a=t(e)}catch(e){if(Z("Language definition for '{}' could not be registered.".replace("{}",n)),!s)throw e;Z(e),a=l}a.name||(a.name=n),i[n]=a,a.rawDefinition=t.bind(null,e),a.aliases&&v(a.aliases,{languageName:n})},unregisterLanguage:function(e){delete i[e];for(const n of Object.keys(a))a[n]===e&&delete a[n]},listLanguages:function(){return Object.keys(i)},getLanguage:A,registerAliases:v,autoDetection:y,inherit:oe,addPlugin:function(e){!function(e){e["before:highlightBlock"]&&!e["before:highlightElement"]&&(e["before:highlightElement"]=n=>{e["before:highlightBlock"](Object.assign({block:n.el},n))}),e["after:highlightBlock"]&&!e["after:highlightElement"]&&(e["after:highlightElement"]=n=>{e["after:highlightBlock"](Object.assign({block:n.el},n))})}(e),r.push(e)},removePlugin:function(e){const n=r.indexOf(e);-1!==n&&r.splice(n,1)}}),e.debugMode=function(){s=!1},e.safeMode=function(){s=!0},e.versionString="11.8.0",e.regex={concat:p,lookahead:u,either:f,optional:b,anyNumberOfTimes:g};for(const e in B)"object"==typeof B[e]&&n(B[e]);return Object.assign(e,B),e},de=ce({});de.newInstance=()=>ce({}),e.exports=de,de.HighlightJS=de,de.default=de},4868:e=>{e.exports=function(e){return{name:"Awk",keywords:{keyword:"BEGIN END if else while do for in break continue delete next nextfile function func exit|10"},contains:[{className:"variable",variants:[{begin:/\$[\w\d#@][\w\d_]*/},{begin:/\$\{(.*?)\}/}]},{className:"string",contains:[e.BACKSLASH_ESCAPE],variants:[{begin:/(u|b)?r?'''/,end:/'''/,relevance:10},{begin:/(u|b)?r?"""/,end:/"""/,relevance:10},{begin:/(u|r|ur)'/,end:/'/,relevance:10},{begin:/(u|r|ur)"/,end:/"/,relevance:10},{begin:/(b|br)'/,end:/'/},{begin:/(b|br)"/,end:/"/},e.APOS_STRING_MODE,e.QUOTE_STRING_MODE]},e.REGEXP_MODE,e.HASH_COMMENT_MODE,e.NUMBER_MODE]}}},8780:e=>{e.exports=function(e){const n=e.regex,t={},i={begin:/\$\{/,end:/\}/,contains:["self",{begin:/:-/,contains:[t]}]};Object.assign(t,{className:"variable",variants:[{begin:n.concat(/\$[\w\d#@][\w\d_]*/,"(?![\\w\\d])(?![$])")},i]});const a={className:"subst",begin:/\$\(/,end:/\)/,contains:[e.BACKSLASH_ESCAPE]},r={begin:/<<-?\s*(?=\w+)/,starts:{contains:[e.END_SAME_AS_BEGIN({begin:/(\w+)/,end:/(\w+)/,className:"string"})]}},s={className:"string",begin:/"/,end:/"/,contains:[e.BACKSLASH_ESCAPE,t,a]};a.contains.push(s);const o={begin:/\$?\(\(/,end:/\)\)/,contains:[{begin:/\d+#[0-9a-f]+/,className:"number"},e.NUMBER_MODE,t]},l=e.SHEBANG({binary:`(${["fish","bash","zsh","sh","csh","ksh","tcsh","dash","scsh"].join("|")})`,relevance:10}),c={className:"function",begin:/\w[\w\d_]*\s*\(\s*\)\s*\{/,returnBegin:!0,contains:[e.inherit(e.TITLE_MODE,{begin:/\w[\w\d_]*/})],relevance:0};return{name:"Bash",aliases:["sh"],keywords:{$pattern:/\b[a-z][a-z0-9._-]+\b/,keyword:["if","then","else","elif","fi","for","while","until","in","do","done","case","esac","function","select"],literal:["true","false"],built_in:["break","cd","continue","eval","exec","exit","export","getopts","hash","pwd","readonly","return","shift","test","times","trap","umask","unset","alias","bind","builtin","caller","command","declare","echo","enable","help","let","local","logout","mapfile","printf","read","readarray","source","type","typeset","ulimit","unalias","set","shopt","autoload","bg","bindkey","bye","cap","chdir","clone","comparguments","compcall","compctl","compdescribe","compfiles","compgroups","compquote","comptags","comptry","compvalues","dirs","disable","disown","echotc","echoti","emulate","fc","fg","float","functions","getcap","getln","history","integer","jobs","kill","limit","log","noglob","popd","print","pushd","pushln","rehash","sched","setcap","setopt","stat","suspend","ttyctl","unfunction","unhash","unlimit","unsetopt","vared","wait","whence","where","which","zcompile","zformat","zftp","zle","zmodload","zparseopts","zprof","zpty","zregexparse","zsocket","zstyle","ztcp","chcon","chgrp","chown","chmod","cp","dd","df","dir","dircolors","ln","ls","mkdir","mkfifo","mknod","mktemp","mv","realpath","rm","rmdir","shred","sync","touch","truncate","vdir","b2sum","base32","base64","cat","cksum","comm","csplit","cut","expand","fmt","fold","head","join","md5sum","nl","numfmt","od","paste","ptx","pr","sha1sum","sha224sum","sha256sum","sha384sum","sha512sum","shuf","sort","split","sum","tac","tail","tr","tsort","unexpand","uniq","wc","arch","basename","chroot","date","dirname","du","echo","env","expr","factor","groups","hostid","id","link","logname","nice","nohup","nproc","pathchk","pinky","printenv","printf","pwd","readlink","runcon","seq","sleep","stat","stdbuf","stty","tee","test","timeout","tty","uname","unlink","uptime","users","who","whoami","yes"]},contains:[l,e.SHEBANG(),c,o,e.HASH_COMMENT_MODE,r,{match:/(\/[a-z._-]+)+/},s,{className:"",begin:/\\"/},{className:"string",begin:/'/,end:/'/},t]}}},612:e=>{e.exports=function(e){const n=e.regex,t=e.COMMENT("//","$",{contains:[{begin:/\\\n/}]}),i="decltype\\(auto\\)",a="[a-zA-Z_]\\w*::",r="("+i+"|"+n.optional(a)+"[a-zA-Z_]\\w*"+n.optional("<[^<>]+>")+")",s={className:"type",variants:[{begin:"\\b[a-z\\d_]*_t\\b"},{match:/\batomic_[a-z]{3,6}\b/}]},o={className:"string",variants:[{begin:'(u8?|U|L)?"',end:'"',illegal:"\\n",contains:[e.BACKSLASH_ESCAPE]},{begin:"(u8?|U|L)?'(\\\\(x[0-9A-Fa-f]{2}|u[0-9A-Fa-f]{4,8}|[0-7]{3}|\\S)|.)",end:"'",illegal:"."},e.END_SAME_AS_BEGIN({begin:/(?:u8?|U|L)?R"([^()\\ ]{0,16})\(/,end:/\)([^()\\ ]{0,16})"/})]},l={className:"number",variants:[{begin:"\\b(0b[01']+)"},{begin:"(-?)\\b([\\d']+(\\.[\\d']*)?|\\.[\\d']+)((ll|LL|l|L)(u|U)?|(u|U)(ll|LL|l|L)?|f|F|b|B)"},{begin:"(-?)(\\b0[xX][a-fA-F0-9']+|(\\b[\\d']+(\\.[\\d']*)?|\\.[\\d']+)([eE][-+]?[\\d']+)?)"}],relevance:0},c={className:"meta",begin:/#\s*[a-z]+\b/,end:/$/,keywords:{keyword:"if else elif endif define undef warning error line pragma _Pragma ifdef ifndef include"},contains:[{begin:/\\\n/,relevance:0},e.inherit(o,{className:"string"}),{className:"string",begin:/<.*?>/},t,e.C_BLOCK_COMMENT_MODE]},d={className:"title",begin:n.optional(a)+e.IDENT_RE,relevance:0},u=n.optional(a)+e.IDENT_RE+"\\s*\\(",g={keyword:["asm","auto","break","case","continue","default","do","else","enum","extern","for","fortran","goto","if","inline","register","restrict","return","sizeof","struct","switch","typedef","union","volatile","while","_Alignas","_Alignof","_Atomic","_Generic","_Noreturn","_Static_assert","_Thread_local","alignas","alignof","noreturn","static_assert","thread_local","_Pragma"],type:["float","double","signed","unsigned","int","short","long","char","void","_Bool","_Complex","_Imaginary","_Decimal32","_Decimal64","_Decimal128","const","static","complex","bool","imaginary"],literal:"true false NULL",built_in:"std string wstring cin cout cerr clog stdin stdout stderr stringstream istringstream ostringstream auto_ptr deque list queue stack vector map set pair bitset multiset multimap unordered_set unordered_map unordered_multiset unordered_multimap priority_queue make_pair array shared_ptr abort terminate abs acos asin atan2 atan calloc ceil cosh cos exit exp fabs floor fmod fprintf fputs free frexp fscanf future isalnum isalpha iscntrl isdigit isgraph islower isprint ispunct isspace isupper isxdigit tolower toupper labs ldexp log10 log malloc realloc memchr memcmp memcpy memset modf pow printf putchar puts scanf sinh sin snprintf sprintf sqrt sscanf strcat strchr strcmp strcpy strcspn strlen strncat strncmp strncpy strpbrk strrchr strspn strstr tanh tan vfprintf vprintf vsprintf endl initializer_list unique_ptr"},b=[c,s,t,e.C_BLOCK_COMMENT_MODE,l,o],p={variants:[{begin:/=/,end:/;/},{begin:/\(/,end:/\)/},{beginKeywords:"new throw return else",end:/;/}],keywords:g,contains:b.concat([{begin:/\(/,end:/\)/,keywords:g,contains:b.concat(["self"]),relevance:0}]),relevance:0},f={begin:"("+r+"[\\*&\\s]+)+"+u,returnBegin:!0,end:/[{;=]/,excludeEnd:!0,keywords:g,illegal:/[^\w\s\*&:<>.]/,contains:[{begin:i,keywords:g,relevance:0},{begin:u,returnBegin:!0,contains:[e.inherit(d,{className:"title.function"})],relevance:0},{relevance:0,match:/,/},{className:"params",begin:/\(/,end:/\)/,keywords:g,relevance:0,contains:[t,e.C_BLOCK_COMMENT_MODE,o,l,s,{begin:/\(/,end:/\)/,keywords:g,relevance:0,contains:["self",t,e.C_BLOCK_COMMENT_MODE,o,l,s]}]},s,t,e.C_BLOCK_COMMENT_MODE,c]};return{name:"C",aliases:["h"],keywords:g,disableAutodetect:!0,illegal:"=]/,contains:[{beginKeywords:"final class struct"},e.TITLE_MODE]}]),exports:{preprocessor:c,strings:o,keywords:g}}}},2177:e=>{e.exports=function(e){const n="a-zA-Z_\\-!.?+*=<>&'",t="[#]?["+n+"]["+n+"0-9/;:$#]*",i="def defonce defprotocol defstruct defmulti defmethod defn- defn defmacro deftype defrecord",a={$pattern:t,built_in:i+" cond apply if-not if-let if not not= =|0 <|0 >|0 <=|0 >=|0 ==|0 +|0 /|0 *|0 -|0 rem quot neg? pos? delay? symbol? keyword? true? false? integer? empty? coll? list? set? ifn? fn? associative? sequential? sorted? counted? reversible? number? decimal? class? distinct? isa? float? rational? reduced? ratio? odd? even? char? seq? vector? string? map? nil? contains? zero? instance? not-every? not-any? libspec? -> ->> .. . inc compare do dotimes mapcat take remove take-while drop letfn drop-last take-last drop-while while intern condp case reduced cycle split-at split-with repeat replicate iterate range merge zipmap declare line-seq sort comparator sort-by dorun doall nthnext nthrest partition eval doseq await await-for let agent atom send send-off release-pending-sends add-watch mapv filterv remove-watch agent-error restart-agent set-error-handler error-handler set-error-mode! error-mode shutdown-agents quote var fn loop recur throw try monitor-enter monitor-exit macroexpand macroexpand-1 for dosync and or when when-not when-let comp juxt partial sequence memoize constantly complement identity assert peek pop doto proxy first rest cons cast coll last butlast sigs reify second ffirst fnext nfirst nnext meta with-meta ns in-ns create-ns import refer keys select-keys vals key val rseq name namespace promise into transient persistent! conj! assoc! dissoc! pop! disj! use class type num float double short byte boolean bigint biginteger bigdec print-method print-dup throw-if printf format load compile get-in update-in pr pr-on newline flush read slurp read-line subvec with-open memfn time re-find re-groups rand-int rand mod locking assert-valid-fdecl alias resolve ref deref refset swap! reset! set-validator! compare-and-set! alter-meta! reset-meta! commute get-validator alter ref-set ref-history-count ref-min-history ref-max-history ensure sync io! new next conj set! to-array future future-call into-array aset gen-class reduce map filter find empty hash-map hash-set sorted-map sorted-map-by sorted-set sorted-set-by vec vector seq flatten reverse assoc dissoc list disj get union difference intersection extend extend-type extend-protocol int nth delay count concat chunk chunk-buffer chunk-append chunk-first chunk-rest max min dec unchecked-inc-int unchecked-inc unchecked-dec-inc unchecked-dec unchecked-negate unchecked-add-int unchecked-add unchecked-subtract-int unchecked-subtract chunk-next chunk-cons chunked-seq? prn vary-meta lazy-seq spread list* str find-keyword keyword symbol gensym force rationalize"},r={begin:t,relevance:0},s={scope:"number",relevance:0,variants:[{match:/[-+]?0[xX][0-9a-fA-F]+N?/},{match:/[-+]?0[0-7]+N?/},{match:/[-+]?[1-9][0-9]?[rR][0-9a-zA-Z]+N?/},{match:/[-+]?[0-9]+\/[0-9]+N?/},{match:/[-+]?[0-9]+((\.[0-9]*([eE][+-]?[0-9]+)?M?)|([eE][+-]?[0-9]+M?|M))/},{match:/[-+]?([1-9][0-9]*|0)N?/}]},o={scope:"character",variants:[{match:/\\o[0-3]?[0-7]{1,2}/},{match:/\\u[0-9a-fA-F]{4}/},{match:/\\(newline|space|tab|formfeed|backspace|return)/},{match:/\\\S/,relevance:0}]},l={scope:"regex",begin:/#"/,end:/"/,contains:[e.BACKSLASH_ESCAPE]},c=e.inherit(e.QUOTE_STRING_MODE,{illegal:null}),d={scope:"punctuation",match:/,/,relevance:0},u=e.COMMENT(";","$",{relevance:0}),g={className:"literal",begin:/\b(true|false|nil)\b/},b={begin:"\\[|(#::?"+t+")?\\{",end:"[\\]\\}]",relevance:0},p={className:"symbol",begin:"[:]{1,2}"+t},f={begin:"\\(",end:"\\)"},m={endsWithParent:!0,relevance:0},E={keywords:a,className:"name",begin:t,relevance:0,starts:m},h=[d,f,o,l,c,u,p,b,s,g,r],_={beginKeywords:i,keywords:{$pattern:t,keyword:i},end:'(\\[|#|\\d|"|:|\\{|\\)|\\(|$)',contains:[{className:"title",begin:t,relevance:0,excludeEnd:!0,endsParent:!0}].concat(h)};return f.contains=[_,E,m],m.contains=h,b.contains=h,{name:"Clojure",aliases:["clj","edn"],illegal:/\S/,contains:[d,f,o,l,c,u,p,b,s,g]}}},6248:e=>{e.exports=function(e){const n=e.regex,t=e.COMMENT("//","$",{contains:[{begin:/\\\n/}]}),i="decltype\\(auto\\)",a="[a-zA-Z_]\\w*::",r="(?!struct)("+i+"|"+n.optional(a)+"[a-zA-Z_]\\w*"+n.optional("<[^<>]+>")+")",s={className:"type",begin:"\\b[a-z\\d_]*_t\\b"},o={className:"string",variants:[{begin:'(u8?|U|L)?"',end:'"',illegal:"\\n",contains:[e.BACKSLASH_ESCAPE]},{begin:"(u8?|U|L)?'(\\\\(x[0-9A-Fa-f]{2}|u[0-9A-Fa-f]{4,8}|[0-7]{3}|\\S)|.)",end:"'",illegal:"."},e.END_SAME_AS_BEGIN({begin:/(?:u8?|U|L)?R"([^()\\ ]{0,16})\(/,end:/\)([^()\\ ]{0,16})"/})]},l={className:"number",variants:[{begin:"\\b(0b[01']+)"},{begin:"(-?)\\b([\\d']+(\\.[\\d']*)?|\\.[\\d']+)((ll|LL|l|L)(u|U)?|(u|U)(ll|LL|l|L)?|f|F|b|B)"},{begin:"(-?)(\\b0[xX][a-fA-F0-9']+|(\\b[\\d']+(\\.[\\d']*)?|\\.[\\d']+)([eE][-+]?[\\d']+)?)"}],relevance:0},c={className:"meta",begin:/#\s*[a-z]+\b/,end:/$/,keywords:{keyword:"if else elif endif define undef warning error line pragma _Pragma ifdef ifndef include"},contains:[{begin:/\\\n/,relevance:0},e.inherit(o,{className:"string"}),{className:"string",begin:/<.*?>/},t,e.C_BLOCK_COMMENT_MODE]},d={className:"title",begin:n.optional(a)+e.IDENT_RE,relevance:0},u=n.optional(a)+e.IDENT_RE+"\\s*\\(",g={type:["bool","char","char16_t","char32_t","char8_t","double","float","int","long","short","void","wchar_t","unsigned","signed","const","static"],keyword:["alignas","alignof","and","and_eq","asm","atomic_cancel","atomic_commit","atomic_noexcept","auto","bitand","bitor","break","case","catch","class","co_await","co_return","co_yield","compl","concept","const_cast|10","consteval","constexpr","constinit","continue","decltype","default","delete","do","dynamic_cast|10","else","enum","explicit","export","extern","false","final","for","friend","goto","if","import","inline","module","mutable","namespace","new","noexcept","not","not_eq","nullptr","operator","or","or_eq","override","private","protected","public","reflexpr","register","reinterpret_cast|10","requires","return","sizeof","static_assert","static_cast|10","struct","switch","synchronized","template","this","thread_local","throw","transaction_safe","transaction_safe_dynamic","true","try","typedef","typeid","typename","union","using","virtual","volatile","while","xor","xor_eq"],literal:["NULL","false","nullopt","nullptr","true"],built_in:["_Pragma"],_type_hints:["any","auto_ptr","barrier","binary_semaphore","bitset","complex","condition_variable","condition_variable_any","counting_semaphore","deque","false_type","future","imaginary","initializer_list","istringstream","jthread","latch","lock_guard","multimap","multiset","mutex","optional","ostringstream","packaged_task","pair","promise","priority_queue","queue","recursive_mutex","recursive_timed_mutex","scoped_lock","set","shared_future","shared_lock","shared_mutex","shared_timed_mutex","shared_ptr","stack","string_view","stringstream","timed_mutex","thread","true_type","tuple","unique_lock","unique_ptr","unordered_map","unordered_multimap","unordered_multiset","unordered_set","variant","vector","weak_ptr","wstring","wstring_view"]},b={className:"function.dispatch",relevance:0,keywords:{_hint:["abort","abs","acos","apply","as_const","asin","atan","atan2","calloc","ceil","cerr","cin","clog","cos","cosh","cout","declval","endl","exchange","exit","exp","fabs","floor","fmod","forward","fprintf","fputs","free","frexp","fscanf","future","invoke","isalnum","isalpha","iscntrl","isdigit","isgraph","islower","isprint","ispunct","isspace","isupper","isxdigit","labs","launder","ldexp","log","log10","make_pair","make_shared","make_shared_for_overwrite","make_tuple","make_unique","malloc","memchr","memcmp","memcpy","memset","modf","move","pow","printf","putchar","puts","realloc","scanf","sin","sinh","snprintf","sprintf","sqrt","sscanf","std","stderr","stdin","stdout","strcat","strchr","strcmp","strcpy","strcspn","strlen","strncat","strncmp","strncpy","strpbrk","strrchr","strspn","strstr","swap","tan","tanh","terminate","to_underlying","tolower","toupper","vfprintf","visit","vprintf","vsprintf"]},begin:n.concat(/\b/,/(?!decltype)/,/(?!if)/,/(?!for)/,/(?!switch)/,/(?!while)/,e.IDENT_RE,n.lookahead(/(<[^<>]+>|)\s*\(/))},p=[b,c,s,t,e.C_BLOCK_COMMENT_MODE,l,o],f={variants:[{begin:/=/,end:/;/},{begin:/\(/,end:/\)/},{beginKeywords:"new throw return else",end:/;/}],keywords:g,contains:p.concat([{begin:/\(/,end:/\)/,keywords:g,contains:p.concat(["self"]),relevance:0}]),relevance:0},m={className:"function",begin:"("+r+"[\\*&\\s]+)+"+u,returnBegin:!0,end:/[{;=]/,excludeEnd:!0,keywords:g,illegal:/[^\w\s\*&:<>.]/,contains:[{begin:i,keywords:g,relevance:0},{begin:u,returnBegin:!0,contains:[d],relevance:0},{begin:/::/,relevance:0},{begin:/:/,endsWithParent:!0,contains:[o,l]},{relevance:0,match:/,/},{className:"params",begin:/\(/,end:/\)/,keywords:g,relevance:0,contains:[t,e.C_BLOCK_COMMENT_MODE,o,l,s,{begin:/\(/,end:/\)/,keywords:g,relevance:0,contains:["self",t,e.C_BLOCK_COMMENT_MODE,o,l,s]}]},s,t,e.C_BLOCK_COMMENT_MODE,c]};return{name:"C++",aliases:["cc","c++","h++","hpp","hh","hxx","cxx"],keywords:g,illegal:"",keywords:g,contains:["self",s]},{begin:e.IDENT_RE+"::",keywords:g},{match:[/\b(?:enum(?:\s+(?:class|struct))?|class|struct|union)/,/\s+/,/\w+/],className:{1:"keyword",3:"title.class"}}])}}},3147:e=>{e.exports=function(e){const n="(_?[ui](8|16|32|64|128))?",t="[a-zA-Z_]\\w*[!?=]?|[-+~]@|<<|>>|[=!]~|===?|<=>|[<>]=?|\\*\\*|[-/+%^&*~|]|//|//=|&[-+*]=?|&\\*\\*|\\[\\][=?]?",i="[A-Za-z_]\\w*(::\\w+)*(\\?|!)?",a={$pattern:"[a-zA-Z_]\\w*[!?=]?",keyword:"abstract alias annotation as as? asm begin break case class def do else elsif end ensure enum extend for fun if include instance_sizeof is_a? lib macro module next nil? of out pointerof private protected rescue responds_to? return require select self sizeof struct super then type typeof union uninitialized unless until verbatim when while with yield __DIR__ __END_LINE__ __FILE__ __LINE__",literal:"false nil true"},r={className:"subst",begin:/#\{/,end:/\}/,keywords:a},s={className:"template-variable",variants:[{begin:"\\{\\{",end:"\\}\\}"},{begin:"\\{%",end:"%\\}"}],keywords:a};function o(e,n){const t=[{begin:e,end:n}];return t[0].contains=t,t}const l={className:"string",contains:[e.BACKSLASH_ESCAPE,r],variants:[{begin:/'/,end:/'/},{begin:/"/,end:/"/},{begin:/`/,end:/`/},{begin:"%[Qwi]?\\(",end:"\\)",contains:o("\\(","\\)")},{begin:"%[Qwi]?\\[",end:"\\]",contains:o("\\[","\\]")},{begin:"%[Qwi]?\\{",end:/\}/,contains:o(/\{/,/\}/)},{begin:"%[Qwi]?<",end:">",contains:o("<",">")},{begin:"%[Qwi]?\\|",end:"\\|"},{begin:/<<-\w+$/,end:/^\s*\w+$/}],relevance:0},c={className:"string",variants:[{begin:"%q\\(",end:"\\)",contains:o("\\(","\\)")},{begin:"%q\\[",end:"\\]",contains:o("\\[","\\]")},{begin:"%q\\{",end:/\}/,contains:o(/\{/,/\}/)},{begin:"%q<",end:">",contains:o("<",">")},{begin:"%q\\|",end:"\\|"},{begin:/<<-'\w+'$/,end:/^\s*\w+$/}],relevance:0},d={begin:"(?!%\\})("+e.RE_STARTERS_RE+"|\\n|\\b(case|if|select|unless|until|when|while)\\b)\\s*",keywords:"case if select unless until when while",contains:[{className:"regexp",contains:[e.BACKSLASH_ESCAPE,r],variants:[{begin:"//[a-z]*",relevance:0},{begin:"/(?!\\/)",end:"/[a-z]*"}]}],relevance:0},u=[s,l,c,{className:"regexp",contains:[e.BACKSLASH_ESCAPE,r],variants:[{begin:"%r\\(",end:"\\)",contains:o("\\(","\\)")},{begin:"%r\\[",end:"\\]",contains:o("\\[","\\]")},{begin:"%r\\{",end:/\}/,contains:o(/\{/,/\}/)},{begin:"%r<",end:">",contains:o("<",">")},{begin:"%r\\|",end:"\\|"}],relevance:0},d,{className:"meta",begin:"@\\[",end:"\\]",contains:[e.inherit(e.QUOTE_STRING_MODE,{className:"string"})]},{className:"variable",begin:"(\\$\\W)|((\\$|@@?)(\\w+))(?=[^@$?])(?![A-Za-z])(?![@$?'])"},e.HASH_COMMENT_MODE,{className:"class",beginKeywords:"class module struct",end:"$|;",illegal:/=/,contains:[e.HASH_COMMENT_MODE,e.inherit(e.TITLE_MODE,{begin:i}),{begin:"<"}]},{className:"class",beginKeywords:"lib enum union",end:"$|;",illegal:/=/,contains:[e.HASH_COMMENT_MODE,e.inherit(e.TITLE_MODE,{begin:i})]},{beginKeywords:"annotation",end:"$|;",illegal:/=/,contains:[e.HASH_COMMENT_MODE,e.inherit(e.TITLE_MODE,{begin:i})],relevance:2},{className:"function",beginKeywords:"def",end:/\B\b/,contains:[e.inherit(e.TITLE_MODE,{begin:t,endsParent:!0})]},{className:"function",beginKeywords:"fun macro",end:/\B\b/,contains:[e.inherit(e.TITLE_MODE,{begin:t,endsParent:!0})],relevance:2},{className:"symbol",begin:e.UNDERSCORE_IDENT_RE+"(!|\\?)?:",relevance:0},{className:"symbol",begin:":",contains:[l,{begin:t}],relevance:0},{className:"number",variants:[{begin:"\\b0b([01_]+)"+n},{begin:"\\b0o([0-7_]+)"+n},{begin:"\\b0x([A-Fa-f0-9_]+)"+n},{begin:"\\b([1-9][0-9_]*[0-9]|[0-9])(\\.[0-9][0-9_]*)?([eE]_?[-+]?[0-9_]*)?(_?f(32|64))?(?!_)"},{begin:"\\b([1-9][0-9_]*|0)"+n}],relevance:0}];return r.contains=u,s.contains=u.slice(1),{name:"Crystal",aliases:["cr"],keywords:a,contains:u}}},3707:e=>{e.exports=function(e){const n={keyword:["abstract","as","base","break","case","catch","class","const","continue","do","else","event","explicit","extern","finally","fixed","for","foreach","goto","if","implicit","in","interface","internal","is","lock","namespace","new","operator","out","override","params","private","protected","public","readonly","record","ref","return","scoped","sealed","sizeof","stackalloc","static","struct","switch","this","throw","try","typeof","unchecked","unsafe","using","virtual","void","volatile","while"].concat(["add","alias","and","ascending","async","await","by","descending","equals","from","get","global","group","init","into","join","let","nameof","not","notnull","on","or","orderby","partial","remove","select","set","unmanaged","value|0","var","when","where","with","yield"]),built_in:["bool","byte","char","decimal","delegate","double","dynamic","enum","float","int","long","nint","nuint","object","sbyte","short","string","ulong","uint","ushort"],literal:["default","false","null","true"]},t=e.inherit(e.TITLE_MODE,{begin:"[a-zA-Z](\\.?\\w)*"}),i={className:"number",variants:[{begin:"\\b(0b[01']+)"},{begin:"(-?)\\b([\\d']+(\\.[\\d']*)?|\\.[\\d']+)(u|U|l|L|ul|UL|f|F|b|B)"},{begin:"(-?)(\\b0[xX][a-fA-F0-9']+|(\\b[\\d']+(\\.[\\d']*)?|\\.[\\d']+)([eE][-+]?[\\d']+)?)"}],relevance:0},a={className:"string",begin:'@"',end:'"',contains:[{begin:'""'}]},r=e.inherit(a,{illegal:/\n/}),s={className:"subst",begin:/\{/,end:/\}/,keywords:n},o=e.inherit(s,{illegal:/\n/}),l={className:"string",begin:/\$"/,end:'"',illegal:/\n/,contains:[{begin:/\{\{/},{begin:/\}\}/},e.BACKSLASH_ESCAPE,o]},c={className:"string",begin:/\$@"/,end:'"',contains:[{begin:/\{\{/},{begin:/\}\}/},{begin:'""'},s]},d=e.inherit(c,{illegal:/\n/,contains:[{begin:/\{\{/},{begin:/\}\}/},{begin:'""'},o]});s.contains=[c,l,a,e.APOS_STRING_MODE,e.QUOTE_STRING_MODE,i,e.C_BLOCK_COMMENT_MODE],o.contains=[d,l,r,e.APOS_STRING_MODE,e.QUOTE_STRING_MODE,i,e.inherit(e.C_BLOCK_COMMENT_MODE,{illegal:/\n/})];const u={variants:[c,l,a,e.APOS_STRING_MODE,e.QUOTE_STRING_MODE]},g={begin:"<",end:">",contains:[{beginKeywords:"in out"},t]},b=e.IDENT_RE+"(<"+e.IDENT_RE+"(\\s*,\\s*"+e.IDENT_RE+")*>)?(\\[\\])?",p={begin:"@"+e.IDENT_RE,relevance:0};return{name:"C#",aliases:["cs","c#"],keywords:n,illegal:/::/,contains:[e.COMMENT("///","$",{returnBegin:!0,contains:[{className:"doctag",variants:[{begin:"///",relevance:0},{begin:"\x3c!--|--\x3e"},{begin:""}]}]}),e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,{className:"meta",begin:"#",end:"$",keywords:{keyword:"if else elif endif define undef warning error line region endregion pragma checksum"}},u,i,{beginKeywords:"class interface",relevance:0,end:/[{;=]/,illegal:/[^\s:,]/,contains:[{beginKeywords:"where class"},t,g,e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE]},{beginKeywords:"namespace",relevance:0,end:/[{;=]/,illegal:/[^\s:]/,contains:[t,e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE]},{beginKeywords:"record",relevance:0,end:/[{;=]/,illegal:/[^\s:]/,contains:[t,g,e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE]},{className:"meta",begin:"^\\s*\\[(?=[\\w])",excludeBegin:!0,end:"\\]",excludeEnd:!0,contains:[{className:"string",begin:/"/,end:/"/}]},{beginKeywords:"new return throw await else",relevance:0},{className:"function",begin:"("+b+"\\s+)+"+e.IDENT_RE+"\\s*(<[^=]+>\\s*)?\\(",returnBegin:!0,end:/\s*[{;=]/,excludeEnd:!0,keywords:n,contains:[{beginKeywords:["public","private","protected","static","internal","protected","abstract","async","extern","override","unsafe","virtual","new","sealed","partial"].join(" "),relevance:0},{begin:e.IDENT_RE+"\\s*(<[^=]+>\\s*)?\\(",returnBegin:!0,contains:[e.TITLE_MODE,g],relevance:0},{match:/\(\)/},{className:"params",begin:/\(/,end:/\)/,excludeBegin:!0,excludeEnd:!0,keywords:n,relevance:0,contains:[u,i,e.C_BLOCK_COMMENT_MODE]},e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE]},p]}}},9534:e=>{e.exports=function(e){return{name:"CSP",case_insensitive:!1,keywords:{$pattern:"[a-zA-Z][a-zA-Z0-9_-]*",keyword:["base-uri","child-src","connect-src","default-src","font-src","form-action","frame-ancestors","frame-src","img-src","manifest-src","media-src","object-src","plugin-types","report-uri","sandbox","script-src","style-src","trusted-types","unsafe-hashes","worker-src"]},contains:[{className:"string",begin:"'",end:"'"},{className:"attribute",begin:"^Content",end:":",excludeEnd:!0}]}}},5064:e=>{const n=["a","abbr","address","article","aside","audio","b","blockquote","body","button","canvas","caption","cite","code","dd","del","details","dfn","div","dl","dt","em","fieldset","figcaption","figure","footer","form","h1","h2","h3","h4","h5","h6","header","hgroup","html","i","iframe","img","input","ins","kbd","label","legend","li","main","mark","menu","nav","object","ol","p","q","quote","samp","section","span","strong","summary","sup","table","tbody","td","textarea","tfoot","th","thead","time","tr","ul","var","video"],t=["any-hover","any-pointer","aspect-ratio","color","color-gamut","color-index","device-aspect-ratio","device-height","device-width","display-mode","forced-colors","grid","height","hover","inverted-colors","monochrome","orientation","overflow-block","overflow-inline","pointer","prefers-color-scheme","prefers-contrast","prefers-reduced-motion","prefers-reduced-transparency","resolution","scan","scripting","update","width","min-width","max-width","min-height","max-height"],i=["active","any-link","blank","checked","current","default","defined","dir","disabled","drop","empty","enabled","first","first-child","first-of-type","fullscreen","future","focus","focus-visible","focus-within","has","host","host-context","hover","indeterminate","in-range","invalid","is","lang","last-child","last-of-type","left","link","local-link","not","nth-child","nth-col","nth-last-child","nth-last-col","nth-last-of-type","nth-of-type","only-child","only-of-type","optional","out-of-range","past","placeholder-shown","read-only","read-write","required","right","root","scope","target","target-within","user-invalid","valid","visited","where"],a=["after","backdrop","before","cue","cue-region","first-letter","first-line","grammar-error","marker","part","placeholder","selection","slotted","spelling-error"],r=["align-content","align-items","align-self","all","animation","animation-delay","animation-direction","animation-duration","animation-fill-mode","animation-iteration-count","animation-name","animation-play-state","animation-timing-function","backface-visibility","background","background-attachment","background-blend-mode","background-clip","background-color","background-image","background-origin","background-position","background-repeat","background-size","block-size","border","border-block","border-block-color","border-block-end","border-block-end-color","border-block-end-style","border-block-end-width","border-block-start","border-block-start-color","border-block-start-style","border-block-start-width","border-block-style","border-block-width","border-bottom","border-bottom-color","border-bottom-left-radius","border-bottom-right-radius","border-bottom-style","border-bottom-width","border-collapse","border-color","border-image","border-image-outset","border-image-repeat","border-image-slice","border-image-source","border-image-width","border-inline","border-inline-color","border-inline-end","border-inline-end-color","border-inline-end-style","border-inline-end-width","border-inline-start","border-inline-start-color","border-inline-start-style","border-inline-start-width","border-inline-style","border-inline-width","border-left","border-left-color","border-left-style","border-left-width","border-radius","border-right","border-right-color","border-right-style","border-right-width","border-spacing","border-style","border-top","border-top-color","border-top-left-radius","border-top-right-radius","border-top-style","border-top-width","border-width","bottom","box-decoration-break","box-shadow","box-sizing","break-after","break-before","break-inside","caption-side","caret-color","clear","clip","clip-path","clip-rule","color","column-count","column-fill","column-gap","column-rule","column-rule-color","column-rule-style","column-rule-width","column-span","column-width","columns","contain","content","content-visibility","counter-increment","counter-reset","cue","cue-after","cue-before","cursor","direction","display","empty-cells","filter","flex","flex-basis","flex-direction","flex-flow","flex-grow","flex-shrink","flex-wrap","float","flow","font","font-display","font-family","font-feature-settings","font-kerning","font-language-override","font-size","font-size-adjust","font-smoothing","font-stretch","font-style","font-synthesis","font-variant","font-variant-caps","font-variant-east-asian","font-variant-ligatures","font-variant-numeric","font-variant-position","font-variation-settings","font-weight","gap","glyph-orientation-vertical","grid","grid-area","grid-auto-columns","grid-auto-flow","grid-auto-rows","grid-column","grid-column-end","grid-column-start","grid-gap","grid-row","grid-row-end","grid-row-start","grid-template","grid-template-areas","grid-template-columns","grid-template-rows","hanging-punctuation","height","hyphens","icon","image-orientation","image-rendering","image-resolution","ime-mode","inline-size","isolation","justify-content","left","letter-spacing","line-break","line-height","list-style","list-style-image","list-style-position","list-style-type","margin","margin-block","margin-block-end","margin-block-start","margin-bottom","margin-inline","margin-inline-end","margin-inline-start","margin-left","margin-right","margin-top","marks","mask","mask-border","mask-border-mode","mask-border-outset","mask-border-repeat","mask-border-slice","mask-border-source","mask-border-width","mask-clip","mask-composite","mask-image","mask-mode","mask-origin","mask-position","mask-repeat","mask-size","mask-type","max-block-size","max-height","max-inline-size","max-width","min-block-size","min-height","min-inline-size","min-width","mix-blend-mode","nav-down","nav-index","nav-left","nav-right","nav-up","none","normal","object-fit","object-position","opacity","order","orphans","outline","outline-color","outline-offset","outline-style","outline-width","overflow","overflow-wrap","overflow-x","overflow-y","padding","padding-block","padding-block-end","padding-block-start","padding-bottom","padding-inline","padding-inline-end","padding-inline-start","padding-left","padding-right","padding-top","page-break-after","page-break-before","page-break-inside","pause","pause-after","pause-before","perspective","perspective-origin","pointer-events","position","quotes","resize","rest","rest-after","rest-before","right","row-gap","scroll-margin","scroll-margin-block","scroll-margin-block-end","scroll-margin-block-start","scroll-margin-bottom","scroll-margin-inline","scroll-margin-inline-end","scroll-margin-inline-start","scroll-margin-left","scroll-margin-right","scroll-margin-top","scroll-padding","scroll-padding-block","scroll-padding-block-end","scroll-padding-block-start","scroll-padding-bottom","scroll-padding-inline","scroll-padding-inline-end","scroll-padding-inline-start","scroll-padding-left","scroll-padding-right","scroll-padding-top","scroll-snap-align","scroll-snap-stop","scroll-snap-type","scrollbar-color","scrollbar-gutter","scrollbar-width","shape-image-threshold","shape-margin","shape-outside","speak","speak-as","src","tab-size","table-layout","text-align","text-align-all","text-align-last","text-combine-upright","text-decoration","text-decoration-color","text-decoration-line","text-decoration-style","text-emphasis","text-emphasis-color","text-emphasis-position","text-emphasis-style","text-indent","text-justify","text-orientation","text-overflow","text-rendering","text-shadow","text-transform","text-underline-position","top","transform","transform-box","transform-origin","transform-style","transition","transition-delay","transition-duration","transition-property","transition-timing-function","unicode-bidi","vertical-align","visibility","voice-balance","voice-duration","voice-family","voice-pitch","voice-range","voice-rate","voice-stress","voice-volume","white-space","widows","width","will-change","word-break","word-spacing","word-wrap","writing-mode","z-index"].reverse();e.exports=function(e){const s=e.regex,o=(e=>({IMPORTANT:{scope:"meta",begin:"!important"},BLOCK_COMMENT:e.C_BLOCK_COMMENT_MODE,HEXCOLOR:{scope:"number",begin:/#(([0-9a-fA-F]{3,4})|(([0-9a-fA-F]{2}){3,4}))\b/},FUNCTION_DISPATCH:{className:"built_in",begin:/[\w-]+(?=\()/},ATTRIBUTE_SELECTOR_MODE:{scope:"selector-attr",begin:/\[/,end:/\]/,illegal:"$",contains:[e.APOS_STRING_MODE,e.QUOTE_STRING_MODE]},CSS_NUMBER_MODE:{scope:"number",begin:e.NUMBER_RE+"(%|em|ex|ch|rem|vw|vh|vmin|vmax|cm|mm|in|pt|pc|px|deg|grad|rad|turn|s|ms|Hz|kHz|dpi|dpcm|dppx)?",relevance:0},CSS_VARIABLE:{className:"attr",begin:/--[A-Za-z][A-Za-z0-9_-]*/}}))(e),l=[e.APOS_STRING_MODE,e.QUOTE_STRING_MODE];return{name:"CSS",case_insensitive:!0,illegal:/[=|'\$]/,keywords:{keyframePosition:"from to"},classNameAliases:{keyframePosition:"selector-tag"},contains:[o.BLOCK_COMMENT,{begin:/-(webkit|moz|ms|o)-(?=[a-z])/},o.CSS_NUMBER_MODE,{className:"selector-id",begin:/#[A-Za-z0-9_-]+/,relevance:0},{className:"selector-class",begin:"\\.[a-zA-Z-][a-zA-Z0-9_-]*",relevance:0},o.ATTRIBUTE_SELECTOR_MODE,{className:"selector-pseudo",variants:[{begin:":("+i.join("|")+")"},{begin:":(:)?("+a.join("|")+")"}]},o.CSS_VARIABLE,{className:"attribute",begin:"\\b("+r.join("|")+")\\b"},{begin:/:/,end:/[;}{]/,contains:[o.BLOCK_COMMENT,o.HEXCOLOR,o.IMPORTANT,o.CSS_NUMBER_MODE,...l,{begin:/(url|data-uri)\(/,end:/\)/,relevance:0,keywords:{built_in:"url data-uri"},contains:[...l,{className:"string",begin:/[^)]/,endsWithParent:!0,excludeEnd:!0}]},o.FUNCTION_DISPATCH]},{begin:s.lookahead(/@/),end:"[{;]",relevance:0,illegal:/:/,contains:[{className:"keyword",begin:/@-?\w[\w]*(-\w+)*/},{begin:/\s/,endsWithParent:!0,excludeEnd:!0,relevance:0,keywords:{$pattern:/[a-z-]+/,keyword:"and or not only",attribute:t.join(" ")},contains:[{begin:/[a-z-]+(?=:)/,className:"attribute"},...l,o.CSS_NUMBER_MODE]}]},{className:"selector-tag",begin:"\\b("+n.join("|")+")\\b"}]}}},6642:e=>{e.exports=function(e){const n={className:"subst",variants:[{begin:"\\$[A-Za-z0-9_]+"}]},t={className:"subst",variants:[{begin:/\$\{/,end:/\}/}],keywords:"true false null this is new super"},i={className:"string",variants:[{begin:"r'''",end:"'''"},{begin:'r"""',end:'"""'},{begin:"r'",end:"'",illegal:"\\n"},{begin:'r"',end:'"',illegal:"\\n"},{begin:"'''",end:"'''",contains:[e.BACKSLASH_ESCAPE,n,t]},{begin:'"""',end:'"""',contains:[e.BACKSLASH_ESCAPE,n,t]},{begin:"'",end:"'",illegal:"\\n",contains:[e.BACKSLASH_ESCAPE,n,t]},{begin:'"',end:'"',illegal:"\\n",contains:[e.BACKSLASH_ESCAPE,n,t]}]};t.contains=[e.C_NUMBER_MODE,i];const a=["Comparable","DateTime","Duration","Function","Iterable","Iterator","List","Map","Match","Object","Pattern","RegExp","Set","Stopwatch","String","StringBuffer","StringSink","Symbol","Type","Uri","bool","double","int","num","Element","ElementList"],r=a.map((e=>`${e}?`));return{name:"Dart",keywords:{keyword:["abstract","as","assert","async","await","base","break","case","catch","class","const","continue","covariant","default","deferred","do","dynamic","else","enum","export","extends","extension","external","factory","false","final","finally","for","Function","get","hide","if","implements","import","in","interface","is","late","library","mixin","new","null","on","operator","part","required","rethrow","return","sealed","set","show","static","super","switch","sync","this","throw","true","try","typedef","var","void","when","while","with","yield"],built_in:a.concat(r).concat(["Never","Null","dynamic","print","document","querySelector","querySelectorAll","window"]),$pattern:/[A-Za-z][A-Za-z0-9_]*\??/},contains:[i,e.COMMENT(/\/\*\*(?!\/)/,/\*\//,{subLanguage:"markdown",relevance:0}),e.COMMENT(/\/{3,} ?/,/$/,{contains:[{subLanguage:"markdown",begin:".",end:"$",relevance:0}]}),e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,{className:"class",beginKeywords:"class interface",end:/\{/,excludeEnd:!0,contains:[{beginKeywords:"extends implements"},e.UNDERSCORE_TITLE_MODE]},e.C_NUMBER_MODE,{className:"meta",begin:"@[A-Za-z]+"},{begin:"=>"}]}}},7731:e=>{e.exports=function(e){const n=e.regex;return{name:"Diff",aliases:["patch"],contains:[{className:"meta",relevance:10,match:n.either(/^@@ +-\d+,\d+ +\+\d+,\d+ +@@/,/^\*\*\* +\d+,\d+ +\*\*\*\*$/,/^--- +\d+,\d+ +----$/)},{className:"comment",variants:[{begin:n.either(/Index: /,/^index/,/={3,}/,/^-{3}/,/^\*{3} /,/^\+{3}/,/^diff --git/),end:/$/},{match:/^\*{15}$/}]},{className:"addition",begin:/^\+/,end:/$/},{className:"deletion",begin:/^-/,end:/$/},{className:"addition",begin:/^!/,end:/$/}]}}},7360:e=>{e.exports=function(e){return{name:"Dockerfile",aliases:["docker"],case_insensitive:!0,keywords:["from","maintainer","expose","env","arg","user","onbuild","stopsignal"],contains:[e.HASH_COMMENT_MODE,e.APOS_STRING_MODE,e.QUOTE_STRING_MODE,e.NUMBER_MODE,{beginKeywords:"run cmd entrypoint volume add copy workdir label healthcheck shell",starts:{end:/[^\\]$/,subLanguage:"bash"}}],illegal:"{e.exports=function(e){const n=e.regex,t="[a-zA-Z_][a-zA-Z0-9_.]*(!|\\?)?",i={$pattern:t,keyword:["after","alias","and","case","catch","cond","defstruct","defguard","do","else","end","fn","for","if","import","in","not","or","quote","raise","receive","require","reraise","rescue","try","unless","unquote","unquote_splicing","use","when","with|0"],literal:["false","nil","true"]},a={className:"subst",begin:/#\{/,end:/\}/,keywords:i},r={match:/\\[\s\S]/,scope:"char.escape",relevance:0},s="[/|([{<\"']",o=[{begin:/"/,end:/"/},{begin:/'/,end:/'/},{begin:/\//,end:/\//},{begin:/\|/,end:/\|/},{begin:/\(/,end:/\)/},{begin:/\[/,end:/\]/},{begin:/\{/,end:/\}/},{begin://}],l=e=>({scope:"char.escape",begin:n.concat(/\\/,e),relevance:0}),c={className:"string",begin:"~[a-z](?="+s+")",contains:o.map((n=>e.inherit(n,{contains:[l(n.end),r,a]})))},d={className:"string",begin:"~[A-Z](?="+s+")",contains:o.map((n=>e.inherit(n,{contains:[l(n.end)]})))},u={className:"regex",variants:[{begin:"~r(?="+s+")",contains:o.map((t=>e.inherit(t,{end:n.concat(t.end,/[uismxfU]{0,7}/),contains:[l(t.end),r,a]})))},{begin:"~R(?="+s+")",contains:o.map((t=>e.inherit(t,{end:n.concat(t.end,/[uismxfU]{0,7}/),contains:[l(t.end)]})))}]},g={className:"string",contains:[e.BACKSLASH_ESCAPE,a],variants:[{begin:/"""/,end:/"""/},{begin:/'''/,end:/'''/},{begin:/~S"""/,end:/"""/,contains:[]},{begin:/~S"/,end:/"/,contains:[]},{begin:/~S'''/,end:/'''/,contains:[]},{begin:/~S'/,end:/'/,contains:[]},{begin:/'/,end:/'/},{begin:/"/,end:/"/}]},b={className:"function",beginKeywords:"def defp defmacro defmacrop",end:/\B\b/,contains:[e.inherit(e.TITLE_MODE,{begin:t,endsParent:!0})]},p=e.inherit(b,{className:"class",beginKeywords:"defimpl defmodule defprotocol defrecord",end:/\bdo\b|$|;/}),f=[g,u,d,c,e.HASH_COMMENT_MODE,p,b,{begin:"::"},{className:"symbol",begin:":(?![\\s:])",contains:[g,{begin:"[a-zA-Z_]\\w*[!?=]?|[-+~]@|<<|>>|=~|===?|<=>|[<>]=?|\\*\\*|[-/+%^&*~`|]|\\[\\]=?"}],relevance:0},{className:"symbol",begin:t+":(?!:)",relevance:0},{className:"title.class",begin:/(\b[A-Z][a-zA-Z0-9_]+)/,relevance:0},{className:"number",begin:"(\\b0o[0-7_]+)|(\\b0b[01_]+)|(\\b0x[0-9a-fA-F_]+)|(-?\\b[0-9][0-9_]*(\\.[0-9_]+([eE][-+]?[0-9]+)?)?)",relevance:0},{className:"variable",begin:"(\\$\\W)|((\\$|@@?)(\\w+))"}];return a.contains=f,{name:"Elixir",aliases:["ex","exs"],keywords:i,contains:f}}},5658:e=>{e.exports=function(e){const n={variants:[e.COMMENT("--","$"),e.COMMENT(/\{-/,/-\}/,{contains:["self"]})]},t={className:"type",begin:"\\b[A-Z][\\w']*",relevance:0},i={begin:"\\(",end:"\\)",illegal:'"',contains:[{className:"type",begin:"\\b[A-Z][\\w]*(\\((\\.\\.|,|\\w+)\\))?"},n]};return{name:"Elm",keywords:["let","in","if","then","else","case","of","where","module","import","exposing","type","alias","as","infix","infixl","infixr","port","effect","command","subscription"],contains:[{beginKeywords:"port effect module",end:"exposing",keywords:"port effect module where command subscription exposing",contains:[i,n],illegal:"\\W\\.|;"},{begin:"import",end:"$",keywords:"import as exposing",contains:[i,n],illegal:"\\W\\.|;"},{begin:"type",end:"$",keywords:"type alias",contains:[t,i,{begin:/\{/,end:/\}/,contains:i.contains},n]},{beginKeywords:"infix infixl infixr",end:"$",contains:[e.C_NUMBER_MODE,n]},{begin:"port",end:"$",keywords:"port",contains:[n]},{className:"string",begin:"'\\\\?.",end:"'",illegal:"."},e.QUOTE_STRING_MODE,e.C_NUMBER_MODE,t,e.inherit(e.TITLE_MODE,{begin:"^[_a-z][\\w']*"}),n,{begin:"->|<-"}],illegal:/;/}}},7569:e=>{e.exports=function(e){const n="[a-z'][a-zA-Z0-9_']*",t="("+n+":"+n+"|"+n+")",i={keyword:"after and andalso|10 band begin bnot bor bsl bzr bxor case catch cond div end fun if let not of orelse|10 query receive rem try when xor",literal:"false true"},a=e.COMMENT("%","$"),r={className:"number",begin:"\\b(\\d+(_\\d+)*#[a-fA-F0-9]+(_[a-fA-F0-9]+)*|\\d+(_\\d+)*(\\.\\d+(_\\d+)*)?([eE][-+]?\\d+)?)",relevance:0},s={begin:"fun\\s+"+n+"/\\d+"},o={begin:t+"\\(",end:"\\)",returnBegin:!0,relevance:0,contains:[{begin:t,relevance:0},{begin:"\\(",end:"\\)",endsWithParent:!0,returnEnd:!0,relevance:0}]},l={begin:/\{/,end:/\}/,relevance:0},c={begin:"\\b_([A-Z][A-Za-z0-9_]*)?",relevance:0},d={begin:"[A-Z][a-zA-Z0-9_]*",relevance:0},u={begin:"#"+e.UNDERSCORE_IDENT_RE,relevance:0,returnBegin:!0,contains:[{begin:"#"+e.UNDERSCORE_IDENT_RE,relevance:0},{begin:/\{/,end:/\}/,relevance:0}]},g={beginKeywords:"fun receive if try case",end:"end",keywords:i};g.contains=[a,s,e.inherit(e.APOS_STRING_MODE,{className:""}),g,o,e.QUOTE_STRING_MODE,r,l,c,d,u];const b=[a,s,g,o,e.QUOTE_STRING_MODE,r,l,c,d,u];o.contains[1].contains=b,l.contains=b,u.contains[1].contains=b;const p={className:"params",begin:"\\(",end:"\\)",contains:b};return{name:"Erlang",aliases:["erl"],keywords:i,illegal:"(",returnBegin:!0,illegal:"\\(|#|//|/\\*|\\\\|:|;",contains:[p,e.inherit(e.TITLE_MODE,{begin:n})],starts:{end:";|\\.",keywords:i,contains:b}},a,{begin:"^-",end:"\\.",relevance:0,excludeEnd:!0,returnBegin:!0,keywords:{$pattern:"-"+e.IDENT_RE,keyword:["-module","-record","-undef","-export","-ifdef","-ifndef","-author","-copyright","-doc","-vsn","-import","-include","-include_lib","-compile","-define","-else","-endif","-file","-behaviour","-behavior","-spec"].map((e=>`${e}|1.5`)).join(" ")},contains:[p]},r,e.QUOTE_STRING_MODE,u,c,d,l,{begin:/\.$/}]}}},6652:e=>{function n(e){return new RegExp(e.replace(/[-/\\^$*+?.()|[\]{}]/g,"\\$&"),"m")}function t(e){return e?"string"==typeof e?e:e.source:null}function i(e){return a("(?=",e,")")}function a(...e){return e.map((e=>t(e))).join("")}function r(...e){const n=function(e){const n=e[e.length-1];return"object"==typeof n&&n.constructor===Object?(e.splice(e.length-1,1),n):{}}(e);return"("+(n.capture?"":"?:")+e.map((e=>t(e))).join("|")+")"}e.exports=function(e){const t={scope:"keyword",match:/\b(yield|return|let|do|match|use)!/},s=["bool","byte","sbyte","int8","int16","int32","uint8","uint16","uint32","int","uint","int64","uint64","nativeint","unativeint","decimal","float","double","float32","single","char","string","unit","bigint","option","voption","list","array","seq","byref","exn","inref","nativeptr","obj","outref","voidptr","Result"],o={keyword:["abstract","and","as","assert","base","begin","class","default","delegate","do","done","downcast","downto","elif","else","end","exception","extern","finally","fixed","for","fun","function","global","if","in","inherit","inline","interface","internal","lazy","let","match","member","module","mutable","namespace","new","of","open","or","override","private","public","rec","return","static","struct","then","to","try","type","upcast","use","val","void","when","while","with","yield"],literal:["true","false","null","Some","None","Ok","Error","infinity","infinityf","nan","nanf"],built_in:["not","ref","raise","reraise","dict","readOnlyDict","set","get","enum","sizeof","typeof","typedefof","nameof","nullArg","invalidArg","invalidOp","id","fst","snd","ignore","lock","using","box","unbox","tryUnbox","printf","printfn","sprintf","eprintf","eprintfn","fprintf","fprintfn","failwith","failwithf"],"variable.constant":["__LINE__","__SOURCE_DIRECTORY__","__SOURCE_FILE__"]},l={variants:[e.COMMENT(/\(\*(?!\))/,/\*\)/,{contains:["self"]}),e.C_LINE_COMMENT_MODE]},c={scope:"variable",begin:/``/,end:/``/},d=/\B('|\^)/,u={scope:"symbol",variants:[{match:a(d,/``.*?``/)},{match:a(d,e.UNDERSCORE_IDENT_RE)}],relevance:0},g=function({includeEqual:e}){let t;t=e?"!%&*+-/<=>@^|~?":"!%&*+-/<>@^|~?";const s=a("[",...Array.from(t).map(n),"]"),o=r(s,/\./),l=a(o,i(o)),c=r(a(l,o,"*"),a(s,"+"));return{scope:"operator",match:r(c,/:\?>/,/:\?/,/:>/,/:=/,/::?/,/\$/),relevance:0}},b=g({includeEqual:!0}),p=g({includeEqual:!1}),f=function(n,t){return{begin:a(n,i(a(/\s*/,r(/\w/,/'/,/\^/,/#/,/``/,/\(/,/{\|/)))),beginScope:t,end:i(r(/\n/,/=/)),relevance:0,keywords:e.inherit(o,{type:s}),contains:[l,u,e.inherit(c,{scope:null}),p]}},m=f(/:/,"operator"),E=f(/\bof\b/,"keyword"),h={begin:[/(^|\s+)/,/type/,/\s+/,/[a-zA-Z_](\w|')*/],beginScope:{2:"keyword",4:"title.class"},end:i(/\(|=|$/),keywords:o,contains:[l,e.inherit(c,{scope:null}),u,{scope:"operator",match:/<|>/},m]},_={scope:"computation-expression",match:/\b[_a-z]\w*(?=\s*\{)/},N={begin:[/^\s*/,a(/#/,r("if","else","endif","line","nowarn","light","r","i","I","load","time","help","quit")),/\b/],beginScope:{2:"meta"},end:i(/\s|$/)},T={variants:[e.BINARY_NUMBER_MODE,e.C_NUMBER_MODE]},O={scope:"string",begin:/"/,end:/"/,contains:[e.BACKSLASH_ESCAPE]},A={scope:"string",begin:/@"/,end:/"/,contains:[{match:/""/},e.BACKSLASH_ESCAPE]},v={scope:"string",begin:/"""/,end:/"""/,relevance:2},y={scope:"subst",begin:/\{/,end:/\}/,keywords:o},S={scope:"string",begin:/\$"/,end:/"/,contains:[{match:/\{\{/},{match:/\}\}/},e.BACKSLASH_ESCAPE,y]},R={scope:"string",begin:/(\$@|@\$)"/,end:/"/,contains:[{match:/\{\{/},{match:/\}\}/},{match:/""/},e.BACKSLASH_ESCAPE,y]},I={scope:"string",begin:/\$"""/,end:/"""/,contains:[{match:/\{\{/},{match:/\}\}/},y],relevance:2},w={scope:"string",match:a(/'/,r(/[^\\']/,/\\(?:.|\d{3}|x[a-fA-F\d]{2}|u[a-fA-F\d]{4}|U[a-fA-F\d]{8})/),/'/)};return y.contains=[R,S,A,O,w,t,l,c,m,_,N,T,u,b],{name:"F#",aliases:["fs","f#"],keywords:o,illegal:/\/\*/,classNameAliases:{"computation-expression":"keyword"},contains:[t,{variants:[I,R,S,v,A,O,w]},l,c,h,{scope:"meta",begin:/\[\]/,relevance:2,contains:[c,v,A,O,w,T]},E,m,_,N,T,u,b]}}},2399:e=>{e.exports=function(e){const n={keyword:["break","case","chan","const","continue","default","defer","else","fallthrough","for","func","go","goto","if","import","interface","map","package","range","return","select","struct","switch","type","var"],type:["bool","byte","complex64","complex128","error","float32","float64","int8","int16","int32","int64","string","uint8","uint16","uint32","uint64","int","uint","uintptr","rune"],literal:["true","false","iota","nil"],built_in:["append","cap","close","complex","copy","imag","len","make","new","panic","print","println","real","recover","delete"]};return{name:"Go",aliases:["golang"],keywords:n,illegal:"{e.exports=function(e){return{name:"Gradle",case_insensitive:!0,keywords:["task","project","allprojects","subprojects","artifacts","buildscript","configurations","dependencies","repositories","sourceSets","description","delete","from","into","include","exclude","source","classpath","destinationDir","includes","options","sourceCompatibility","targetCompatibility","group","flatDir","doLast","doFirst","flatten","todir","fromdir","ant","def","abstract","break","case","catch","continue","default","do","else","extends","final","finally","for","if","implements","instanceof","native","new","private","protected","public","return","static","switch","synchronized","throw","throws","transient","try","volatile","while","strictfp","package","import","false","null","super","this","true","antlrtask","checkstyle","codenarc","copy","boolean","byte","char","class","double","float","int","interface","long","short","void","compile","runTime","file","fileTree","abs","any","append","asList","asWritable","call","collect","compareTo","count","div","dump","each","eachByte","eachFile","eachLine","every","find","findAll","flatten","getAt","getErr","getIn","getOut","getText","grep","immutable","inject","inspect","intersect","invokeMethods","isCase","join","leftShift","minus","multiply","newInputStream","newOutputStream","newPrintWriter","newReader","newWriter","next","plus","pop","power","previous","print","println","push","putAt","read","readBytes","readLines","reverse","reverseEach","round","size","sort","splitEachLine","step","subMap","times","toInteger","toList","tokenize","upto","waitForOrKill","withPrintWriter","withReader","withStream","withWriter","withWriterAppend","write","writeLine"],contains:[e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,e.APOS_STRING_MODE,e.QUOTE_STRING_MODE,e.NUMBER_MODE,e.REGEXP_MODE]}}},4658:e=>{function n(e,n={}){return n.variants=e,n}e.exports=function(e){const t=e.regex,i="[A-Za-z0-9_$]+",a=n([e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,e.COMMENT("/\\*\\*","\\*/",{relevance:0,contains:[{begin:/\w+@/,relevance:0},{className:"doctag",begin:"@[A-Za-z]+"}]})]),r={className:"regexp",begin:/~?\/[^\/\n]+\//,contains:[e.BACKSLASH_ESCAPE]},s=n([e.BINARY_NUMBER_MODE,e.C_NUMBER_MODE]),o=n([{begin:/"""/,end:/"""/},{begin:/'''/,end:/'''/},{begin:"\\$/",end:"/\\$",relevance:10},e.APOS_STRING_MODE,e.QUOTE_STRING_MODE],{className:"string"}),l={match:[/(class|interface|trait|enum|extends|implements)/,/\s+/,e.UNDERSCORE_IDENT_RE],scope:{1:"keyword",3:"title.class"}};return{name:"Groovy",keywords:{"variable.language":"this super",literal:"true false null",type:["byte","short","char","int","long","boolean","float","double","void"],keyword:["def","as","in","assert","trait","abstract","static","volatile","transient","public","private","protected","synchronized","final","class","interface","enum","if","else","for","while","switch","case","break","default","continue","throw","throws","try","catch","finally","implements","extends","new","import","package","return","instanceof"]},contains:[e.SHEBANG({binary:"groovy",relevance:10}),a,o,r,s,l,{className:"meta",begin:"@[A-Za-z]+",relevance:0},{className:"attr",begin:i+"[ \t]*:",relevance:0},{begin:/\?/,end:/:/,relevance:0,contains:[a,o,r,s,"self"]},{className:"symbol",begin:"^[ \t]*"+t.lookahead(i+":"),excludeBegin:!0,end:i+":",relevance:0}],illegal:/#|<\//}}},1407:e=>{e.exports=function(e){const n=e.regex,t={$pattern:/[\w.\/]+/,built_in:["action","bindattr","collection","component","concat","debugger","each","each-in","get","hash","if","in","input","link-to","loc","log","lookup","mut","outlet","partial","query-params","render","template","textarea","unbound","unless","view","with","yield"]},i=/\[\]|\[[^\]]+\]/,a=/[^\s!"#%&'()*+,.\/;<=>@\[\\\]^`{|}~]+/,r=n.either(/""|"[^"]+"/,/''|'[^']+'/,i,a),s=n.concat(n.optional(/\.|\.\/|\//),r,n.anyNumberOfTimes(n.concat(/(\.|\/)/,r))),o=n.concat("(",i,"|",a,")(?==)"),l={begin:s},c=e.inherit(l,{keywords:{$pattern:/[\w.\/]+/,literal:["true","false","undefined","null"]}}),d={begin:/\(/,end:/\)/},u={className:"attr",begin:o,relevance:0,starts:{begin:/=/,end:/=/,starts:{contains:[e.NUMBER_MODE,e.QUOTE_STRING_MODE,e.APOS_STRING_MODE,c,d]}}},g={contains:[e.NUMBER_MODE,e.QUOTE_STRING_MODE,e.APOS_STRING_MODE,{begin:/as\s+\|/,keywords:{keyword:"as"},end:/\|/,contains:[{begin:/\w+/}]},u,c,d],returnEnd:!0},b=e.inherit(l,{className:"name",keywords:t,starts:e.inherit(g,{end:/\)/})});d.contains=[b];const p=e.inherit(l,{keywords:t,className:"name",starts:e.inherit(g,{end:/\}\}/})}),f=e.inherit(l,{keywords:t,className:"name"}),m=e.inherit(l,{className:"name",keywords:t,starts:e.inherit(g,{end:/\}\}/})});return{name:"Handlebars",aliases:["hbs","html.hbs","html.handlebars","htmlbars"],case_insensitive:!0,subLanguage:"xml",contains:[{begin:/\\\{\{/,skip:!0},{begin:/\\\\(?=\{\{)/,skip:!0},e.COMMENT(/\{\{!--/,/--\}\}/),e.COMMENT(/\{\{!/,/\}\}/),{className:"template-tag",begin:/\{\{\{\{(?!\/)/,end:/\}\}\}\}/,contains:[p],starts:{end:/\{\{\{\{\//,returnEnd:!0,subLanguage:"xml"}},{className:"template-tag",begin:/\{\{\{\{\//,end:/\}\}\}\}/,contains:[f]},{className:"template-tag",begin:/\{\{#/,end:/\}\}/,contains:[p]},{className:"template-tag",begin:/\{\{(?=else\}\})/,end:/\}\}/,keywords:"else"},{className:"template-tag",begin:/\{\{(?=else if)/,end:/\}\}/,keywords:"else if"},{className:"template-tag",begin:/\{\{\//,end:/\}\}/,contains:[f]},{className:"template-variable",begin:/\{\{\{/,end:/\}\}\}/,contains:[m]},{className:"template-variable",begin:/\{\{/,end:/\}\}/,contains:[m]}]}}},7077:e=>{e.exports=function(e){const n={variants:[e.COMMENT("--","$"),e.COMMENT(/\{-/,/-\}/,{contains:["self"]})]},t={className:"meta",begin:/\{-#/,end:/#-\}/},i={className:"meta",begin:"^#",end:"$"},a={className:"type",begin:"\\b[A-Z][\\w']*",relevance:0},r={begin:"\\(",end:"\\)",illegal:'"',contains:[t,i,{className:"type",begin:"\\b[A-Z][\\w]*(\\((\\.\\.|,|\\w+)\\))?"},e.inherit(e.TITLE_MODE,{begin:"[_a-z][\\w']*"}),n]},s="([0-9]_*)+",o="([0-9a-fA-F]_*)+",l={className:"number",relevance:0,variants:[{match:`\\b(${s})(\\.(${s}))?([eE][+-]?(${s}))?\\b`},{match:`\\b0[xX]_*(${o})(\\.(${o}))?([pP][+-]?(${s}))?\\b`},{match:"\\b0[oO](([0-7]_*)+)\\b"},{match:"\\b0[bB](([01]_*)+)\\b"}]};return{name:"Haskell",aliases:["hs"],keywords:"let in if then else case of where do module import hiding qualified type data newtype deriving class instance as default infix infixl infixr foreign export ccall stdcall cplusplus jvm dotnet safe unsafe family forall mdo proc rec",contains:[{beginKeywords:"module",end:"where",keywords:"module where",contains:[r,n],illegal:"\\W\\.|;"},{begin:"\\bimport\\b",end:"$",keywords:"import qualified as hiding",contains:[r,n],illegal:"\\W\\.|;"},{className:"class",begin:"^(\\s*)?(class|instance)\\b",end:"where",keywords:"class family instance where",contains:[a,r,n]},{className:"class",begin:"\\b(data|(new)?type)\\b",end:"$",keywords:"data family type newtype deriving",contains:[t,a,r,{begin:/\{/,end:/\}/,contains:r.contains},n]},{beginKeywords:"default",end:"$",contains:[a,r,n]},{beginKeywords:"infix infixl infixr",end:"$",contains:[e.C_NUMBER_MODE,n]},{begin:"\\bforeign\\b",end:"$",keywords:"foreign import export ccall stdcall cplusplus jvm dotnet safe unsafe",contains:[a,e.QUOTE_STRING_MODE,n]},{className:"meta",begin:"#!\\/usr\\/bin\\/env runhaskell",end:"$"},t,i,{scope:"string",begin:/'(?=\\?.')/,end:/'/,contains:[{scope:"char.escape",match:/\\./}]},e.QUOTE_STRING_MODE,l,a,e.inherit(e.TITLE_MODE,{begin:"^[_a-z][\\w']*"}),n,{begin:"->|<-"}]}}},4762:e=>{e.exports=function(e){const n=e.regex,t={className:"number",relevance:0,variants:[{begin:/([+-]+)?[\d]+_[\d_]+/},{begin:e.NUMBER_RE}]},i=e.COMMENT();i.variants=[{begin:/;/,end:/$/},{begin:/#/,end:/$/}];const a={className:"variable",variants:[{begin:/\$[\w\d"][\w\d_]*/},{begin:/\$\{(.*?)\}/}]},r={className:"literal",begin:/\bon|off|true|false|yes|no\b/},s={className:"string",contains:[e.BACKSLASH_ESCAPE],variants:[{begin:"'''",end:"'''",relevance:10},{begin:'"""',end:'"""',relevance:10},{begin:'"',end:'"'},{begin:"'",end:"'"}]},o={begin:/\[/,end:/\]/,contains:[i,r,a,s,t,"self"],relevance:0},l=n.either(/[A-Za-z0-9_-]+/,/"(\\"|[^"])*"/,/'[^']*'/);return{name:"TOML, also INI",aliases:["toml"],case_insensitive:!0,illegal:/\S/,contains:[i,{className:"section",begin:/\[+/,end:/\]+/},{begin:n.concat(l,"(\\s*\\.\\s*",l,")*",n.lookahead(/\s*=\s*[^#\s]/)),className:"attr",starts:{end:/$/,contains:[i,o,r,a,s,t]}}]}}},8257:e=>{var n="[0-9](_*[0-9])*",t=`\\.(${n})`,i="[0-9a-fA-F](_*[0-9a-fA-F])*",a={className:"number",variants:[{begin:`(\\b(${n})((${t})|\\.)?|(${t}))[eE][+-]?(${n})[fFdD]?\\b`},{begin:`\\b(${n})((${t})[fFdD]?\\b|\\.([fFdD]\\b)?)`},{begin:`(${t})[fFdD]?\\b`},{begin:`\\b(${n})[fFdD]\\b`},{begin:`\\b0[xX]((${i})\\.?|(${i})?\\.(${i}))[pP][+-]?(${n})[fFdD]?\\b`},{begin:"\\b(0|[1-9](_*[0-9])*)[lL]?\\b"},{begin:`\\b0[xX](${i})[lL]?\\b`},{begin:"\\b0(_*[0-7])*[lL]?\\b"},{begin:"\\b0[bB][01](_*[01])*[lL]?\\b"}],relevance:0};function r(e,n,t){return-1===t?"":e.replace(n,(i=>r(e,n,t-1)))}e.exports=function(e){const n=e.regex,t="[À-ʸa-zA-Z_$][À-ʸa-zA-Z_$0-9]*",i=t+r("(?:<"+t+"~~~(?:\\s*,\\s*"+t+"~~~)*>)?",/~~~/g,2),s={keyword:["synchronized","abstract","private","var","static","if","const ","for","while","strictfp","finally","protected","import","native","final","void","enum","else","break","transient","catch","instanceof","volatile","case","assert","package","default","public","try","switch","continue","throws","protected","public","private","module","requires","exports","do","sealed","yield","permits"],literal:["false","true","null"],type:["char","boolean","long","float","int","byte","short","double"],built_in:["super","this"]},o={className:"meta",begin:"@"+t,contains:[{begin:/\(/,end:/\)/,contains:["self"]}]},l={className:"params",begin:/\(/,end:/\)/,keywords:s,relevance:0,contains:[e.C_BLOCK_COMMENT_MODE],endsParent:!0};return{name:"Java",aliases:["jsp"],keywords:s,illegal:/<\/|#/,contains:[e.COMMENT("/\\*\\*","\\*/",{relevance:0,contains:[{begin:/\w+@/,relevance:0},{className:"doctag",begin:"@[A-Za-z]+"}]}),{begin:/import java\.[a-z]+\./,keywords:"import",relevance:2},e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,{begin:/"""/,end:/"""/,className:"string",contains:[e.BACKSLASH_ESCAPE]},e.APOS_STRING_MODE,e.QUOTE_STRING_MODE,{match:[/\b(?:class|interface|enum|extends|implements|new)/,/\s+/,t],className:{1:"keyword",3:"title.class"}},{match:/non-sealed/,scope:"keyword"},{begin:[n.concat(/(?!else)/,t),/\s+/,t,/\s+/,/=(?!=)/],className:{1:"type",3:"variable",5:"operator"}},{begin:[/record/,/\s+/,t],className:{1:"keyword",3:"title.class"},contains:[l,e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE]},{beginKeywords:"new throw return else",relevance:0},{begin:["(?:"+i+"\\s+)",e.UNDERSCORE_IDENT_RE,/\s*(?=\()/],className:{2:"title.function"},keywords:s,contains:[{className:"params",begin:/\(/,end:/\)/,keywords:s,relevance:0,contains:[o,e.APOS_STRING_MODE,e.QUOTE_STRING_MODE,a,e.C_BLOCK_COMMENT_MODE]},e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE]},a,o]}}},978:e=>{const n="[A-Za-z$_][0-9A-Za-z$_]*",t=["as","in","of","if","for","while","finally","var","new","function","do","return","void","else","break","catch","instanceof","with","throw","case","default","try","switch","continue","typeof","delete","let","yield","const","class","debugger","async","await","static","import","from","export","extends"],i=["true","false","null","undefined","NaN","Infinity"],a=["Object","Function","Boolean","Symbol","Math","Date","Number","BigInt","String","RegExp","Array","Float32Array","Float64Array","Int8Array","Uint8Array","Uint8ClampedArray","Int16Array","Int32Array","Uint16Array","Uint32Array","BigInt64Array","BigUint64Array","Set","Map","WeakSet","WeakMap","ArrayBuffer","SharedArrayBuffer","Atomics","DataView","JSON","Promise","Generator","GeneratorFunction","AsyncFunction","Reflect","Proxy","Intl","WebAssembly"],r=["Error","EvalError","InternalError","RangeError","ReferenceError","SyntaxError","TypeError","URIError"],s=["setInterval","setTimeout","clearInterval","clearTimeout","require","exports","eval","isFinite","isNaN","parseFloat","parseInt","decodeURI","decodeURIComponent","encodeURI","encodeURIComponent","escape","unescape"],o=["arguments","this","super","console","window","document","localStorage","sessionStorage","module","global"],l=[].concat(s,a,r);e.exports=function(e){const c=e.regex,d=n,u={begin:/<[A-Za-z0-9\\._:-]+/,end:/\/[A-Za-z0-9\\._:-]+>|\/>/,isTrulyOpeningTag:(e,n)=>{const t=e[0].length+e.index,i=e.input[t];if("<"===i||","===i)return void n.ignoreMatch();let a;">"===i&&(((e,{after:n})=>{const t="",k={match:[/const|var|let/,/\s+/,d,/\s*/,/=\s*/,/(async\s*)?/,c.lookahead(x)],keywords:"async",className:{1:"keyword",3:"title.function"},contains:[S]};return{name:"JavaScript",aliases:["js","jsx","mjs","cjs"],keywords:g,exports:{PARAMS_CONTAINS:y,CLASS_REFERENCE:I},illegal:/#(?![$_A-z])/,contains:[e.SHEBANG({label:"shebang",binary:"node",relevance:5}),{label:"use_strict",className:"meta",relevance:10,begin:/^\s*['"]use (strict|asm)['"]/},e.APOS_STRING_MODE,e.QUOTE_STRING_MODE,h,_,N,T,O,{match:/\$\d+/},m,I,{className:"attr",begin:d+c.lookahead(":"),relevance:0},k,{begin:"("+e.RE_STARTERS_RE+"|\\b(case|return|throw)\\b)\\s*",keywords:"return throw case",relevance:0,contains:[O,e.REGEXP_MODE,{className:"function",begin:x,returnBegin:!0,end:"\\s*=>",contains:[{className:"params",variants:[{begin:e.UNDERSCORE_IDENT_RE,relevance:0},{className:null,begin:/\(\s*\)/,skip:!0},{begin:/\(/,end:/\)/,excludeBegin:!0,excludeEnd:!0,keywords:g,contains:y}]}]},{begin:/,/,relevance:0},{match:/\s+/,relevance:0},{variants:[{begin:"<>",end:""},{match:/<[A-Za-z0-9\\._:-]+\s*\/>/},{begin:u.begin,"on:begin":u.isTrulyOpeningTag,end:u.end}],subLanguage:"xml",contains:[{begin:u.begin,end:u.end,skip:!0,contains:["self"]}]}]},w,{beginKeywords:"while if switch catch for"},{begin:"\\b(?!function)"+e.UNDERSCORE_IDENT_RE+"\\([^()]*(\\([^()]*(\\([^()]*\\)[^()]*)*\\)[^()]*)*\\)\\s*\\{",returnBegin:!0,label:"func.def",contains:[S,e.inherit(e.TITLE_MODE,{begin:d,className:"title.function"})]},{match:/\.\.\./,relevance:0},M,{match:"\\$"+d,relevance:0},{match:[/\bconstructor(?=\s*\()/],className:{1:"title.function"},contains:[S]},C,{relevance:0,match:/\b[A-Z][A-Z_0-9]+\b/,className:"variable.constant"},R,D,{match:/\$[(.]/}]}}},14:e=>{e.exports=function(e){const n=["true","false","null"],t={scope:"literal",beginKeywords:n.join(" ")};return{name:"JSON",keywords:{literal:n},contains:[{className:"attr",begin:/"(\\.|[^\\"\r\n])*"(?=\s*:)/,relevance:1.01},{match:/[{}[\],:]/,className:"punctuation",relevance:0},e.QUOTE_STRING_MODE,t,e.C_NUMBER_MODE,e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE],illegal:"\\S"}}},5812:e=>{var n="[0-9](_*[0-9])*",t=`\\.(${n})`,i="[0-9a-fA-F](_*[0-9a-fA-F])*",a={className:"number",variants:[{begin:`(\\b(${n})((${t})|\\.)?|(${t}))[eE][+-]?(${n})[fFdD]?\\b`},{begin:`\\b(${n})((${t})[fFdD]?\\b|\\.([fFdD]\\b)?)`},{begin:`(${t})[fFdD]?\\b`},{begin:`\\b(${n})[fFdD]\\b`},{begin:`\\b0[xX]((${i})\\.?|(${i})?\\.(${i}))[pP][+-]?(${n})[fFdD]?\\b`},{begin:"\\b(0|[1-9](_*[0-9])*)[lL]?\\b"},{begin:`\\b0[xX](${i})[lL]?\\b`},{begin:"\\b0(_*[0-7])*[lL]?\\b"},{begin:"\\b0[bB][01](_*[01])*[lL]?\\b"}],relevance:0};e.exports=function(e){const n={keyword:"abstract as val var vararg get set class object open private protected public noinline crossinline dynamic final enum if else do while for when throw try catch finally import package is in fun override companion reified inline lateinit init interface annotation data sealed internal infix operator out by constructor super tailrec where const inner suspend typealias external expect actual",built_in:"Byte Short Char Int Long Boolean Float Double Void Unit Nothing",literal:"true false null"},t={className:"symbol",begin:e.UNDERSCORE_IDENT_RE+"@"},i={className:"subst",begin:/\$\{/,end:/\}/,contains:[e.C_NUMBER_MODE]},r={className:"variable",begin:"\\$"+e.UNDERSCORE_IDENT_RE},s={className:"string",variants:[{begin:'"""',end:'"""(?=[^"])',contains:[r,i]},{begin:"'",end:"'",illegal:/\n/,contains:[e.BACKSLASH_ESCAPE]},{begin:'"',end:'"',illegal:/\n/,contains:[e.BACKSLASH_ESCAPE,r,i]}]};i.contains.push(s);const o={className:"meta",begin:"@(?:file|property|field|get|set|receiver|param|setparam|delegate)\\s*:(?:\\s*"+e.UNDERSCORE_IDENT_RE+")?"},l={className:"meta",begin:"@"+e.UNDERSCORE_IDENT_RE,contains:[{begin:/\(/,end:/\)/,contains:[e.inherit(s,{className:"string"}),"self"]}]},c=a,d=e.COMMENT("/\\*","\\*/",{contains:[e.C_BLOCK_COMMENT_MODE]}),u={variants:[{className:"type",begin:e.UNDERSCORE_IDENT_RE},{begin:/\(/,end:/\)/,contains:[]}]},g=u;return g.variants[1].contains=[u],u.variants[1].contains=[g],{name:"Kotlin",aliases:["kt","kts"],keywords:n,contains:[e.COMMENT("/\\*\\*","\\*/",{relevance:0,contains:[{className:"doctag",begin:"@[A-Za-z]+"}]}),e.C_LINE_COMMENT_MODE,d,{className:"keyword",begin:/\b(break|continue|return|this)\b/,starts:{contains:[{className:"symbol",begin:/@\w+/}]}},t,o,l,{className:"function",beginKeywords:"fun",end:"[(]|$",returnBegin:!0,excludeEnd:!0,keywords:n,relevance:5,contains:[{begin:e.UNDERSCORE_IDENT_RE+"\\s*\\(",returnBegin:!0,relevance:0,contains:[e.UNDERSCORE_TITLE_MODE]},{className:"type",begin://,keywords:"reified",relevance:0},{className:"params",begin:/\(/,end:/\)/,endsParent:!0,keywords:n,relevance:0,contains:[{begin:/:/,end:/[=,\/]/,endsWithParent:!0,contains:[u,e.C_LINE_COMMENT_MODE,d],relevance:0},e.C_LINE_COMMENT_MODE,d,o,l,s,e.C_NUMBER_MODE]},d]},{begin:[/class|interface|trait/,/\s+/,e.UNDERSCORE_IDENT_RE],beginScope:{3:"title.class"},keywords:"class interface trait",end:/[:\{(]|$/,excludeEnd:!0,illegal:"extends implements",contains:[{beginKeywords:"public protected internal private constructor"},e.UNDERSCORE_TITLE_MODE,{className:"type",begin://,excludeBegin:!0,excludeEnd:!0,relevance:0},{className:"type",begin:/[,:]\s*/,end:/[<\(,){\s]|$/,excludeBegin:!0,returnEnd:!0},o,l]},s,{className:"meta",begin:"^#!/usr/bin/env",end:"$",illegal:"\n"},c]}}},4210:e=>{const n=["a","abbr","address","article","aside","audio","b","blockquote","body","button","canvas","caption","cite","code","dd","del","details","dfn","div","dl","dt","em","fieldset","figcaption","figure","footer","form","h1","h2","h3","h4","h5","h6","header","hgroup","html","i","iframe","img","input","ins","kbd","label","legend","li","main","mark","menu","nav","object","ol","p","q","quote","samp","section","span","strong","summary","sup","table","tbody","td","textarea","tfoot","th","thead","time","tr","ul","var","video"],t=["any-hover","any-pointer","aspect-ratio","color","color-gamut","color-index","device-aspect-ratio","device-height","device-width","display-mode","forced-colors","grid","height","hover","inverted-colors","monochrome","orientation","overflow-block","overflow-inline","pointer","prefers-color-scheme","prefers-contrast","prefers-reduced-motion","prefers-reduced-transparency","resolution","scan","scripting","update","width","min-width","max-width","min-height","max-height"],i=["active","any-link","blank","checked","current","default","defined","dir","disabled","drop","empty","enabled","first","first-child","first-of-type","fullscreen","future","focus","focus-visible","focus-within","has","host","host-context","hover","indeterminate","in-range","invalid","is","lang","last-child","last-of-type","left","link","local-link","not","nth-child","nth-col","nth-last-child","nth-last-col","nth-last-of-type","nth-of-type","only-child","only-of-type","optional","out-of-range","past","placeholder-shown","read-only","read-write","required","right","root","scope","target","target-within","user-invalid","valid","visited","where"],a=["after","backdrop","before","cue","cue-region","first-letter","first-line","grammar-error","marker","part","placeholder","selection","slotted","spelling-error"],r=["align-content","align-items","align-self","all","animation","animation-delay","animation-direction","animation-duration","animation-fill-mode","animation-iteration-count","animation-name","animation-play-state","animation-timing-function","backface-visibility","background","background-attachment","background-blend-mode","background-clip","background-color","background-image","background-origin","background-position","background-repeat","background-size","block-size","border","border-block","border-block-color","border-block-end","border-block-end-color","border-block-end-style","border-block-end-width","border-block-start","border-block-start-color","border-block-start-style","border-block-start-width","border-block-style","border-block-width","border-bottom","border-bottom-color","border-bottom-left-radius","border-bottom-right-radius","border-bottom-style","border-bottom-width","border-collapse","border-color","border-image","border-image-outset","border-image-repeat","border-image-slice","border-image-source","border-image-width","border-inline","border-inline-color","border-inline-end","border-inline-end-color","border-inline-end-style","border-inline-end-width","border-inline-start","border-inline-start-color","border-inline-start-style","border-inline-start-width","border-inline-style","border-inline-width","border-left","border-left-color","border-left-style","border-left-width","border-radius","border-right","border-right-color","border-right-style","border-right-width","border-spacing","border-style","border-top","border-top-color","border-top-left-radius","border-top-right-radius","border-top-style","border-top-width","border-width","bottom","box-decoration-break","box-shadow","box-sizing","break-after","break-before","break-inside","caption-side","caret-color","clear","clip","clip-path","clip-rule","color","column-count","column-fill","column-gap","column-rule","column-rule-color","column-rule-style","column-rule-width","column-span","column-width","columns","contain","content","content-visibility","counter-increment","counter-reset","cue","cue-after","cue-before","cursor","direction","display","empty-cells","filter","flex","flex-basis","flex-direction","flex-flow","flex-grow","flex-shrink","flex-wrap","float","flow","font","font-display","font-family","font-feature-settings","font-kerning","font-language-override","font-size","font-size-adjust","font-smoothing","font-stretch","font-style","font-synthesis","font-variant","font-variant-caps","font-variant-east-asian","font-variant-ligatures","font-variant-numeric","font-variant-position","font-variation-settings","font-weight","gap","glyph-orientation-vertical","grid","grid-area","grid-auto-columns","grid-auto-flow","grid-auto-rows","grid-column","grid-column-end","grid-column-start","grid-gap","grid-row","grid-row-end","grid-row-start","grid-template","grid-template-areas","grid-template-columns","grid-template-rows","hanging-punctuation","height","hyphens","icon","image-orientation","image-rendering","image-resolution","ime-mode","inline-size","isolation","justify-content","left","letter-spacing","line-break","line-height","list-style","list-style-image","list-style-position","list-style-type","margin","margin-block","margin-block-end","margin-block-start","margin-bottom","margin-inline","margin-inline-end","margin-inline-start","margin-left","margin-right","margin-top","marks","mask","mask-border","mask-border-mode","mask-border-outset","mask-border-repeat","mask-border-slice","mask-border-source","mask-border-width","mask-clip","mask-composite","mask-image","mask-mode","mask-origin","mask-position","mask-repeat","mask-size","mask-type","max-block-size","max-height","max-inline-size","max-width","min-block-size","min-height","min-inline-size","min-width","mix-blend-mode","nav-down","nav-index","nav-left","nav-right","nav-up","none","normal","object-fit","object-position","opacity","order","orphans","outline","outline-color","outline-offset","outline-style","outline-width","overflow","overflow-wrap","overflow-x","overflow-y","padding","padding-block","padding-block-end","padding-block-start","padding-bottom","padding-inline","padding-inline-end","padding-inline-start","padding-left","padding-right","padding-top","page-break-after","page-break-before","page-break-inside","pause","pause-after","pause-before","perspective","perspective-origin","pointer-events","position","quotes","resize","rest","rest-after","rest-before","right","row-gap","scroll-margin","scroll-margin-block","scroll-margin-block-end","scroll-margin-block-start","scroll-margin-bottom","scroll-margin-inline","scroll-margin-inline-end","scroll-margin-inline-start","scroll-margin-left","scroll-margin-right","scroll-margin-top","scroll-padding","scroll-padding-block","scroll-padding-block-end","scroll-padding-block-start","scroll-padding-bottom","scroll-padding-inline","scroll-padding-inline-end","scroll-padding-inline-start","scroll-padding-left","scroll-padding-right","scroll-padding-top","scroll-snap-align","scroll-snap-stop","scroll-snap-type","scrollbar-color","scrollbar-gutter","scrollbar-width","shape-image-threshold","shape-margin","shape-outside","speak","speak-as","src","tab-size","table-layout","text-align","text-align-all","text-align-last","text-combine-upright","text-decoration","text-decoration-color","text-decoration-line","text-decoration-style","text-emphasis","text-emphasis-color","text-emphasis-position","text-emphasis-style","text-indent","text-justify","text-orientation","text-overflow","text-rendering","text-shadow","text-transform","text-underline-position","top","transform","transform-box","transform-origin","transform-style","transition","transition-delay","transition-duration","transition-property","transition-timing-function","unicode-bidi","vertical-align","visibility","voice-balance","voice-duration","voice-family","voice-pitch","voice-range","voice-rate","voice-stress","voice-volume","white-space","widows","width","will-change","word-break","word-spacing","word-wrap","writing-mode","z-index"].reverse(),s=i.concat(a);e.exports=function(e){const o=(e=>({IMPORTANT:{scope:"meta",begin:"!important"},BLOCK_COMMENT:e.C_BLOCK_COMMENT_MODE,HEXCOLOR:{scope:"number",begin:/#(([0-9a-fA-F]{3,4})|(([0-9a-fA-F]{2}){3,4}))\b/},FUNCTION_DISPATCH:{className:"built_in",begin:/[\w-]+(?=\()/},ATTRIBUTE_SELECTOR_MODE:{scope:"selector-attr",begin:/\[/,end:/\]/,illegal:"$",contains:[e.APOS_STRING_MODE,e.QUOTE_STRING_MODE]},CSS_NUMBER_MODE:{scope:"number",begin:e.NUMBER_RE+"(%|em|ex|ch|rem|vw|vh|vmin|vmax|cm|mm|in|pt|pc|px|deg|grad|rad|turn|s|ms|Hz|kHz|dpi|dpcm|dppx)?",relevance:0},CSS_VARIABLE:{className:"attr",begin:/--[A-Za-z][A-Za-z0-9_-]*/}}))(e),l=s,c="[\\w-]+",d="("+c+"|@\\{"+c+"\\})",u=[],g=[],b=function(e){return{className:"string",begin:"~?"+e+".*?"+e}},p=function(e,n,t){return{className:e,begin:n,relevance:t}},f={$pattern:/[a-z-]+/,keyword:"and or not only",attribute:t.join(" ")},m={begin:"\\(",end:"\\)",contains:g,keywords:f,relevance:0};g.push(e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,b("'"),b('"'),o.CSS_NUMBER_MODE,{begin:"(url|data-uri)\\(",starts:{className:"string",end:"[\\)\\n]",excludeEnd:!0}},o.HEXCOLOR,m,p("variable","@@?"+c,10),p("variable","@\\{"+c+"\\}"),p("built_in","~?`[^`]*?`"),{className:"attribute",begin:c+"\\s*:",end:":",returnBegin:!0,excludeEnd:!0},o.IMPORTANT,{beginKeywords:"and not"},o.FUNCTION_DISPATCH);const E=g.concat({begin:/\{/,end:/\}/,contains:u}),h={beginKeywords:"when",endsWithParent:!0,contains:[{beginKeywords:"and not"}].concat(g)},_={begin:d+"\\s*:",returnBegin:!0,end:/[;}]/,relevance:0,contains:[{begin:/-(webkit|moz|ms|o)-/},o.CSS_VARIABLE,{className:"attribute",begin:"\\b("+r.join("|")+")\\b",end:/(?=:)/,starts:{endsWithParent:!0,illegal:"[<=$]",relevance:0,contains:g}}]},N={className:"keyword",begin:"@(import|media|charset|font-face|(-[a-z]+-)?keyframes|supports|document|namespace|page|viewport|host)\\b",starts:{end:"[;{}]",keywords:f,returnEnd:!0,contains:g,relevance:0}},T={className:"variable",variants:[{begin:"@"+c+"\\s*:",relevance:15},{begin:"@"+c}],starts:{end:"[;}]",returnEnd:!0,contains:E}},O={variants:[{begin:"[\\.#:&\\[>]",end:"[;{}]"},{begin:d,end:/\{/}],returnBegin:!0,returnEnd:!0,illegal:"[<='$\"]",relevance:0,contains:[e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,h,p("keyword","all\\b"),p("variable","@\\{"+c+"\\}"),{begin:"\\b("+n.join("|")+")\\b",className:"selector-tag"},o.CSS_NUMBER_MODE,p("selector-tag",d,0),p("selector-id","#"+d),p("selector-class","\\."+d,0),p("selector-tag","&",0),o.ATTRIBUTE_SELECTOR_MODE,{className:"selector-pseudo",begin:":("+i.join("|")+")"},{className:"selector-pseudo",begin:":(:)?("+a.join("|")+")"},{begin:/\(/,end:/\)/,relevance:0,contains:E},{begin:"!important"},o.FUNCTION_DISPATCH]},A={begin:c+":(:)?"+`(${l.join("|")})`,returnBegin:!0,contains:[O]};return u.push(e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,N,T,A,_,O,h,o.FUNCTION_DISPATCH),{name:"Less",case_insensitive:!0,illegal:"[=>'/<($\"]",contains:u}}},1943:e=>{e.exports=function(e){const n="[a-zA-Z_\\-+\\*\\/<=>&#][a-zA-Z0-9_\\-+*\\/<=>&#!]*",t="\\|[^]*?\\|",i="(-|\\+)?\\d+(\\.\\d+|\\/\\d+)?((d|e|f|l|s|D|E|F|L|S)(\\+|-)?\\d+)?",a={className:"literal",begin:"\\b(t{1}|nil)\\b"},r={className:"number",variants:[{begin:i,relevance:0},{begin:"#(b|B)[0-1]+(/[0-1]+)?"},{begin:"#(o|O)[0-7]+(/[0-7]+)?"},{begin:"#(x|X)[0-9a-fA-F]+(/[0-9a-fA-F]+)?"},{begin:"#(c|C)\\("+i+" +"+i,end:"\\)"}]},s=e.inherit(e.QUOTE_STRING_MODE,{illegal:null}),o=e.COMMENT(";","$",{relevance:0}),l={begin:"\\*",end:"\\*"},c={className:"symbol",begin:"[:&]"+n},d={begin:n,relevance:0},u={begin:t},g={contains:[r,s,l,c,{begin:"\\(",end:"\\)",contains:["self",a,s,r,d]},d],variants:[{begin:"['`]\\(",end:"\\)"},{begin:"\\(quote ",end:"\\)",keywords:{name:"quote"}},{begin:"'"+t}]},b={variants:[{begin:"'"+n},{begin:"#'"+n+"(::"+n+")*"}]},p={begin:"\\(\\s*",end:"\\)"},f={endsWithParent:!0,relevance:0};return p.contains=[{className:"name",variants:[{begin:n,relevance:0},{begin:t}]},f],f.contains=[g,b,p,a,r,s,o,l,c,u,d],{name:"Lisp",illegal:/\S/,contains:[r,e.SHEBANG(),a,s,o,g,b,p,d]}}},4981:e=>{e.exports=function(e){const n="\\[=*\\[",t="\\]=*\\]",i={begin:n,end:t,contains:["self"]},a=[e.COMMENT("--(?!"+n+")","$"),e.COMMENT("--"+n,t,{contains:[i],relevance:10})];return{name:"Lua",keywords:{$pattern:e.UNDERSCORE_IDENT_RE,literal:"true false nil",keyword:"and break do else elseif end for goto if in local not or repeat return then until while",built_in:"_G _ENV _VERSION __index __newindex __mode __call __metatable __tostring __len __gc __add __sub __mul __div __mod __pow __concat __unm __eq __lt __le assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall arg self coroutine resume yield status wrap create running debug getupvalue debug sethook getmetatable gethook setmetatable setlocal traceback setfenv getinfo setupvalue getlocal getregistry getfenv io lines write close flush open output type read stderr stdin input stdout popen tmpfile math log max acos huge ldexp pi cos tanh pow deg tan cosh sinh random randomseed frexp ceil floor rad abs sqrt modf asin min mod fmod log10 atan2 exp sin atan os exit setlocale date getenv difftime remove time clock tmpname rename execute package preload loadlib loaded loaders cpath config path seeall string sub upper len gfind rep find match char dump gmatch reverse byte format gsub lower table setn insert getn foreachi maxn foreach concat sort remove"},contains:a.concat([{className:"function",beginKeywords:"function",end:"\\)",contains:[e.inherit(e.TITLE_MODE,{begin:"([_a-zA-Z]\\w*\\.)*([_a-zA-Z]\\w*:)?[_a-zA-Z]\\w*"}),{className:"params",begin:"\\(",endsWithParent:!0,contains:a}].concat(a)},e.C_NUMBER_MODE,e.APOS_STRING_MODE,e.QUOTE_STRING_MODE,{className:"string",begin:n,end:t,contains:[i],relevance:5}])}}},7903:e=>{e.exports=function(e){const n={className:"variable",variants:[{begin:"\\$\\("+e.UNDERSCORE_IDENT_RE+"\\)",contains:[e.BACKSLASH_ESCAPE]},{begin:/\$[@%{e.exports=function(e){const n={begin:/<\/?[A-Za-z_]/,end:">",subLanguage:"xml",relevance:0},t={variants:[{begin:/\[.+?\]\[.*?\]/,relevance:0},{begin:/\[.+?\]\(((data|javascript|mailto):|(?:http|ftp)s?:\/\/).*?\)/,relevance:2},{begin:e.regex.concat(/\[.+?\]\(/,/[A-Za-z][A-Za-z0-9+.-]*/,/:\/\/.*?\)/),relevance:2},{begin:/\[.+?\]\([./?&#].*?\)/,relevance:1},{begin:/\[.*?\]\(.*?\)/,relevance:0}],returnBegin:!0,contains:[{match:/\[(?=\])/},{className:"string",relevance:0,begin:"\\[",end:"\\]",excludeBegin:!0,returnEnd:!0},{className:"link",relevance:0,begin:"\\]\\(",end:"\\)",excludeBegin:!0,excludeEnd:!0},{className:"symbol",relevance:0,begin:"\\]\\[",end:"\\]",excludeBegin:!0,excludeEnd:!0}]},i={className:"strong",contains:[],variants:[{begin:/_{2}(?!\s)/,end:/_{2}/},{begin:/\*{2}(?!\s)/,end:/\*{2}/}]},a={className:"emphasis",contains:[],variants:[{begin:/\*(?![*\s])/,end:/\*/},{begin:/_(?![_\s])/,end:/_/,relevance:0}]},r=e.inherit(i,{contains:[]}),s=e.inherit(a,{contains:[]});i.contains.push(s),a.contains.push(r);let o=[n,t];return[i,a,r,s].forEach((e=>{e.contains=e.contains.concat(o)})),o=o.concat(i,a),{name:"Markdown",aliases:["md","mkdown","mkd"],contains:[{className:"section",variants:[{begin:"^#{1,6}",end:"$",contains:o},{begin:"(?=^.+?\\n[=-]{2,}$)",contains:[{begin:"^[=-]*$"},{begin:"^",end:"\\n",contains:o}]}]},n,{className:"bullet",begin:"^[ \t]*([*+-]|(\\d+\\.))(?=\\s+)",end:"\\s+",excludeEnd:!0},i,a,{className:"quote",begin:"^>\\s+",contains:o,end:"$"},{className:"code",variants:[{begin:"(`{3,})[^`](.|\\n)*?\\1`*[ ]*"},{begin:"(~{3,})[^~](.|\\n)*?\\1~*[ ]*"},{begin:"```",end:"```+[ ]*$"},{begin:"~~~",end:"~~~+[ ]*$"},{begin:"`.+?`"},{begin:"(?=^( {4}|\\t))",contains:[{begin:"^( {4}|\\t)",end:"(\\n)$"}],relevance:0}]},{begin:"^[-\\*]{3,}",end:"$"},t,{begin:/^\[[^\n]+\]:/,returnBegin:!0,contains:[{className:"symbol",begin:/\[/,end:/\]/,excludeBegin:!0,excludeEnd:!0},{className:"link",begin:/:\s*/,end:/$/,excludeBegin:!0}]}]}}},4028:e=>{e.exports=function(e){const n=e.regex,t={className:"variable",variants:[{begin:/\$\d+/},{begin:/\$\{\w+\}/},{begin:n.concat(/[$@]/,e.UNDERSCORE_IDENT_RE)}]},i={endsWithParent:!0,keywords:{$pattern:/[a-z_]{2,}|\/dev\/poll/,literal:["on","off","yes","no","true","false","none","blocked","debug","info","notice","warn","error","crit","select","break","last","permanent","redirect","kqueue","rtsig","epoll","poll","/dev/poll"]},relevance:0,illegal:"=>",contains:[e.HASH_COMMENT_MODE,{className:"string",contains:[e.BACKSLASH_ESCAPE,t],variants:[{begin:/"/,end:/"/},{begin:/'/,end:/'/}]},{begin:"([a-z]+):/",end:"\\s",endsWithParent:!0,excludeEnd:!0,contains:[t]},{className:"regexp",contains:[e.BACKSLASH_ESCAPE,t],variants:[{begin:"\\s\\^",end:"\\s|\\{|;",returnEnd:!0},{begin:"~\\*?\\s+",end:"\\s|\\{|;",returnEnd:!0},{begin:"\\*(\\.[a-z\\-]+)+"},{begin:"([a-z\\-]+\\.)+\\*"}]},{className:"number",begin:"\\b\\d{1,3}\\.\\d{1,3}\\.\\d{1,3}\\.\\d{1,3}(:\\d{1,5})?\\b"},{className:"number",begin:"\\b\\d+[kKmMgGdshdwy]?\\b",relevance:0},t]};return{name:"Nginx config",aliases:["nginxconf"],contains:[e.HASH_COMMENT_MODE,{beginKeywords:"upstream location",end:/;|\{/,contains:i.contains,keywords:{section:"upstream location"}},{className:"section",begin:n.concat(e.UNDERSCORE_IDENT_RE+n.lookahead(/\s+\{/)),relevance:0},{begin:n.lookahead(e.UNDERSCORE_IDENT_RE+"\\s"),end:";|\\{",contains:[{className:"attribute",begin:e.UNDERSCORE_IDENT_RE,starts:i}],relevance:0}],illegal:"[^\\s\\}\\{]"}}},2446:e=>{e.exports=function(e){const n=/[a-zA-Z@][a-zA-Z0-9_]*/,t={$pattern:n,keyword:["@interface","@class","@protocol","@implementation"]};return{name:"Objective-C",aliases:["mm","objc","obj-c","obj-c++","objective-c++"],keywords:{"variable.language":["this","super"],$pattern:n,keyword:["while","export","sizeof","typedef","const","struct","for","union","volatile","static","mutable","if","do","return","goto","enum","else","break","extern","asm","case","default","register","explicit","typename","switch","continue","inline","readonly","assign","readwrite","self","@synchronized","id","typeof","nonatomic","IBOutlet","IBAction","strong","weak","copy","in","out","inout","bycopy","byref","oneway","__strong","__weak","__block","__autoreleasing","@private","@protected","@public","@try","@property","@end","@throw","@catch","@finally","@autoreleasepool","@synthesize","@dynamic","@selector","@optional","@required","@encode","@package","@import","@defs","@compatibility_alias","__bridge","__bridge_transfer","__bridge_retained","__bridge_retain","__covariant","__contravariant","__kindof","_Nonnull","_Nullable","_Null_unspecified","__FUNCTION__","__PRETTY_FUNCTION__","__attribute__","getter","setter","retain","unsafe_unretained","nonnull","nullable","null_unspecified","null_resettable","class","instancetype","NS_DESIGNATED_INITIALIZER","NS_UNAVAILABLE","NS_REQUIRES_SUPER","NS_RETURNS_INNER_POINTER","NS_INLINE","NS_AVAILABLE","NS_DEPRECATED","NS_ENUM","NS_OPTIONS","NS_SWIFT_UNAVAILABLE","NS_ASSUME_NONNULL_BEGIN","NS_ASSUME_NONNULL_END","NS_REFINED_FOR_SWIFT","NS_SWIFT_NAME","NS_SWIFT_NOTHROW","NS_DURING","NS_HANDLER","NS_ENDHANDLER","NS_VALUERETURN","NS_VOIDRETURN"],literal:["false","true","FALSE","TRUE","nil","YES","NO","NULL"],built_in:["dispatch_once_t","dispatch_queue_t","dispatch_sync","dispatch_async","dispatch_once"],type:["int","float","char","unsigned","signed","short","long","double","wchar_t","unichar","void","bool","BOOL","id|0","_Bool"]},illegal:"/,end:/$/,illegal:"\\n"},e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE]},{className:"class",begin:"("+t.keyword.join("|")+")\\b",end:/(\{|$)/,excludeEnd:!0,keywords:t,contains:[e.UNDERSCORE_TITLE_MODE]},{begin:"\\."+e.UNDERSCORE_IDENT_RE,relevance:0}]}}},2482:e=>{e.exports=function(e){const n=e.regex,t=/[dualxmsipngr]{0,12}/,i={$pattern:/[\w.]+/,keyword:["abs","accept","alarm","and","atan2","bind","binmode","bless","break","caller","chdir","chmod","chomp","chop","chown","chr","chroot","close","closedir","connect","continue","cos","crypt","dbmclose","dbmopen","defined","delete","die","do","dump","each","else","elsif","endgrent","endhostent","endnetent","endprotoent","endpwent","endservent","eof","eval","exec","exists","exit","exp","fcntl","fileno","flock","for","foreach","fork","format","formline","getc","getgrent","getgrgid","getgrnam","gethostbyaddr","gethostbyname","gethostent","getlogin","getnetbyaddr","getnetbyname","getnetent","getpeername","getpgrp","getpriority","getprotobyname","getprotobynumber","getprotoent","getpwent","getpwnam","getpwuid","getservbyname","getservbyport","getservent","getsockname","getsockopt","given","glob","gmtime","goto","grep","gt","hex","if","index","int","ioctl","join","keys","kill","last","lc","lcfirst","length","link","listen","local","localtime","log","lstat","lt","ma","map","mkdir","msgctl","msgget","msgrcv","msgsnd","my","ne","next","no","not","oct","open","opendir","or","ord","our","pack","package","pipe","pop","pos","print","printf","prototype","push","q|0","qq","quotemeta","qw","qx","rand","read","readdir","readline","readlink","readpipe","recv","redo","ref","rename","require","reset","return","reverse","rewinddir","rindex","rmdir","say","scalar","seek","seekdir","select","semctl","semget","semop","send","setgrent","sethostent","setnetent","setpgrp","setpriority","setprotoent","setpwent","setservent","setsockopt","shift","shmctl","shmget","shmread","shmwrite","shutdown","sin","sleep","socket","socketpair","sort","splice","split","sprintf","sqrt","srand","stat","state","study","sub","substr","symlink","syscall","sysopen","sysread","sysseek","system","syswrite","tell","telldir","tie","tied","time","times","tr","truncate","uc","ucfirst","umask","undef","unless","unlink","unpack","unshift","untie","until","use","utime","values","vec","wait","waitpid","wantarray","warn","when","while","write","x|0","xor","y|0"].join(" ")},a={className:"subst",begin:"[$@]\\{",end:"\\}",keywords:i},r={begin:/->\{/,end:/\}/},s={variants:[{begin:/\$\d/},{begin:n.concat(/[$%@](\^\w\b|#\w+(::\w+)*|\{\w+\}|\w+(::\w*)*)/,"(?![A-Za-z])(?![@$%])")},{begin:/[$%@][^\s\w{]/,relevance:0}]},o=[e.BACKSLASH_ESCAPE,a,s],l=[/!/,/\//,/\|/,/\?/,/'/,/"/,/#/],c=(e,i,a="\\1")=>{const r="\\1"===a?a:n.concat(a,i);return n.concat(n.concat("(?:",e,")"),i,/(?:\\.|[^\\\/])*?/,r,/(?:\\.|[^\\\/])*?/,a,t)},d=(e,i,a)=>n.concat(n.concat("(?:",e,")"),i,/(?:\\.|[^\\\/])*?/,a,t),u=[s,e.HASH_COMMENT_MODE,e.COMMENT(/^=\w/,/=cut/,{endsWithParent:!0}),r,{className:"string",contains:o,variants:[{begin:"q[qwxr]?\\s*\\(",end:"\\)",relevance:5},{begin:"q[qwxr]?\\s*\\[",end:"\\]",relevance:5},{begin:"q[qwxr]?\\s*\\{",end:"\\}",relevance:5},{begin:"q[qwxr]?\\s*\\|",end:"\\|",relevance:5},{begin:"q[qwxr]?\\s*<",end:">",relevance:5},{begin:"qw\\s+q",end:"q",relevance:5},{begin:"'",end:"'",contains:[e.BACKSLASH_ESCAPE]},{begin:'"',end:'"'},{begin:"`",end:"`",contains:[e.BACKSLASH_ESCAPE]},{begin:/\{\w+\}/,relevance:0},{begin:"-?\\w+\\s*=>",relevance:0}]},{className:"number",begin:"(\\b0[0-7_]+)|(\\b0x[0-9a-fA-F_]+)|(\\b[1-9][0-9_]*(\\.[0-9_]+)?)|[0_]\\b",relevance:0},{begin:"(\\/\\/|"+e.RE_STARTERS_RE+"|\\b(split|return|print|reverse|grep)\\b)\\s*",keywords:"split return print reverse grep",relevance:0,contains:[e.HASH_COMMENT_MODE,{className:"regexp",variants:[{begin:c("s|tr|y",n.either(...l,{capture:!0}))},{begin:c("s|tr|y","\\(","\\)")},{begin:c("s|tr|y","\\[","\\]")},{begin:c("s|tr|y","\\{","\\}")}],relevance:2},{className:"regexp",variants:[{begin:/(m|qr)\/\//,relevance:0},{begin:d("(?:m|qr)?",/\//,/\//)},{begin:d("m|qr",n.either(...l,{capture:!0}),/\1/)},{begin:d("m|qr",/\(/,/\)/)},{begin:d("m|qr",/\[/,/\]/)},{begin:d("m|qr",/\{/,/\}/)}]}]},{className:"function",beginKeywords:"sub",end:"(\\s*\\(.*?\\))?[;{]",excludeEnd:!0,relevance:5,contains:[e.TITLE_MODE]},{begin:"-\\w\\b",relevance:0},{begin:"^__DATA__$",end:"^__END__$",subLanguage:"mojolicious",contains:[{begin:"^@@.*",end:"$",className:"comment"}]}];return a.contains=u,r.contains=u,{name:"Perl",aliases:["pl","pm"],keywords:i,contains:u}}},9814:e=>{e.exports=function(e){const n=e.COMMENT("--","$"),t="\\$([a-zA-Z_]?|[a-zA-Z_][a-zA-Z_0-9]*)\\$",i="BIGINT INT8 BIGSERIAL SERIAL8 BIT VARYING VARBIT BOOLEAN BOOL BOX BYTEA CHARACTER CHAR VARCHAR CIDR CIRCLE DATE DOUBLE PRECISION FLOAT8 FLOAT INET INTEGER INT INT4 INTERVAL JSON JSONB LINE LSEG|10 MACADDR MACADDR8 MONEY NUMERIC DEC DECIMAL PATH POINT POLYGON REAL FLOAT4 SMALLINT INT2 SMALLSERIAL|10 SERIAL2|10 SERIAL|10 SERIAL4|10 TEXT TIME ZONE TIMETZ|10 TIMESTAMP TIMESTAMPTZ|10 TSQUERY|10 TSVECTOR|10 TXID_SNAPSHOT|10 UUID XML NATIONAL NCHAR INT4RANGE|10 INT8RANGE|10 NUMRANGE|10 TSRANGE|10 TSTZRANGE|10 DATERANGE|10 ANYELEMENT ANYARRAY ANYNONARRAY ANYENUM ANYRANGE CSTRING INTERNAL RECORD PG_DDL_COMMAND VOID UNKNOWN OPAQUE REFCURSOR NAME OID REGPROC|10 REGPROCEDURE|10 REGOPER|10 REGOPERATOR|10 REGCLASS|10 REGTYPE|10 REGROLE|10 REGNAMESPACE|10 REGCONFIG|10 REGDICTIONARY|10 ",a=i.trim().split(" ").map((function(e){return e.split("|")[0]})).join("|"),r="ARRAY_AGG AVG BIT_AND BIT_OR BOOL_AND BOOL_OR COUNT EVERY JSON_AGG JSONB_AGG JSON_OBJECT_AGG JSONB_OBJECT_AGG MAX MIN MODE STRING_AGG SUM XMLAGG CORR COVAR_POP COVAR_SAMP REGR_AVGX REGR_AVGY REGR_COUNT REGR_INTERCEPT REGR_R2 REGR_SLOPE REGR_SXX REGR_SXY REGR_SYY STDDEV STDDEV_POP STDDEV_SAMP VARIANCE VAR_POP VAR_SAMP PERCENTILE_CONT PERCENTILE_DISC ROW_NUMBER RANK DENSE_RANK PERCENT_RANK CUME_DIST NTILE LAG LEAD FIRST_VALUE LAST_VALUE NTH_VALUE NUM_NONNULLS NUM_NULLS ABS CBRT CEIL CEILING DEGREES DIV EXP FLOOR LN LOG MOD PI POWER RADIANS ROUND SCALE SIGN SQRT TRUNC WIDTH_BUCKET RANDOM SETSEED ACOS ACOSD ASIN ASIND ATAN ATAND ATAN2 ATAN2D COS COSD COT COTD SIN SIND TAN TAND BIT_LENGTH CHAR_LENGTH CHARACTER_LENGTH LOWER OCTET_LENGTH OVERLAY POSITION SUBSTRING TREAT TRIM UPPER ASCII BTRIM CHR CONCAT CONCAT_WS CONVERT CONVERT_FROM CONVERT_TO DECODE ENCODE INITCAP LEFT LENGTH LPAD LTRIM MD5 PARSE_IDENT PG_CLIENT_ENCODING QUOTE_IDENT|10 QUOTE_LITERAL|10 QUOTE_NULLABLE|10 REGEXP_MATCH REGEXP_MATCHES REGEXP_REPLACE REGEXP_SPLIT_TO_ARRAY REGEXP_SPLIT_TO_TABLE REPEAT REPLACE REVERSE RIGHT RPAD RTRIM SPLIT_PART STRPOS SUBSTR TO_ASCII TO_HEX TRANSLATE OCTET_LENGTH GET_BIT GET_BYTE SET_BIT SET_BYTE TO_CHAR TO_DATE TO_NUMBER TO_TIMESTAMP AGE CLOCK_TIMESTAMP|10 DATE_PART DATE_TRUNC ISFINITE JUSTIFY_DAYS JUSTIFY_HOURS JUSTIFY_INTERVAL MAKE_DATE MAKE_INTERVAL|10 MAKE_TIME MAKE_TIMESTAMP|10 MAKE_TIMESTAMPTZ|10 NOW STATEMENT_TIMESTAMP|10 TIMEOFDAY TRANSACTION_TIMESTAMP|10 ENUM_FIRST ENUM_LAST ENUM_RANGE AREA CENTER DIAMETER HEIGHT ISCLOSED ISOPEN NPOINTS PCLOSE POPEN RADIUS WIDTH BOX BOUND_BOX CIRCLE LINE LSEG PATH POLYGON ABBREV BROADCAST HOST HOSTMASK MASKLEN NETMASK NETWORK SET_MASKLEN TEXT INET_SAME_FAMILY INET_MERGE MACADDR8_SET7BIT ARRAY_TO_TSVECTOR GET_CURRENT_TS_CONFIG NUMNODE PLAINTO_TSQUERY PHRASETO_TSQUERY WEBSEARCH_TO_TSQUERY QUERYTREE SETWEIGHT STRIP TO_TSQUERY TO_TSVECTOR JSON_TO_TSVECTOR JSONB_TO_TSVECTOR TS_DELETE TS_FILTER TS_HEADLINE TS_RANK TS_RANK_CD TS_REWRITE TSQUERY_PHRASE TSVECTOR_TO_ARRAY TSVECTOR_UPDATE_TRIGGER TSVECTOR_UPDATE_TRIGGER_COLUMN XMLCOMMENT XMLCONCAT XMLELEMENT XMLFOREST XMLPI XMLROOT XMLEXISTS XML_IS_WELL_FORMED XML_IS_WELL_FORMED_DOCUMENT XML_IS_WELL_FORMED_CONTENT XPATH XPATH_EXISTS XMLTABLE XMLNAMESPACES TABLE_TO_XML TABLE_TO_XMLSCHEMA TABLE_TO_XML_AND_XMLSCHEMA QUERY_TO_XML QUERY_TO_XMLSCHEMA QUERY_TO_XML_AND_XMLSCHEMA CURSOR_TO_XML CURSOR_TO_XMLSCHEMA SCHEMA_TO_XML SCHEMA_TO_XMLSCHEMA SCHEMA_TO_XML_AND_XMLSCHEMA DATABASE_TO_XML DATABASE_TO_XMLSCHEMA DATABASE_TO_XML_AND_XMLSCHEMA XMLATTRIBUTES TO_JSON TO_JSONB ARRAY_TO_JSON ROW_TO_JSON JSON_BUILD_ARRAY JSONB_BUILD_ARRAY JSON_BUILD_OBJECT JSONB_BUILD_OBJECT JSON_OBJECT JSONB_OBJECT JSON_ARRAY_LENGTH JSONB_ARRAY_LENGTH JSON_EACH JSONB_EACH JSON_EACH_TEXT JSONB_EACH_TEXT JSON_EXTRACT_PATH JSONB_EXTRACT_PATH JSON_OBJECT_KEYS JSONB_OBJECT_KEYS JSON_POPULATE_RECORD JSONB_POPULATE_RECORD JSON_POPULATE_RECORDSET JSONB_POPULATE_RECORDSET JSON_ARRAY_ELEMENTS JSONB_ARRAY_ELEMENTS JSON_ARRAY_ELEMENTS_TEXT JSONB_ARRAY_ELEMENTS_TEXT JSON_TYPEOF JSONB_TYPEOF JSON_TO_RECORD JSONB_TO_RECORD JSON_TO_RECORDSET JSONB_TO_RECORDSET JSON_STRIP_NULLS JSONB_STRIP_NULLS JSONB_SET JSONB_INSERT JSONB_PRETTY CURRVAL LASTVAL NEXTVAL SETVAL COALESCE NULLIF GREATEST LEAST ARRAY_APPEND ARRAY_CAT ARRAY_NDIMS ARRAY_DIMS ARRAY_FILL ARRAY_LENGTH ARRAY_LOWER ARRAY_POSITION ARRAY_POSITIONS ARRAY_PREPEND ARRAY_REMOVE ARRAY_REPLACE ARRAY_TO_STRING ARRAY_UPPER CARDINALITY STRING_TO_ARRAY UNNEST ISEMPTY LOWER_INC UPPER_INC LOWER_INF UPPER_INF RANGE_MERGE GENERATE_SERIES GENERATE_SUBSCRIPTS CURRENT_DATABASE CURRENT_QUERY CURRENT_SCHEMA|10 CURRENT_SCHEMAS|10 INET_CLIENT_ADDR INET_CLIENT_PORT INET_SERVER_ADDR INET_SERVER_PORT ROW_SECURITY_ACTIVE FORMAT_TYPE TO_REGCLASS TO_REGPROC TO_REGPROCEDURE TO_REGOPER TO_REGOPERATOR TO_REGTYPE TO_REGNAMESPACE TO_REGROLE COL_DESCRIPTION OBJ_DESCRIPTION SHOBJ_DESCRIPTION TXID_CURRENT TXID_CURRENT_IF_ASSIGNED TXID_CURRENT_SNAPSHOT TXID_SNAPSHOT_XIP TXID_SNAPSHOT_XMAX TXID_SNAPSHOT_XMIN TXID_VISIBLE_IN_SNAPSHOT TXID_STATUS CURRENT_SETTING SET_CONFIG BRIN_SUMMARIZE_NEW_VALUES BRIN_SUMMARIZE_RANGE BRIN_DESUMMARIZE_RANGE GIN_CLEAN_PENDING_LIST SUPPRESS_REDUNDANT_UPDATES_TRIGGER LO_FROM_BYTEA LO_PUT LO_GET LO_CREAT LO_CREATE LO_UNLINK LO_IMPORT LO_EXPORT LOREAD LOWRITE GROUPING CAST ".trim().split(" ").map((function(e){return e.split("|")[0]})).join("|");return{name:"PostgreSQL",aliases:["postgres","postgresql"],supersetOf:"sql",case_insensitive:!0,keywords:{keyword:"ABORT ALTER ANALYZE BEGIN CALL CHECKPOINT|10 CLOSE CLUSTER COMMENT COMMIT COPY CREATE DEALLOCATE DECLARE DELETE DISCARD DO DROP END EXECUTE EXPLAIN FETCH GRANT IMPORT INSERT LISTEN LOAD LOCK MOVE NOTIFY PREPARE REASSIGN|10 REFRESH REINDEX RELEASE RESET REVOKE ROLLBACK SAVEPOINT SECURITY SELECT SET SHOW START TRUNCATE UNLISTEN|10 UPDATE VACUUM|10 VALUES AGGREGATE COLLATION CONVERSION|10 DATABASE DEFAULT PRIVILEGES DOMAIN TRIGGER EXTENSION FOREIGN WRAPPER|10 TABLE FUNCTION GROUP LANGUAGE LARGE OBJECT MATERIALIZED VIEW OPERATOR CLASS FAMILY POLICY PUBLICATION|10 ROLE RULE SCHEMA SEQUENCE SERVER STATISTICS SUBSCRIPTION SYSTEM TABLESPACE CONFIGURATION DICTIONARY PARSER TEMPLATE TYPE USER MAPPING PREPARED ACCESS METHOD CAST AS TRANSFORM TRANSACTION OWNED TO INTO SESSION AUTHORIZATION INDEX PROCEDURE ASSERTION ALL ANALYSE AND ANY ARRAY ASC ASYMMETRIC|10 BOTH CASE CHECK COLLATE COLUMN CONCURRENTLY|10 CONSTRAINT CROSS DEFERRABLE RANGE DESC DISTINCT ELSE EXCEPT FOR FREEZE|10 FROM FULL HAVING ILIKE IN INITIALLY INNER INTERSECT IS ISNULL JOIN LATERAL LEADING LIKE LIMIT NATURAL NOT NOTNULL NULL OFFSET ON ONLY OR ORDER OUTER OVERLAPS PLACING PRIMARY REFERENCES RETURNING SIMILAR SOME SYMMETRIC TABLESAMPLE THEN TRAILING UNION UNIQUE USING VARIADIC|10 VERBOSE WHEN WHERE WINDOW WITH BY RETURNS INOUT OUT SETOF|10 IF STRICT CURRENT CONTINUE OWNER LOCATION OVER PARTITION WITHIN BETWEEN ESCAPE EXTERNAL INVOKER DEFINER WORK RENAME VERSION CONNECTION CONNECT TABLES TEMP TEMPORARY FUNCTIONS SEQUENCES TYPES SCHEMAS OPTION CASCADE RESTRICT ADD ADMIN EXISTS VALID VALIDATE ENABLE DISABLE REPLICA|10 ALWAYS PASSING COLUMNS PATH REF VALUE OVERRIDING IMMUTABLE STABLE VOLATILE BEFORE AFTER EACH ROW PROCEDURAL ROUTINE NO HANDLER VALIDATOR OPTIONS STORAGE OIDS|10 WITHOUT INHERIT DEPENDS CALLED INPUT LEAKPROOF|10 COST ROWS NOWAIT SEARCH UNTIL ENCRYPTED|10 PASSWORD CONFLICT|10 INSTEAD INHERITS CHARACTERISTICS WRITE CURSOR ALSO STATEMENT SHARE EXCLUSIVE INLINE ISOLATION REPEATABLE READ COMMITTED SERIALIZABLE UNCOMMITTED LOCAL GLOBAL SQL PROCEDURES RECURSIVE SNAPSHOT ROLLUP CUBE TRUSTED|10 INCLUDE FOLLOWING PRECEDING UNBOUNDED RANGE GROUPS UNENCRYPTED|10 SYSID FORMAT DELIMITER HEADER QUOTE ENCODING FILTER OFF FORCE_QUOTE FORCE_NOT_NULL FORCE_NULL COSTS BUFFERS TIMING SUMMARY DISABLE_PAGE_SKIPPING RESTART CYCLE GENERATED IDENTITY DEFERRED IMMEDIATE LEVEL LOGGED UNLOGGED OF NOTHING NONE EXCLUDE ATTRIBUTE USAGE ROUTINES TRUE FALSE NAN INFINITY ALIAS BEGIN CONSTANT DECLARE END EXCEPTION RETURN PERFORM|10 RAISE GET DIAGNOSTICS STACKED|10 FOREACH LOOP ELSIF EXIT WHILE REVERSE SLICE DEBUG LOG INFO NOTICE WARNING ASSERT OPEN SUPERUSER NOSUPERUSER CREATEDB NOCREATEDB CREATEROLE NOCREATEROLE INHERIT NOINHERIT LOGIN NOLOGIN REPLICATION NOREPLICATION BYPASSRLS NOBYPASSRLS ",built_in:"CURRENT_TIME CURRENT_TIMESTAMP CURRENT_USER CURRENT_CATALOG|10 CURRENT_DATE LOCALTIME LOCALTIMESTAMP CURRENT_ROLE|10 CURRENT_SCHEMA|10 SESSION_USER PUBLIC FOUND NEW OLD TG_NAME|10 TG_WHEN|10 TG_LEVEL|10 TG_OP|10 TG_RELID|10 TG_RELNAME|10 TG_TABLE_NAME|10 TG_TABLE_SCHEMA|10 TG_NARGS|10 TG_ARGV|10 TG_EVENT|10 TG_TAG|10 ROW_COUNT RESULT_OID|10 PG_CONTEXT|10 RETURNED_SQLSTATE COLUMN_NAME CONSTRAINT_NAME PG_DATATYPE_NAME|10 MESSAGE_TEXT TABLE_NAME SCHEMA_NAME PG_EXCEPTION_DETAIL|10 PG_EXCEPTION_HINT|10 PG_EXCEPTION_CONTEXT|10 SQLSTATE SQLERRM|10 SUCCESSFUL_COMPLETION WARNING DYNAMIC_RESULT_SETS_RETURNED IMPLICIT_ZERO_BIT_PADDING NULL_VALUE_ELIMINATED_IN_SET_FUNCTION PRIVILEGE_NOT_GRANTED PRIVILEGE_NOT_REVOKED STRING_DATA_RIGHT_TRUNCATION DEPRECATED_FEATURE NO_DATA NO_ADDITIONAL_DYNAMIC_RESULT_SETS_RETURNED SQL_STATEMENT_NOT_YET_COMPLETE CONNECTION_EXCEPTION CONNECTION_DOES_NOT_EXIST CONNECTION_FAILURE SQLCLIENT_UNABLE_TO_ESTABLISH_SQLCONNECTION SQLSERVER_REJECTED_ESTABLISHMENT_OF_SQLCONNECTION TRANSACTION_RESOLUTION_UNKNOWN PROTOCOL_VIOLATION TRIGGERED_ACTION_EXCEPTION FEATURE_NOT_SUPPORTED INVALID_TRANSACTION_INITIATION LOCATOR_EXCEPTION INVALID_LOCATOR_SPECIFICATION INVALID_GRANTOR INVALID_GRANT_OPERATION INVALID_ROLE_SPECIFICATION DIAGNOSTICS_EXCEPTION STACKED_DIAGNOSTICS_ACCESSED_WITHOUT_ACTIVE_HANDLER CASE_NOT_FOUND CARDINALITY_VIOLATION DATA_EXCEPTION ARRAY_SUBSCRIPT_ERROR CHARACTER_NOT_IN_REPERTOIRE DATETIME_FIELD_OVERFLOW DIVISION_BY_ZERO ERROR_IN_ASSIGNMENT ESCAPE_CHARACTER_CONFLICT INDICATOR_OVERFLOW INTERVAL_FIELD_OVERFLOW INVALID_ARGUMENT_FOR_LOGARITHM INVALID_ARGUMENT_FOR_NTILE_FUNCTION INVALID_ARGUMENT_FOR_NTH_VALUE_FUNCTION INVALID_ARGUMENT_FOR_POWER_FUNCTION INVALID_ARGUMENT_FOR_WIDTH_BUCKET_FUNCTION INVALID_CHARACTER_VALUE_FOR_CAST INVALID_DATETIME_FORMAT INVALID_ESCAPE_CHARACTER INVALID_ESCAPE_OCTET INVALID_ESCAPE_SEQUENCE NONSTANDARD_USE_OF_ESCAPE_CHARACTER INVALID_INDICATOR_PARAMETER_VALUE INVALID_PARAMETER_VALUE INVALID_REGULAR_EXPRESSION INVALID_ROW_COUNT_IN_LIMIT_CLAUSE INVALID_ROW_COUNT_IN_RESULT_OFFSET_CLAUSE INVALID_TABLESAMPLE_ARGUMENT INVALID_TABLESAMPLE_REPEAT INVALID_TIME_ZONE_DISPLACEMENT_VALUE INVALID_USE_OF_ESCAPE_CHARACTER MOST_SPECIFIC_TYPE_MISMATCH NULL_VALUE_NOT_ALLOWED NULL_VALUE_NO_INDICATOR_PARAMETER NUMERIC_VALUE_OUT_OF_RANGE SEQUENCE_GENERATOR_LIMIT_EXCEEDED STRING_DATA_LENGTH_MISMATCH STRING_DATA_RIGHT_TRUNCATION SUBSTRING_ERROR TRIM_ERROR UNTERMINATED_C_STRING ZERO_LENGTH_CHARACTER_STRING FLOATING_POINT_EXCEPTION INVALID_TEXT_REPRESENTATION INVALID_BINARY_REPRESENTATION BAD_COPY_FILE_FORMAT UNTRANSLATABLE_CHARACTER NOT_AN_XML_DOCUMENT INVALID_XML_DOCUMENT INVALID_XML_CONTENT INVALID_XML_COMMENT INVALID_XML_PROCESSING_INSTRUCTION INTEGRITY_CONSTRAINT_VIOLATION RESTRICT_VIOLATION NOT_NULL_VIOLATION FOREIGN_KEY_VIOLATION UNIQUE_VIOLATION CHECK_VIOLATION EXCLUSION_VIOLATION INVALID_CURSOR_STATE INVALID_TRANSACTION_STATE ACTIVE_SQL_TRANSACTION BRANCH_TRANSACTION_ALREADY_ACTIVE HELD_CURSOR_REQUIRES_SAME_ISOLATION_LEVEL INAPPROPRIATE_ACCESS_MODE_FOR_BRANCH_TRANSACTION INAPPROPRIATE_ISOLATION_LEVEL_FOR_BRANCH_TRANSACTION NO_ACTIVE_SQL_TRANSACTION_FOR_BRANCH_TRANSACTION READ_ONLY_SQL_TRANSACTION SCHEMA_AND_DATA_STATEMENT_MIXING_NOT_SUPPORTED NO_ACTIVE_SQL_TRANSACTION IN_FAILED_SQL_TRANSACTION IDLE_IN_TRANSACTION_SESSION_TIMEOUT INVALID_SQL_STATEMENT_NAME TRIGGERED_DATA_CHANGE_VIOLATION INVALID_AUTHORIZATION_SPECIFICATION INVALID_PASSWORD DEPENDENT_PRIVILEGE_DESCRIPTORS_STILL_EXIST DEPENDENT_OBJECTS_STILL_EXIST INVALID_TRANSACTION_TERMINATION SQL_ROUTINE_EXCEPTION FUNCTION_EXECUTED_NO_RETURN_STATEMENT MODIFYING_SQL_DATA_NOT_PERMITTED PROHIBITED_SQL_STATEMENT_ATTEMPTED READING_SQL_DATA_NOT_PERMITTED INVALID_CURSOR_NAME EXTERNAL_ROUTINE_EXCEPTION CONTAINING_SQL_NOT_PERMITTED MODIFYING_SQL_DATA_NOT_PERMITTED PROHIBITED_SQL_STATEMENT_ATTEMPTED READING_SQL_DATA_NOT_PERMITTED EXTERNAL_ROUTINE_INVOCATION_EXCEPTION INVALID_SQLSTATE_RETURNED NULL_VALUE_NOT_ALLOWED TRIGGER_PROTOCOL_VIOLATED SRF_PROTOCOL_VIOLATED EVENT_TRIGGER_PROTOCOL_VIOLATED SAVEPOINT_EXCEPTION INVALID_SAVEPOINT_SPECIFICATION INVALID_CATALOG_NAME INVALID_SCHEMA_NAME TRANSACTION_ROLLBACK TRANSACTION_INTEGRITY_CONSTRAINT_VIOLATION SERIALIZATION_FAILURE STATEMENT_COMPLETION_UNKNOWN DEADLOCK_DETECTED SYNTAX_ERROR_OR_ACCESS_RULE_VIOLATION SYNTAX_ERROR INSUFFICIENT_PRIVILEGE CANNOT_COERCE GROUPING_ERROR WINDOWING_ERROR INVALID_RECURSION INVALID_FOREIGN_KEY INVALID_NAME NAME_TOO_LONG RESERVED_NAME DATATYPE_MISMATCH INDETERMINATE_DATATYPE COLLATION_MISMATCH INDETERMINATE_COLLATION WRONG_OBJECT_TYPE GENERATED_ALWAYS UNDEFINED_COLUMN UNDEFINED_FUNCTION UNDEFINED_TABLE UNDEFINED_PARAMETER UNDEFINED_OBJECT DUPLICATE_COLUMN DUPLICATE_CURSOR DUPLICATE_DATABASE DUPLICATE_FUNCTION DUPLICATE_PREPARED_STATEMENT DUPLICATE_SCHEMA DUPLICATE_TABLE DUPLICATE_ALIAS DUPLICATE_OBJECT AMBIGUOUS_COLUMN AMBIGUOUS_FUNCTION AMBIGUOUS_PARAMETER AMBIGUOUS_ALIAS INVALID_COLUMN_REFERENCE INVALID_COLUMN_DEFINITION INVALID_CURSOR_DEFINITION INVALID_DATABASE_DEFINITION INVALID_FUNCTION_DEFINITION INVALID_PREPARED_STATEMENT_DEFINITION INVALID_SCHEMA_DEFINITION INVALID_TABLE_DEFINITION INVALID_OBJECT_DEFINITION WITH_CHECK_OPTION_VIOLATION INSUFFICIENT_RESOURCES DISK_FULL OUT_OF_MEMORY TOO_MANY_CONNECTIONS CONFIGURATION_LIMIT_EXCEEDED PROGRAM_LIMIT_EXCEEDED STATEMENT_TOO_COMPLEX TOO_MANY_COLUMNS TOO_MANY_ARGUMENTS OBJECT_NOT_IN_PREREQUISITE_STATE OBJECT_IN_USE CANT_CHANGE_RUNTIME_PARAM LOCK_NOT_AVAILABLE OPERATOR_INTERVENTION QUERY_CANCELED ADMIN_SHUTDOWN CRASH_SHUTDOWN CANNOT_CONNECT_NOW DATABASE_DROPPED SYSTEM_ERROR IO_ERROR UNDEFINED_FILE DUPLICATE_FILE SNAPSHOT_TOO_OLD CONFIG_FILE_ERROR LOCK_FILE_EXISTS FDW_ERROR FDW_COLUMN_NAME_NOT_FOUND FDW_DYNAMIC_PARAMETER_VALUE_NEEDED FDW_FUNCTION_SEQUENCE_ERROR FDW_INCONSISTENT_DESCRIPTOR_INFORMATION FDW_INVALID_ATTRIBUTE_VALUE FDW_INVALID_COLUMN_NAME FDW_INVALID_COLUMN_NUMBER FDW_INVALID_DATA_TYPE FDW_INVALID_DATA_TYPE_DESCRIPTORS FDW_INVALID_DESCRIPTOR_FIELD_IDENTIFIER FDW_INVALID_HANDLE FDW_INVALID_OPTION_INDEX FDW_INVALID_OPTION_NAME FDW_INVALID_STRING_LENGTH_OR_BUFFER_LENGTH FDW_INVALID_STRING_FORMAT FDW_INVALID_USE_OF_NULL_POINTER FDW_TOO_MANY_HANDLES FDW_OUT_OF_MEMORY FDW_NO_SCHEMAS FDW_OPTION_NAME_NOT_FOUND FDW_REPLY_HANDLE FDW_SCHEMA_NOT_FOUND FDW_TABLE_NOT_FOUND FDW_UNABLE_TO_CREATE_EXECUTION FDW_UNABLE_TO_CREATE_REPLY FDW_UNABLE_TO_ESTABLISH_CONNECTION PLPGSQL_ERROR RAISE_EXCEPTION NO_DATA_FOUND TOO_MANY_ROWS ASSERT_FAILURE INTERNAL_ERROR DATA_CORRUPTED INDEX_CORRUPTED "},illegal:/:==|\W\s*\(\*|(^|\s)\$[a-z]|\{\{|[a-z]:\s*$|\.\.\.|TO:|DO:/,contains:[{className:"keyword",variants:[{begin:/\bTEXT\s*SEARCH\b/},{begin:/\b(PRIMARY|FOREIGN|FOR(\s+NO)?)\s+KEY\b/},{begin:/\bPARALLEL\s+(UNSAFE|RESTRICTED|SAFE)\b/},{begin:/\bSTORAGE\s+(PLAIN|EXTERNAL|EXTENDED|MAIN)\b/},{begin:/\bMATCH\s+(FULL|PARTIAL|SIMPLE)\b/},{begin:/\bNULLS\s+(FIRST|LAST)\b/},{begin:/\bEVENT\s+TRIGGER\b/},{begin:/\b(MAPPING|OR)\s+REPLACE\b/},{begin:/\b(FROM|TO)\s+(PROGRAM|STDIN|STDOUT)\b/},{begin:/\b(SHARE|EXCLUSIVE)\s+MODE\b/},{begin:/\b(LEFT|RIGHT)\s+(OUTER\s+)?JOIN\b/},{begin:/\b(FETCH|MOVE)\s+(NEXT|PRIOR|FIRST|LAST|ABSOLUTE|RELATIVE|FORWARD|BACKWARD)\b/},{begin:/\bPRESERVE\s+ROWS\b/},{begin:/\bDISCARD\s+PLANS\b/},{begin:/\bREFERENCING\s+(OLD|NEW)\b/},{begin:/\bSKIP\s+LOCKED\b/},{begin:/\bGROUPING\s+SETS\b/},{begin:/\b(BINARY|INSENSITIVE|SCROLL|NO\s+SCROLL)\s+(CURSOR|FOR)\b/},{begin:/\b(WITH|WITHOUT)\s+HOLD\b/},{begin:/\bWITH\s+(CASCADED|LOCAL)\s+CHECK\s+OPTION\b/},{begin:/\bEXCLUDE\s+(TIES|NO\s+OTHERS)\b/},{begin:/\bFORMAT\s+(TEXT|XML|JSON|YAML)\b/},{begin:/\bSET\s+((SESSION|LOCAL)\s+)?NAMES\b/},{begin:/\bIS\s+(NOT\s+)?UNKNOWN\b/},{begin:/\bSECURITY\s+LABEL\b/},{begin:/\bSTANDALONE\s+(YES|NO|NO\s+VALUE)\b/},{begin:/\bWITH\s+(NO\s+)?DATA\b/},{begin:/\b(FOREIGN|SET)\s+DATA\b/},{begin:/\bSET\s+(CATALOG|CONSTRAINTS)\b/},{begin:/\b(WITH|FOR)\s+ORDINALITY\b/},{begin:/\bIS\s+(NOT\s+)?DOCUMENT\b/},{begin:/\bXML\s+OPTION\s+(DOCUMENT|CONTENT)\b/},{begin:/\b(STRIP|PRESERVE)\s+WHITESPACE\b/},{begin:/\bNO\s+(ACTION|MAXVALUE|MINVALUE)\b/},{begin:/\bPARTITION\s+BY\s+(RANGE|LIST|HASH)\b/},{begin:/\bAT\s+TIME\s+ZONE\b/},{begin:/\bGRANTED\s+BY\b/},{begin:/\bRETURN\s+(QUERY|NEXT)\b/},{begin:/\b(ATTACH|DETACH)\s+PARTITION\b/},{begin:/\bFORCE\s+ROW\s+LEVEL\s+SECURITY\b/},{begin:/\b(INCLUDING|EXCLUDING)\s+(COMMENTS|CONSTRAINTS|DEFAULTS|IDENTITY|INDEXES|STATISTICS|STORAGE|ALL)\b/},{begin:/\bAS\s+(ASSIGNMENT|IMPLICIT|PERMISSIVE|RESTRICTIVE|ENUM|RANGE)\b/}]},{begin:/\b(FORMAT|FAMILY|VERSION)\s*\(/},{begin:/\bINCLUDE\s*\(/,keywords:"INCLUDE"},{begin:/\bRANGE(?!\s*(BETWEEN|UNBOUNDED|CURRENT|[-0-9]+))/},{begin:/\b(VERSION|OWNER|TEMPLATE|TABLESPACE|CONNECTION\s+LIMIT|PROCEDURE|RESTRICT|JOIN|PARSER|COPY|START|END|COLLATION|INPUT|ANALYZE|STORAGE|LIKE|DEFAULT|DELIMITER|ENCODING|COLUMN|CONSTRAINT|TABLE|SCHEMA)\s*=/},{begin:/\b(PG_\w+?|HAS_[A-Z_]+_PRIVILEGE)\b/,relevance:10},{begin:/\bEXTRACT\s*\(/,end:/\bFROM\b/,returnEnd:!0,keywords:{type:"CENTURY DAY DECADE DOW DOY EPOCH HOUR ISODOW ISOYEAR MICROSECONDS MILLENNIUM MILLISECONDS MINUTE MONTH QUARTER SECOND TIMEZONE TIMEZONE_HOUR TIMEZONE_MINUTE WEEK YEAR"}},{begin:/\b(XMLELEMENT|XMLPI)\s*\(\s*NAME/,keywords:{keyword:"NAME"}},{begin:/\b(XMLPARSE|XMLSERIALIZE)\s*\(\s*(DOCUMENT|CONTENT)/,keywords:{keyword:"DOCUMENT CONTENT"}},{beginKeywords:"CACHE INCREMENT MAXVALUE MINVALUE",end:e.C_NUMBER_RE,returnEnd:!0,keywords:"BY CACHE INCREMENT MAXVALUE MINVALUE"},{className:"type",begin:/\b(WITH|WITHOUT)\s+TIME\s+ZONE\b/},{className:"type",begin:/\bINTERVAL\s+(YEAR|MONTH|DAY|HOUR|MINUTE|SECOND)(\s+TO\s+(MONTH|HOUR|MINUTE|SECOND))?\b/},{begin:/\bRETURNS\s+(LANGUAGE_HANDLER|TRIGGER|EVENT_TRIGGER|FDW_HANDLER|INDEX_AM_HANDLER|TSM_HANDLER)\b/,keywords:{keyword:"RETURNS",type:"LANGUAGE_HANDLER TRIGGER EVENT_TRIGGER FDW_HANDLER INDEX_AM_HANDLER TSM_HANDLER"}},{begin:"\\b("+r+")\\s*\\("},{begin:"\\.("+a+")\\b"},{begin:"\\b("+a+")\\s+PATH\\b",keywords:{keyword:"PATH",type:i.replace("PATH ","")}},{className:"type",begin:"\\b("+a+")\\b"},{className:"string",begin:"'",end:"'",contains:[{begin:"''"}]},{className:"string",begin:"(e|E|u&|U&)'",end:"'",contains:[{begin:"\\\\."}],relevance:10},e.END_SAME_AS_BEGIN({begin:t,end:t,contains:[{subLanguage:["pgsql","perl","python","tcl","r","lua","java","php","ruby","bash","scheme","xml","json"],endsWithParent:!0}]}),{begin:'"',end:'"',contains:[{begin:'""'}]},e.C_NUMBER_MODE,e.C_BLOCK_COMMENT_MODE,n,{className:"meta",variants:[{begin:"%(ROW)?TYPE",relevance:10},{begin:"\\$\\d+"},{begin:"^#\\w",end:"$"}]},{className:"symbol",begin:"<<\\s*[a-zA-Z_][a-zA-Z_0-9$]*\\s*>>",relevance:10}]}}},2656:e=>{e.exports=function(e){const n=e.regex,t=/(?![A-Za-z0-9])(?![$])/,i=n.concat(/[a-zA-Z_\x7f-\xff][a-zA-Z0-9_\x7f-\xff]*/,t),a=n.concat(/(\\?[A-Z][a-z0-9_\x7f-\xff]+|\\?[A-Z]+(?=[A-Z][a-z0-9_\x7f-\xff])){1,}/,t),r={scope:"variable",match:"\\$+"+i},s={scope:"subst",variants:[{begin:/\$\w+/},{begin:/\{\$/,end:/\}/}]},o=e.inherit(e.APOS_STRING_MODE,{illegal:null}),l="[ \t\n]",c={scope:"string",variants:[e.inherit(e.QUOTE_STRING_MODE,{illegal:null,contains:e.QUOTE_STRING_MODE.contains.concat(s)}),o,{begin:/<<<[ \t]*(?:(\w+)|"(\w+)")\n/,end:/[ \t]*(\w+)\b/,contains:e.QUOTE_STRING_MODE.contains.concat(s),"on:begin":(e,n)=>{n.data._beginMatch=e[1]||e[2]},"on:end":(e,n)=>{n.data._beginMatch!==e[1]&&n.ignoreMatch()}},e.END_SAME_AS_BEGIN({begin:/<<<[ \t]*'(\w+)'\n/,end:/[ \t]*(\w+)\b/})]},d={scope:"number",variants:[{begin:"\\b0[bB][01]+(?:_[01]+)*\\b"},{begin:"\\b0[oO][0-7]+(?:_[0-7]+)*\\b"},{begin:"\\b0[xX][\\da-fA-F]+(?:_[\\da-fA-F]+)*\\b"},{begin:"(?:\\b\\d+(?:_\\d+)*(\\.(?:\\d+(?:_\\d+)*))?|\\B\\.\\d+)(?:[eE][+-]?\\d+)?"}],relevance:0},u=["false","null","true"],g=["__CLASS__","__DIR__","__FILE__","__FUNCTION__","__COMPILER_HALT_OFFSET__","__LINE__","__METHOD__","__NAMESPACE__","__TRAIT__","die","echo","exit","include","include_once","print","require","require_once","array","abstract","and","as","binary","bool","boolean","break","callable","case","catch","class","clone","const","continue","declare","default","do","double","else","elseif","empty","enddeclare","endfor","endforeach","endif","endswitch","endwhile","enum","eval","extends","final","finally","float","for","foreach","from","global","goto","if","implements","instanceof","insteadof","int","integer","interface","isset","iterable","list","match|0","mixed","new","never","object","or","private","protected","public","readonly","real","return","string","switch","throw","trait","try","unset","use","var","void","while","xor","yield"],b=["Error|0","AppendIterator","ArgumentCountError","ArithmeticError","ArrayIterator","ArrayObject","AssertionError","BadFunctionCallException","BadMethodCallException","CachingIterator","CallbackFilterIterator","CompileError","Countable","DirectoryIterator","DivisionByZeroError","DomainException","EmptyIterator","ErrorException","Exception","FilesystemIterator","FilterIterator","GlobIterator","InfiniteIterator","InvalidArgumentException","IteratorIterator","LengthException","LimitIterator","LogicException","MultipleIterator","NoRewindIterator","OutOfBoundsException","OutOfRangeException","OuterIterator","OverflowException","ParentIterator","ParseError","RangeException","RecursiveArrayIterator","RecursiveCachingIterator","RecursiveCallbackFilterIterator","RecursiveDirectoryIterator","RecursiveFilterIterator","RecursiveIterator","RecursiveIteratorIterator","RecursiveRegexIterator","RecursiveTreeIterator","RegexIterator","RuntimeException","SeekableIterator","SplDoublyLinkedList","SplFileInfo","SplFileObject","SplFixedArray","SplHeap","SplMaxHeap","SplMinHeap","SplObjectStorage","SplObserver","SplPriorityQueue","SplQueue","SplStack","SplSubject","SplTempFileObject","TypeError","UnderflowException","UnexpectedValueException","UnhandledMatchError","ArrayAccess","BackedEnum","Closure","Fiber","Generator","Iterator","IteratorAggregate","Serializable","Stringable","Throwable","Traversable","UnitEnum","WeakReference","WeakMap","Directory","__PHP_Incomplete_Class","parent","php_user_filter","self","static","stdClass"],p={keyword:g,literal:(e=>{const n=[];return e.forEach((e=>{n.push(e),e.toLowerCase()===e?n.push(e.toUpperCase()):n.push(e.toLowerCase())})),n})(u),built_in:b},f=e=>e.map((e=>e.replace(/\|\d+$/,""))),m={variants:[{match:[/new/,n.concat(l,"+"),n.concat("(?!",f(b).join("\\b|"),"\\b)"),a],scope:{1:"keyword",4:"title.class"}}]},E=n.concat(i,"\\b(?!\\()"),h={variants:[{match:[n.concat(/::/,n.lookahead(/(?!class\b)/)),E],scope:{2:"variable.constant"}},{match:[/::/,/class/],scope:{2:"variable.language"}},{match:[a,n.concat(/::/,n.lookahead(/(?!class\b)/)),E],scope:{1:"title.class",3:"variable.constant"}},{match:[a,n.concat("::",n.lookahead(/(?!class\b)/))],scope:{1:"title.class"}},{match:[a,/::/,/class/],scope:{1:"title.class",3:"variable.language"}}]},_={scope:"attr",match:n.concat(i,n.lookahead(":"),n.lookahead(/(?!::)/))},N={relevance:0,begin:/\(/,end:/\)/,keywords:p,contains:[_,r,h,e.C_BLOCK_COMMENT_MODE,c,d,m]},T={relevance:0,match:[/\b/,n.concat("(?!fn\\b|function\\b|",f(g).join("\\b|"),"|",f(b).join("\\b|"),"\\b)"),i,n.concat(l,"*"),n.lookahead(/(?=\()/)],scope:{3:"title.function.invoke"},contains:[N]};N.contains.push(T);const O=[_,h,e.C_BLOCK_COMMENT_MODE,c,d,m];return{case_insensitive:!1,keywords:p,contains:[{begin:n.concat(/#\[\s*/,a),beginScope:"meta",end:/]/,endScope:"meta",keywords:{literal:u,keyword:["new","array"]},contains:[{begin:/\[/,end:/]/,keywords:{literal:u,keyword:["new","array"]},contains:["self",...O]},...O,{scope:"meta",match:a}]},e.HASH_COMMENT_MODE,e.COMMENT("//","$"),e.COMMENT("/\\*","\\*/",{contains:[{scope:"doctag",match:"@[A-Za-z]+"}]}),{match:/__halt_compiler\(\);/,keywords:"__halt_compiler",starts:{scope:"comment",end:e.MATCH_NOTHING_RE,contains:[{match:/\?>/,scope:"meta",endsParent:!0}]}},{scope:"meta",variants:[{begin:/<\?php/,relevance:10},{begin:/<\?=/},{begin:/<\?/,relevance:.1},{begin:/\?>/}]},{scope:"variable.language",match:/\$this\b/},r,T,h,{match:[/const/,/\s/,i],scope:{1:"keyword",3:"variable.constant"}},m,{scope:"function",relevance:0,beginKeywords:"fn function",end:/[;{]/,excludeEnd:!0,illegal:"[$%\\[]",contains:[{beginKeywords:"use"},e.UNDERSCORE_TITLE_MODE,{begin:"=>",endsParent:!0},{scope:"params",begin:"\\(",end:"\\)",excludeBegin:!0,excludeEnd:!0,keywords:p,contains:["self",r,h,e.C_BLOCK_COMMENT_MODE,c,d]}]},{scope:"class",variants:[{beginKeywords:"enum",illegal:/[($"]/},{beginKeywords:"class interface trait",illegal:/[:($"]/}],relevance:0,end:/\{/,excludeEnd:!0,contains:[{beginKeywords:"extends implements"},e.UNDERSCORE_TITLE_MODE]},{beginKeywords:"namespace",relevance:0,end:";",illegal:/[.']/,contains:[e.inherit(e.UNDERSCORE_TITLE_MODE,{scope:"title.class"})]},{beginKeywords:"use",relevance:0,end:";",contains:[{match:/\b(as|const|function)\b/,scope:"keyword"},e.UNDERSCORE_TITLE_MODE]},c,d]}}},2437:e=>{e.exports=function(e){return{name:"Plain text",aliases:["text","txt"],disableAutodetect:!0}}},5040:e=>{e.exports=function(e){const n={$pattern:/-?[A-z\.\-]+\b/,keyword:"if else foreach return do while until elseif begin for trap data dynamicparam end break throw param continue finally in switch exit filter try process catch hidden static parameter",built_in:"ac asnp cat cd CFS chdir clc clear clhy cli clp cls clv cnsn compare copy cp cpi cpp curl cvpa dbp del diff dir dnsn ebp echo|0 epal epcsv epsn erase etsn exsn fc fhx fl ft fw gal gbp gc gcb gci gcm gcs gdr gerr ghy gi gin gjb gl gm gmo gp gps gpv group gsn gsnp gsv gtz gu gv gwmi h history icm iex ihy ii ipal ipcsv ipmo ipsn irm ise iwmi iwr kill lp ls man md measure mi mount move mp mv nal ndr ni nmo npssc nsn nv ogv oh popd ps pushd pwd r rbp rcjb rcsn rd rdr ren ri rjb rm rmdir rmo rni rnp rp rsn rsnp rujb rv rvpa rwmi sajb sal saps sasv sbp sc scb select set shcm si sl sleep sls sort sp spjb spps spsv start stz sujb sv swmi tee trcm type wget where wjb write"},t={begin:"`[\\s\\S]",relevance:0},i={className:"variable",variants:[{begin:/\$\B/},{className:"keyword",begin:/\$this/},{begin:/\$[\w\d][\w\d_:]*/}]},a={className:"string",variants:[{begin:/"/,end:/"/},{begin:/@"/,end:/^"@/}],contains:[t,i,{className:"variable",begin:/\$[A-z]/,end:/[^A-z]/}]},r={className:"string",variants:[{begin:/'/,end:/'/},{begin:/@'/,end:/^'@/}]},s=e.inherit(e.COMMENT(null,null),{variants:[{begin:/#/,end:/$/},{begin:/<#/,end:/#>/}],contains:[{className:"doctag",variants:[{begin:/\.(synopsis|description|example|inputs|outputs|notes|link|component|role|functionality)/},{begin:/\.(parameter|forwardhelptargetname|forwardhelpcategory|remotehelprunspace|externalhelp)\s+\S+/}]}]}),o={className:"built_in",variants:[{begin:"(".concat("Add|Clear|Close|Copy|Enter|Exit|Find|Format|Get|Hide|Join|Lock|Move|New|Open|Optimize|Pop|Push|Redo|Remove|Rename|Reset|Resize|Search|Select|Set|Show|Skip|Split|Step|Switch|Undo|Unlock|Watch|Backup|Checkpoint|Compare|Compress|Convert|ConvertFrom|ConvertTo|Dismount|Edit|Expand|Export|Group|Import|Initialize|Limit|Merge|Mount|Out|Publish|Restore|Save|Sync|Unpublish|Update|Approve|Assert|Build|Complete|Confirm|Deny|Deploy|Disable|Enable|Install|Invoke|Register|Request|Restart|Resume|Start|Stop|Submit|Suspend|Uninstall|Unregister|Wait|Debug|Measure|Ping|Repair|Resolve|Test|Trace|Connect|Disconnect|Read|Receive|Send|Write|Block|Grant|Protect|Revoke|Unblock|Unprotect|Use|ForEach|Sort|Tee|Where",")+(-)[\\w\\d]+")}]},l={className:"class",beginKeywords:"class enum",end:/\s*[{]/,excludeEnd:!0,relevance:0,contains:[e.TITLE_MODE]},c={className:"function",begin:/function\s+/,end:/\s*\{|$/,excludeEnd:!0,returnBegin:!0,relevance:0,contains:[{begin:"function",relevance:0,className:"keyword"},{className:"title",begin:/\w[\w\d]*((-)[\w\d]+)*/,relevance:0},{begin:/\(/,end:/\)/,className:"params",relevance:0,contains:[i]}]},d={begin:/using\s/,end:/$/,returnBegin:!0,contains:[a,r,{className:"keyword",begin:/(using|assembly|command|module|namespace|type)/}]},u={variants:[{className:"operator",begin:"(".concat("-and|-as|-band|-bnot|-bor|-bxor|-casesensitive|-ccontains|-ceq|-cge|-cgt|-cle|-clike|-clt|-cmatch|-cne|-cnotcontains|-cnotlike|-cnotmatch|-contains|-creplace|-csplit|-eq|-exact|-f|-file|-ge|-gt|-icontains|-ieq|-ige|-igt|-ile|-ilike|-ilt|-imatch|-in|-ine|-inotcontains|-inotlike|-inotmatch|-ireplace|-is|-isnot|-isplit|-join|-le|-like|-lt|-match|-ne|-not|-notcontains|-notin|-notlike|-notmatch|-or|-regex|-replace|-shl|-shr|-split|-wildcard|-xor",")\\b")},{className:"literal",begin:/(-){1,2}[\w\d-]+/,relevance:0}]},g={className:"function",begin:/\[.*\]\s*[\w]+[ ]??\(/,end:/$/,returnBegin:!0,relevance:0,contains:[{className:"keyword",begin:"(".concat(n.keyword.toString().replace(/\s/g,"|"),")\\b"),endsParent:!0,relevance:0},e.inherit(e.TITLE_MODE,{endsParent:!0})]},b=[g,s,t,e.NUMBER_MODE,a,r,o,i,{className:"literal",begin:/\$(null|true|false)\b/},{className:"selector-tag",begin:/@\B/,relevance:0}],p={begin:/\[/,end:/\]/,excludeBegin:!0,excludeEnd:!0,relevance:0,contains:[].concat("self",b,{begin:"("+["string","char","byte","int","long","bool","decimal","single","double","DateTime","xml","array","hashtable","void"].join("|")+")",className:"built_in",relevance:0},{className:"type",begin:/[\.\w\d]+/,relevance:0})};return g.contains.unshift(p),{name:"PowerShell",aliases:["pwsh","ps","ps1"],case_insensitive:!0,keywords:n,contains:b.concat(l,c,d,u,p)}}},7546:e=>{e.exports=function(e){const n="[ \\t\\f]*",t=n+"[:=]"+n,i="[ \\t\\f]+",a="([^\\\\:= \\t\\f\\n]|\\\\.)+",r={end:"("+t+"|"+i+")",relevance:0,starts:{className:"string",end:/$/,relevance:0,contains:[{begin:"\\\\\\\\"},{begin:"\\\\\\n"}]}};return{name:".properties",disableAutodetect:!0,case_insensitive:!0,illegal:/\S/,contains:[e.COMMENT("^\\s*[!#]","$"),{returnBegin:!0,variants:[{begin:a+t},{begin:a+i}],contains:[{className:"attr",begin:a,endsParent:!0}],starts:r},{className:"attr",begin:a+n+"$"}]}}},5559:e=>{e.exports=function(e){const n={match:[/(message|enum|service)\s+/,e.IDENT_RE],scope:{1:"keyword",2:"title.class"}};return{name:"Protocol Buffers",aliases:["proto"],keywords:{keyword:["package","import","option","optional","required","repeated","group","oneof"],type:["double","float","int32","int64","uint32","uint64","sint32","sint64","fixed32","fixed64","sfixed32","sfixed64","bool","string","bytes"],literal:["true","false"]},contains:[e.QUOTE_STRING_MODE,e.NUMBER_MODE,e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,n,{className:"function",beginKeywords:"rpc",end:/[{;]/,excludeEnd:!0,keywords:"rpc returns"},{begin:/^\s*[A-Z_]+(?=\s*=[^\n]+;$)/}]}}},8245:e=>{e.exports=function(e){const n=e.regex,t=/[\p{XID_Start}_]\p{XID_Continue}*/u,i=["and","as","assert","async","await","break","case","class","continue","def","del","elif","else","except","finally","for","from","global","if","import","in","is","lambda","match","nonlocal|10","not","or","pass","raise","return","try","while","with","yield"],a={$pattern:/[A-Za-z]\w+|__\w+__/,keyword:i,built_in:["__import__","abs","all","any","ascii","bin","bool","breakpoint","bytearray","bytes","callable","chr","classmethod","compile","complex","delattr","dict","dir","divmod","enumerate","eval","exec","filter","float","format","frozenset","getattr","globals","hasattr","hash","help","hex","id","input","int","isinstance","issubclass","iter","len","list","locals","map","max","memoryview","min","next","object","oct","open","ord","pow","print","property","range","repr","reversed","round","set","setattr","slice","sorted","staticmethod","str","sum","super","tuple","type","vars","zip"],literal:["__debug__","Ellipsis","False","None","NotImplemented","True"],type:["Any","Callable","Coroutine","Dict","List","Literal","Generic","Optional","Sequence","Set","Tuple","Type","Union"]},r={className:"meta",begin:/^(>>>|\.\.\.) /},s={className:"subst",begin:/\{/,end:/\}/,keywords:a,illegal:/#/},o={begin:/\{\{/,relevance:0},l={className:"string",contains:[e.BACKSLASH_ESCAPE],variants:[{begin:/([uU]|[bB]|[rR]|[bB][rR]|[rR][bB])?'''/,end:/'''/,contains:[e.BACKSLASH_ESCAPE,r],relevance:10},{begin:/([uU]|[bB]|[rR]|[bB][rR]|[rR][bB])?"""/,end:/"""/,contains:[e.BACKSLASH_ESCAPE,r],relevance:10},{begin:/([fF][rR]|[rR][fF]|[fF])'''/,end:/'''/,contains:[e.BACKSLASH_ESCAPE,r,o,s]},{begin:/([fF][rR]|[rR][fF]|[fF])"""/,end:/"""/,contains:[e.BACKSLASH_ESCAPE,r,o,s]},{begin:/([uU]|[rR])'/,end:/'/,relevance:10},{begin:/([uU]|[rR])"/,end:/"/,relevance:10},{begin:/([bB]|[bB][rR]|[rR][bB])'/,end:/'/},{begin:/([bB]|[bB][rR]|[rR][bB])"/,end:/"/},{begin:/([fF][rR]|[rR][fF]|[fF])'/,end:/'/,contains:[e.BACKSLASH_ESCAPE,o,s]},{begin:/([fF][rR]|[rR][fF]|[fF])"/,end:/"/,contains:[e.BACKSLASH_ESCAPE,o,s]},e.APOS_STRING_MODE,e.QUOTE_STRING_MODE]},c="[0-9](_?[0-9])*",d=`(\\b(${c}))?\\.(${c})|\\b(${c})\\.`,u=`\\b|${i.join("|")}`,g={className:"number",relevance:0,variants:[{begin:`(\\b(${c})|(${d}))[eE][+-]?(${c})[jJ]?(?=${u})`},{begin:`(${d})[jJ]?`},{begin:`\\b([1-9](_?[0-9])*|0+(_?0)*)[lLjJ]?(?=${u})`},{begin:`\\b0[bB](_?[01])+[lL]?(?=${u})`},{begin:`\\b0[oO](_?[0-7])+[lL]?(?=${u})`},{begin:`\\b0[xX](_?[0-9a-fA-F])+[lL]?(?=${u})`},{begin:`\\b(${c})[jJ](?=${u})`}]},b={className:"comment",begin:n.lookahead(/# type:/),end:/$/,keywords:a,contains:[{begin:/# type:/},{begin:/#/,end:/\b\B/,endsWithParent:!0}]},p={className:"params",variants:[{className:"",begin:/\(\s*\)/,skip:!0},{begin:/\(/,end:/\)/,excludeBegin:!0,excludeEnd:!0,keywords:a,contains:["self",r,g,l,e.HASH_COMMENT_MODE]}]};return s.contains=[l,g,r],{name:"Python",aliases:["py","gyp","ipython"],unicodeRegex:!0,keywords:a,illegal:/(<\/|\?)|=>/,contains:[r,g,{begin:/\bself\b/},{beginKeywords:"if",relevance:0},l,b,e.HASH_COMMENT_MODE,{match:[/\bdef/,/\s+/,t],scope:{1:"keyword",3:"title.function"},contains:[p]},{variants:[{match:[/\bclass/,/\s+/,t,/\s*/,/\(\s*/,t,/\s*\)/]},{match:[/\bclass/,/\s+/,t]}],scope:{1:"keyword",3:"title.class",6:"title.class.inherited"}},{className:"meta",begin:/^[\t ]*@/,end:/(?=#)|$/,contains:[g,p,l]}]}}},7905:e=>{e.exports=function(e){const n=e.regex,t="([a-zA-Z_]\\w*[!?=]?|[-+~]@|<<|>>|=~|===?|<=>|[<>]=?|\\*\\*|[-/+%^&*~`|]|\\[\\]=?)",i=n.either(/\b([A-Z]+[a-z0-9]+)+/,/\b([A-Z]+[a-z0-9]+)+[A-Z]+/),a=n.concat(i,/(::\w+)*/),r={"variable.constant":["__FILE__","__LINE__","__ENCODING__"],"variable.language":["self","super"],keyword:["alias","and","begin","BEGIN","break","case","class","defined","do","else","elsif","end","END","ensure","for","if","in","module","next","not","or","redo","require","rescue","retry","return","then","undef","unless","until","when","while","yield","include","extend","prepend","public","private","protected","raise","throw"],built_in:["proc","lambda","attr_accessor","attr_reader","attr_writer","define_method","private_constant","module_function"],literal:["true","false","nil"]},s={className:"doctag",begin:"@[A-Za-z]+"},o={begin:"#<",end:">"},l=[e.COMMENT("#","$",{contains:[s]}),e.COMMENT("^=begin","^=end",{contains:[s],relevance:10}),e.COMMENT("^__END__",e.MATCH_NOTHING_RE)],c={className:"subst",begin:/#\{/,end:/\}/,keywords:r},d={className:"string",contains:[e.BACKSLASH_ESCAPE,c],variants:[{begin:/'/,end:/'/},{begin:/"/,end:/"/},{begin:/`/,end:/`/},{begin:/%[qQwWx]?\(/,end:/\)/},{begin:/%[qQwWx]?\[/,end:/\]/},{begin:/%[qQwWx]?\{/,end:/\}/},{begin:/%[qQwWx]?/},{begin:/%[qQwWx]?\//,end:/\//},{begin:/%[qQwWx]?%/,end:/%/},{begin:/%[qQwWx]?-/,end:/-/},{begin:/%[qQwWx]?\|/,end:/\|/},{begin:/\B\?(\\\d{1,3})/},{begin:/\B\?(\\x[A-Fa-f0-9]{1,2})/},{begin:/\B\?(\\u\{?[A-Fa-f0-9]{1,6}\}?)/},{begin:/\B\?(\\M-\\C-|\\M-\\c|\\c\\M-|\\M-|\\C-\\M-)[\x20-\x7e]/},{begin:/\B\?\\(c|C-)[\x20-\x7e]/},{begin:/\B\?\\?\S/},{begin:n.concat(/<<[-~]?'?/,n.lookahead(/(\w+)(?=\W)[^\n]*\n(?:[^\n]*\n)*?\s*\1\b/)),contains:[e.END_SAME_AS_BEGIN({begin:/(\w+)/,end:/(\w+)/,contains:[e.BACKSLASH_ESCAPE,c]})]}]},u="[0-9](_?[0-9])*",g={className:"number",relevance:0,variants:[{begin:`\\b([1-9](_?[0-9])*|0)(\\.(${u}))?([eE][+-]?(${u})|r)?i?\\b`},{begin:"\\b0[dD][0-9](_?[0-9])*r?i?\\b"},{begin:"\\b0[bB][0-1](_?[0-1])*r?i?\\b"},{begin:"\\b0[oO][0-7](_?[0-7])*r?i?\\b"},{begin:"\\b0[xX][0-9a-fA-F](_?[0-9a-fA-F])*r?i?\\b"},{begin:"\\b0(_?[0-7])+r?i?\\b"}]},b={variants:[{match:/\(\)/},{className:"params",begin:/\(/,end:/(?=\))/,excludeBegin:!0,endsParent:!0,keywords:r}]},p=[d,{variants:[{match:[/class\s+/,a,/\s+<\s+/,a]},{match:[/\b(class|module)\s+/,a]}],scope:{2:"title.class",4:"title.class.inherited"},keywords:r},{match:[/(include|extend)\s+/,a],scope:{2:"title.class"},keywords:r},{relevance:0,match:[a,/\.new[. (]/],scope:{1:"title.class"}},{relevance:0,match:/\b[A-Z][A-Z_0-9]+\b/,className:"variable.constant"},{relevance:0,match:i,scope:"title.class"},{match:[/def/,/\s+/,t],scope:{1:"keyword",3:"title.function"},contains:[b]},{begin:e.IDENT_RE+"::"},{className:"symbol",begin:e.UNDERSCORE_IDENT_RE+"(!|\\?)?:",relevance:0},{className:"symbol",begin:":(?!\\s)",contains:[d,{begin:t}],relevance:0},g,{className:"variable",begin:"(\\$\\W)|((\\$|@@?)(\\w+))(?=[^@$?])(?![A-Za-z])(?![@$?'])"},{className:"params",begin:/\|/,end:/\|/,excludeBegin:!0,excludeEnd:!0,relevance:0,keywords:r},{begin:"("+e.RE_STARTERS_RE+"|unless)\\s*",keywords:"unless",contains:[{className:"regexp",contains:[e.BACKSLASH_ESCAPE,c],illegal:/\n/,variants:[{begin:"/",end:"/[a-z]*"},{begin:/%r\{/,end:/\}[a-z]*/},{begin:"%r\\(",end:"\\)[a-z]*"},{begin:"%r!",end:"![a-z]*"},{begin:"%r\\[",end:"\\][a-z]*"}]}].concat(o,l),relevance:0}].concat(o,l);c.contains=p,b.contains=p;const f=[{begin:/^\s*=>/,starts:{end:"$",contains:p}},{className:"meta.prompt",begin:"^([>?]>|[\\w#]+\\(\\w+\\):\\d+:\\d+[>*]|(\\w+-)?\\d+\\.\\d+\\.\\d+(p\\d+)?[^\\d][^>]+>)(?=[ ])",starts:{end:"$",keywords:r,contains:p}}];return l.unshift(o),{name:"Ruby",aliases:["rb","gemspec","podspec","thor","irb"],keywords:r,illegal:/\/\*/,contains:[e.SHEBANG({binary:"ruby"})].concat(f).concat(l).concat(p)}}},9880:e=>{e.exports=function(e){const n=e.regex,t={className:"title.function.invoke",relevance:0,begin:n.concat(/\b/,/(?!let\b)/,e.IDENT_RE,n.lookahead(/\s*\(/))},i="([ui](8|16|32|64|128|size)|f(32|64))?",a=["drop ","Copy","Send","Sized","Sync","Drop","Fn","FnMut","FnOnce","ToOwned","Clone","Debug","PartialEq","PartialOrd","Eq","Ord","AsRef","AsMut","Into","From","Default","Iterator","Extend","IntoIterator","DoubleEndedIterator","ExactSizeIterator","SliceConcatExt","ToString","assert!","assert_eq!","bitflags!","bytes!","cfg!","col!","concat!","concat_idents!","debug_assert!","debug_assert_eq!","env!","panic!","file!","format!","format_args!","include_bytes!","include_str!","line!","local_data_key!","module_path!","option_env!","print!","println!","select!","stringify!","try!","unimplemented!","unreachable!","vec!","write!","writeln!","macro_rules!","assert_ne!","debug_assert_ne!"],r=["i8","i16","i32","i64","i128","isize","u8","u16","u32","u64","u128","usize","f32","f64","str","char","bool","Box","Option","Result","String","Vec"];return{name:"Rust",aliases:["rs"],keywords:{$pattern:e.IDENT_RE+"!?",type:r,keyword:["abstract","as","async","await","become","box","break","const","continue","crate","do","dyn","else","enum","extern","false","final","fn","for","if","impl","in","let","loop","macro","match","mod","move","mut","override","priv","pub","ref","return","self","Self","static","struct","super","trait","true","try","type","typeof","unsafe","unsized","use","virtual","where","while","yield"],literal:["true","false","Some","None","Ok","Err"],built_in:a},illegal:""},t]}}},729:e=>{e.exports=function(e){const n=e.regex,t={className:"subst",variants:[{begin:"\\$[A-Za-z0-9_]+"},{begin:/\$\{/,end:/\}/}]},i={className:"string",variants:[{begin:'"""',end:'"""'},{begin:'"',end:'"',illegal:"\\n",contains:[e.BACKSLASH_ESCAPE]},{begin:'[a-z]+"',end:'"',illegal:"\\n",contains:[e.BACKSLASH_ESCAPE,t]},{className:"string",begin:'[a-z]+"""',end:'"""',contains:[t],relevance:10}]},a={className:"type",begin:"\\b[A-Z][A-Za-z0-9_]*",relevance:0},r={className:"title",begin:/[^0-9\n\t "'(),.`{}\[\]:;][^\n\t "'(),.`{}\[\]:;]+|[^0-9\n\t "'(),.`{}\[\]:;=]/,relevance:0},s={className:"class",beginKeywords:"class object trait type",end:/[:={\[\n;]/,excludeEnd:!0,contains:[e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,{beginKeywords:"extends with",relevance:10},{begin:/\[/,end:/\]/,excludeBegin:!0,excludeEnd:!0,relevance:0,contains:[a]},{className:"params",begin:/\(/,end:/\)/,excludeBegin:!0,excludeEnd:!0,relevance:0,contains:[a]},r]},o={className:"function",beginKeywords:"def",end:n.lookahead(/[:={\[(\n;]/),contains:[r]};return{name:"Scala",keywords:{literal:"true false null",keyword:"type yield lazy override def with val var sealed abstract private trait object if then forSome for while do throw finally protected extends import final return else break new catch super class case package default try this match continue throws implicit export enum given transparent"},contains:[e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,i,a,o,s,e.C_NUMBER_MODE,{begin:[/^\s*/,"extension",/\s+(?=[[(])/],beginScope:{2:"keyword"}},{begin:[/^\s*/,/end/,/\s+/,/(extension\b)?/],beginScope:{2:"keyword",4:"keyword"}},{match:/\.inline\b/},{begin:/\binline(?=\s)/,keywords:"inline"},{begin:[/\(\s*/,/using/,/\s+(?!\))/],beginScope:{2:"keyword"}},{className:"meta",begin:"@[A-Za-z]+"}]}}},1062:e=>{const n=["a","abbr","address","article","aside","audio","b","blockquote","body","button","canvas","caption","cite","code","dd","del","details","dfn","div","dl","dt","em","fieldset","figcaption","figure","footer","form","h1","h2","h3","h4","h5","h6","header","hgroup","html","i","iframe","img","input","ins","kbd","label","legend","li","main","mark","menu","nav","object","ol","p","q","quote","samp","section","span","strong","summary","sup","table","tbody","td","textarea","tfoot","th","thead","time","tr","ul","var","video"],t=["any-hover","any-pointer","aspect-ratio","color","color-gamut","color-index","device-aspect-ratio","device-height","device-width","display-mode","forced-colors","grid","height","hover","inverted-colors","monochrome","orientation","overflow-block","overflow-inline","pointer","prefers-color-scheme","prefers-contrast","prefers-reduced-motion","prefers-reduced-transparency","resolution","scan","scripting","update","width","min-width","max-width","min-height","max-height"],i=["active","any-link","blank","checked","current","default","defined","dir","disabled","drop","empty","enabled","first","first-child","first-of-type","fullscreen","future","focus","focus-visible","focus-within","has","host","host-context","hover","indeterminate","in-range","invalid","is","lang","last-child","last-of-type","left","link","local-link","not","nth-child","nth-col","nth-last-child","nth-last-col","nth-last-of-type","nth-of-type","only-child","only-of-type","optional","out-of-range","past","placeholder-shown","read-only","read-write","required","right","root","scope","target","target-within","user-invalid","valid","visited","where"],a=["after","backdrop","before","cue","cue-region","first-letter","first-line","grammar-error","marker","part","placeholder","selection","slotted","spelling-error"],r=["align-content","align-items","align-self","all","animation","animation-delay","animation-direction","animation-duration","animation-fill-mode","animation-iteration-count","animation-name","animation-play-state","animation-timing-function","backface-visibility","background","background-attachment","background-blend-mode","background-clip","background-color","background-image","background-origin","background-position","background-repeat","background-size","block-size","border","border-block","border-block-color","border-block-end","border-block-end-color","border-block-end-style","border-block-end-width","border-block-start","border-block-start-color","border-block-start-style","border-block-start-width","border-block-style","border-block-width","border-bottom","border-bottom-color","border-bottom-left-radius","border-bottom-right-radius","border-bottom-style","border-bottom-width","border-collapse","border-color","border-image","border-image-outset","border-image-repeat","border-image-slice","border-image-source","border-image-width","border-inline","border-inline-color","border-inline-end","border-inline-end-color","border-inline-end-style","border-inline-end-width","border-inline-start","border-inline-start-color","border-inline-start-style","border-inline-start-width","border-inline-style","border-inline-width","border-left","border-left-color","border-left-style","border-left-width","border-radius","border-right","border-right-color","border-right-style","border-right-width","border-spacing","border-style","border-top","border-top-color","border-top-left-radius","border-top-right-radius","border-top-style","border-top-width","border-width","bottom","box-decoration-break","box-shadow","box-sizing","break-after","break-before","break-inside","caption-side","caret-color","clear","clip","clip-path","clip-rule","color","column-count","column-fill","column-gap","column-rule","column-rule-color","column-rule-style","column-rule-width","column-span","column-width","columns","contain","content","content-visibility","counter-increment","counter-reset","cue","cue-after","cue-before","cursor","direction","display","empty-cells","filter","flex","flex-basis","flex-direction","flex-flow","flex-grow","flex-shrink","flex-wrap","float","flow","font","font-display","font-family","font-feature-settings","font-kerning","font-language-override","font-size","font-size-adjust","font-smoothing","font-stretch","font-style","font-synthesis","font-variant","font-variant-caps","font-variant-east-asian","font-variant-ligatures","font-variant-numeric","font-variant-position","font-variation-settings","font-weight","gap","glyph-orientation-vertical","grid","grid-area","grid-auto-columns","grid-auto-flow","grid-auto-rows","grid-column","grid-column-end","grid-column-start","grid-gap","grid-row","grid-row-end","grid-row-start","grid-template","grid-template-areas","grid-template-columns","grid-template-rows","hanging-punctuation","height","hyphens","icon","image-orientation","image-rendering","image-resolution","ime-mode","inline-size","isolation","justify-content","left","letter-spacing","line-break","line-height","list-style","list-style-image","list-style-position","list-style-type","margin","margin-block","margin-block-end","margin-block-start","margin-bottom","margin-inline","margin-inline-end","margin-inline-start","margin-left","margin-right","margin-top","marks","mask","mask-border","mask-border-mode","mask-border-outset","mask-border-repeat","mask-border-slice","mask-border-source","mask-border-width","mask-clip","mask-composite","mask-image","mask-mode","mask-origin","mask-position","mask-repeat","mask-size","mask-type","max-block-size","max-height","max-inline-size","max-width","min-block-size","min-height","min-inline-size","min-width","mix-blend-mode","nav-down","nav-index","nav-left","nav-right","nav-up","none","normal","object-fit","object-position","opacity","order","orphans","outline","outline-color","outline-offset","outline-style","outline-width","overflow","overflow-wrap","overflow-x","overflow-y","padding","padding-block","padding-block-end","padding-block-start","padding-bottom","padding-inline","padding-inline-end","padding-inline-start","padding-left","padding-right","padding-top","page-break-after","page-break-before","page-break-inside","pause","pause-after","pause-before","perspective","perspective-origin","pointer-events","position","quotes","resize","rest","rest-after","rest-before","right","row-gap","scroll-margin","scroll-margin-block","scroll-margin-block-end","scroll-margin-block-start","scroll-margin-bottom","scroll-margin-inline","scroll-margin-inline-end","scroll-margin-inline-start","scroll-margin-left","scroll-margin-right","scroll-margin-top","scroll-padding","scroll-padding-block","scroll-padding-block-end","scroll-padding-block-start","scroll-padding-bottom","scroll-padding-inline","scroll-padding-inline-end","scroll-padding-inline-start","scroll-padding-left","scroll-padding-right","scroll-padding-top","scroll-snap-align","scroll-snap-stop","scroll-snap-type","scrollbar-color","scrollbar-gutter","scrollbar-width","shape-image-threshold","shape-margin","shape-outside","speak","speak-as","src","tab-size","table-layout","text-align","text-align-all","text-align-last","text-combine-upright","text-decoration","text-decoration-color","text-decoration-line","text-decoration-style","text-emphasis","text-emphasis-color","text-emphasis-position","text-emphasis-style","text-indent","text-justify","text-orientation","text-overflow","text-rendering","text-shadow","text-transform","text-underline-position","top","transform","transform-box","transform-origin","transform-style","transition","transition-delay","transition-duration","transition-property","transition-timing-function","unicode-bidi","vertical-align","visibility","voice-balance","voice-duration","voice-family","voice-pitch","voice-range","voice-rate","voice-stress","voice-volume","white-space","widows","width","will-change","word-break","word-spacing","word-wrap","writing-mode","z-index"].reverse();e.exports=function(e){const s=(e=>({IMPORTANT:{scope:"meta",begin:"!important"},BLOCK_COMMENT:e.C_BLOCK_COMMENT_MODE,HEXCOLOR:{scope:"number",begin:/#(([0-9a-fA-F]{3,4})|(([0-9a-fA-F]{2}){3,4}))\b/},FUNCTION_DISPATCH:{className:"built_in",begin:/[\w-]+(?=\()/},ATTRIBUTE_SELECTOR_MODE:{scope:"selector-attr",begin:/\[/,end:/\]/,illegal:"$",contains:[e.APOS_STRING_MODE,e.QUOTE_STRING_MODE]},CSS_NUMBER_MODE:{scope:"number",begin:e.NUMBER_RE+"(%|em|ex|ch|rem|vw|vh|vmin|vmax|cm|mm|in|pt|pc|px|deg|grad|rad|turn|s|ms|Hz|kHz|dpi|dpcm|dppx)?",relevance:0},CSS_VARIABLE:{className:"attr",begin:/--[A-Za-z][A-Za-z0-9_-]*/}}))(e),o=a,l=i,c="@[a-z-]+",d={className:"variable",begin:"(\\$[a-zA-Z-][a-zA-Z0-9_-]*)\\b",relevance:0};return{name:"SCSS",case_insensitive:!0,illegal:"[=/|']",contains:[e.C_LINE_COMMENT_MODE,e.C_BLOCK_COMMENT_MODE,s.CSS_NUMBER_MODE,{className:"selector-id",begin:"#[A-Za-z0-9_-]+",relevance:0},{className:"selector-class",begin:"\\.[A-Za-z0-9_-]+",relevance:0},s.ATTRIBUTE_SELECTOR_MODE,{className:"selector-tag",begin:"\\b("+n.join("|")+")\\b",relevance:0},{className:"selector-pseudo",begin:":("+l.join("|")+")"},{className:"selector-pseudo",begin:":(:)?("+o.join("|")+")"},d,{begin:/\(/,end:/\)/,contains:[s.CSS_NUMBER_MODE]},s.CSS_VARIABLE,{className:"attribute",begin:"\\b("+r.join("|")+")\\b"},{begin:"\\b(whitespace|wait|w-resize|visible|vertical-text|vertical-ideographic|uppercase|upper-roman|upper-alpha|underline|transparent|top|thin|thick|text|text-top|text-bottom|tb-rl|table-header-group|table-footer-group|sw-resize|super|strict|static|square|solid|small-caps|separate|se-resize|scroll|s-resize|rtl|row-resize|ridge|right|repeat|repeat-y|repeat-x|relative|progress|pointer|overline|outside|outset|oblique|nowrap|not-allowed|normal|none|nw-resize|no-repeat|no-drop|newspaper|ne-resize|n-resize|move|middle|medium|ltr|lr-tb|lowercase|lower-roman|lower-alpha|loose|list-item|line|line-through|line-edge|lighter|left|keep-all|justify|italic|inter-word|inter-ideograph|inside|inset|inline|inline-block|inherit|inactive|ideograph-space|ideograph-parenthesis|ideograph-numeric|ideograph-alpha|horizontal|hidden|help|hand|groove|fixed|ellipsis|e-resize|double|dotted|distribute|distribute-space|distribute-letter|distribute-all-lines|disc|disabled|default|decimal|dashed|crosshair|collapse|col-resize|circle|char|center|capitalize|break-word|break-all|bottom|both|bolder|bold|block|bidi-override|below|baseline|auto|always|all-scroll|absolute|table|table-cell)\\b"},{begin:/:/,end:/[;}{]/,relevance:0,contains:[s.BLOCK_COMMENT,d,s.HEXCOLOR,s.CSS_NUMBER_MODE,e.QUOTE_STRING_MODE,e.APOS_STRING_MODE,s.IMPORTANT,s.FUNCTION_DISPATCH]},{begin:"@(page|font-face)",keywords:{$pattern:c,keyword:"@page @font-face"}},{begin:"@",end:"[{;]",returnBegin:!0,keywords:{$pattern:/[a-z-]+/,keyword:"and or not only",attribute:t.join(" ")},contains:[{begin:c,className:"keyword"},{begin:/[a-z-]+(?=:)/,className:"attribute"},d,e.QUOTE_STRING_MODE,e.APOS_STRING_MODE,s.HEXCOLOR,s.CSS_NUMBER_MODE]},s.FUNCTION_DISPATCH]}}},7874:e=>{e.exports=function(e){return{name:"Shell Session",aliases:["console","shellsession"],contains:[{className:"meta.prompt",begin:/^\s{0,3}[/~\w\d[\]()@-]*[>%$#][ ]?/,starts:{end:/[^\\](?=\s*$)/,subLanguage:"bash"}}]}}},8935:e=>{e.exports=function(e){const n=e.regex,t=e.COMMENT("--","$"),i=["true","false","unknown"],a=["bigint","binary","blob","boolean","char","character","clob","date","dec","decfloat","decimal","float","int","integer","interval","nchar","nclob","national","numeric","real","row","smallint","time","timestamp","varchar","varying","varbinary"],r=["abs","acos","array_agg","asin","atan","avg","cast","ceil","ceiling","coalesce","corr","cos","cosh","count","covar_pop","covar_samp","cume_dist","dense_rank","deref","element","exp","extract","first_value","floor","json_array","json_arrayagg","json_exists","json_object","json_objectagg","json_query","json_table","json_table_primitive","json_value","lag","last_value","lead","listagg","ln","log","log10","lower","max","min","mod","nth_value","ntile","nullif","percent_rank","percentile_cont","percentile_disc","position","position_regex","power","rank","regr_avgx","regr_avgy","regr_count","regr_intercept","regr_r2","regr_slope","regr_sxx","regr_sxy","regr_syy","row_number","sin","sinh","sqrt","stddev_pop","stddev_samp","substring","substring_regex","sum","tan","tanh","translate","translate_regex","treat","trim","trim_array","unnest","upper","value_of","var_pop","var_samp","width_bucket"],s=["create table","insert into","primary key","foreign key","not null","alter table","add constraint","grouping sets","on overflow","character set","respect nulls","ignore nulls","nulls first","nulls last","depth first","breadth first"],o=r,l=["abs","acos","all","allocate","alter","and","any","are","array","array_agg","array_max_cardinality","as","asensitive","asin","asymmetric","at","atan","atomic","authorization","avg","begin","begin_frame","begin_partition","between","bigint","binary","blob","boolean","both","by","call","called","cardinality","cascaded","case","cast","ceil","ceiling","char","char_length","character","character_length","check","classifier","clob","close","coalesce","collate","collect","column","commit","condition","connect","constraint","contains","convert","copy","corr","corresponding","cos","cosh","count","covar_pop","covar_samp","create","cross","cube","cume_dist","current","current_catalog","current_date","current_default_transform_group","current_path","current_role","current_row","current_schema","current_time","current_timestamp","current_path","current_role","current_transform_group_for_type","current_user","cursor","cycle","date","day","deallocate","dec","decimal","decfloat","declare","default","define","delete","dense_rank","deref","describe","deterministic","disconnect","distinct","double","drop","dynamic","each","element","else","empty","end","end_frame","end_partition","end-exec","equals","escape","every","except","exec","execute","exists","exp","external","extract","false","fetch","filter","first_value","float","floor","for","foreign","frame_row","free","from","full","function","fusion","get","global","grant","group","grouping","groups","having","hold","hour","identity","in","indicator","initial","inner","inout","insensitive","insert","int","integer","intersect","intersection","interval","into","is","join","json_array","json_arrayagg","json_exists","json_object","json_objectagg","json_query","json_table","json_table_primitive","json_value","lag","language","large","last_value","lateral","lead","leading","left","like","like_regex","listagg","ln","local","localtime","localtimestamp","log","log10","lower","match","match_number","match_recognize","matches","max","member","merge","method","min","minute","mod","modifies","module","month","multiset","national","natural","nchar","nclob","new","no","none","normalize","not","nth_value","ntile","null","nullif","numeric","octet_length","occurrences_regex","of","offset","old","omit","on","one","only","open","or","order","out","outer","over","overlaps","overlay","parameter","partition","pattern","per","percent","percent_rank","percentile_cont","percentile_disc","period","portion","position","position_regex","power","precedes","precision","prepare","primary","procedure","ptf","range","rank","reads","real","recursive","ref","references","referencing","regr_avgx","regr_avgy","regr_count","regr_intercept","regr_r2","regr_slope","regr_sxx","regr_sxy","regr_syy","release","result","return","returns","revoke","right","rollback","rollup","row","row_number","rows","running","savepoint","scope","scroll","search","second","seek","select","sensitive","session_user","set","show","similar","sin","sinh","skip","smallint","some","specific","specifictype","sql","sqlexception","sqlstate","sqlwarning","sqrt","start","static","stddev_pop","stddev_samp","submultiset","subset","substring","substring_regex","succeeds","sum","symmetric","system","system_time","system_user","table","tablesample","tan","tanh","then","time","timestamp","timezone_hour","timezone_minute","to","trailing","translate","translate_regex","translation","treat","trigger","trim","trim_array","true","truncate","uescape","union","unique","unknown","unnest","update","upper","user","using","value","values","value_of","var_pop","var_samp","varbinary","varchar","varying","versioning","when","whenever","where","width_bucket","window","with","within","without","year","add","asc","collation","desc","final","first","last","view"].filter((e=>!r.includes(e))),c={begin:n.concat(/\b/,n.either(...o),/\s*\(/),relevance:0,keywords:{built_in:o}};return{name:"SQL",case_insensitive:!0,illegal:/[{}]|<\//,keywords:{$pattern:/\b[\w\.]+/,keyword:function(e,{exceptions:n,when:t}={}){const i=t;return n=n||[],e.map((e=>e.match(/\|\d+$/)||n.includes(e)?e:i(e)?`${e}|0`:e))}(l,{when:e=>e.length<3}),literal:i,type:a,built_in:["current_catalog","current_date","current_default_transform_group","current_path","current_role","current_schema","current_transform_group_for_type","current_user","session_user","system_time","system_user","current_time","localtime","current_timestamp","localtimestamp"]},contains:[{begin:n.either(...s),relevance:0,keywords:{$pattern:/[\w\.]+/,keyword:l.concat(s),literal:i,type:a}},{className:"type",begin:n.either("double precision","large object","with timezone","without timezone")},c,{className:"variable",begin:/@[a-z0-9][a-z0-9_]*/},{className:"string",variants:[{begin:/'/,end:/'/,contains:[{begin:/''/}]}]},{begin:/"/,end:/"/,contains:[{begin:/""/}]},e.C_NUMBER_MODE,e.C_BLOCK_COMMENT_MODE,t,{className:"operator",begin:/[-+*/=%^~]|&&?|\|\|?|!=?|<(?:=>?|<|>)?|>[>=]?/,relevance:0}]}}},7690:e=>{function n(e){return e?"string"==typeof e?e:e.source:null}function t(e){return i("(?=",e,")")}function i(...e){return e.map((e=>n(e))).join("")}function a(...e){const t=function(e){const n=e[e.length-1];return"object"==typeof n&&n.constructor===Object?(e.splice(e.length-1,1),n):{}}(e);return"("+(t.capture?"":"?:")+e.map((e=>n(e))).join("|")+")"}const r=e=>i(/\b/,e,/\w$/.test(e)?/\b/:/\B/),s=["Protocol","Type"].map(r),o=["init","self"].map(r),l=["Any","Self"],c=["actor","any","associatedtype","async","await",/as\?/,/as!/,"as","break","case","catch","class","continue","convenience","default","defer","deinit","didSet","distributed","do","dynamic","else","enum","extension","fallthrough",/fileprivate\(set\)/,"fileprivate","final","for","func","get","guard","if","import","indirect","infix",/init\?/,/init!/,"inout",/internal\(set\)/,"internal","in","is","isolated","nonisolated","lazy","let","mutating","nonmutating",/open\(set\)/,"open","operator","optional","override","postfix","precedencegroup","prefix",/private\(set\)/,"private","protocol",/public\(set\)/,"public","repeat","required","rethrows","return","set","some","static","struct","subscript","super","switch","throws","throw",/try\?/,/try!/,"try","typealias",/unowned\(safe\)/,/unowned\(unsafe\)/,"unowned","var","weak","where","while","willSet"],d=["false","nil","true"],u=["assignment","associativity","higherThan","left","lowerThan","none","right"],g=["#colorLiteral","#column","#dsohandle","#else","#elseif","#endif","#error","#file","#fileID","#fileLiteral","#filePath","#function","#if","#imageLiteral","#keyPath","#line","#selector","#sourceLocation","#warn_unqualified_access","#warning"],b=["abs","all","any","assert","assertionFailure","debugPrint","dump","fatalError","getVaList","isKnownUniquelyReferenced","max","min","numericCast","pointwiseMax","pointwiseMin","precondition","preconditionFailure","print","readLine","repeatElement","sequence","stride","swap","swift_unboxFromSwiftValueWithType","transcode","type","unsafeBitCast","unsafeDowncast","withExtendedLifetime","withUnsafeMutablePointer","withUnsafePointer","withVaList","withoutActuallyEscaping","zip"],p=a(/[/=\-+!*%<>&|^~?]/,/[\u00A1-\u00A7]/,/[\u00A9\u00AB]/,/[\u00AC\u00AE]/,/[\u00B0\u00B1]/,/[\u00B6\u00BB\u00BF\u00D7\u00F7]/,/[\u2016-\u2017]/,/[\u2020-\u2027]/,/[\u2030-\u203E]/,/[\u2041-\u2053]/,/[\u2055-\u205E]/,/[\u2190-\u23FF]/,/[\u2500-\u2775]/,/[\u2794-\u2BFF]/,/[\u2E00-\u2E7F]/,/[\u3001-\u3003]/,/[\u3008-\u3020]/,/[\u3030]/),f=a(p,/[\u0300-\u036F]/,/[\u1DC0-\u1DFF]/,/[\u20D0-\u20FF]/,/[\uFE00-\uFE0F]/,/[\uFE20-\uFE2F]/),m=i(p,f,"*"),E=a(/[a-zA-Z_]/,/[\u00A8\u00AA\u00AD\u00AF\u00B2-\u00B5\u00B7-\u00BA]/,/[\u00BC-\u00BE\u00C0-\u00D6\u00D8-\u00F6\u00F8-\u00FF]/,/[\u0100-\u02FF\u0370-\u167F\u1681-\u180D\u180F-\u1DBF]/,/[\u1E00-\u1FFF]/,/[\u200B-\u200D\u202A-\u202E\u203F-\u2040\u2054\u2060-\u206F]/,/[\u2070-\u20CF\u2100-\u218F\u2460-\u24FF\u2776-\u2793]/,/[\u2C00-\u2DFF\u2E80-\u2FFF]/,/[\u3004-\u3007\u3021-\u302F\u3031-\u303F\u3040-\uD7FF]/,/[\uF900-\uFD3D\uFD40-\uFDCF\uFDF0-\uFE1F\uFE30-\uFE44]/,/[\uFE47-\uFEFE\uFF00-\uFFFD]/),h=a(E,/\d/,/[\u0300-\u036F\u1DC0-\u1DFF\u20D0-\u20FF\uFE20-\uFE2F]/),_=i(E,h,"*"),N=i(/[A-Z]/,h,"*"),T=["autoclosure",i(/convention\(/,a("swift","block","c"),/\)/),"discardableResult","dynamicCallable","dynamicMemberLookup","escaping","frozen","GKInspectable","IBAction","IBDesignable","IBInspectable","IBOutlet","IBSegueAction","inlinable","main","nonobjc","NSApplicationMain","NSCopying","NSManaged",i(/objc\(/,_,/\)/),"objc","objcMembers","propertyWrapper","requires_stored_property_inits","resultBuilder","testable","UIApplicationMain","unknown","usableFromInline"],O=["iOS","iOSApplicationExtension","macOS","macOSApplicationExtension","macCatalyst","macCatalystApplicationExtension","watchOS","watchOSApplicationExtension","tvOS","tvOSApplicationExtension","swift"];e.exports=function(e){const n={match:/\s+/,relevance:0},p=e.COMMENT("/\\*","\\*/",{contains:["self"]}),E=[e.C_LINE_COMMENT_MODE,p],A={match:[/\./,a(...s,...o)],className:{2:"keyword"}},v={match:i(/\./,a(...c)),relevance:0},y=c.filter((e=>"string"==typeof e)).concat(["_|0"]),S={variants:[{className:"keyword",match:a(...c.filter((e=>"string"!=typeof e)).concat(l).map(r),...o)}]},R={$pattern:a(/\b\w+/,/#\w+/),keyword:y.concat(g),literal:d},I=[A,v,S],w=[{match:i(/\./,a(...b)),relevance:0},{className:"built_in",match:i(/\b/,a(...b),/(?=\()/)}],C={match:/->/,relevance:0},L=[C,{className:"operator",relevance:0,variants:[{match:m},{match:`\\.(\\.|${f})+`}]}],M="([0-9]_*)+",D="([0-9a-fA-F]_*)+",x={className:"number",relevance:0,variants:[{match:`\\b(${M})(\\.(${M}))?([eE][+-]?(${M}))?\\b`},{match:`\\b0x(${D})(\\.(${D}))?([pP][+-]?(${M}))?\\b`},{match:/\b0o([0-7]_*)+\b/},{match:/\b0b([01]_*)+\b/}]},k=(e="")=>({className:"subst",variants:[{match:i(/\\/,e,/[0\\tnr"']/)},{match:i(/\\/,e,/u\{[0-9a-fA-F]{1,8}\}/)}]}),P=(e="")=>({className:"subst",match:i(/\\/,e,/[\t ]*(?:[\r\n]|\r\n)/)}),U=(e="")=>({className:"subst",label:"interpol",begin:i(/\\/,e,/\(/),end:/\)/}),B=(e="")=>({begin:i(e,/"""/),end:i(/"""/,e),contains:[k(e),P(e),U(e)]}),F=(e="")=>({begin:i(e,/"/),end:i(/"/,e),contains:[k(e),U(e)]}),j={className:"string",variants:[B(),B("#"),B("##"),B("###"),F(),F("#"),F("##"),F("###")]},H={match:i(/`/,_,/`/)},G=[H,{className:"variable",match:/\$\d+/},{className:"variable",match:`\\$${h}+`}],$=[{match:/(@|#(un)?)available/,className:"keyword",starts:{contains:[{begin:/\(/,end:/\)/,keywords:O,contains:[...L,x,j]}]}},{className:"keyword",match:i(/@/,a(...T))},{className:"meta",match:i(/@/,_)}],z={match:t(/\b[A-Z]/),relevance:0,contains:[{className:"type",match:i(/(AV|CA|CF|CG|CI|CL|CM|CN|CT|MK|MP|MTK|MTL|NS|SCN|SK|UI|WK|XC)/,h,"+")},{className:"type",match:N,relevance:0},{match:/[?!]+/,relevance:0},{match:/\.\.\./,relevance:0},{match:i(/\s+&\s+/,t(N)),relevance:0}]},V={begin://,keywords:R,contains:[...E,...I,...$,C,z]};z.contains.push(V);const K={begin:/\(/,end:/\)/,relevance:0,keywords:R,contains:["self",{match:i(_,/\s*:/),keywords:"_|0",relevance:0},...E,...I,...w,...L,x,j,...G,...$,z]},W={begin://,contains:[...E,z]},q={begin:/\(/,end:/\)/,keywords:R,contains:[{begin:a(t(i(_,/\s*:/)),t(i(_,/\s+/,_,/\s*:/))),end:/:/,relevance:0,contains:[{className:"keyword",match:/\b_\b/},{className:"params",match:_}]},...E,...I,...L,x,j,...$,z,K],endsParent:!0,illegal:/["']/},X={match:[/func/,/\s+/,a(H.match,_,m)],className:{1:"keyword",3:"title.function"},contains:[W,q,n],illegal:[/\[/,/%/]},Y={match:[/\b(?:subscript|init[?!]?)/,/\s*(?=[<(])/],className:{1:"keyword"},contains:[W,q,n],illegal:/\[|%/},Z={match:[/operator/,/\s+/,m],className:{1:"keyword",3:"title"}},Q={begin:[/precedencegroup/,/\s+/,N],className:{1:"keyword",3:"title"},contains:[z],keywords:[...u,...d],end:/}/};for(const e of j.variants){const n=e.contains.find((e=>"interpol"===e.label));n.keywords=R;const t=[...I,...w,...L,x,j,...G];n.contains=[...t,{begin:/\(/,end:/\)/,contains:["self",...t]}]}return{name:"Swift",keywords:R,contains:[...E,X,Y,{beginKeywords:"struct protocol class extension enum actor",end:"\\{",excludeEnd:!0,keywords:R,contains:[e.inherit(e.TITLE_MODE,{className:"title.class",begin:/[A-Za-z$_][\u00C0-\u02B80-9A-Za-z$_]*/}),...I]},Z,Q,{beginKeywords:"import",end:/$/,contains:[...E],relevance:0},...I,...w,...L,x,j,...G,...$,z,K]}}},8987:e=>{const n="[A-Za-z$_][0-9A-Za-z$_]*",t=["as","in","of","if","for","while","finally","var","new","function","do","return","void","else","break","catch","instanceof","with","throw","case","default","try","switch","continue","typeof","delete","let","yield","const","class","debugger","async","await","static","import","from","export","extends"],i=["true","false","null","undefined","NaN","Infinity"],a=["Object","Function","Boolean","Symbol","Math","Date","Number","BigInt","String","RegExp","Array","Float32Array","Float64Array","Int8Array","Uint8Array","Uint8ClampedArray","Int16Array","Int32Array","Uint16Array","Uint32Array","BigInt64Array","BigUint64Array","Set","Map","WeakSet","WeakMap","ArrayBuffer","SharedArrayBuffer","Atomics","DataView","JSON","Promise","Generator","GeneratorFunction","AsyncFunction","Reflect","Proxy","Intl","WebAssembly"],r=["Error","EvalError","InternalError","RangeError","ReferenceError","SyntaxError","TypeError","URIError"],s=["setInterval","setTimeout","clearInterval","clearTimeout","require","exports","eval","isFinite","isNaN","parseFloat","parseInt","decodeURI","decodeURIComponent","encodeURI","encodeURIComponent","escape","unescape"],o=["arguments","this","super","console","window","document","localStorage","sessionStorage","module","global"],l=[].concat(s,a,r);function c(e){const c=e.regex,d=n,u={begin:/<[A-Za-z0-9\\._:-]+/,end:/\/[A-Za-z0-9\\._:-]+>|\/>/,isTrulyOpeningTag:(e,n)=>{const t=e[0].length+e.index,i=e.input[t];if("<"===i||","===i)return void n.ignoreMatch();let a;">"===i&&(((e,{after:n})=>{const t="",k={match:[/const|var|let/,/\s+/,d,/\s*/,/=\s*/,/(async\s*)?/,c.lookahead(x)],keywords:"async",className:{1:"keyword",3:"title.function"},contains:[S]};return{name:"JavaScript",aliases:["js","jsx","mjs","cjs"],keywords:g,exports:{PARAMS_CONTAINS:y,CLASS_REFERENCE:I},illegal:/#(?![$_A-z])/,contains:[e.SHEBANG({label:"shebang",binary:"node",relevance:5}),{label:"use_strict",className:"meta",relevance:10,begin:/^\s*['"]use (strict|asm)['"]/},e.APOS_STRING_MODE,e.QUOTE_STRING_MODE,h,_,N,T,O,{match:/\$\d+/},m,I,{className:"attr",begin:d+c.lookahead(":"),relevance:0},k,{begin:"("+e.RE_STARTERS_RE+"|\\b(case|return|throw)\\b)\\s*",keywords:"return throw case",relevance:0,contains:[O,e.REGEXP_MODE,{className:"function",begin:x,returnBegin:!0,end:"\\s*=>",contains:[{className:"params",variants:[{begin:e.UNDERSCORE_IDENT_RE,relevance:0},{className:null,begin:/\(\s*\)/,skip:!0},{begin:/\(/,end:/\)/,excludeBegin:!0,excludeEnd:!0,keywords:g,contains:y}]}]},{begin:/,/,relevance:0},{match:/\s+/,relevance:0},{variants:[{begin:"<>",end:""},{match:/<[A-Za-z0-9\\._:-]+\s*\/>/},{begin:u.begin,"on:begin":u.isTrulyOpeningTag,end:u.end}],subLanguage:"xml",contains:[{begin:u.begin,end:u.end,skip:!0,contains:["self"]}]}]},w,{beginKeywords:"while if switch catch for"},{begin:"\\b(?!function)"+e.UNDERSCORE_IDENT_RE+"\\([^()]*(\\([^()]*(\\([^()]*\\)[^()]*)*\\)[^()]*)*\\)\\s*\\{",returnBegin:!0,label:"func.def",contains:[S,e.inherit(e.TITLE_MODE,{begin:d,className:"title.function"})]},{match:/\.\.\./,relevance:0},M,{match:"\\$"+d,relevance:0},{match:[/\bconstructor(?=\s*\()/],className:{1:"title.function"},contains:[S]},C,{relevance:0,match:/\b[A-Z][A-Z_0-9]+\b/,className:"variable.constant"},R,D,{match:/\$[(.]/}]}}e.exports=function(e){const a=c(e),r=n,s=["any","void","number","boolean","string","object","never","symbol","bigint","unknown"],d={beginKeywords:"namespace",end:/\{/,excludeEnd:!0,contains:[a.exports.CLASS_REFERENCE]},u={beginKeywords:"interface",end:/\{/,excludeEnd:!0,keywords:{keyword:"interface extends",built_in:s},contains:[a.exports.CLASS_REFERENCE]},g={$pattern:n,keyword:t.concat(["type","namespace","interface","public","private","protected","implements","declare","abstract","readonly","enum","override"]),literal:i,built_in:l.concat(s),"variable.language":o},b={className:"meta",begin:"@"+r},p=(e,n,t)=>{const i=e.contains.findIndex((e=>e.label===n));if(-1===i)throw new Error("can not find mode to replace");e.contains.splice(i,1,t)};return Object.assign(a.keywords,g),a.exports.PARAMS_CONTAINS.push(b),a.contains=a.contains.concat([b,d,u]),p(a,"shebang",e.SHEBANG()),p(a,"use_strict",{className:"meta",relevance:10,begin:/^\s*['"]use strict['"]/}),a.contains.find((e=>"func.def"===e.label)).relevance=0,Object.assign(a,{name:"TypeScript",aliases:["ts","tsx","mts","cts"]}),a}},4610:e=>{e.exports=function(e){const n=e.regex,t=n.concat(/[\p{L}_]/u,n.optional(/[\p{L}0-9_.-]*:/u),/[\p{L}0-9_.-]*/u),i={className:"symbol",begin:/&[a-z]+;|&#[0-9]+;|&#x[a-f0-9]+;/},a={begin:/\s/,contains:[{className:"keyword",begin:/#?[a-z_][a-z1-9_-]+/,illegal:/\n/}]},r=e.inherit(a,{begin:/\(/,end:/\)/}),s=e.inherit(e.APOS_STRING_MODE,{className:"string"}),o=e.inherit(e.QUOTE_STRING_MODE,{className:"string"}),l={endsWithParent:!0,illegal:/`]+/}]}]}]};return{name:"HTML, XML",aliases:["html","xhtml","rss","atom","xjb","xsd","xsl","plist","wsf","svg"],case_insensitive:!0,unicodeRegex:!0,contains:[{className:"meta",begin://,relevance:10,contains:[a,o,s,r,{begin:/\[/,end:/\]/,contains:[{className:"meta",begin://,contains:[a,r,o,s]}]}]},e.COMMENT(//,{relevance:10}),{begin://,relevance:10},i,{className:"meta",end:/\?>/,variants:[{begin:/<\?xml/,relevance:10,contains:[o]},{begin:/<\?[a-z][a-z0-9]+/}]},{className:"tag",begin:/)/,end:/>/,keywords:{name:"style"},contains:[l],starts:{end:/<\/style>/,returnEnd:!0,subLanguage:["css","xml"]}},{className:"tag",begin:/)/,end:/>/,keywords:{name:"script"},contains:[l],starts:{end:/<\/script>/,returnEnd:!0,subLanguage:["javascript","handlebars","xml"]}},{className:"tag",begin:/<>|<\/>/},{className:"tag",begin:n.concat(//,/>/,/\s/)))),end:/\/?>/,contains:[{className:"name",begin:t,relevance:0,starts:l}]},{className:"tag",begin:n.concat(/<\//,n.lookahead(n.concat(t,/>/))),contains:[{className:"name",begin:t,relevance:0},{begin:/>/,relevance:0,endsParent:!0}]}]}}},1392:e=>{e.exports=function(e){const n="true false yes no null",t="[\\w#;/?:@&=+$,.~*'()[\\]]+",i={className:"string",relevance:0,variants:[{begin:/'/,end:/'/},{begin:/"/,end:/"/},{begin:/\S+/}],contains:[e.BACKSLASH_ESCAPE,{className:"template-variable",variants:[{begin:/\{\{/,end:/\}\}/},{begin:/%\{/,end:/\}/}]}]},a=e.inherit(i,{variants:[{begin:/'/,end:/'/},{begin:/"/,end:/"/},{begin:/[^\s,{}[\]]+/}]}),r={end:",",endsWithParent:!0,excludeEnd:!0,keywords:n,relevance:0},s={begin:/\{/,end:/\}/,contains:[r],illegal:"\\n",relevance:0},o={begin:"\\[",end:"\\]",contains:[r],illegal:"\\n",relevance:0},l=[{className:"attr",variants:[{begin:"\\w[\\w :\\/.-]*:(?=[ \t]|$)"},{begin:'"\\w[\\w :\\/.-]*":(?=[ \t]|$)'},{begin:"'\\w[\\w :\\/.-]*':(?=[ \t]|$)"}]},{className:"meta",begin:"^---\\s*$",relevance:10},{className:"string",begin:"[\\|>]([1-9]?[+-])?[ ]*\\n( +)[^ ][^\\n]*\\n(\\2[^\\n]+\\n?)*"},{begin:"<%[%=-]?",end:"[%-]?%>",subLanguage:"ruby",excludeBegin:!0,excludeEnd:!0,relevance:0},{className:"type",begin:"!\\w+!"+t},{className:"type",begin:"!<"+t+">"},{className:"type",begin:"!"+t},{className:"type",begin:"!!"+t},{className:"meta",begin:"&"+e.UNDERSCORE_IDENT_RE+"$"},{className:"meta",begin:"\\*"+e.UNDERSCORE_IDENT_RE+"$"},{className:"bullet",begin:"-(?=[ ]|$)",relevance:0},e.HASH_COMMENT_MODE,{beginKeywords:n,keywords:{literal:n}},{className:"number",begin:"\\b[0-9]{4}(-[0-9][0-9]){0,2}([Tt \\t][0-9][0-9]?(:[0-9][0-9]){2})?(\\.[0-9]*)?([ \\t])*(Z|[-+][0-9][0-9]?(:[0-9][0-9])?)?\\b"},{className:"number",begin:e.C_NUMBER_RE+"\\b",relevance:0},s,o,i],c=[...l];return c.pop(),c.push(a),r.contains=c,{name:"YAML",case_insensitive:!0,aliases:["yml"],contains:l}}}},n={};function t(i){var a=n[i];if(void 0!==a)return a.exports;var r=n[i]={exports:{}};return e[i].call(r.exports,r,r.exports,t),r.exports}var i={};return(()=>{"use strict";var e=i;Object.defineProperty(e,"__esModule",{value:!0}),e.defaultDiff2HtmlUIConfig=e.Diff2HtmlUI=void 0;const n=t(1761),a=t(4169);Object.defineProperty(e,"defaultDiff2HtmlUIConfig",{enumerable:!0,get:function(){return a.defaultDiff2HtmlUIConfig}});class r extends a.Diff2HtmlUI{constructor(e,t,i={}){super(e,t,i,n.hljs)}}e.Diff2HtmlUI=r})(),i})())); \ No newline at end of file diff --git a/docs/docsgen/source/_static/diff2html.min.css b/docs/docsgen/source/_static/diff2html.min.css new file mode 100644 index 0000000..1534347 --- /dev/null +++ b/docs/docsgen/source/_static/diff2html.min.css @@ -0,0 +1 @@ +:root{--d2h-bg-color:#fff;--d2h-border-color:#ddd;--d2h-dim-color:rgba(0,0,0,.3);--d2h-line-border-color:#eee;--d2h-file-header-bg-color:#f7f7f7;--d2h-file-header-border-color:#d8d8d8;--d2h-empty-placeholder-bg-color:#f1f1f1;--d2h-empty-placeholder-border-color:#e1e1e1;--d2h-selected-color:#c8e1ff;--d2h-ins-bg-color:#dfd;--d2h-ins-border-color:#b4e2b4;--d2h-ins-highlight-bg-color:#97f295;--d2h-ins-label-color:#399839;--d2h-del-bg-color:#fee8e9;--d2h-del-border-color:#e9aeae;--d2h-del-highlight-bg-color:#ffb6ba;--d2h-del-label-color:#c33;--d2h-change-del-color:#fdf2d0;--d2h-change-ins-color:#ded;--d2h-info-bg-color:#f8fafd;--d2h-info-border-color:#d5e4f2;--d2h-change-label-color:#d0b44c;--d2h-moved-label-color:#3572b0;--d2h-dark-color:#e6edf3;--d2h-dark-bg-color:#0d1117;--d2h-dark-border-color:#30363d;--d2h-dark-dim-color:#6e7681;--d2h-dark-line-border-color:#21262d;--d2h-dark-file-header-bg-color:#161b22;--d2h-dark-file-header-border-color:#30363d;--d2h-dark-empty-placeholder-bg-color:hsla(215,8%,47%,.1);--d2h-dark-empty-placeholder-border-color:#30363d;--d2h-dark-selected-color:rgba(56,139,253,.1);--d2h-dark-ins-bg-color:rgba(46,160,67,.15);--d2h-dark-ins-border-color:rgba(46,160,67,.4);--d2h-dark-ins-highlight-bg-color:rgba(46,160,67,.4);--d2h-dark-ins-label-color:#3fb950;--d2h-dark-del-bg-color:rgba(248,81,73,.1);--d2h-dark-del-border-color:rgba(248,81,73,.4);--d2h-dark-del-highlight-bg-color:rgba(248,81,73,.4);--d2h-dark-del-label-color:#f85149;--d2h-dark-change-del-color:rgba(210,153,34,.2);--d2h-dark-change-ins-color:rgba(46,160,67,.25);--d2h-dark-info-bg-color:rgba(56,139,253,.1);--d2h-dark-info-border-color:rgba(56,139,253,.4);--d2h-dark-change-label-color:#d29922;--d2h-dark-moved-label-color:#3572b0}.d2h-wrapper{text-align:left}.d2h-file-header{background-color:#f7f7f7;background-color:var(--d2h-file-header-bg-color);border-bottom:1px solid #d8d8d8;border-bottom:1px solid var(--d2h-file-header-border-color);display:-webkit-box;display:-ms-flexbox;display:flex;font-family:Source Sans Pro,Helvetica Neue,Helvetica,Arial,sans-serif;height:35px;padding:5px 10px}.d2h-file-header.d2h-sticky-header{position:sticky;top:0;z-index:1}.d2h-file-stats{display:-webkit-box;display:-ms-flexbox;display:flex;font-size:14px;margin-left:auto}.d2h-lines-added{border:1px solid #b4e2b4;border:1px solid var(--d2h-ins-border-color);border-radius:5px 0 0 5px;color:#399839;color:var(--d2h-ins-label-color);padding:2px;text-align:right;vertical-align:middle}.d2h-lines-deleted{border:1px solid #e9aeae;border:1px solid var(--d2h-del-border-color);border-radius:0 5px 5px 0;color:#c33;color:var(--d2h-del-label-color);margin-left:1px;padding:2px;text-align:left;vertical-align:middle}.d2h-file-name-wrapper{-webkit-box-align:center;-ms-flex-align:center;align-items:center;display:-webkit-box;display:-ms-flexbox;display:flex;font-size:15px;width:100%}.d2h-file-name{overflow-x:hidden;text-overflow:ellipsis;white-space:nowrap}.d2h-file-wrapper{border:1px solid #ddd;border:1px solid var(--d2h-border-color);border-radius:3px;margin-bottom:1em}.d2h-file-collapse{-webkit-box-pack:end;-ms-flex-pack:end;-webkit-box-align:center;-ms-flex-align:center;align-items:center;border:1px solid #ddd;border:1px solid var(--d2h-border-color);border-radius:3px;cursor:pointer;display:none;font-size:12px;justify-content:flex-end;padding:4px 8px}.d2h-file-collapse.d2h-selected{background-color:#c8e1ff;background-color:var(--d2h-selected-color)}.d2h-file-collapse-input{margin:0 4px 0 0}.d2h-diff-table{border-collapse:collapse;font-family:Menlo,Consolas,monospace;font-size:13px;width:100%}.d2h-files-diff{display:-webkit-box;display:-ms-flexbox;display:flex;width:100%}.d2h-file-diff{overflow-y:hidden}.d2h-file-diff.d2h-d-none,.d2h-files-diff.d2h-d-none{display:none}.d2h-file-side-diff{display:inline-block;overflow-x:scroll;overflow-y:hidden;width:50%}.d2h-code-line{padding:0 8em;width:calc(100% - 16em)}.d2h-code-line,.d2h-code-side-line{display:inline-block;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;white-space:nowrap}.d2h-code-side-line{padding:0 4.5em;width:calc(100% - 9em)}.d2h-code-line-ctn{word-wrap:normal;background:none;display:inline-block;padding:0;-webkit-user-select:text;-moz-user-select:text;-ms-user-select:text;user-select:text;vertical-align:middle;white-space:pre;width:100%}.d2h-code-line del,.d2h-code-side-line del{background-color:#ffb6ba;background-color:var(--d2h-del-highlight-bg-color)}.d2h-code-line del,.d2h-code-line ins,.d2h-code-side-line del,.d2h-code-side-line ins{border-radius:.2em;display:inline-block;margin-top:-1px;-webkit-text-decoration:none;text-decoration:none}.d2h-code-line ins,.d2h-code-side-line ins{background-color:#97f295;background-color:var(--d2h-ins-highlight-bg-color);text-align:left}.d2h-code-line-prefix{word-wrap:normal;background:none;display:inline;padding:0;white-space:pre}.line-num1{float:left}.line-num1,.line-num2{-webkit-box-sizing:border-box;box-sizing:border-box;overflow:hidden;padding:0 .5em;text-overflow:ellipsis;width:3.5em}.line-num2{float:right}.d2h-code-linenumber{background-color:#fff;background-color:var(--d2h-bg-color);border:solid #eee;border:solid var(--d2h-line-border-color);border-width:0 1px;-webkit-box-sizing:border-box;box-sizing:border-box;color:rgba(0,0,0,.3);color:var(--d2h-dim-color);cursor:pointer;display:inline-block;position:absolute;text-align:right;width:7.5em}.d2h-code-linenumber:after{content:"\200b"}.d2h-code-side-linenumber{background-color:#fff;background-color:var(--d2h-bg-color);border:solid #eee;border:solid var(--d2h-line-border-color);border-width:0 1px;-webkit-box-sizing:border-box;box-sizing:border-box;color:rgba(0,0,0,.3);color:var(--d2h-dim-color);cursor:pointer;display:inline-block;overflow:hidden;padding:0 .5em;position:absolute;text-align:right;text-overflow:ellipsis;width:4em}.d2h-code-side-linenumber:after{content:"\200b"}.d2h-code-side-emptyplaceholder,.d2h-emptyplaceholder{background-color:#f1f1f1;background-color:var(--d2h-empty-placeholder-bg-color);border-color:#e1e1e1;border-color:var(--d2h-empty-placeholder-border-color)}.d2h-code-line-prefix,.d2h-code-linenumber,.d2h-code-side-linenumber,.d2h-emptyplaceholder{-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none}.d2h-code-linenumber,.d2h-code-side-linenumber{direction:rtl}.d2h-del{background-color:#fee8e9;background-color:var(--d2h-del-bg-color);border-color:#e9aeae;border-color:var(--d2h-del-border-color)}.d2h-ins{background-color:#dfd;background-color:var(--d2h-ins-bg-color);border-color:#b4e2b4;border-color:var(--d2h-ins-border-color)}.d2h-info{background-color:#f8fafd;background-color:var(--d2h-info-bg-color);border-color:#d5e4f2;border-color:var(--d2h-info-border-color);color:rgba(0,0,0,.3);color:var(--d2h-dim-color)}.d2h-file-diff .d2h-del.d2h-change{background-color:#fdf2d0;background-color:var(--d2h-change-del-color)}.d2h-file-diff .d2h-ins.d2h-change{background-color:#ded;background-color:var(--d2h-change-ins-color)}.d2h-file-list-wrapper{margin-bottom:10px}.d2h-file-list-wrapper a{-webkit-text-decoration:none;text-decoration:none}.d2h-file-list-wrapper a,.d2h-file-list-wrapper a:visited{color:#3572b0;color:var(--d2h-moved-label-color)}.d2h-file-list-header{text-align:left}.d2h-file-list-title{font-weight:700}.d2h-file-list-line{display:-webkit-box;display:-ms-flexbox;display:flex;text-align:left}.d2h-file-list{display:block;list-style:none;margin:0;padding:0}.d2h-file-list>li{border-bottom:1px solid #ddd;border-bottom:1px solid var(--d2h-border-color);margin:0;padding:5px 10px}.d2h-file-list>li:last-child{border-bottom:none}.d2h-file-switch{cursor:pointer;display:none;font-size:10px}.d2h-icon{fill:currentColor;margin-right:10px;vertical-align:middle}.d2h-deleted{color:#c33;color:var(--d2h-del-label-color)}.d2h-added{color:#399839;color:var(--d2h-ins-label-color)}.d2h-changed{color:#d0b44c;color:var(--d2h-change-label-color)}.d2h-moved{color:#3572b0;color:var(--d2h-moved-label-color)}.d2h-tag{background-color:#fff;background-color:var(--d2h-bg-color);display:-webkit-box;display:-ms-flexbox;display:flex;font-size:10px;margin-left:5px;padding:0 2px}.d2h-deleted-tag{border:1px solid #c33;border:1px solid var(--d2h-del-label-color)}.d2h-added-tag{border:1px solid #399839;border:1px solid var(--d2h-ins-label-color)}.d2h-changed-tag{border:1px solid #d0b44c;border:1px solid var(--d2h-change-label-color)}.d2h-moved-tag{border:1px solid #3572b0;border:1px solid var(--d2h-moved-label-color)}.d2h-dark-color-scheme{background-color:#0d1117;background-color:var(--d2h-dark-bg-color);color:#e6edf3;color:var(--d2h-dark-color)}.d2h-dark-color-scheme .d2h-file-header{background-color:#161b22;background-color:var(--d2h-dark-file-header-bg-color);border-bottom:#30363d;border-bottom:var(--d2h-dark-file-header-border-color)}.d2h-dark-color-scheme .d2h-lines-added{border:1px solid rgba(46,160,67,.4);border:1px solid var(--d2h-dark-ins-border-color);color:#3fb950;color:var(--d2h-dark-ins-label-color)}.d2h-dark-color-scheme .d2h-lines-deleted{border:1px solid rgba(248,81,73,.4);border:1px solid var(--d2h-dark-del-border-color);color:#f85149;color:var(--d2h-dark-del-label-color)}.d2h-dark-color-scheme .d2h-code-line del,.d2h-dark-color-scheme .d2h-code-side-line del{background-color:rgba(248,81,73,.4);background-color:var(--d2h-dark-del-highlight-bg-color)}.d2h-dark-color-scheme .d2h-code-line ins,.d2h-dark-color-scheme .d2h-code-side-line ins{background-color:rgba(46,160,67,.4);background-color:var(--d2h-dark-ins-highlight-bg-color)}.d2h-dark-color-scheme .d2h-diff-tbody{border-color:#30363d;border-color:var(--d2h-dark-border-color)}.d2h-dark-color-scheme .d2h-code-side-linenumber{background-color:#0d1117;background-color:var(--d2h-dark-bg-color);border-color:#21262d;border-color:var(--d2h-dark-line-border-color);color:#6e7681;color:var(--d2h-dark-dim-color)}.d2h-dark-color-scheme .d2h-files-diff .d2h-code-side-emptyplaceholder,.d2h-dark-color-scheme .d2h-files-diff .d2h-emptyplaceholder{background-color:hsla(215,8%,47%,.1);background-color:var(--d2h-dark-empty-placeholder-bg-color);border-color:#30363d;border-color:var(--d2h-dark-empty-placeholder-border-color)}.d2h-dark-color-scheme .d2h-code-linenumber{background-color:#0d1117;background-color:var(--d2h-dark-bg-color);border-color:#21262d;border-color:var(--d2h-dark-line-border-color);color:#6e7681;color:var(--d2h-dark-dim-color)}.d2h-dark-color-scheme .d2h-del{background-color:rgba(248,81,73,.1);background-color:var(--d2h-dark-del-bg-color);border-color:rgba(248,81,73,.4);border-color:var(--d2h-dark-del-border-color)}.d2h-dark-color-scheme .d2h-ins{background-color:rgba(46,160,67,.15);background-color:var(--d2h-dark-ins-bg-color);border-color:rgba(46,160,67,.4);border-color:var(--d2h-dark-ins-border-color)}.d2h-dark-color-scheme .d2h-info{background-color:rgba(56,139,253,.1);background-color:var(--d2h-dark-info-bg-color);border-color:rgba(56,139,253,.4);border-color:var(--d2h-dark-info-border-color);color:#6e7681;color:var(--d2h-dark-dim-color)}.d2h-dark-color-scheme .d2h-file-diff .d2h-del.d2h-change{background-color:rgba(210,153,34,.2);background-color:var(--d2h-dark-change-del-color)}.d2h-dark-color-scheme .d2h-file-diff .d2h-ins.d2h-change{background-color:rgba(46,160,67,.25);background-color:var(--d2h-dark-change-ins-color)}.d2h-dark-color-scheme .d2h-file-wrapper{border:1px solid #30363d;border:1px solid var(--d2h-dark-border-color)}.d2h-dark-color-scheme .d2h-file-collapse{border:1px solid #0d1117;border:1px solid var(--d2h-dark-bg-color)}.d2h-dark-color-scheme .d2h-file-collapse.d2h-selected{background-color:rgba(56,139,253,.1);background-color:var(--d2h-dark-selected-color)}.d2h-dark-color-scheme .d2h-file-list-wrapper a,.d2h-dark-color-scheme .d2h-file-list-wrapper a:visited{color:#3572b0;color:var(--d2h-dark-moved-label-color)}.d2h-dark-color-scheme .d2h-file-list>li{border-bottom:1px solid #0d1117;border-bottom:1px solid var(--d2h-dark-bg-color)}.d2h-dark-color-scheme .d2h-deleted{color:#f85149;color:var(--d2h-dark-del-label-color)}.d2h-dark-color-scheme .d2h-added{color:#3fb950;color:var(--d2h-dark-ins-label-color)}.d2h-dark-color-scheme .d2h-changed{color:#d29922;color:var(--d2h-dark-change-label-color)}.d2h-dark-color-scheme .d2h-moved{color:#3572b0;color:var(--d2h-dark-moved-label-color)}.d2h-dark-color-scheme .d2h-tag{background-color:#0d1117;background-color:var(--d2h-dark-bg-color)}.d2h-dark-color-scheme .d2h-deleted-tag{border:1px solid #f85149;border:1px solid var(--d2h-dark-del-label-color)}.d2h-dark-color-scheme .d2h-added-tag{border:1px solid #3fb950;border:1px solid var(--d2h-dark-ins-label-color)}.d2h-dark-color-scheme .d2h-changed-tag{border:1px solid #d29922;border:1px solid var(--d2h-dark-change-label-color)}.d2h-dark-color-scheme .d2h-moved-tag{border:1px solid #3572b0;border:1px solid var(--d2h-dark-moved-label-color)}@media (prefers-color-scheme:dark){.d2h-auto-color-scheme{background-color:#0d1117;background-color:var(--d2h-dark-bg-color);color:#e6edf3;color:var(--d2h-dark-color)}.d2h-auto-color-scheme .d2h-file-header{background-color:#161b22;background-color:var(--d2h-dark-file-header-bg-color);border-bottom:#30363d;border-bottom:var(--d2h-dark-file-header-border-color)}.d2h-auto-color-scheme .d2h-lines-added{border:1px solid rgba(46,160,67,.4);border:1px solid var(--d2h-dark-ins-border-color);color:#3fb950;color:var(--d2h-dark-ins-label-color)}.d2h-auto-color-scheme .d2h-lines-deleted{border:1px solid rgba(248,81,73,.4);border:1px solid var(--d2h-dark-del-border-color);color:#f85149;color:var(--d2h-dark-del-label-color)}.d2h-auto-color-scheme .d2h-code-line del,.d2h-auto-color-scheme .d2h-code-side-line del{background-color:rgba(248,81,73,.4);background-color:var(--d2h-dark-del-highlight-bg-color)}.d2h-auto-color-scheme .d2h-code-line ins,.d2h-auto-color-scheme .d2h-code-side-line ins{background-color:rgba(46,160,67,.4);background-color:var(--d2h-dark-ins-highlight-bg-color)}.d2h-auto-color-scheme .d2h-diff-tbody{border-color:#30363d;border-color:var(--d2h-dark-border-color)}.d2h-auto-color-scheme .d2h-code-side-linenumber{background-color:#0d1117;background-color:var(--d2h-dark-bg-color);border-color:#21262d;border-color:var(--d2h-dark-line-border-color);color:#6e7681;color:var(--d2h-dark-dim-color)}.d2h-auto-color-scheme .d2h-files-diff .d2h-code-side-emptyplaceholder,.d2h-auto-color-scheme .d2h-files-diff .d2h-emptyplaceholder{background-color:hsla(215,8%,47%,.1);background-color:var(--d2h-dark-empty-placeholder-bg-color);border-color:#30363d;border-color:var(--d2h-dark-empty-placeholder-border-color)}.d2h-auto-color-scheme .d2h-code-linenumber{background-color:#0d1117;background-color:var(--d2h-dark-bg-color);border-color:#21262d;border-color:var(--d2h-dark-line-border-color);color:#6e7681;color:var(--d2h-dark-dim-color)}.d2h-auto-color-scheme .d2h-del{background-color:rgba(248,81,73,.1);background-color:var(--d2h-dark-del-bg-color);border-color:rgba(248,81,73,.4);border-color:var(--d2h-dark-del-border-color)}.d2h-auto-color-scheme .d2h-ins{background-color:rgba(46,160,67,.15);background-color:var(--d2h-dark-ins-bg-color);border-color:rgba(46,160,67,.4);border-color:var(--d2h-dark-ins-border-color)}.d2h-auto-color-scheme .d2h-info{background-color:rgba(56,139,253,.1);background-color:var(--d2h-dark-info-bg-color);border-color:rgba(56,139,253,.4);border-color:var(--d2h-dark-info-border-color);color:#6e7681;color:var(--d2h-dark-dim-color)}.d2h-auto-color-scheme .d2h-file-diff .d2h-del.d2h-change{background-color:rgba(210,153,34,.2);background-color:var(--d2h-dark-change-del-color)}.d2h-auto-color-scheme .d2h-file-diff .d2h-ins.d2h-change{background-color:rgba(46,160,67,.25);background-color:var(--d2h-dark-change-ins-color)}.d2h-auto-color-scheme .d2h-file-wrapper{border:1px solid #30363d;border:1px solid var(--d2h-dark-border-color)}.d2h-auto-color-scheme .d2h-file-collapse{border:1px solid #0d1117;border:1px solid var(--d2h-dark-bg-color)}.d2h-auto-color-scheme .d2h-file-collapse.d2h-selected{background-color:rgba(56,139,253,.1);background-color:var(--d2h-dark-selected-color)}.d2h-auto-color-scheme .d2h-file-list-wrapper a,.d2h-auto-color-scheme .d2h-file-list-wrapper a:visited{color:#3572b0;color:var(--d2h-dark-moved-label-color)}.d2h-auto-color-scheme .d2h-file-list>li{border-bottom:1px solid #0d1117;border-bottom:1px solid var(--d2h-dark-bg-color)}.d2h-dark-color-scheme .d2h-deleted{color:#f85149;color:var(--d2h-dark-del-label-color)}.d2h-auto-color-scheme .d2h-added{color:#3fb950;color:var(--d2h-dark-ins-label-color)}.d2h-auto-color-scheme .d2h-changed{color:#d29922;color:var(--d2h-dark-change-label-color)}.d2h-auto-color-scheme .d2h-moved{color:#3572b0;color:var(--d2h-dark-moved-label-color)}.d2h-auto-color-scheme .d2h-tag{background-color:#0d1117;background-color:var(--d2h-dark-bg-color)}.d2h-auto-color-scheme .d2h-deleted-tag{border:1px solid #f85149;border:1px solid var(--d2h-dark-del-label-color)}.d2h-auto-color-scheme .d2h-added-tag{border:1px solid #3fb950;border:1px solid var(--d2h-dark-ins-label-color)}.d2h-auto-color-scheme .d2h-changed-tag{border:1px solid #d29922;border:1px solid var(--d2h-dark-change-label-color)}.d2h-auto-color-scheme .d2h-moved-tag{border:1px solid #3572b0;border:1px solid var(--d2h-dark-moved-label-color)}} \ No newline at end of file diff --git a/docs/docsgen/source/_static/onnx-horizontal-color.png b/docs/docsgen/source/_static/onnx-horizontal-color.png new file mode 100644 index 0000000000000000000000000000000000000000..e93090bdd3bc81b9f8f8ac16049bd3194e5fd5a5 GIT binary patch literal 70339 zcmeFZg;!Pm7B;#8X#|u~kdg)g5u_VQDS=Ig2nd_*E(JtsK|mU$J2%}32uPQ7Nk}8z z_qToJJ?Gpz?jP`t@iB%&S$nUw=9>LHb1pw9D@xzSBE^D0Ah%^@o~b|}m>CcVay|wc zIC6{Q9xwO{(^f{y0Rmy8LHt2-%@lElK&T+H&&1SR6E|mUEOl4C7|x_aBhMy_axx~` zC$%)>KT49{`uNjTaeGqnIel;b7m`=|c)g99gB}J4?ezG1(~~=|b7JqXsUM7yRO|UZ z38<^awX~5yn_0A0YbK<^`hjaC$M)8ORI;tKgaaPo+&&qxX>HTrZb-pj5q{#*_dStw z{}kdM#YxEL@S4}B?^L*H=M9$v3J~_6pNfsxO|hF_AVE|Sh{yT!BT9X1=;q%LMt@P# ze}4(Um-+VssrbmSe?K4;r@DRfi?29F2GY$h${26BZvG9yqW-@(`J;^g!_|epC3WwS ztct(`abAih+_~kBIV@`3uEzM@cAc8%H>Q7yjXuQHa84&WClfiTiDAg)A8~p7jlz0- z6u;FI!!Tsv#q{gzl? zObjH*tDP)3K5@s@-THgF?;H9EG%s3N^K`o)tvdvR&XEf8WH#I+t1c(PH+Mk$%3hP6 zCPFLAnxJ_k7u#w^eU+@SxPyB0Il)ls`YTCOZlbMccsAXquN~|JtBC%+{=*0~BH}hP zEt8efyVF7W)c%979ro0>@7|n^_O@Fwr?Sj{P`sXkv!?LoFb>WeuACg&+}=-&A(acy z>h=HqfC_w&N}C%Rh_4uP5T6nrfG6F29YvJtc3#^0n+jFC zZnB$$Uh?4Jf-U8=(!Th=$>nlOG(Kzl6=Mx;?qLt*x;BO(v;X5^OSYn)Y&Qa>4gu#t zK4Xli)MD|E2pC!cdMjh6e){*izUBlnepqY6qLVWP?Y8j|HwjG?6*EH&6-SwB%i4eB zMh0e(%9vj^qrhBfGHz0$vN$qK83`Fy_we6C14f8LvMN2|>NMN=LIf%|j=J^`Sk@Pq z+#g@Dwit{0KW9+?b4CINb)u7jQ^q;^sHf4~hvOGcN_z<$Zc5jzP0$p%*#w@1yy?G5*KY zQW6-&)lJOLWo~lzng{S1C!D%s=3d$xrKRowmKKd1p%=$6WpOmn!=aYvN(k2A=KkZ4k4YIC$cxRYi{$?!(+~*! z$mfxhwm1)NLw`2%&sef4u2gyh=f~}Q86Tl-_n)sO18pF_8jS_jPX^K5Y2dL1e`L&k zr|q=_2jkPaJCXeE2Wy_&23WaHLH}xZ;3-i1D_|)4(5KAylzP2yjfaO<%QifjU0}1L zp}wCYw8t@XZ=#1BlqkArTKFSZxZ1~aQ<@NN3US_qI!7B3X_-Qv|Hx7al!MvcJnAW# z$G<9$u>>yJA&$|%%4U@kp}m9g@AW)D;-LFP3(wUX(?6P21tC0QQQY7x_%7yMt)rL! zh`aQ^M)(maO8QAYa-?i`goUojfbF&FU>VgA_%`=!T9=szUB;Nezjkisf5XN2@m`Gf zuf>eghVHj!QQ*q%KQ`dw`Yh+PJk;>nJ2wvLiv)s|4g}-`e(Q(G83rv3);j%ivNvDG z^*vaztxXIojB0k+ME>U#VFqx@M_RbPhCyF@#6V%_H8;YFJ`y-z6yD)W*) z$|iIZ-L~N4+^uSDcIRdnbFI)Os{ejoMDPt50V6U$CKIY9wgBpWXY|c9A}fVRR;JM9 zU%R9lgHQ^L%!mnyjB$4*TFJ`pC*NLhtn~gL?F~qPtI=?Tj0|@f8(b`2Uo>5}P+qiL z$JF^M;;U>VHR9&d$y&Lpj&8qZX?6V@T8;^#qzAw)v&1p7-D|El4$n?$q~9s6u%+q% zU1XP0!$CH0?`&OTLn$QTv9>6$0e*8+->`rUguv52lKCxHe0+IuP!qh){~WlKiyWCs z@L?kXwg00K?oa3MW_qlRh~*N@xk-v0&p{LU=V#y=JXO!D2(!#yW!XDyib^R2Qu%1W zV!vQAtnV}!1PCAWBIgpky-AsbAQjg>k~N=j{6f$$yF)`mvwFSUB$!)NUW*WQi5l+O zWCCxTIU)1n#%_3;5L{&+`}bDO6u_}b=ww;*r_ROhADw>vs=6wCecFGWKaHvu2V9$w zSS46>@O7P_WNGy1oA{W&DN?i!i1<7*gZqIWb3@U(?lFVU@13iQ`4+#Xs5)P78emKh ztlak1_g!oXl2hFXcl}Nr`#0*E!~QurCO}0>uB@pTJfOfMTdooytf`f%s8zY8`(kON zu5{W>D|w^<9*|1b9H_0_WMUjo=wQI-bE4qo;WR4F6Y8hj1ag0~@k;}63CsLnCn z6)7p6zq@|1d;Q|M9-l_t6%#l2Q(*j?1AT=+jle}@m%GrtLIDrw^;3yJT%ou+h%89e(+%s1mw2Tx+Scxe5XJiqV#o#R^I(I!rn;VR8g28wMJn<}Ud)68!z{ z4XrN*_FO{)1K1e#J}0ZDhun2<>ZLMAE}OQ~dk&mo7kPlvm$YJG$feRlWr&zJS~2 ziMns)@?Ixp)101S_=8{rhH``JNsHoux`2^ayD(wzLLg~w8h-0DjFwL~iggARf)IF) z5w$-h;fZSbuVkEG#Tmmp|A&7qE-5GhVy^se^~@20Sz?0{u|Qsw6P-Zo-LCsLIKx ztDI3_O@}%(l)YB|>vvT)iauI4Q2v|djm%!-_J2~IEl>p zrm4uel08{k6GpDnm>EtN@;@>)>wz%f2r4S|sMNP9u)M0-X*kxecFHeod8UC!A$U?V zt-j_X88Gs0=~|e(IXrIZUo2!PO?7(`B>b+oD9nZUt;RU*I98UHRYm)0ffQ#O_^sm( zB+V};U1V4Kre4-(V6dvY+sdc5+yz7Kflx|6pvybkjXtSstD zaeVaxM#@5aSHApP+@=FWNjFssezVl(?K?KiRy$Qi=;V*;Q*C{eZnWjCy7NQ!1vR`B zzx5k#Q$E(&0{7qh42U7aN_3koVU#!sV* zFPwFjMt!}s@N&Jb_EYdC#^Dy@tpV5~SlHOuNA!P%@On;l`^dU-zPlm#(47Xm+A5CU z0$y2freO}BytWmQ*qN?zV9>2QdabFiUv$r zPkxSV`&Xb-xXv$3OQkMZ&(_~3CT0u3CSmiRsM8r&?5w7+n2`sb*P!=jl)Y&abHSJ8C{_n~7`TS&4KI6$I&?JI7%$mZY zqLs?6_H5xprR$I9_kvvZS9%}w@Q}sC#I$ZrRiPW=RmJy865QbbA2Fh&mNc^Fff5+y zIzF9^f;z=K)?4&HD=RCBt!HYZry9}!#DhP$RC#^orK>cRAt7}1V~NLs?2TTJIUmVCw!)EPU5x<{R1X#Qx2)W=M26Xt zl;%0m`^f4f=Z%vE9^bXE^Y(gLWyLOhT+=VKb`{V7E>KQ2+1mQH;je!}JOCBq5*sl1 zZri=Gp_soOgXORuRheyWlr9B5Mr zRo7=tpr@w?TtSg_b(UNb;_35V9O}h99mfPuU~ugkD-W)w@gU_jX)V^(*49j0;NDM0?E64K0ZcCJB8yQD zE`~{+_+3whO@c$vRPF}OxzI&)G=mDhz}sFR(+@n&^6kLH<;8dI?DT+-AMXqY-!s!< zH6?zgqp8W*X3d*|S`FOgP-K%!A+WsJ9D^FW6}M9sRt{PdS5IMPkI!5Qk=*5W9)H#M zlJgDM$!SIkNF0Ird#1BiSVKj+O^uUdMECCJitker?)+B}k9)YVj@dUF*u-W1GD~DT zT4>ffLib=+7#=QNacP`CoBwjVeYJY;u7NZp^%dxv++!^~1;-MqmaJ2H)crSr1N%0- z57n8l3&Nvpt0oG>Q0v6*D7n|G?Dl%msyiij>y z?ek=>Ki{r9Z$t(?GTU}GGhyD^|K^Bd8!;9|iG+zC5m~=nalUNP8z23AgM$7Vh%wsd zAQS63iOvF5A|bDkT6NW`w{bNzkAemm-j2mw<>OOdHuP{(r=KQ7}{Jb`5cPS zerX+68Qm$~H-y&$bCzuSU0x&uB#T{NdKp40v6|;5FQ(_+-Xv`p9$ zQlk-Bw~r^(u@-zmU-aT76;a(h;+0~VP0`|19K5Y3D*JUQ#c0oheM`flw84fdnda8e zJrr}4<4IY90!MTLl}%{y3wm{ zoX3StQ9|cB^>fy*1zs`BR|y0ZIujU~p&hcuCYK(KKP=g1ts*fY1FNr^jq-~hW-G<^ zjw&!@&0QBqp(+cbdL^5;`+enXMOL<=#L5npCme71@BFmb!|*nzK>9+V{EYUl6&>zGkTMstThJI`c{4f4avmSoTZfpU5nF1Wzot zdpLQ9{sB0x5r$QLFJ6KYIj=312TIvq$LqZL)%-7>6PTKiH3N* ztZb+|i(r&KUyP)^Jx~>u{NC)QxuB?ERNXDcLbgXo8|MTvmaMYoOM83JpKn7pIi3(i?u*QY+wbr9x9|K#?|opf4wB4@%JVFrw!RKiTM5wlBORDpum_NCA+ zbbf#A6uCH?2?iNJUtFIH9hi{9C`@5UavGLWw(?e$>FttS^SvjPuw!dmNkl@2nI-n2 z{g_)-FOfe*h$kd2i}&8*l0_Ca_R}&WR{|imzP`i3yv4hzFApWu?n7B>gfo{JQ6e(B3;{t+p{UZ=yuvSGE}Nb1!SBZRu14Ah#PR*#*)w%* zgyK6UEbgDF;1!F{-BL`YJbbw@6?oY0@U>-{cnx5CmfCX<)^{qegHNbx%Z6r4g90hs z*NP=D4EpsTM0SWmYr7!sy5Gy=;T6DBTWiXV`!EYO*=4;u!+ToSbY8d6f3?|vbcq52 zCuZx%^>TZxKE>4-Pm^6~>S`O!g!yCDAxP@+V#~;&0umQCG;it03W~0*ibcDh{=|3z zrg;h6RlJ16u;P2}$$953<6t4#Vy2@|l(!qVT>y&h(XlA_QgC&CClJrGFB^#yg+ z=;-KlWy8t3M$U5U8NVr-*0Z5(&p;T5GrmQia9VQ=#PIk zLC+}$wA=SNhJ%r3va{Hs#$wQZ^h-K6$DL)&E_}+liSt%NqT~Tp0y?b=r|;r>mGUt^ zy(EZvYrZrR$eC8&+PM^ekPNE6$#@zOHlm(JKlHKAADw+QnJyiEsbWp@X#}+Avw!@F z_qNsLOT&WY2ke7^Jx(_Dds5a@`hvd$*e|>;j%e5$4}PyzHJ*moV&m0GdUB4Bq&Z{hDG(IT?!ewpbX-{37hC0Cd|8fHUKm6p=1YYP6vV#j@=~ z?*LvQD{e)ly2UHQ=8)*k6=6yXm&AMiY(c6_A-d?hg4s z)WMsA7ze|JA$R=3Tku4(H$%rP^ut`_p<)#TfYAHs%#h*rlZMjm)$lbrzmxf*1$Hc3 z5?fB6DjZWETdSKnTJC**QAG;-JGW=$u;Q|ZOX=Ah~(mcbT>4G@?JlnN3m zF?+CXF*+OKc(DRp@Gt?Py(8i z4Lp6}E%JjM9WB_-<)TL8@2UKn&;cVvZJxWn!e025ea%?E+lt+Qn{EP^GO*?*DJPQP z^f`^~ee|9?ISY2wc`P^4W)@d9H+%gDneATpTHzQj{xVHI9p+^%!eB5m!G)Aa11^&c zGX>CUkFWizNtNz}Ht^piF=hqi16{&PGL1b6Kyrt61fEX$Z8I;;O)X3l=jDCxtnKS) zeDIB4D>XQ!Q)e_)Us;OD!>xyS%1|B9BctFa~;*C``#fdz#?aSI!%z$b7zJ(Xnyc zF?5$CX@!rx_7N^`ypv@m=dGbN9S%wTpI_7Au4QI<1mt^LiZUMaZ_p|Fto+&PJl&|j z_x46-#yBu1+amIZe(Uy*o*MuuJ2tCxBHH9zWl+Hh@05N$ zWQb)!HCebpfKQPx>w^#2_+{Gt`Fjn&&xwwdUznr_eDQEb6NfG3U1x0@??Lu|yDGI$ub_%FoY@h|u ztgaBgvvssJ)zDZtRpkV@R*a?IG;abtQk+ zd<#NJNb)^IBaJ?>(dAL&P7gDx>C!{oV9cB^%!gZ&ClplQd(|-e`IqQ7 ztC{t6bC9t@LPGA%$c^jOIkP4T@Q^2apM{TQe>vn0aTfO9l_{_JTzD%{Vfer{J}U|q zKO4j_g(f_4wcFNIs+Vf@6X;j}!<)v3A)o4n(IK@D$xf5WpOJ9=MNa5%~Cu^=z=4V@)1#-v-Am?QEz>9UK`PD*Hf+G~iPc8UfRPQ>bk_m?lP0QE& zkec#S(--G7iQcA=cV#>FI5T9YMJ@&wJ)9#fYL70E44j@FH;2bEoUdMAl)3;)OQKA> z?|Ce7>+jzi4YTcqCoX*O+?ajW#CPgl+I?H~XY1irHIAm}c;1fL-ThrBhjQDImtETl z3$KomA!-tdKLqbE5b}!25m_e3SHAGFSs)-}$nR+=bBpJJI5tn5xEsB?+k8~EPZDa? zlz#l0`3VszGXvYVP0gcpQIdRNw+Vl)GdcOPWY!bzP|zQSsc!(1$0hN4+~;y!Bxwo* zwbo%(@nivSwUx79kUT3p+k;%Hj#S(g%mN;Oguqb#SX;d(E(_+IYLZL&Cl`1laB=rZTUQ{{zM zL?P(9`$17p1Z}D}fk#BS$8D)2RDV0XAZEV@V8lJp7txQ|D}n~_ZZ_Y+Vf7}@tSe2G z72cS$o4Dr_MSG_143w7;oVR$WgJ(}3^o#N{eMSo)4NBum`3PCd*gWj;=km4s`eA;x z>MCrA7tEYq9Mcgp0K(0V+HOK`GcdOb8Fmo*`7ZmDuF0UM0_91GmL;{d2*55KPI396 z3!BgifSg3I-XZk9vZMsxdaiLM-zyO{S1}M@7%<9VrtI~wc-(AQeA2Go2ZB4{Kd+!E zHK5;^t8xf7hZ|QJ-3zY=-QE(aH@?8#6ltVXcFZpvRJVpxc_-{xRB<6|T&hkGXYTv` zV$n&GtjZAz&?M*p39A|qE8sIOJAYR^)Jf6NUDv42DJfj|Y%OV8>PH~Z4#2^O+I_3Z z=S%wI_OIUzFi2pK!4p|v@nId>eW3)6CT_Tnvk$o4};*?e)=W0?SCE@fv< zBEBY1eqm`S;P#@&(`YvUI<#Z)ANZ!HQlg&Uuo@{%ar!nD-Kxu+uj%@dWGp*PK}qSj zHpYQ%%>K1lMO@QG&uf`sM?nDBhjsw_q@{KG>rzLzf+bw!>Z?9|!R;h~!QjXu?{CLN zqKS$CUgTNgh<3o>_5h~V<1M+!k&*JYD+M~)t?Em3(LgO3NJ*$Al{m)2`qCp;Rru!LHs9dn@*cTT24kH`aPL#)hf=9O}JAP?qw)Q&}-RVaiFzvywF58vw$E~fc z;_SySV|Kln$Lzyw@|F`n9BSNV@bK9#tn!Wl9@=!Z(bTw##{}2@9k^Fh!YFz52(lKh z+0RFo4R_zUP5J3kK@ik!H}np}{;?RpS!LW9E}M&UD2Qy2-lpt9&z?*d>U zJ{2))uh>3HPpvTgNR9uFfvy7CL`T2(*sOm=xc5PNO0U&)QZDbr=IL8&xubn42OXqT zJ^j7OC|N4XN-ttjK>(v^?Ms zsU(&r_W?1XFFWkWJ7#}%tl7g9Qt1a~VKYA0m*e(iE;fJa9h0qza-}<%dHAXT=B)!s z#<0R(Gk8bRc^}TbCXi2CM%qd@JoIw>-Qt($kTu$39>|m#H;D%GUELQmn04c$o6=-` z$2!neG=ia=M9ow61G2Y>`OIW5bFzB=7O@~h*NN|+Hdh6K{;C1RUF^~~%6oh3lBXRc zl=cs16jdI!Ta$2?GTjO9>&rGgpQEM#e2Q+p5)aiqZ;#B>1TGk02u%QI-?m2;;NI6~ zW)&|bM0xGO;$1Xho4FS4yXQTlOO|}O(?)qV*G4Ss=nn%_{l=&ONb;Y&&_vNqig}_+ z<8giQx@N;W`%6tk!kxbF^H~;)Lz4kImwwmXU5@zVw?pL~u`eLv08rK}&NzjOWQgSYTEG^@#+S&kQ^zKRLGiX-j@ zXc4hOJRL@oBH#2hg_HXiw)5<9e>*k>$^m&pJ{wyRdU}CDS#pSluAW}6)l@JT$p_YX zCH-nBh~dOe?FC(NT1+{SpJX|ijq@LH#0AwrbvxgxnmQ9}ITMK`OP(K{j0mw1Ef5G3Y|%beA!v~CE4;640*SntCRW^_5r-zj{eKRN zg;sjcr%40~f1IioAyD%C^}GoVW=EQ0syk);@%LF=AZr09r>`0xqBGROtW-}dz4y;PYDt2|3*VqRJUBEXu^91fY+9BCP^Y~LpKP{CY!#?SImWn`k4eUB(0gxu?^i$5A_Yk)jju;C)Ik~1kpZ>=s2~6`&=wldUpY0oTA%m zx~XHhU(jLm`*EX&8iw6NS#wokTBHG1JNwa1U-Y-!%)EukTagN?BizX5WGxk?NGU@u z+!T)~-xH+)km|JhIrk~V@rM+K`}Z)kZ#`iN;9(|k6SLVS@AR+Y=c$ak*+meWVP(89 zCzmTrE&_upkJA-HCnv<9%6o~&d&y`0$SG?dQQz2fPX^DpD#&=O58T5|fy$E>&wYGK zV;rU(8`oV=76Qw;rqU(#7Z3o{^$9$O38w1zB!<{`wwoDQ0;3;qex~F_8>qsmxxO2v z3Iro%+Kmee3l#(?)dGqg{g+F`Orl{3VcRgN&3vsCI{G&yx!s)Jt~#Kp>t1Lb&6~aN zgrTn|=7^G-@Tl${yhT(w<|4(N4{$i#7G$x`yE>HTKRPM;tuo6HgVdFk*FDNTaeS27 zX@%FJ88CNVMG3Ov=7g=r{j-f?QzhcSV~@J+rfUuUJ$|46Ak;4u1QW8~WEnb<)-sTA z(yc}c`}y&;toMWlzITWh@KSp&St!$=nuyg~j-KYs-}Y=)7Ce1wSOpm{6d!=-ydOJ&T}kO-#sFJ5I!wMUFV z!O*u`xu=FWTdmi!9wQ6JUk@AYq|(qn9I4xB^zadAx*pWFf=evHOc$5Za|q5=rDRbC zm{5qN5`uSr`42EN&$?_J&t=(8QmHLd63F@0@ki4>CF+4V#;+^-TB+sK4)62%nUy`1 zfUa=!MBNTDAgv&ifBJPYvg>W}=Cg)b-q{V(Phgh;jO0%G?a6XOK+nZ~wJ7eo5|`cP z(0^%H+-9aBfuRJ*)MWY>Lq^QAD2H)w%<^&K5ct> zhJet5XKU$AlYaRM2>^d$lXzF?_qd%OZRzjy)oAwAG`8}CjuFZzCj5BTZFQ^Cdc6(r zT8hD7l_$?Ah*UM{QV=ZpU~xhWEAZ;G(|y`M7;*n3H3^rDawXgU?r+g3T#F0XMl86K$^ z-t+Qq;$XV{Tk*91mM94hbnPpf=(tHr*OKX8*}i8V)tcLpQEVTceL7g-$h@oJ!GqPP zXax2d1YU|n)@AgltE=Ny?300Ap$cVN&2ZcrK=LFC3*GNaJk{6Kc8AB-xuv%<=}d5! z=EgVx9Ga}sqor@tGy@W?2YA}S>5f-K4RUknw2;ZMY{Lq;YcnVImr$JrxyrmJW80Kr zPRWJBQ%vHpi0oPdUu)$V%#zT}a#zVbfpTg{)}!IrMUQ9#qH3!`u@q;v2q~DOS;1mC zkdDm9mD@!YuY!-9)pIY{-~wh%llgJf{%SYSVWDpBQt@+gGOkVI>1+04ZEpDbF&MBu zdLIMjn~?(uOJSI$^i`<=i}#sa0>Mx%2kX+eVD<(a7!A9tPrQtGgODpDnn`o|B;QCdU})+}nD^ZLy1$fU~$ z6|Gygr)b49WPpx)UmC!-lgpaWgK5MHNsYts20YQh!J(mG87F}`QL20|1r|@CIPCaD zVK`JH4V5sc4wDjgIo;2Am$1%IgCD+Ox{GLe2HcE}OUyTO%nv$KuF4`!Q^#PU6@ z!CvaiZUuM}Sp&F#%4Y;VSM%7ST|Bd(Qr>n3Wv}|rlO4n;+0%N%7f46A*ixa2Y7kqH zSH2dZqob}qDG=`X^IHj_#|Kb2P5TltqoP7fN=gQLSjK`Fl7t)WT7hIsr7`x%`$IUX1#3hr2LjBk-~I(d;&Yq4-Qi z>H&YT90xrj;cc%R2PhhByLq>akV0z3Fi2mLX$icrWMQr=s8#*;xe?#OOq$tNrwHr# zI#HBVhy?FTFR`bPB4BRX5N!&eL7g4C+k*sKj^F|=+XdtH!u#J5I_>ExKK{M(n*FsB z0QX&*J$x})u$W4H;Eh%+EFxAG4;PohMF!PVt-|TOuczjOb>?ZqFc~xe5Kak`}8{ z!E)kzc%lft?k1kd`Gk(uOqi1AB7slr)DJ!hgY(I8B@)eSE@OZd>WDPll5Fs4Z z`EQ@|@>HPNsGXFKkE>8?X}#YwnaB6N>jLCSeF!uF@XI&5alhUJ00#Rt`b1v5r(*GB za3XXt8NucJdI3$j)yh5Vp5K58JDzsv@3bcqC{TZX^;noYg?)`jLEX&h4n&Vwsa3sN`@|h!--n=^>RIIv+W< zT%w>Q;rtL8_OCElIN?chAz;mCB-yR~^7!v$2Ox_wgt|5${m_2Em{v=`pUxaUA$$9t z8(_ztp?5GBwUeBwB!AM@vnu|(`-hcc$pT=>{&dZ+L#j#8X}Ih4xCw)5Xa)Ea`QZcg zM1H{Fi`HKrO`SKU9nbk(!>2p8<3d!2f)RHpsFB@XK17A*NAly;Y5-Gblqo|2$%Q2o@C=E?ZP8uSh(PCRP+gKdoVHGZKLc9Fim!I|A*$;Tpq2t$6#zMR*|a! zPuvGt(|Ft9F+ONO-IN5X_BpRNIj^5qmDXipc`H_Gqx<&ex!(P0&K)Wz$zU>K$@Y2& z?#)gintzu;a)AT^FmCJHKq_Cr*A;{78Ef($?BCZIt$F%}s}`(7s-lKNE!)kjJ6ABHti?*kb`x&R zTa@FYzU*QvG9vvW-=HO1pGHRQ{(D2nCn)+8x}+y0;kl)#Kur3uD?#Y)PEQ^`L-&RM z!I+K>{Llq4>`YPH&Cr1%7Q{GB`~17Qq>3&0#$OTjb%LZ5jo2t*U6p&l=T3WE{F2gp zYutN_Qm$l<9$O#tr{A4sUqJBK>V~M$ts31^Ko!X(VK@oxz&DBJWx^6MHK&5i%mV_- zG>Yid(o4>X!DKj1UX~$eeHtOqLy8A%XA+3L9lQL^nHJD`-xfdV*6{XtTJ_P&1fJ-# zM?wO0mg2D;-Uh}MQ(hj~dZF;dz;fn9P*I32_m z*wmgi6S{(ppy>ka*p(vu3LaU!~%38-z6 zY`NRM)`;xkSHLIh#IUItkjJkho><2aAtmhu+CQ+$eXppMKqY^)+iEI3m>y=e^$*j} zY1&8<6Qe)#`@EK_Eh&Myu8*ox!r$~hl-JzCd#w4&!~rUPsllSw1(}*8l(RVZvxyPBL(Ew}QrA$`5oZ{K7*L{jm9cZpmq z=$>r5G!Y<%OnNwTPw2oJ8Mx-=*v~BI0wUP^)i4&)L3}N=6TLLAGR7 zcTr#X5z`fWFeESet}d~UbjV(30=(QqPN302$TwrbN`X<_O`!6ZU%Q?6wa1~wc1E$MJ=522PMI|N=%X0omAkVUVB zpmjWxoINl0QNJUVPTB6a3eDYC-otI0_rLp3WTFFus6DWM;Y`mu?6k+>@&(!B%DiB~ z1P=tkd)i#`9nnb-#Ad0+PfXa%y17S9gp${K%2oPw2U66ieKHs`HGp*kP#AcGv4@OR zlZ9u&hzUsgr$-IMCpG6|Y9+I&9~@Jv0pN3|%j!{J*INS;fREN3?DC#N-_lKpqYj7u z(&6}YWGrcI(6)%`nnEB4ar6VwI0>3rJDg!sOD{Z@ch{1#-^viI{yNGHs(xPqn^{n) z_@X-ROLtc=(RP?Ro?mbALr+;dI%MJv5836Yg5^PegB(tM7B5JLpHyBcY=B+zkxKM)jQx`0fV;4QLZ2$VpGMZi)>@@RGEvyZui`=jH zpaXUtbc>fS`dqFK-C0*b5wM>dmIR$%EXc9)pcR!F=BhLDQMM&#tuLOzgn=p4@h2y7 z?e`35!MbAe&uBzb@&+6$_Dh|S(8NJtzxWt`hiKh%w>;QnPkLZ=h1z2&9owOQI}v%-Cpo0*vj=5yJ= zwqQP2)UMs91k8B)C;gi`2PWU9e$aKv!Z-kCOy!l{KLC|*Zf0%d7J`@+rZq=>rpd}c zIw)k8m$rcBs1OEmsRPl_u@^`{TKggLreX#}YYFFQ5>9>f`x2~9t1&gFN@=RlR>z#B z;|W9oK%uS*)6vtj+R*Uac1ZDfTvZwlAW$qjI|5W3Pg8=~9H&6si7|#@dW-{|!tfxM zLV2#I2&e}rPYc6l-8Krrv}`8&1YdPGL7s&j=}eVK@l}Y4({@ea!0O^sBUTHsDgcXr zvMea(mNed3MdHmSjFqus3~HHq4o}9m%9M~{gs`&tUoXI(M~5*{8=8fM#ll!&WuF2Z zl<`JwYYWP6|3CA>O@){7SqP_(4hq4{3U- zgk_3|8Ze!|DaxNvo0(R>Cm3Zn=@_Trnu0+ILcrG=|0U$d#4}W3RrDx!rpUa$7McjD zDSoxlQ#HU_O}~lI(Lh`hY@;)eY;Vb__qg43lK?yD>ql=Y%&3Bx^i=)h|CTw+8RLk} z@#*0{sO!ZZ!<8_Qql^i*gjxjQSNUkp191lL%MD^ZlCgV?oTtk9MD3&di#2+H2}j3+ z80327eUV<05#?`loko87uKv^W!u5hIQvsMYZ7pmL!AfRzH;fdjdk_p!cx;wdiZeCD z{D-FK10k7U(>T#%;mF~?YYf5>Fu#117JN!8SEvWyrT~vJ#(wpcL3q!lCUp7f-uc{> zb9%F%X_f)xAh3+`Jt)%*7vcrmk34$ce|YDpGm-s(H|>_P7YmVV_!8)X(DyhG%cp!a zm_B>TPDN&@N58JG(ae7i-t#qk?zzYJVqSmxo{y&Ew?Ye$DH|$!G5_VIw2{V(^sra^ z6pF{~`+9p1%VIdcMm-D%Gum{JzvV$mK=fz@%LU>*Pf=Dvd1lQv&_qt={jPbdO6%V> z9>I%6?u#?#`FMT%#N}%q+_LzDomi8=@=aYlZ|~HM5C~v;DOoxv0knW z|A`k!W^tURtv`)v9zA;WlWphAChfOhfLQ#mUeYs7V5o-$W6Etc=?F{?foD?oSUsxx zke%YTqJ%B@W@7gZM(R+v_bbn#9OrMrJ|rf17W|)Os=nLg%`%R$a<`D0Ddi*C-_W14pGyH z)99;OyoEf^wm2sp*T%zuYO@x3Xv9y1Y0eLWEV6uX+_JnUJuWkCKe4}wcGl&Pv zg27BtWZ33&^kiOUgV7qq*g=W=iOM#87a%4Y4SG&nvDy6rh5xfLJGkuQPsILPI2fWo z2nW52a*ymNfPhyFH-pyvg`WtPOlKbRdz88!vYL|THrxIi%r~Y#v?lQ6@))`h z;vszjvr1%`bQn(T6PeQb`g(oP9t{HG7Q8^R+L-;kYo#a-{`3m5$B=Xl*wOj zddmUIyx%pWc2ArM8Akinjg@lOH}Qd!(F)rUUOwEn1TPC-HTSQu<$eakUnG%fZzdoHnaQYo)bI<546fmVLks3vsuiI#0|6rKcx=*90^#o@u1 zcx9Mlbj9~xOv#Re*K$>fa<^Qk-w zJWe53@ypztW(~0@%KTL0yGfcM)+gGlHW)1AV8-YLbS^8Tnv}0G6>GbIQ?h z$Qb^9k7g>e&p}b7|JQMT(3f?-vJsJtVXrgKzGFX@7ZNl_dWPi+@6`0zkiq|<@SJ!c zES>eEOgwH4wD-s7rrN%_^M-4fq7Kf-g5?XF>n-+GKQo{6Dj%I*ZN*%jQG4zlzh82h zWu+>)8O+h$pYoPNzV6_ZymWT%7&o#5o!@nSLz*s{J zLHUGyY5BN>F6P8gl-l+&ua6L#1j^^T%=IS^Zgtb&?FwTgcTE)XQ2*Y0N@~kp>3+Cg zY?hD26x*{{d|j&~c)rQG4^O1@+!fOlap$mqI79*MiBo|mvPR2Ejy2Rvurippd=F{D zgl&u!28&Y_rmggpP$itT+-dk^kqZ)gRflr6opc;?d+&t)Za)yI%#^IqaPUg1v=^}ZERhBsVl1R&%4KCV`taf^VUKy;&6 zMw1#61CgLfn&3ZD(TzTXbRUG#M9a`7-~>+R_1WmH?p$B(D8Yi!8ut4H7Hd|M33AoS zCgF+0c`^G49p$PqhXNfX??&M*wlJa+y~zY?ByWrNf0$ z`7e;TfS^j33e>c>hHNF*{TQO6Kg6m#Nin-}2?=dgl{uIJM{Y_lEE)i$((qsh_zF^zu)X7M7$sS9FO0G~eTlp)>(y*(0VRYdq?vPHfj|3SMbz*olzM%|dNI88MB1&5H^#DoK z^x?-x&-NtE%*k|N1R(d66B7_9FXQV#!6$t+jH~ticxle{ z+AA2D2(bdq9Y{lNGn3x#7)gjypOZt2>wZ*p=YrkQH1I6sYk{c9R*^fpUwDf*O8GFm zn1JkC0qv<8Qdd`R3zV&1vEs(107*a;?X6d0GroGedxtZg@QE7Rad$Rw@*v-Fl61b! z_4C&G_ANLN4(8d3OY$QY-#WdXj!%fJ3BZ~aqNF_I%rART?NNrJBP|2&^%;rQy?Hr8 z*xDm|)(~s*yeV0=aMRb)d$Lxp|Rz(B!1PX&F0%1!Q6 zSoYCyxUT3`7VXhX$w^ha?AH%7owZHz6s}pi{@gN7`!t`y%B%0`!h{ENwAZ7Joh*&I zhk`jjS^1muCFs;Uc%tam3s&OvpWV0FkrE{lmu&w=OWd?>(B#Nay(Wf4-hCDm6BF6o zEVK12U_T%dEN{CF9UNiJw*+9Um}Sp{#oH4#3|oG&0JxO`H-k*9&d#)NvuPFl_U+qW z?>dK;)aeQ+TxlWhMc-{^0o_NfM5s*bP8%NB$$R$|RxZg$co?hDrhNx;W?;hB0sQNR zbK**IpohZNz%z=QEiUZ7I~~~-QeYIXVG52pDVdox($W>LHa7SLgsFan4h0-|29%UV zD%*+g9R3M*PNLHTtssjLPW?U9-0#$?k&C{T=37@&)C^g90r@dkn#DpL^@Vui@oqjz z?9lpmLm2E)wFuKU0;)i7)R{S$#iIck1MX{pha&J*_Lp9+^x5cE4DsW3h7ZpWA;FxN zUx<+RT|_aqdyi6=<4kQdGo@~lG5fH+wY9m`aTSB?OOGjdc97n^a}kp;WUQIroCzH=R3g#3a9 z(~pheqAR3D{>Dnqv%9SRF;=MS>%_z18+)FQj|J2|_RwZv!(`^$o;?Pt(bn4fO4%hk z_SAI5%CI4Y1hy?95-9`%1tV$r4~6GjKffFzY~H83Iy&;#Wbr84o-fOuZtk2*?|eJ6 zBN(vGuGB-jpgMID!6quAU9RRNQr>RgX+rxq4LEUkE@PqQbB3T$NSb6~G>!o{bDdSpN7k&+xrA7vVygJ$&1yc`IyZR31M zi;SxymEQM!`CLts*X7f9yD;~Go%%=b_@VC;w-AeKAfO)?FXFC&t|{NSJO*zy2A|i= zNn9k2wR}7${=SuXFoP%a1c0;-oEXg{cy2( zBO12XknT2{X^o!2=apkE`-&)G&@klh?lR#ci?KWTra%XGanYSL#92st7jD4!EWGrU zc5QkvGI6N>o@XGBW|qw2YU1?s4(1PbQ3LOe_Jhe~DlF{PZI)OGVQtU0l-SEtab8-j z&d}yC7(?J$E&`mJKpPt&ufA(Hm=u+QiTme-4Kh5~;tbn|(b0uP?ZC0$J5*^loqSxy zK$h2aPm$cqqWpoOqCWRK5TdUeAoOYM#LemIFAdiKTQ?AKJ7?Tqw+~Q4d#19IzbsC} z3jFhc9zRiJZ-O1B7htKEs5zJYo^nK|LIo^43|kraO{HLmMxt5Y019h&E_pVUFY{Go zDlRxT_r$R7Q>d1}3ri##DI*uMId1u6?R;f)8rG$Qn=sorZa0~wN&NvftN~ENsNL|> z;Qb01?$6;BFT6Y(bqZ+$&J;t`ak{aU{ z5i6RLCDYJ4*gF?DR0mrT9u%XZ{K$U%kwHqVE3%f|MU<$i7aDgAzoj2y`_P&p<*iFX zidK@0mFhxu?6@#41{DI(1Tf2X_oOE{32jFFNwQlzicT%3uZ4ikqGh)`5BBLY-+X}G~kL7rb7GZM54(Y0&Z+NnL8D8}*AveNWy?~P(22K|t%S9(@maa|wrb;S$ zVt59|3%>npF}cYL(wS8^_V{!n)&-h7576RbOFZO{OKkUY7NwDiLV5^&2A+3n;S%E> zaVDDCM}1kcEq(nlbqDtk&eTjxnQykUBno_`g|%7Dhtzah7GN8l)ryK1GxPx}F%T%k{?D1=a;0@}QmzQrC zx_6V`eDcmbZ)tI25v>7GPQ5Oob7{$MFLzGAfSck86n!?P)<4WV4@MbFn!I?;%p&aN z^7Mm`D=CaHGiFnz{v{bG9^%Gc?$O<=eh4VFTwdirc0YHU~{+3+4J)yievANcq(Q(faPsCbrY=0T%Jjh3|IP zxgUk&8MdO?=&pmlL`j&BjtJB()dGR!uF72zJNDbm@J9?2oc%quojPy9>I3^*czqBd zB{y_N4YOBQSO42^v&6`9vt(Nc+hdPOlAk1V>__kW&T1fF~pK#bV2<|k=f)^Ay@qzkH<7G>==APHHrxKHySBabDB%@Ng#xo3Av(PLX4V)VcQNollU?L zG7kdt3gNWW)WH$6t=*yHO+9&G7}<{LXuaJY@&UZcLu41&uY-$PD#c5k78nHyqpX?B z!7R*~JV>}z_5z#dHY7x;Va={E)fX!}ljxY+b;k1;WT+3j5(d=}>do_c(l>=$l^eOg zn$cen_iI}Qa<7e~T4h)t)Mz+*RNo7FuxHMcOiSU+u`>I{gR=xr)_t?_U_lROh+0MU zFSXx<9Ch2QD%vy{2h7oWh=Y`T&d0wXyKvY-1E)6Rh@;W5D}6Hf=VO9F#(J2dW6ts4 z`Pk8V-(xJL_HW;!D~@%Hj5>sq`bz>9;y2d??tH4w1Oo%Kg_6f`Lpr`5RFX~*Jru`$ z%bzB*TfimH%Bb+$U1n0$SCP8vdi^|pyNM}lULs7+#4Y72Vk|}VS`jv3eQ3EEe}%{4 zKIp)HUW6iS6^AU|aeJi9G-V;%K47=1<27CSn?!x9S6HxtuLSgRs>9*4uy4+}Y8v2J z)&UEHMZkcyjRd`DlaW?hVM-=WSbm?GBqS1t&Bh)_BfsUPekPDg!bb|Glx;FccZaKj z6mNn776SJenE?Y-Knry?pMvJS`;7iO-y&>aw|katdtk6J1w1LZtVNCo0gOon3U056 z+2%9Wh7^?zxUiL3`5q73{PYXkG??^k-pMF}@?p*_7LTL-q@G2s8n7$k?kjIA(tJ<% zMmN9Hu+;+p&A*EKl|-7MEII-j+V+scJ+y9JkF0nK>!0K&dxgxL@I{ptc@V&#aBfy| zg3GE=6e7)oyHwS{ZU3N?>n!8mK5W{d9PCRMpLbil0XrY5uskQ1iZ*&D=&SvO?jiVH z7%CT&FoTU?#fHvOr51T&gH30s^W!X{ukAXxgSiFwV>m^x15Y5CyPY#IUj0njGVA!b zBr@_joHWohEIi!r`r&3vz;V$8JOA~fd}@-b`s+XKn4ljQb+PWog0}V!^!vIdOoa5+ z;P;6QGOWm&4q8PhRHGuiq&O|_crZ4J4Xu#>03RU0!4G)nxUXHHG^lk-~oYN zQX_P&RMq=R?BmvV7HArlmu&}zcI>R%Y(N!%U<6k8#i??4e{!=<-t_b0h4Ta{l^O~3 z>ldpNCMr_(`RQ;I7Rc|e2-2(#_3F}Iy3SmO$jD6Mpt=kS1-YMgRlhVVv8OygRm^n| z!Vm<5X zlumW!#veVm&OO0ONkYsGP)j)t(Nra8fP|)}M7n`?9Cd{hn)zl4hCNqa33{M}hj~_U z&fg)c|D*U0CfNKk?Gu^Vn5Hn z1%{@Sm0>vt_g}IdD%Y$meG`^x)iFtORU6G%j)fU4YsG}nh493G&CiG5;Q>AMf@>~v zXvd|=Q3a4vDtNbbpR0cyyHm@-!oZLUxCx&=6W5jDn_JyH6~jMH@y$Hv+VCvR+W_Nq zUVw_Qw}o5^8AJM&?|tu}(cv(MML^fEDP=8cQT%(b8JI57-)}^0n8m*(vYm98h(9vs zr3?}#z!TQhF?4SjWaI_MSJJY7d~}rO*Dv2ElIi;=I|2OQT-fiQ;qWuCupGFTtcp37 zPGwAYU%-$?)^v--$&G`n^S79LU_%O#j`|Z3L_AB4CLw+jDjGKk#fh zjYLHh*aZ98KcY#xD^;H36vwb6*qil>)8aKdw)UO;MvHIOCi2m zt03;+4=bPipAbHCe)P<3d!Y+#SWb>S+>|dZCFSz6TdFq0Y{b^qB}=iI@^VKt)6crY z9gnwUMNZ}@S*)S&s-*evfmd*U=H_-s7rorJ$z&KIm#R4Y8ofYO9m;P};(6D~%IfDA zT5>3>&nsLNuTTyfBx7=Z^oCn|^6uTczEkI*B~$R7{UU_-`hoi&=_Ysp?G-QNiCIbb z)zLY>NHIV&n9RV_rFo4=zdYZ4tIW_g4=FT3m@m$r9sdde`G?g$8y_E&-Q6bchQ`Lv z3k&%uX&8!y*IX|kpk=Hjpd~pYqxK{byVf4!x&2z^!WdP59dRg*t>d>qlmYlNj{4Uh z{nkcjJisiR(JWZFtkjhlLXNm!ulZ}OL<(-jt9pz#2z5WL{`ozJ_A`JwC_b3W+g*V1 zx}2wR@VS=sl||DogN?g)gS7JP3@AE|lpYqMA{NJ{T-W}-{S=HaznJ3qP$(QJDUQO1 zg^LRsgoCFWGthGK_3Q|75>P0GE}Pw>C|)VO3|vUIM}8ye^kAs$AB#zkyNA9OC%;uP zc5;vuE8s{5z#Y)TspFMcd;JSlxIMV8+_S&ZOMA;Ay<%x? zZH*#SL4NWo>@e3mwn=U2_dwEulBkrWMfiWT0OHp-vL#@bx$~32%L$f_&{JuvRok1eZxxC$?w=z8=<0D6sZE-tyvo;`M4gS>n)9+3C;s@85kX zG)7#*9ML&=!a{nyUn`rtBB@aZ4~M@0P~m&SB{H(Z_q~_aE6U)ln$fzo8rmp=dG0`t-$u zwLx$^=q|OIkO7z2<}gC<@#Zwwr2;I2+d@L~tKrbx%VpDFwX#QJRw^V{RcnQ=Llboa zaWFf6k&6Ovl7NgX=LFtZe7$p%3|o%1S8%1f#Sh;J-Nxc&my+-)nvRi?@>a8#Dn$D~ zLc7W=dW(xw^Ck2znd(!ltL0P_AZMzq!e=H&GP=m9z&|mITwb?ZSnuuBP6esEdiWsl z+n+b40&aj9SPO@}jb_c@lW9TCp={_|g3JJd(& z-p-MK^$LbIvehRu$sUyNVHCpdf)o2C^WGn^0`0RKhRC>`KMbzP8yBzO1__%M^V|Z6 zTbwvxHs`w4&SBHvfmjI;Z7!8$y2b*6X)&V1SLW8Hy*%2L!HD~v;R&CZ$H_hTMay666e0m#foC{z6-L=Ord0 zPCy%ZATDV8K*>F;{G~^bH9LvDo}R8wJFrnXv`@-38w!w)fNE0);N`z4)6TBymOz;v z+SLObdrkmmu652%jg574#riMuSng3P)GOIn>GxT`$8qVHvOZctCt3PGV7e1G0^UqO zMrLmHczD$5`^dxeCf(|(S+T{COW>Vgt6fpji+isUXEZ@(a=`_=-4V1EN1`+Mao3NA zvzdqQVB31!mi|%-A~hNQ>DRNLU!s76%gA8P)h4|DgS$g~hQo4JU%JTm$MmI!%pLIG zSvFBU-+!FMbv=K+-Q2A`@iz7CQ@79V06Td7jO_Tb#v7ar8{*{r>gU&SygB2ah9Vrg z$Tv1jJ)gHh^N$0GrEVo*GBROI#$<-ri2mly09Zi=Y+D1G*p3NO0_4|W7B8nb%J#6Z zuwS4(F;fV4TSmHcMV$E1x7i%+iVE9m=i=n_EvPPUne!PsE!b8rjO!xny!m_^ug3J4 zhy}8}>5p2k1_pH6#Tbj>n3~d*b318}C!3W;_?(=W)fu#Lbr%~4_AQXm*?RDOgxYsw zyj8Q}6^E@Rvm3QOl`KIu+`6aDg|bTVfCG@fZq@%kFjt}LZYOOOr}&^iR=?QNj`v}k z$L-F_sYQOXy4h8{m2y7MvJslogacVU-gk>+AA5JBq;rH@r4Bt8J5ojLo4batyn@mf zF$R)j(a-Y&<+#u*J(LE^@=uArJ^o%#pXFFCwmW}Ffu!^;+HNI|x*$0d_muLLHT7$6 z^yIHuJ&}F}mAc)2hgPnLAL~m$TLJN06zS9V0V(j(#lb<{)bzEfsFMZlca4M=$qMTG zi=umvTU5XzJ$?H0m!d6~<}QKTPS6DVd7+`*RD(Oq_UaYn+qq!M@>ysA?JVxUuftyO zWxFeh9h(-FuV2~s)GD34eUUKV;2!L>Sus*ckXlJ_f1xV@93Vx-;0YA0lNqfDOjQR2 zAsNyzgZ}ztT|`~y%3ztrM+wk&{gUI0ZhiZX%XWQ0>jH5z$@lyrTLs%dEl?H8saTm= zpY%MbMn$k*2rt zd)T!*_;tbDSvTOZ`w`%n}3EJlw>_y++tK3{|i%L`uHKZ;cz<;Boxp2fy2k2V(mq~doFuUoI& z5j-5MCpN_}V)LrAu7vrayxAFF^&*vA@(|W*ySp{Nm%R<#JUt`3y6(YWz!q>*IUih# z>tABrIrObSF4a6U^SqqpZvAqr$(?d{5-}rQ&;;OJ)16EiG;{eyg?0 zxMgQxwl-HuoFvI96Mebks3BMC#8kwQf=5ZzVo1RJ^ClcTJRy!C10P&LU>eUex-a}Pjam><3gKq(b zu(Wu7#Ft+HRWjonO z`B0UZ7bi8JX+!ea^t?0m}wm#2HO2fS2r}(a<5rPCxoJ14H|_$-Zp8RKNtO z)h6@&Ldhv9DR&YEL2-+D{}kJ95ReP>mn@~B@X|Aid>7ku(qst4&EUnHLr+Lzp&;`e z@>QzOa~g*Bv#U0Pl-N3v^UAZYz$xq4)yxBqyqj`?=ZK@lYwM@4^s{Hr!f#L!P3*%u z!SpEYy>u6L({SX{4{ET32n51%W?^Z@X{uq55-uRMeXy=trSEvqlzwvhVr(Jm98`^c zZ>)cg^atzKFX`Sr6CzA-G?!Q2{zo4+oee<)=S%-)x0h!-7w$i@|9<*0}5xzy2PE?tuMh{*E zR(^p!!N##l40|tY_H4$v%Uhu<=i9V{Mcva0&8UNF0udQu%j6CwS0b$^U5cVIq+fV6 zLtt5~)Td9=pnv>b|5BR1^wMR5ulE^TXHE|$DO3|oq6%D`I3-yv=~Auvb%56-WAWDk zv6t*v220ZmG&cV_`O(Mc2}k_OS3WW9{w24_8{D0Rbp1nUf`hkDA12F?2gvN_3 zDuV0*4=Xh!8Pf+$ZpfOTmd}WFReZ*Grx#R(*+yKa!4i2G^xtF&_H?LNGW(Vy)M9Ln ze&C{J$(g{MteY?x9X5D$eEfynztZSiU!UD%poz!X$)UUdhQHnr&WX#II%F0!sm#vwzNuM>Q4rnxN~@to%-{an=~W{uBj!A>%UiM%-u8lFwJKp zI({yVE7Qv?`t|F(TU9uTaHysg48$;^mXsq8^%cRw_2RcTk#2cK^W1|@)*VFy%v%z^ z<-iJiOGE+AngR-$;N1prDv zIy%B|z3J)0D1vM)v}@t(tA2o$S@N~NnNwdp+3XkIYt^+Exr+ZCFb>U=axw)kJp2 zUX3!sj3sFK`zipEsofaQETeO^e#ywI$jV1o2sRfut$7nChbMoRAWG)@_iTugDcS)w zR}+Bqn)-uvkW1;{4^5La_bbim%wX(;H{NUDvB&^vGgB6>?w@?)e}Q_fweJs5GZ3*K z@zpa`5)oOO96gKpwESi}gm?ZFa$I2bYR3E(APU<#oufjm5juSKJAf2j9w~=~ToooO zcE3fa@BUbsFrOZvztjqmFAgb%21fp)@5WlNgdT$_B9=AE zFKTJ-FMd+@kP9PDQ*WV%bZlROaBKx6SM7f@&87p%;`h;8K!depha-t0Ml-J{xY0GZ z@BQMOl(hQ+}5@B1Ht-oVqr0RR6b zm)q(r6_J_M>i`#qfMXX>6=(LAD(brB!LJL4>ydEFD@EPoVIIKpyWFE^XZ}^5n+YL< z+IoeBg@(uVEgz;lCc`J#9jZ1{o@Jf#1ha%)uYUthA1v@b*iBdnCWM0nLaqOO>}e1| z435ndaa#>eMN1`9kT6tcHico8Ap%Z10;H-zcxI%f0l5Bij^eFvqpo3c8`ha1>rX%~ zy=9pukr=dEn>^fE^7V@X>o-#?m#Up3EPhi%v&~*dZ!4Ofo<4=VAm;UJ!vhO!HmJ_$ zw@%P}0fdY>pjbYK!&NSb3FVLk=07-k2js}cBioWx#^3K`k9=g!FJ1`&H}Q}}bXaNN z+c(PW^g8*Mxc4OT?K}#re%9Xy#&7KH0n9p%SP_>5?1`RCR+7`cWt5%t(&N629a>e_ zWOcu1;IO5ozv2oRv)V4FP*dbfNlDpqcXB%ZNqg#tgIiV$M6dF}dV?m1-ytF~F!+BK zYciHCKbV`~u}OE(H)TUS0xVFrIL!Zg^XV|7hLvOEy6C`S;}6eB2!&3G}TMlE#*nPA`^r(48|&9Nce53KVR=PE&Qpy(THhxHC`z z$H`sTUDj*f=;ccjn4yMl~B!FawuL=Dtpi?bmq;y z$nA;Ch(5o?#g(&*T;EoVHeBZ+0Y$4c@|;j%QjOvgHC=AsYr!_~m1~Uy*%|lkyc8x1 ztZ$S(&{T`+unOjEPZE5as9E&o+qY7$-2s)_by$dTi@)?c_Wen;>ssC4yz_P0r=+1D z7Q+pDkzrtoV6b%763O{32sPTr zpD%=SJ0Xs`<=SlCkmQEa-A!Nfu*8x zH~%;$9Nb9jjrjN?>)P9}PLMI|i~MrwF*lVNpm`*8INi1ouKo{^ZmemDxWp#G*OFq6 z+~-oV>8rxN9N>Ac$V za;OXowh=_9g0~}~+5;KxOXQ8Rai6{~Cnrm?2h^&KpiA~#*)$se0ZTGqmfruY?cGfa z2G!NAzwr9&E9KA3eJ9W=wZjc_RCrag z4>>JN_5MP-MhHo^JjIvb!}fvepqk80K}csr8B_yMbwObun%&#I6H2eeeRc~#0J`PL z8BDhJ7rF8+#q#gbmgxt_lF&$372u(|q1RYt3&tW50GNX1AL735F;w2+n*z8;0)81) zGV`J0^m4F!8flol9P}crRM0xKK_xP!aByTu?~^(q{91_RZ-G%0$(;m)`gCOmlN5e} zOuKKF78mqu)ImW*N500Sj7RzQ8}Rk_Kg!#P(yJx>3*hToH$#As4ni@`@)sE){z3S9 zIxEIm-+-KT5LDdERjhR#Z_-NI2Wq*eaDycUgJ&k8jX`os5l`H)I6!}pDBP>_QVmn;{-)Iu=0Vt2K|tZk0->d zb;kv+65D(Y2oFk0Ud#k#WVovp=$6Ui_FNC>u)YBYVRYo#cG|)JA<$8&$;p3QdmNfy zGL4Q*fKylTX-S~VH~=g)j9*u858g8P2g#KHMh@u>nP6{{aID!9cqk29D0X( zpoBVcenf^%+a;HJ#37ptkKnIwAA^#q`rtlLritb%GFUIk>1a(}UPg$^&aC!Wp1o?x z1cR#)&cVQ1)B8qqIrr$HQZN+X{S{^8Za_?xz3i2!FUxkGmPvCsWaYY;?YM><214Ap zGI=}a;4&~%B=EY+ES-WG=!p(R_JIv2n&5tDtk0E?9%O8&K`yOzFYWXKayJ4;Qr!OZ zX!H!*Hcb@HZ68&ucsBh69Kxvd;b%v$@*!LNOc{HTj6bAoanBGdYg|s6U^lBS0m1U+ zC!9jV@@x#RavKx9ZS)kG9H62)uX)tsu^ZxG(6$FuiweT(#&}B(vm3Eeq)ZQRdtT@=|obp*MyAH1s zWd;t4*~KS|B6r;F2K@Z2=u0H!DezDD+n&R*lg)pX17={TIK4qphEk$2_t33==%D-` zYB0#JL3_zj_x0N3qFl;NI5nZ_FlT%p~T#*O=c11wkf$j(-8pO$LfvnQ-Sma=?snD(XBdTi|mRGlNAsE?uN zGW}9;FIxX_8RiTJaqO;ND?Tjw@v)3~ahDD1!{0s7jbvnshfP&0XQD65`7)}zZ6B(t1T1HMCibR{+D6rCXm0l4QmeiA7Ya+3z;TdIfchw~8Vt~} zw|6##(lCvc+8?;yatV)0WudpA71x~_|0F;QtLu$V{g99J1pPRn6Gg5A5wccNVZ~vc zxPy?Q5f{hgKWG=!y_$S?cx7wG%lxjFRz2-aW%L#mmNd9grV~1-jwkdBilCUlg>Mh? z*6vM>s#9(vUabz0V|`z^qjGq>|~G8wFYyDV05ilin}cQ zqzrlP8gL=m)hGu|YN$QCLRPL~$f-UWYNh+S{q#*>y2;-odT2_GmT5?x@LNvBM1=^= zgxFW5s&8{@Fx=Ga8_CTdE^oMaT~-c&%=q)Jzm`U}RBmCR#nDU!fJD~;$hl{Cv%Cf} zkB4lu$l-}v16ZKFL%_#zFdYNNn*!2%(UrF;2_^v)+RQpc`6Bi_jwZzzG+9nZIw@}7*5P?x;>yU0J5UGdGKxG9k(Gh`uExTz4aym?HD4Ww)N3vc@k&MZc zt&RP%svZl85?dv@Y(*SIN^kMP17>$YdbX7G^j-Af-`a|GUrW=^kD&z9gT9GMLCQfe z`+vh!$Q-D-Hm_XSJRbC0s#R8!hdO(1@vTahxLGpj-+||>B$#MHhr?hI2tFYki?t#Z zvi_f~SGC1``M5v-&mb&e@Lzti1$BmzP^QN{;;XA zCRLqOUjxh|*C}3Iy8|LKjkvg)xnDT<-#_i-ej3q-py11YX@1cb)LlWwm7V>vK#{ZM zaHs#g9;5!jkx$LY*xH-=0G~u@nkv0@GZwu1pn3ETU2KP-`h5t>Y4ah~FjARpCWJQ> zE)xS+`mWOhA))*;eRJ^SzAV+#cG(}NvE0#N+!CG}OymK7t@+AG3NBIei%5tUx3pNU z&9qp4>pQZu!b|-2Z%mGOGX+>jr<&idzL|r;_2Zy~#0tBoB(HD6z~TYLgkmC$N7jb( z0c0SD>t(^u!4Q^zqG3C@aCd8WjEBwfz1KOZHtC7+N)#j#3!*0bsj;5UnJoFd$Z6hK1@`d&- zLYuNR8s(&{q^6OJkL2eLs<|;{zDgcQ`t|DxT4M&xp|!73=0*Fm=7PHW3k%JwR5DMt z;}i5st_GYOKX!A|pTtiV77&~Yw*%!4boh`m*zBe;>6ozNt`3BSqat=M#v1#>4gYx05rzQBPC^pG~~!p zY~WF#(ExhXoTtJBXh-@0}Mdg?v+<#>Hd%Yf>F zgZr5em(dXa4`XHeFMbO_ZqyvMC05oT?qvsx56mI3vr5{Asm4(Ri>i_+NSOFj_dO{x1cpc&;D!Pdxb;W zFWNvaEpJdmWq~2r{q7uo_2*&lfb6fkD9gWLqTi}&efq_-TIJ+QJ;L}dYXSyzK6K>l z{#&HfVmV<1yyTv=<9O2{KeT=~?re>O0`V^RLt@%T=yVGwACO=ytaHbOS;S=Oo&i@W z^*gzVY}8MhJiV++!IqQ7;)~mU44e1 z4sKW~17|@T3<{&1?)U%IHuZ+fI2E{W@JE=;VPN;)1++y}fd-!bo+mX1tykg+^Svz` zzb`WPw48oukp#TA+n3yym0`dsuGT0gUt6URgdk1m;C}MSFTuB*To5$^*x<#JtA|h_ zq+JV8>T{6ryjxTNgPPoNuUXu-va z2Hmsi6;9Ret0*7%<-)FxzY#s|!mmsLxem!14pO&OU!uE(?U49}9#tF7uf z-AL64AKc#zFiFQS7v)dAyvB@V!Eo%EWr{~b@Gp8n@pN((88I0{eJUY6$AWmWc>gbRZhe(0Yn?lrjJG^Mf z_W7)wi77#OzCJg+sjcw(c$Jk(5P{f)_4mCW^_KNsEcM;cQI|vi_gT!`HcaTK_3F$d zD?83~ih@hF3Zac=f~P-RWxriwV_<5WiWe6B^*FQY*E)@YH4v-Zq+SzdP+Q&H%px!7 z#qjk#8Gt7T8Oh}4Lj{xJ(}mBMj5c#hasN=)LE#;i_21Y+sPgh_k4P^CR^2tbpc?*; zySEBA>gHdc7nik(&z$WC-woC_rV&qZyq&;ov)h!O{&obuJqMiZMM7?m2`H=}p1?Cu z5uJaew5Xz#lQzlRp{Kx*U*j(k{P*ku<83lq&rHgDeyJ})pH<_dj@zPIO*2q#yaBUb z;KT|>H>e7|hBG*eNS*>vUkwt!Pq_o-z47t!huRare;1=j>vj)5$AUCVkd3dcDeLsx zo}|i0NK_h#ba~hY>7*&J!1LtXlu8?_CwE-d&gb41Yp7DL#NIA}vaN=;w%S(%oku`t z?}sP1Uaj^Xl${X+6TO8M=^6ltzcaEJ7}042Y1!?irM`~G&!?5{j(z(^(vR_i zbU#(ypH+a?3+GoHldfK(lzGvS|pZHD5VN0bD;}@JM~S5F(ZX&iw@>0o+%F-=vzhTv{Rvm}i{X zHDJU@(dGD_=QrpXB1CLKnRs;U%%>VEc^Y3y;!=B`6`sJ(tV)>}ItcDSTAo}$KcB$Y zp8`y=-&BY}iQyE2uZ?Eg=^raFi=xP#Tv+dZDs-U@L)SqfUGi2ie>EkJ=8@Qxn;sRn zB=Lp5Q$HRMpbF;cBrs<>a6o690mJg!OEZ|*d^R5t9iN>PZOQdtM88l}oDZI0FLegP zT&QQF0a)MG0-QKU(G|!RqJS-+)wNQZ{6M$HJV2KyV7fc%+Zc{be32Q*aN|vttMA`2tT-PoyGsvc`7kL#?v@)J0``^847> zZ>J`PHg-i@jEbYb$Gx~Ka;VUMA5>~j@@s~QwCn*RQ(Z08ykTW-KE(~<8WaR>G@Vgi zJd)X9YW(q`5Y*AF1d+b8+2B+8am+7!ZTO^wT^bz*jkH4uDkK=;8&eBX+qy z!j=aH!?4U7I<#R-(Vco95^R&5H8k)BX#j$R$P0t{akyX{_JO*_=@F-_m;_BM=;0YF zHwYU9j`Jl^0X=hV2H=qA+QVzzm6Av}%7?>98U4E}?(TwI`f1YUe9ciS#`;Sa{)-X5 z{tAMc)dO>}8+00PD`~dZ_$dhgNl7Ki=ZoVq0u64f#|wSG9~mrlEz~gVfYw6G&84fO zqobBZd~JAT3K8nsIW_I!mQ(~}WByfC>VVbE+}?qMNnB^{6YT6o#=9cX)e}C4F&L@> zHd%6E{5I9T-=VI9Ka4j;40!F5!!A7=c4J-F1zoW3+%sKVfH2Uy)g}SCzb3r*bsKcB zejk}_2JivBBP?NGL{uQ+V^MxjLtVHKfOM&?&+~wzLyfR3sRG7*hx*1O2u$)9QBdc` zom1g|HuXX-1WE&+f^J%eP{W3WslQFp4s>K^&J)$Wl$)UTvT#&0c!45jV+f`2iwamA z1e0cG^!H>QJ%84B@_4{~IQmfA1i05Xl#~X9=w!grK=cr3^?Nx}9I-9sRGam}Ay2{!tATAPE>7|<4LUKHKINrZx<$159yrA3d9`6>@MbtOqq&2{a$8k!a-e%JbR zXcPyE>;rmqROR{bv$FVm+n&H`(ui_41ia6m+|bZeV`OGVqdr|6-}Wg_P%n;nU8Nv@ z0zzZbrdBu*e}g{{2#`_W1Jh~MH0MrcbVq;_+_}l>aF-SC?dzA!#?v_Me+-8+qeVnT zzX}E70S5Ga>Bc)J{i~whx5fV{s@5Ln-Yg9vVLGY;a&J?w*}QvU17OI@4VtL(=@wgT z3<@-ZtOU~?5a{=FE15~KrtxkjgSc~O7^88F+=O7USkjb#b4M!}#538MJ*(z-*kies zq!biG2I>X;20Eh-5^qrf=R@VA2Xlhk-g|lF4vE-oYe4yA5q_om`?huHLAMmQ;2Kuqi)5?s3x)d83343upQp#Ttb zIA1u=NE{F+{o#X8pZPsu*LgzAZW6NBo$(QG&7Xo$S5VIk{PZZ*JViIS6&XPdAH@a$e=g4#;S#;C;9@kgucs<2+?_DLERdmPbyY3=!;HVFsp4O?g4~KM@19@1 z&22ZF!+a@sFet}JVqL(y{Zc`_y{~WED7>PiB=5?A4zb=q!)vC?0RE3&iE=BM#XL#r zLSzD2c<7+R-2%oyRh^+Q&+StG!lY9v2L)F6qOYH-Hut;Ft*wEm*UmsRad_kn!jPCv zRK)xL2=F$~u@a0Vd%t~)=wLZ-zJn3=zv#N1!vd-Z@P%df*d=eBv{#&jc@XA~5QO5^ zBj9hKhDHi95mMf+_EYgavo-8A3RP3Ky~07G)aCUL@G}zJ`_G|fzu6$9IHoSYw1^){ z(d&PJlXU+-Q0Gz6_&B}ghZ4{|^A1I5-*x*C&=u)_=UfA53J^BBskb4bRi_x7-4T z?ZcEvU%gE|iumFVl$s`~+U~f6CtIgjpBSA^Vo>;EzU%gqGfN1h?fY1%gf21%0C^fT zWav&$PamJ3odQR0IEnwF6N!Fj(&B$A`$vC5OH=tc;A}M@Wi|dLT7fp+$+u&vdg|&bqTacYmzvFspvR zIxvM<(4rIt;3Z8S()EGI^D1Iwj9h9!vxk>pvHU1_egy~w!VyG1FFoo zGPO>rq30z zU6n!y7Z`JZaQyI^J=@|x=to;RbLo(%cKW@y5Wx`*0}mAt4?`i#C85Y*>RZ9Ndx=_K zcnC!rNDtT#G}M(K_Y^qeoQB4E%|!3SS%b5FtUaDaLbF`LIuG#}%!w|-1T!t!7B zaEhO^YAmk-mHqtJH+peCmC3MKs;D!hZp#rmpnZm%pG(oU*9BWg>Z*NW!jO8wF0F!X zm$mLm`1v1#36BsbsN&ZD16177Q4vT9k5U!LFE8GUW3ets_JDFs7!POB2o-N;C2JA= zu8)PIHaHT_ktEukJ9wmt)4!(PH|T{f!CKWoOSeuY<=ps#%idM@!$G$JbJMyr6XoEy zKN;N2bY8%9$Pr`+#9{s zlj4$+#-yUsBQSK65*SI5a?Y$eszezb3hfHa*Co55I_Mq+=$QruhN3T?As`hP(KXm_ z~F(r`D8X9c1x`up+NFQ*ywVy*>eF#_%o90 zDs6Rj^@M0h`%*}aZTUhmKfRh}(VN`W&dW)T9%7F$c;rQfp4cJAEuPH)IcNOxFg%8$1GU5FsJIwi>elMh)kbvEA_pVk5e@XinO1-6{goym3fSN5DW8VtuN{BV#0u2+^lJ8e17@&sLQ z=Eu3zamP_){)*7buxi5-jS&ACk_9sh_L*tiT(LB8e~)ynauq{qwO{j%{tnbpsxg!6 zdppcp8QqaPOZ4fKHK6Nn%!tnVnI2_;q;aG*V1#syKznw*f%m43>EIeuI^_!p%C|>o|^z=04%psVktVMs95eTK5z3?W`FD?>U zIZR_7xJ&?6ClQX6pe5Y|=KMWqrOzDLuDK~535UMb^y8ajPoSIMn46Cs{~q=SZ1BVx z0vfG(;G+y(`|;qWz_uSaTaZ3Ui?P00ON)tu@!7rBriL&=77l0J!IyLP zY7^kT*v|;Pq8j&Pj$mxLck5INBu_zI;r%Zy!U6d9FHq$LeIHlUh$BEdX^COk5fVkn z0ptm%S_BrBJhC_J4D|*U?!oBqI(^(FCect?)ib(}c@qT%NGVSOjjCA^*GJdq`+Wdq zZZdVTY5LfX26RE5a}*aM35Gcu6t&DegE&h_xa@Wi8JS%-`;Dj zz4kr{tNC?+8LxOeJ6rq7xbxIICZ_}erpO*X{8_?bXJPS^a2z?AV4fyICppY@cd;_S zTXX{G#h{@pc*e-j0CRM?#rxHtfjIi!$w^{*9=2N(V~9h`#s(k~5+qj7TEOC4m6C}@ zrPz+c!cAh^3`?mR7~e#}M5K@{W{#2Efo?y2X_ogBjir)&AS`QD_!D zuwVYyagJ4x!X+rU{nQ$~6|f*OMe==6Cxx?5&@H{i&=<`?Gfv|fg@8n!Xi z=suPh0&*MDXf2B5dRXa~ZSO!tu4e^hNt{1!y4UQlDypUmdk;31yCoITgGFI4y{BB| zZ$!LK9)W)Kb+4NkV|FUNW7bvb@a_+bW=g!>rf%nRtuAcl`!ch3HT5JH=(0;y=Fe6g zs5}VXGPP;xsweh5>Y>P_7<}FC$Lx}~ZV$Pm}YU zP62e&5l^rNG-RHkeb{-F`fN#P{Z*%tmZrMpld@LJLObaPDF@si?7 z6F=nI$-HC4<9DRF+mgT6nqO0)JC>iz1nQr05Z`!{&eLKX)u_Loae zgDms=x^gEUa7My2Gj~TdfPajWn5C7O*gkx(DDnP-$<1kl1P!^jhy_5nlE2DFP6M`p zhFFATRNQ9sWGyM$pXs^ML;Z(LSlwn%Olee!@`GiCRHvB?$tj1=K4n|F9xvIxx_%7| z0XJWvSm?IlO!BsBQq~s%moxV$1sNzke~AHxy6%OWbbK)AY2XEh%~TxWno0p@ha_C3 zQ5Mwpoths^OUL!?w3;)`R~Rg9ai%C9o0eE|k+q5^dk6ha_T zzuC5Wn1q?8iScosxw(Q0pYU}q6^l2_Xttz;5dr&_`8H30DYJb1?1}=qC_AVu`Tb)q z_sozR`Vyrna2W}nersfqFZ&3Fh}8SufUfUCs`9&>#7YgnyniM<|8U90$*SwSjtN$h zgeNZj6Ml3x>Ctw&6606pdV!j1j>JA#F&O2ORv<=mh;AuEY!(==E;?`E%Ve9+OMVq>>U%A@D0PTYJvdm|m2Ku^yNDe` zyNRN@`8<{<)7xP0M<4Z2V9g7cl&&!?B(_B^f;e`kEX#dGMy(*Xo=rh&A+IAaDSq&0 z&-S(*-eA{9*?DVO#_u_r^p-EMX;Jl-4MYz3`4 zNtFGM$$E3YUSEM-==cD|3M%W=PuWwq_ureg$7oAeyoi;v1|#}v_IOT*Mn=GEr#gW{ zY0yWOwH4`GEh_@czOqf7r&7s&bdAi!TRaV z+la0Fg?@hahLH?mQO|#1koUSi3+$Nr?FinBTMK|!VGeCgTVMIAYif*To{n_}WrQ79 zZLKZ1?rf5P0N~r0sb|dun;IRB(a{<8)e9GUQi{wZAEdyci4slrXvLkZgA1=pssM9I z{E!)H+U*it*XyXRuP2IwVU4d^A`R<(EFtm6`$h4{?9B#X{g5(MiO{9$(rNYeOZ`ubSnPe6MRq(c)ME639E|$dvFB!8aLln@Q`Jrc;^1HdI458 zxx-hYz06Q0PLwYtV_WEN$K|oob9QWuE)J)XO>#ol0LmM0|zD_h5j7=8lSi-E+(d)3LpKD|9xS2`egWV}^Jfa^X1{ z9#HqoYOhIFZUQPKq{NTEVjQmc>hqd3%*-Sb&(tVp>(%dXVvaXgMx@Tp-Uh=io0D?O zIxM{Q^ONIUy-;l)VwTm0YNy139VTKWU$|M_oa?em`6vROUa>rUua^1~Au$-+EG>lbn*$P(DWN3bZFp@A>Ul zv|5#5YV_l@+ck48ueF>-Wy1p~g*-J8aRC4q&$q^od96YfO)J<0`2$2>t0 zZ?S#uMjbcMkGh{UdHa2?8DmW`t=t95gD6hD^`b1SXY6ABt6^GRw!w zS@%UfJOrPVQ`)blgU?TII5!rmN_7}Q{pQa?`KtUn_v#^98e^UyN|nh=ZB1US%gD&| zZofa~d7EziF%rqytMmlQfk_abz1_t!T3!Sd@<^+$WeV($S=q2Oop5SL7HQqCq!8Gr zn5{K)F__vq+g)6Y*#1}@pu)n+t^wfv?zBub>V5@2!RYzE#rvzb2SLg-3S>*H=WpsaNIF^R!>c^Kya@FSnOBTFxpAT3hr$Hdd6{I=<4(08REFIPA8;G)qDdrwAF;9W}LV z{gFTKlLvv6fyCoqU^Ue&ZX+Tx`|D>@kbP8~jnzKowzwGVkJZAQXbYH$0!(9bP^{z^ zUd@~$?}j4#na;JRD%ja;7@$= zY3n98|INvppkmp(x+;lxLj<)ztc1!?5ne5^H5^*qjuO=KM|%u}s^8Roa!2TA5p++0 z$^H$aoz>D@UkBw=DKS`&=3=h4`4(`86>BneIO?bFy0&iD3{Ib8qSV^yVmp40e6Imw zf%f<#LHCy>4E7GnPno~SS8UnTZ(n_}6lwhgO@`FqH8mFLuq!jDCbm{+}3(o>8`7Q{%l7M8((4HozD}LO^>>`*d<@&wOjq)_r z^)9^I*RZNmhJ``a_RbF5FA#^}carEP>^EIL;!<1(D&0}&2hYyOPp`CiJbPhd6cf%8 zK2GWVTkkEi{Dk8mOQY3^M_uGri$aMNFnq};E3k&M{FG z-5w)#bBJkwfVRy7A}t?l#2F|#Y)Ee@Q+?M|e~(Le7-==-!HRQyD893$lo^92p&By{ zv7uM@GbgeapvWyK_^F8coZ>ikYGcmH+%QvdK>qwqpnhDQOmtYK-&N;ZFzx-!FuW|u zh;jy2TUODs>A7sIIKcgxKJ)Y)$jmncQcSRm8qGSP&DuxBy}iBJ<73Co4#F{+S%{N_ zJRU;;7>tQ%C^*AdB@!lkZf~4F{1Y19lWWe(RqxEY7rOw?5dPOX?T1Ie^qJ|P^`q*@mS6H!WToSJa3!6Kh+rPINW&? zbP;ZJYKWuipvwgF(g9EklsQdZwY_2$-@ZahD4|efoI(JXLZ@`iG`!r2-%zV@mqY0y zFWvDLjjGDN0$tZ3}YVy;g?~`bqpX+maOQ8Vl z&beX)(xeALS5U&{ZtlGck{=Z&IJ+)x@CxjoNHSU4rX3z{jz$pON1)gYM6=wXsVqe< zxqQNzq?W^@QceZFprB(zoX-P=8m&>FZ&w?T6qq|d#?XrAy}ufJa(o;H%trH2O47r_ z!_Fz)1O#$9!Tr?#pOhThZ()S(TwuloD{HH+R3BTl^hnz)LOosymk_%Ql$x+yK%N}x zU#x&w*pm%*G;~xNJr`x2KP4Nt&r=+*__a7Gze$L}URXa5kBueCP)c*)xPoBoe|Dvo zoFZM^ zT>%g-b(q6u484Y+hPNzG5qL#3U=$%SVF386m_y($aSwkqzLrI>t8h9)(Y^Z&oRuh( zMRjp4`s2q$`Q`D<%n+&uENZ%%=657BLyg$f#~96WcmIpWi?lk;(-D>E49WV2uauRz zG$$p&AJmy({W~!8u~kIsB~eo@{_t5AJw7T?WB&}X4gToW*$L$TjwpLflOT7aw}@5` z0pVy{QZlh>-l3YM-@~wuEJ}#w<>Gpb7>V`ccbVqrj%pY;6bd(~rJ-Au6Kl{fZuHYo&XHhnq_m{^?tGqRQI&Se@XYnPle4|3PeLb z=-7PNlT~woiev_OSbs4|4uwBHC-~>Sg+$pf<;)PNcV43EfJ0&9uGt41ZxDb4cjU*u ze@}kaUr9#w2W0DUG*x#Z<2_ z_iAETrUy_upGu74)UfHe0W+8u6U;JxtGTI(Op|}!FpUkg#)AJmt$++0iVA>kWG(?- zBl$IEzcMe3S}cU!rzTvL(2zm&(;KiHc9+=VDNeYJO_*sJag4Ro(fJ2b2MVayj(^Kh z($hl&uqx6(w_6@G&VN5K{o(7}IgKLavrjgTy~NLTvSPHAq@EIqmsiGG``B~(@`0c@ zM=6u5C{;%O@F6h{&OAKjlJ)3R8c8tUiX#E9J~Rurg+jgyT;`sLANepA_q(R{Wy8)T z(;0$#nkpSZ4=>3Wmw-)NlT4}+;;x-+YRiv^34A5t#=y|lM zCbf>-DjvgbwgQCM^Hvt?Vli>XTA0tf_H9*X1=`;ciu8KrElXdr@g<9*) znk#7=7i!os)j`u z5OV*W?f$zPuB#PaxKM#mBO95hhg=NyGvFdbu1xdUiYsx@A6!7Ku+CJ)c!x`@$`4as z`h&bi-}DVa8rAa6K4EHjtuu2)D+u2U94 zns?PuAItCUm3lXbL0~gomK%%QA3ipKY7v><)E}_7fI{qWKy<5m^U-hVsCw&5VLJz? z#aQ6Y3=O_HjN?~`2LxQ*0q;G+8}*q^N*wXs`!Jm!9aBBsG0&Nq>AYSB)zC+O+ARol z1&dzn%-8#wZMF>j=63DVhZ7+N2#=cB^N)PGC)xUKzI%in%Ma9hdaZPqH&o5oA|7cy)TNKdO4&tfS z?SINjtb~?Yd%k=T=$Jv%RFX;?GUK$K0G|}*Ewq$Tt`c?R>>h{?J~Cls)1{_)9A4c6 zp^!{DgOHq<_a&uN0(J^t-EPnTxGE9zCRRcZldO&DmD7Jm(og;Q^GAQY5d|-XMPUio z`7sLH{GLb{dy0ZN^b^4^! zW2psW**Ne4UzO1|f}UaMkdAM)oK1s^g98*2%0WX;7HWrPQlke4ky9q@IryUxJPJWG zP)>>zM~bA1DkI>CK&*%koLp$v=AEHR&(S?q# zkm+tQnBehjLDaBk7wf<%iG{P#{zAZUdzNF_R!vVw$7=j6uDrYv+E%0d|4XX1Kt$r& zbSal&O(LwVeLYFxymAL}{DOoR04FgMg*!cfxTz2~i_IfVOT*}2{X2Te|;)qxmgb2A|2$28O1VAD~E&pJ1W(HxBd{eqx z7}g=`@4{g81~pVBHa4H3ncTgmYPON!89}dFI9`|2!eGMLYmhRs!^TmC1~$NFuOf0) z2N%R<>lCawdXA7rc1nw21Df1(xre|Id;4&tA28WDZOMY%hwtd4JF+-O!8F3&RWqs? zdV@_JT*QL>d2ziGzCMGNe0uC>-Ap-@( z4cJY;d6L2OY^In0!-m_`H73LbLRWt}-}X_4Uz6(=)@}TiMM}*RE?NK!Q=2WZDgfzW z%T69CT#1fLI5QE0Q6S(icj?4fj7os5sCop_mVW5@Zp!syk!0C$xJGXuFVQwGpR!E` zoTyq%EVWx7%7!E0#IL{yM&lX8m4>JRAR`6untl9G*`;>3x+N%hI1(dCB|m_KmVumEg59GY20OI*GQ zqySRTcYy%;ONd26KK#XK^SH<_PEwH=bME)jR*u$sL9;`LO5^ksR0@wZ5;)i(6|W){ zdDs;U1{ zqNtv$krWK%*CEL?j(WAuWSD8|_4$^BlB^d<#d(TS6#ZtL4}s)RwpYZ=&d<+eVOf9o zw)n&7=G#c&7K<31A4IK+O6bjWhNz^bm}=4Yg*5F99(8bzc3d#GPed5juA}mDh~d9K ze*Eaze}XU_vCilYA=s;o|2ZPb3YAtJ?@~Se4F<)v#>XcP2D_De12wjsFzY|C1LeP} z=+kYGXh8r6%R)Q2s$=&wgBmSX&2-QS=xevTeS{RTpVG?K7W4-bY4gV35|3Zr&=afwQayf@+Sj}i_wpY zi|s~Vz+HyaZBFKA!sn`RFPb_+NwLSba~`TC_r!DShC;fq^pXNfogQy#s>8BL8%G2W zBiy&Ejp=c-Krbj%I~+_u5yc)43l2Cf!i)Pxr2AyH5D@tbE)y@X4A!Kh$f?7T;^|uC z#z+pWa-+evAfnzN=5^BuV{xF%vxc|fcdy@Oa*KDY-?m|E1t z0b_Z3&IGT^xA%So25`4pX$1}g_uu}$9@A*8i5r4gr;bb*+pFJGrR?``M8=Q>z!Z7(jZj`6bcbrDrGS7RVQOxpQ z6@b&@x7K{{k%{rs{lD*9k8vrWVa;WW#5?dQBBins=SjT_1A6;gMY;*XBF(x|&{WGI zP?qtHnawr%Co)H{qx`$Pc+NI7uW@f{)@|N-jco!H>>^{oNB6)HN%sD{yUDzN>MqPt zVgg#Wzg_?+^xmNL_@FrHzHBn0V60OAq(i7bd=ZR^VWlnj>2##DHqIhzYv10!){#@= z-|$c3d_UuGN@I3sFM|nOe?U_;>kz3lL%tvuOqRtpox5n7MIKQy0?qp~6#s}E*mwpu z=nqi!o!QFyKmKSrW1W&4Fmv)-CjB^+#XyulC(}ys3c)a#qrfXyC0b+eoF2qk`a>MY zbSu6T)q;A_y9E-nWhdSL`Pf^Dmb9aXhHQOakj4#KB&m-Xdi+cXzl`U7wjfrRYp%sK zkMF@ry7tH}9>6pL92}Q7lUl@D0gsS}CjGZ`T#6<6liF z4BT=7frs={YZSpe46?Nm{I@5WW@Y6UhUezoD)(`xjS=)U+B4;K&^&8Ue`#*rboSE& zFE6j&Od*luNTYu6+zOw2)iF)OPvNT#k+U}VZ$a=B7nmdZ=&j$?Oi`|^)Yd%s@(uJg z5X?2#eW!lL`-h9MHWab8E=jOgfK`J?{iRrZScv33aXJ6vZ!I2YZB;V?Qc42Iq^q4x zHKYuq*R6q=nwj-LDZy*EXVXj>S}jQSik@35MO;u{oL?tuX#}3$LKLxBeUxeSTs~`O zaT+M~ZB9M+Jy-ACZK*gNZhSbvhC@5oPhlwF{kZ|1KbgyRV=%`I1GtaHzNFwTm@BnTBYrpRyZp&#pK3})j0 z*h3_@|Ke;Mpr@U!n{UnyX9y&P4G+`yfp!<`YWs!dw}|v_SUI)ldvbw8X#@y~%H=Pk z4~kN8Tr<0g3!r*yKJ`AyVp-)=KN#wsoSAvoG`Pv}E&!3J^UUDY7|ku^IKBV!M^W!$ zROJaY96IZ%IlP?^dQb2kMuz5fL>;VJRyni@y@*I^;&+Geu zJbH==bOI7Za{$Ee_dT^V+QB;5`{^`SkXsAh{C-&iQM#tT>3raIzTtOx<=w7-a#>PSvd)EJ0EkAc4x8pSk5_&! aRY1sWW^;0+)A`g>rh5B}lGE>IS%T?nYp*0My? ziRsb@Iw=kyDQr80?{jh_Iyk(|K+j2I%`^Xy7%0mHK$NUALq%F@Ada_z%&dT6bs%zf z6X_KG^>I!(ncbS5k6{2mO2d7C~YqGtQLPPx5oMK)>WaRtw0!s5`?cU?*5n z|5NSmawzq;>ph2#fVr-| zKPb@g>XC(IaJh8xNH3`8!ZaFIFN-pwL%(J^04>B1Q3df4A*J$&oh=2I$*eqmsHz-) z>nm1)x;5h!D@5o^2M7*W#9X>4RH7txe^n8LL{l~pFccY5*#&qBBeT8W1R)1ToZ+^F zSnAUGtHciV62xO7%kKL9M`bhijrJZ-)v}&+>HiPnrnC2~#10N{M|csGrMzZ=Y}4xM zGE%Bpw;mfCrcJC`Hk_M^0^pGU@82<4kN_DZOpAao$vuUB`|*IDHml6#+>kwbfHC52 z4$N)-x8WbvAiH}YdKda|hzQzw9+>*b7wGPE*h3S+^S7OLU3!s(FOZFJo{by6y6d-zh=d8A;fOeqhME?!GUd3LxUuQFz@UgodVY@2A zr&98?iryh`4XD?9yE2Q4rC!yQiM$E9d-loh1XKVC2XT5-YUSnQN{kF!5Fb6R_96>Vg#YJI4#u#hLUrXkZp+G($Xo4t_=bg@mA{Tms5l8hMl2_nm z4!ci^Bv06%yJ!U234`BQWbxQ4!-~Nv`pTUh9YX!LwoUQ? z&O6>hC}*Jt)a=ZmWCdFG1-v`r_%+`_wf2p0Mtw}L6$0NnrXg5lflwMb+jHH0!28b>SCCy5L?jfy@*YruxtP$5ujS-Y4?6Ux^vg& z;Xx@~%4;+Tu8AqZyA9~3e-jn{EeMTqcVHgS5@>o6PjrCa$qasU^(xo0SeddHxXrp! z_%%6Tv+wP^*>nViA~p3OvV0O3YIho4oJ8uMXg{|os~4d@zWUanOmJY(jI(BxRe@<< z#tg+J2d}?gelZN(CWxY?jo0FPI68#c68t#_@B#T>Ata~X5bcb>(q2JknvYy&{ej;V z78MCFs)HfP~ z6z?52!$W)3B`_LSa-=7SyBpVrilC4rA4OFE{<8cK6ctj1UoP+(Q)8pH@D>HwQ3u{O zw*}1w7+Fw2o%@))T-<-|(E<>qbTgz3LGME@xJVnRHy3_4eIt6;tUJz2-mrpPrBUH8 zY)OM}=XA>ryou@aA)|b+xBDf!hODW)rAd4_)xe+pTjX?Uvxd6Am}nuaE>yQ{Wzp%o zTi@v)tjVWdiJ|Z{Z+_+j9=JV&{-vxH3!OI?Hx0gl-oLwVfe*(v#wq;_pg8{y3SPU@ z%0#%dZbTv-@y;c%TJ(a}N)MwlwaTqEw2iadkmB%nx25MLs(JpbTLHsELEyWM0h_K$-ysI7SvQIyZ zgos}L74UGdiLI92&N`-RW?_``hu3XJB)Bg5=VQh6*nvO3B6A^j(cfPM|3dfENy^q5 zYTpuUMgw_w@k8YL2V3>j$ljlg2~#J|7|{$?njDlaEV{4Rg;cm$NhQl1!?-9tln7jm z&w1t!!c$L}{MA;SaRz|qk4w?s;a;f6CgX~U!M+sh=$GH`0ItPc^|7a$Hu+qZ7N%!k zQ9~P5yjKWkev@P^hqK+yaBGu}t4On)=7A)4_e>8^q%h(JeV#n#D{R2W^cVemgy5cz zsh(8Xxqq#Gp7Dl8S|7S>DYqgxkcT6!K%%F{X6;m zij$)nAM!yG8L+`e71ZOXUV~@Vim{NSoxxG~RWI)3NbAW9UNQ{p{Q{mm#li?BC8e`Z z_StbU$IT;Gb85>}kf#w38lZD({4SeB^lR(=Kb;Vj^4j>OZV2QPM?EAcjZhQ|pe~w^ z{Yr}sr3HB4r~D7_pn5=3zqw8Ogz8!!Dnb=RR%a}AOZJo9190ADDsDq|b_-{qakYI9 zI6uJL==D593&^>0RwjwWo4dO0v?BX&RW7HY$;Jiha=kWUWE(&B9gEaU&aY7l2t_=g2WW@J+>go@P$y zwt{jQGAq#`+(o*U`|43I&hHX-c?h1Zs$2#prlh5L1)dx~p0m?tW@bM9WUF~6SCj&R zhh1))P{l4mjwRKala>Ii0&>7T0aBqNt0v$W?^pOrSgCt>aT4c3c%!gFZI-MzeX@Qw zl(;sAA~v|tV8dEZe*R*wvvVG}1QC`{;Z1YQY|E3~zTWTe@V6RQR#tLdkF@l`bOYdf z>p$$Kbm6=MVUI2erXtEm>E=4}k`g2L@5x1KDVy#uoW@AY!E?tz)YVZJowI#ui5?_b zZ@FaKeXLZrM?fPf28#d&i*kWOV=JqEs6~y@^+6I0!y4+)v=2;6kjDvwKMPE`7xY}T zmtIXM&75NrG9?XSRu~_zaWK3OtvK(E7F0aUMa~)4&haOJ=-<{&`aSA+3cy$XoxQbhh%QH_);dDtn%c zm}xA9*;&oBXw`R0NrHqRvcoj;-Nb9&RavQhlcKB_$gLh)K#?#u|H|j<&4|DA)aW|a zeRCiy6ueE!XWzU68Vh(YqzNq?9XX0$)%h^dkU@+YKtcRY$NeE(@Cb3CZx(|z8U*pT zX>xZ_K=dkp$A53L!MOFPc<&j3qEHgo*?g|ey&1Bx94eNz$zeZLOP0RA!xhlI0&EPh zcF!Zrxa$-YMVg7xkdV7?;ljUh_q)&pOcO#I7PBq~vMxJGdRX2#fN?%x@b%{Peci>b zZA?g2U`g*saL_2K&=X$r7M7GYfIa@PeN|G5;XtR&3*6b4_G-O6(%^?Hg!JU+6Neu) z(izrT48cK!CSUopVZA@8p%QkndsRAn^|7ya|3{zcmGq*^j1j%>nrw+C!HXApKzI66 zVr2~RKT~AEnv-8proy9Zw8;kZCxk1OzWsrLAC8oObbym;fTofq1YQ|SYfp~PLLp43 z|3VVvMf8`z;`DxMTzBFfz5w<(V7zd<0*6-nK|M7#Hvi~2Yb%3{Wx|ssuW0w)iUA=zzM_|B*2r<3KHbKH&}|mwsY&e8YFY{$f{GULq}- zFSghXy#M8J>KA)Gs#mDjaXfDeHej@r5W<}6pku9+ejC4!NEB^{AEit9UkrKCy>9^| z4XY_it%4*wm>$l}|zyWpGUON+nex{sxLHcjo%rD%qba7`?oxxhMy+|?l|4Kk6}{ez zxT$lLFV>bE;yHqEwQIPNE~`N;_$t*XSPhPYNBOoR(8 z$r}hPkMw$#Wc~rxL9$E-M!0kxa9!xjzv;f+6}K6}03M(KEJaT5A*3p+5arW`Zms=f{+2sT9vzXa7o%Mgk0?KHmX=jS0D-27f}Hpg8T&J3wb#xy~@T8LdTk6osLAM|UCb zBgV=F;(SFgAf`dn8C^2tm@2!axmxs_V-BF8X7N$LLkBo_R}pZK(F{o|2#LaB^{%F% z<_radW^()Am)Cjqc{y}R#9RM?h;KyXsUzo5*sT>;1Z8OT~ZAQrj1TY%oYUFTyw`(l-bv5=Fo&QY9zadBQ+ zzS$ZIOwfsf^HZ0c`wUU6z{FTDck>d$@9TS7^=tDovuW?ES5GZy`2hBchN5Ld;Oqxp z(`L4>n_fI(s5Vo&nvR_iNNb8d)wm)7y((cGoPE545MuMUo&i8wzH_=}IPFHImll*a z+k#|!tITuKjQ|DQKod=jO+cH23LeWL<%x%c2iWW-ws5rx1-zZ8W06O)OD_UZSiCOMX;I$^{$)YDG zhnP}@+Q!Gci^Y8JFJH>xF^;k03dwe*Yn+)YUcg0%7~h%DbBnEuV4oXRDv}+1h2FL5 z<3oIL=?^wUF+V>eub`l}4O^hgIF67Ij42Jdlz8h*MmE&`mB9+4cF|3;_IAB z5e89xj5EJwR|Nd}vm*gNaP1fIuz}lDXI!Z$P&;F^Kv;q|y%nyutc30lf~5Mr-}MB6 zh$iM2k(vxdb$0{6+(M&2xTvSePR-1jKF3|^;RhFMl-a15O4<_n3x%#j~ZW(H;O(lY&&%>dV9mK$nKzSb@ zv`5p~5$P%>Ccb?!g`(?Th#fz?fIV^8CYBJ&p3Q1^Yf^%%nQd|9BKTX~-hBNlp?A1v zdpahd17g+bwSMFF&(}1No>&d#Nz}JqLJ#=dg%W?~ z^5UX*3LHWeqggQg?W@I*Q4p1ahN;lP3(?WwCeO%wik zQnBQk{E;tD*@yZh!I3x?VRLh@=yQtK=U|HT%?qu|!D}gD4CaKM+1X-v+g+hA(hz@l z^^K&pFP_gtCc@YJWlOXh3YXLK$Z#R3#{akK;_%I{FKpiU{{(J+k?#~3sM*IyCKw^; zB^o}KYV_FSu}T*vVT!7p-6>pLa(-+mEFvP_I7$-(|<8NDV z4vCUbm2h)=1RNwXF-@@inP&>(Tk$ZAmRH0&=&?N5ySctO_lg}8=Hx5ZLNlQKk;8}% z7Rlu!(y3J-hH@Gxj6-Mq@)`@$y)nAef&%Hn?w4be;w4&{T1TZ2wxad81F+|bYvQ6!b4899u9iLJwRS!iWUrjw11E{fZe zkVs2y)-xd_QOnv2G~5rBa@Uo2ERWBy5Y9}#RiP?j227|Do7!Ec9H4Q~?}-rY(@;gf zl&s=T!1+)9b_+d;NTv3yZaAYO)7%J$_(PC97no2(t24=4ueq_8MI?JeMqpeU7e{$i zxG=8ZrwL(Xi0(9*q4?<>e_T3xI`Fz)G+RMV!G#}FVNnr9`((}1M^JZ8vVZ2KNe)X} z_dsI;6^ek#;GI%jD=G{XzoX}VEXF9G;@JTXaipB0;m?V`T(ItJ6Rd!Ns2b?=DVQ8ehJZCdPc_$5w+q+IzWLG zjE_U2v>7`yrnoJhAFypqTqE3KF$ubyf;6fdVPiZs#0$V1=lw$sF?smw%$Wv=pIN|5 z$g%%v{|6K9PT5Kcy`QN#d8%N}?Dv*47tlTwyW=c4Vym@bwg+6j+^5~zo*zSSRW77$ z>wK>~vB!h(fM8FPgtL0zn5{e!LiKEtCkz+b*#as+n-mzP!@0UkM3JbXvi~fqJ+kuw z75uavaI5UQICKmXJb8DC^2N~itBT$+T7&0hn30;yg9$eW(Vbl~81A;2|%RY9$SrqQIi1^t*{rbB6Gg2bR>hU zL_M@p`;Mt_7HNWfFO)L)WZBQJemF}Ph*o<(f&Re=ZmlA8lz+NVc^UbSAdt5caLRkV z!OH~($cIiwKRgV>*FAhP*9TIRuEaxGI+ktPSeJZO0CV+=gtMD~m!4QaQ$c{v1}hKY zT}TWSNkP8-q*6MjT}~L=E#~AY7FiiI|NeiD?5fz*N%N0aEHIbc?~|6Jz;F0!s2Byq zx_CVGJ9GFLn9b$$L+$h@n_r1mjO&jO3N~=SBxh&eKd#&oIkj|LZa)dNytfU1MV_i^ zZROVDU`Uc%@Dt5^aVch|yu>^f$N83xk!Dr)(X3s}yA) z&3M%AfemfQ&mTx}d{7yK?h~TS@Qf7kq2R+xP(YOnsP%SsI{k^-8Qz`7VXg|hx18DB zgBfH$5*&3IZ4EBVJdD-RKBzPv`zQxpwi;G%a%2^kJJz z=K2xHBQLbG;_lh{beb36TxQ+aln=-H01*!sT+`t7A1SED86gG5kZp$4vh)S)lwzRa zDS);p&f=l#f*5e#_4BI;D@mG<5g*kMZH5E!Pvv`|pcmyEHjXE=>D=BIt3|Q;fB-@c%Ha_I z(E&@m1CHBzu15UF4`e3|)t3pS)HEiNjeBD&pbH8aQ z7MWyW&cI%xB9gL@@lXEV)0G?YPO!GJKj;w!A*h(UE{R!Qg|E^DwD0pR^RcoZ#nqhw zZrWHDKqY21SEni+_))8~9o`@5)v-R=MioUvhG5VJTlE@u=ThD~%ed;U0$p4FL;Ct* zXxKfqn2>t|W{(qGH5>nGKr~CmoGImoN=mP`f|`+X0q#zRS0|I}&2LA-@zK>UdBFXS zz+bEhA=A0zqu85b2`<)Fq}lb9MJmwCcAHR8so>z{CGW}F)#!Tyrcv5w*HviHBo;8> zI%W&^@ealB{zU8)(DXoo9IUR;IXynzuCq5l3M8=zb8wQ8iA34FV0-wIj!7*Lo;(t| z4AdqTJ=ojhU_|bEzJ7wBG=STI8EUqFXm$z37k69Ew5NC#ia~zO+h+l5+rD5Kw4Z-9 zIOaKHY+}+4O+=TFkPv(C#Sz*AB&zLM?iaNwx!kkLKEAai8SBn>Z$m6Vy8?vj6>xKw zXnuBY#DI?)!dyPQB2jXmTjMKnU1msBjR}cX>!qX3D5t?8$W><=qO09RtA0#a{|}<1 z0YLTk?c+#8sWZ`a6Y3sMVBFOuAOsn!@32i5T(GGVJ&qdJozyeWnNR-Qt^EUap9cos zQeE5Kt961v*40a7c!&H88E~g-s>qZ!R}dS8EM{Rjw!K!H z;7*U8e!r5Ea-Rdy=yvZTvc=m6m^_a}p3Xc}Zwnv7ji zqt-pj{a<;@q;F)T$=k7|T)w;3f(~4B^wM>YEy0nk13{I-cWqC8Nc(gWJ0oQM@?rAp z^O;TucNY^)5ND-wRIyenIiB8vykz7haXZq8$D}A8vFS#$hqL#B$-!&f;qnCp=bsX+?f7}&9Y;*UD50PK zKS9DqT(2)3&-9JFM4UYVPUd>OXC{5P`ZBTjcC{I=41;!MhfV!(XKPbavDTB;_KuKwt;k~_1wV4{1?t!NuVHsrsBl$QGS8Wry+MA+Wc9IA+_)V0amYGD0kiC zA;?a*+GK)24Wctal!b)_=$duBYE>?r;U_-U8?d)!X$wT&4#QIC?emfpkm z26gZ~<-iw#3lv{ULt~;IG^ay~R8#v^`Xl(_n?BEv+yT-K$t_~ zkASVU3r!g^Ds_+)s#a55o)g1Sr`5B*zhTk22H`(f6_!JVOPZ(E&bVM56&-!?N$Frf z#rFVH7zHa9MAiQY0bmnB0f)TZ=j1l^?yDv5GDV3vDG_k0UYYcIk2?xt@$8bD@tX?^ z>B>>tfy!czG|VBzxutAOh;i>EE|%T|kGaC1#~|fAB*M)T_R|g5`@;oJc3TXE138vK zG9OU`pkL6dnxgl>3~Z_@hBT+$9`F+d$15$M{L-Cou^~f4L!+ZQq2+rB?hXFt2WpPL zxcvI~sb6h;2rq|Y{#cozCh^J^FoXd#2lj1iU$BxY$X&U>nM$%iy)7$xAPB?{QM$?B z?355y(kV~xtX@5Mv|WNMj*6tUZa8qzs3>|Qr!4(z#HMQf075Q)EeN%7Xn4>DA}adp zj!K|(lIAIYt)u)M@lQyjv$D44aPBsa1jQ8Q|C8}1gShL&sC)hb@{7NK+!{n%dq-4p ziAC_#VbSJQE6S5M_jh_p(X0p4HCS>{AjQvAeX+w8&OMP&Be?0}5F9%P^Bg@xc~_<)u83UTef&m(n>P5lZWM~j)px#%vl4q#-^aZ za)JNpyw6Ff%%iIgAn-Zu8(|sA4RN;5xg6Hm5BdKMCIBL_-KK6h?3gZUZa&NK^mFs{ zoQIN-B1FLQM*#FdSW3-}T7TLlOHSpl=oy+yC;@s+dwtu_Yh|+>lOkK-P)=KuRM#J+b8S#PuAt| zu?yDx%8IG#hLu-J3K|3}AeET(4USg|j3>up4)yiR^!TYLU2sq)QxW$&Z0Pgoe+5!=XQ~`!KQOvkpo>tR+6$+9o%)!NdNU-ev zjUu=9u=OM`n)RO-!1UK>jx>BP)+X=i#OS;}wDZ+?+4!j0{xAH55DVwc*J>~r@3pk0 z2f{VnddmXzjEdDgW^vR2j-Mwy?<{2Q@RkZjFZ|$R)J>jV^CH%vK05=YPO`BD_vY-ucvX1KxHABYGZms1EaEx2Cy z8eIm%^<|p%$Ih=7;5Uao+PNz$2mpU|C1POH_Y4Yk8_SB6RU_WJqe+MYSdZGRIach^ z54E00T4gTR`$*m9L`j1wL6S8I*R;e_wh;nBS0dReYq*`kG23*owg%v4EOb<$>1}9& zdT_+SyoqfD@BH0Z7ZfUOeoasM6Ci4sthfh?mTc`f;567M6Pu$}12BZrCx;6gy>FcM zBcH$2giB6boSapT7GLioo)cccSuoY zueyyLwz-ZRVg}*Cd%;`9I><0?4sY9+evdtlkSSM9u^@P$L(^@;u$4@jgt+}D*j>As z54UcS<`RspetDf_suPhotD6tOlR;hvv4?Og-B8hDS1*uezD(wPFi@OH>bYno$~P&f zyH`^)OD|hPTpYEwWJBj!6Mt!lFv)!NI{g?+<2YbySz{ zkB>nh?_#)qm^o=GhK~oye^~-pQ_h2xF^EjH76tzW)5A(wsKv)G+nzfba&I91WN{}S zRzH#f2mV(}h^Wrw)oa#_^)y*lmqx9VDO z@7;>)x&GhEvd4PdkH{Z=y+CdWuk=1s_^TXnq+Ze>qtdf2qp$uzbL~BB6Xww<$};74 z{u|y6lIL_UvT)cyV~aj(%=u~dFtrHK;`-M``K{%Q8*DnGy z#s%7Gpux9%uO5~z=uyKQr1N}lZ}}kWdLkW;R}D&u($XQlwYTnu{^*NNsj{Kfnw)EF zTbVm-YigXYHUw^Y0r60knOaIm23atdPV7m0!sHip2!tlVKfkVY&3s2Nd{BLY0kbil ziS@nV(cY3WLF^W6L4@a~x|DlP*ivrEEVS5rgcCfu+-k5uQC5AjX6J~nAGG~j~ zvhx45_tk$mJ;> zW;cTVSyn64SpT!PcRD*Wk!pARP!nssj04=xL?p@U3lAgLn)OysPXYI0EmgVHs~tvX zLSN0|Cs0jtS4VXKy6YcPrG9%S4n#0eoB)9E(Cn@#e~&A8TvF=>NOqG$CB|}_sG)^b z@a3IWhbQ>k$!T|F$QPJ00KcUg-v`D ygi&0tB z+!^_&%z z77v47o|js2S$t^_7wc!*@Sd29kB{CfwV1CHBu8}h^`S?0-{502=rrzRvv%V^;9NQ@ z;O>wTv1gsahZM2-prl5x-h%Mnn(+tuSE#2pW~2S5O%>D&-&THjP(s30Gu%LOO{lv$ zy#L}5sv%sX9UY%cx3G2#m`ffeDY+n(|HF^)p{7ZYkih!GR5~Mt zL6nW8+kgVNIXH6FA19<`S0} zAR@s{@4A_^*^PPj(%U!hb9CrB*3`ya4EVC29hx4{e_ubsMxz9O)xneJhJu^~fyQIl zxG)i~J z3oxL-Fb>%%(ZLiod3X7|j22!R+$BT(DlL^x#a4D69+vJgRs`_TR~1VMo#Xp_xxajt zUL-HxDAnS8nuS}kwDT$Xi-#FdoIeM4BJtsRSK=qr#w5|yIFR@tthwOrR+b=L<;V{` zaF|V}%Z<+n9=q$?j9;r#tfxv#G4QM`Cd7spfSR;+`%PN1rQve7tAQ_8WPRn8!;5G4 zOC<%36F4;~VU?mkfJu+L`sWXC^CRpTBB8)bLjrX%h3V;0Rp>Ee;@uUsa3CL59+Z`3 zedZw!khD!b$d66t=zD1oE$LD62pcJyE{67CELn8}1rRt8`&-z!bzcw-BEp1qe>r+( z3IE)E_BtvYbEE(H=+;0&2a>t8V>fUPSS0oPVuUlFnuTE&&e*N|U@6BlUmsK(tZ00s zH{`X?f9euWs5T#gAB15PVGd!EGH)&V_0$i}R2m7?e|>6@)NgY^ul`S*S@|BhKI}2| zO+1K^;N3VhxZsXlvAt~6wzBf8!{#!?zo8)D!QGdBKn=fn7q@mdKlrQ>%h-!Aiz$Bf@wtL$xi|&!li8{ht7+Ih+e*jpcHyA zW}vv#AN^5348TbQ0n-QMguPo%PZ08j<91;SxRVCK?tGj!RkWN=l`RS9eiA3K;N^$F zw`wEq9LSC=o;z*3pJAkebY=C9hvPz~{|5Tv*ytUy1LbRW`}8$CeM*Bcu2K3V1O1ba z?z0k0BI9!4!tWAsCr&e5dfZU&(;H1InUVVXdOJ^_y9)Qz-aQ}Z#ky^<8LuiT*Ke6s z5NOhk1dz7@4e}{r&T4Iv{0d6H%Di^Mg_PNb_cpMjQZ**EhI!yX3c}phhmm3%1j?~7 zHVq#n6rlK4vZ^OprMo8uMmE>R>q{(CbTmu4RMf3$Zor&zAxnRtL1s6aJ_bC6zhzX$ z<#PM9h~L-znQGp0xf43c6aAT)tg=Vcq4bXhQ+5o*h|~_=oCP)%gXi8c&}Z2Eiv@i< z@lXD)KtWFz6qLK17P$1Ax3?>r@%_uhGBy48&SC!1*1{7nX2u%tA200JooweXoGgAg z`M&t9`A;@6$oOV>8$`t9DIr89t*%thGUsJ-T=)= z>o+ksmtS5RFQ1YpzC{mrG?87II4;LJ=PR2mX$vn=i3{od zyzLTHsm8!b0c&{=d=s!bCO>Q>-$oZDgn|%@hjypiu{uS%W2beEe8CR_0m-68F=U|H zcSI$jz;}5ynto)jzb${!Z>=gjDrpJPJ~pM^mgpWnhZg}i6nto+n3K0dIg&8I>N8DW z7pj!*;jIOo2S@{VMANxKTIV*OOif|E_0DGwA0&$C)&+K#JIm(40jfDl0xV`zFAbpI z?$~is;%Ta?THWj&z6$ItA0d4 zLeYavm6D3ZMC=ah}?Q@lHbL9mW|xae6lZjGZ4%W&C3@&-1B9>h1J_2N>r5_m_OB* zQYf>TA+OaPsugLX#QSW%DlPPR-hki~a{0Z?vE;G7OWc#J3Q9dE+!YW+Z$ceP(y|jU zoKpKO^qqI!Ls|sUd7W$T2pNY8fdxZYcyO_!>Jf-30G`^>kLc=xb@BZ1-{Skjp32(+ zoMdFQTfjc^gh$5#YoRqKt~ez#+7FaL+Y>;(XmInXPJFD5xYqlnnH0Gb&o&3J%>P8% zgUHF0Cq0y2LV69+yo!#v)_tZe5(djjH3q-vah4(mve}YAE+;UqOb_VVZP&Qo1X?;4 zoIswaa9|e~odA5wm#&L2`(LUtlEsPDhB8Owl?H8y+a(14SS?`lr}f8%(AakCT;4wx z2pdaMQ&YQ8_zZ1^3t@ZXc69_B)#x(wMv+VbspxV^ffXzo_vV-~#|3NSAUsGtYi}0N z4xjdv6^G3Lfws)6*s%PWhm7LVz6vn>J4)F%uB%o*$VeOZI0g3G3*bOk4Yv`~dJ>v3 zQ#x8&{Y^E<==f;e7$^bw;LK`_d_br10FKv>1Kc|rlytNdUv7f~YJtCEZW3fn0eP0J zmeX=cpDKezX0|bj@soH*+$G4{a5p&H+M3zDU0jH-iortJruePW5k8TatFx8#L@_Gg zoZ~g}@mJle`tF8C!H3^TX929m1TU)>FZRM$MM(T`FNzaz{jafBIOVEp(?Ba;Q%6x$ zIBy6bw8RB$tbkM;KTz%UvBl)+w@?;|;S19+8C^>tn=Y>P!=m0ot-x=EG%jAnl zghr{)lD197%9yk~{vo$kNV%U%ZCH0JsV*hIi6GgXY1+ICyNs~>Uuaz+nY7V-!+XvKB<e44)LNz*@6Coky?eIKz!iwXrk9A5a>xgf} zl!gX19-^P;1DPC`>T>h+ea)d<6$(E|gaWMeemf{RJHmK`eJ__-V89n>xzutDd=KhueN zCTns=vRRE^%P~2yek|ZE7$`KX%*@xwjj#gcjcDAsztH@u{@pn_PoPtHoEe>y+|i+# zaeHp#h0LtH*H`o6mx7Fu+kLDww#`@%eXsjG|ErClRrlRr9l^V!?d>Y%ZrhgmbuKJB znuL%63Lk_Hmkn!T*Y`Jw0IgG7uJH|#Dd`e2f|mGWZfk?#LMSOwdt6p42U{qCXQZaw zqwtZprkIp&H` zYzPaK)0`cpOD$G7*xltWciooq|(I_U)dTTe?D1L6EmqJO_Xb) zlU3HuHA|POBMPynH>vWA*toD$sNLOgi_fgh^Vb`krxljZ5+Z)Su084~kwZjP*+26V z&_HS*_psqE)*+$uTd$lBks^h2_nI%M=UIoFhxZlzF2!iAXI~vU_K~KeH!d3NDZwK2 z?QO=?#nWTnSk?>`nYCTQ>;)Kjt(FMfJ??ST?FS z(15|K(`l3a!99alFGYn1{B)S!Lz2Z1u%r*(XwyQ=k;lyyTRK8lj%HiA`vj7aJ9M_Q z@~_Eg`LAp8MVfmF=y5-9pB1@ukY68l4`0J>^4T|BM6e)=sVx}SQ>$C3gT%*U<@rO~ z3vN`W%x+7leE@Vj+jJX7M<`S*`U#RJ%=4P`tzzBi)1mzJZjJj$QkW#A;$!w1xQ&sdUmVB z4ZvxxG~(_v!NT5uA9+~Jkcz?RGTGsH6){v<8>C*}` z_D{jZn|ATZ2%)%C2EO-O7Uo%z!$nRAM0VB^9P=&09wIF>-7>TCQ6$d0jE9lLqO954 z)XkYY5ia$U6<^M}T6R_Nf_n0OYGmXf%;alB0qU_wPUlGSQ{BDbWN-~+k=UPuRXT|Y6;_2+mm&kn*}T@&8&XNl@) zfoKi=`^)~7CuB*C)qXr zqQg`56BN3y5|X%*-ak!m*Vo`jLR+6G;U0bJwK4rI(ICHsts&IZp4R7k6&3qXsY|f0Em#DYN#z4_$zC9zOwa@ZkH)dhu33Nd*HY#9 zLh^TH3WNLn@>GuDOY!O#(|+8(-9^Luh6=q03hw5wiigEVtc8lN$Vz=6d|3}CeGlXN zncY4F(pd&~rL|;GqV#H6xY89x!&Sq%LI5Xp6P4m8iJh+)u|ZhBXP zJ91NHA6$#-K@WQq2Oj!$OEU4?HIa^Hq2_!adHibTBRMzQiE7B!{&7Z)l)%eshz=JY zG;_^@n$1}7Dh0b{qc|gGm+afCS8XL$gN{Dxk!;x=+qX^Opjze>h@e?(J3Rr&+L$R+ z*+i=+#x3E%@U#jbGIbW2^V@s`MAg^V=Xe)=TNl?=`>p164pC&v>Qa_}Ml2p)j=>{~ zy3^w|reChdNp4F&a}IY_mvu#iNH!Uz48AYK@^hV%LHen4G8x{Lr5v^wx0uli%}-mm zU!!F(vQd08N}W@@n^P~dvwM%@JwZQMeN%_Yg0Kja56z{z6Pdy|YrU+bkXv1bOx^i%U5hbE+c&!vZ0%zhDC^IQTrf^fRhBo;tg~Mp1-D+bdHwL|q*BU^e zf&4-j*)aD$+L$S`Usxz5Mbs__S$_R zi?ZU6J*@FC589i6Sy|p#x^X7;9oZ1Spb16CiP4q7DZdS}S9PBj#V0Rm=r$jAbieUI zpvtALV)#n zYLUx#*GI-#)JB*nswW{iZ`XuPC4SM;F+>&NKRx6n=q)n! zZ+1KCAaS@IM3W%ECd$E1MJdU6aM2DIVO`Fat-2$s%teyE-II5R^ab;(gYgf<+p3{LO5=1Etvm7M7xQ1#i&SOxom4K< z&=QV%75`@Ne(Y5&@N9acD=fsb0e0vsKgQHP#I*icbi!_=dE}omA}O~zZ`EYex~^As zIyT|j3nxZRzUfPgM6`@G9ZsAT%UNIS-yH^|Zn%m~^~*P3h)J0oBuOirnhrP=ZX4@K zP5t2eZsEkrwt5M~YSssFs`-{po_J|$_d??n?so5gEysS}%ksD1GS)$jDVJk`hrl~( zcrw@YyucHfJR8pRw6YtAaZHb8=^#P@dlfcmsj(xdn~8;$ok2vaBp01RGC^jyQ2{DT zTI#Z0!bn8d1!ChXFb7+g^lvXh?0ZWNPHV=9sCY^E0lPqr5T2k$60KUi2xUg50aEEM z+HdIht0GU9^kM4|=g2ZGvS3xS(bB4u{mnULvC5b2K|%Chd`111)RgjV*PS+~AVMER zem=XYCR@Q$nmyG2c+NNmfVj|8L1_&xDVgLAdjCdD!Ne)+E zv$%BzYve4)99;YT4VmJTIIiN@f!;?_*G^V8@_n0`yu`=c*E}oc%*vgA_Bayp0W}{K zH|5UlktHQpD{X5LitEs4=^VVdxHytE`9g+Ww~WO zEXYG3&_yCR>RMmm+|u)Jr>-Yr&KHJ7sVU9lI)Xj4TXJn~`gcahATrM&Pt>n_3m)7c z5tVp4Pvf=GNYYkz_^XQFW22kj57atEN7}lGe3L~;#3RXljq+un0!?bA0Z9&Edy9_5c%G1V;Q@zNaDSMF zoDZ2j5EEp>!=OZA#@py*zJ6e?Ictl4(sZ5?eu?2}(NWZ*dg-VW6YEd=ppMZUk6({Y z*vhK4q}&fz?We!d)?4M*#^OIdI9`YLzGXqT>=c<7-=iCKWaKMGag)CBA!HiqZ?ilB zDBcG{nNr6;$6oM9M@QGMShL_i#+x0e@UQ*oE0I!(?QpcwdDi0XtaczB2&Ju zR)KYs>~AyI)&9sYi05{46cd&NRHVz)iuYwS_U-4}!8g!0jUHb7yH?TCbY|a?X`VSq z-P$OhWd;J#gE1100*`5NihZH3HM1|ph3BVGA>C;Ntj@l<(1*$2glW>6_kI+e*-o(#xFfQ%@w&6RmqWb+Pyg# zm4a>>8w2OOD|-p?$6yvTeSa{RZrD{>_s?Lu4YL5j`Ym|5M%J_?l!QeKnb~vhq~>r``wRx$kyrfs^yB1u};c_)NX9 zZoAiOFbc?%29W2@by+W)$&+_{HAg@YZ^kf9mf0EFdei{<-GHSe&vp_7j0ppf9L}0F z`Gp>?wzqtzVKhY)Fbn)tFwA8+cr#$4&6Uxz{LH#KgZgJ+{mFyncUNV_fSc~T;}o`I z7GGV)@Ksr{;S+ywGiLLD$f&!lCVe{gav?#YWJTPo_E6}{UxHYWB>2-6637}VuM;w~ zDU1Ny97jGV2d@H3n8e2_LiJcsSoc+F%;dO@ANB?{zpw!A`B!;tX3j(q1R~IPp=Ys* zE!#{cN3Qg!NS;QPge86j-gb{p4*qDimi{X7Lg)A^KXiI}x`3Ei1pQjQ)9MdU&G@WQ z0miuvAG}t-k;@8;9k5GNj_ok7qKhqHBz@i8-3{nPPs=NRTth+3^T1hKPp>_AG9#C6 zrhJh0d;t0fU;yg@*8~y?c;b?VOS$P|AZ4pVs!BL9t;H)ahi9w zttOdSJq8dtXU!U@Kdl8;5jNXrOU4KnihF2Ev|2d!91m}5D&SPp%GNe-S+mA*PRZ7q z51>@)YijBmv#2g#`Ij`gLIF<>k@AV@0tr9guuUV@DQopGZQ088uM!X!l zJm8JpxAo1o-inQz9#!L_XLT;Yn}uS*XP3JWn?fWuzt>4_Re%JJ{fCS%8%c98hkQj`T^fqaLA+7rfNB< z%CoDCIIRSRI;WIEK~(;6(0Qk1BVcaxPa%2evF1KWR`q=CB2d}1PfkH0$a)Z)YklR4Jz`F=W zm2ExIXaSeSXgdqciE~Fu<-N3>FPo?MgGL3wl;e~gL)c|m@~X=7a3`(_T+r$tc~$Pz z&I(=sz4`Z8nZ$B|APNJH0BYHxich5{b8R5Be!%Tw-s*5M=gLHl*mJ4mQ^r?&AM&a< zTdfD{{t|_IueMBocSt6}bFm(zM4mN`B(S_KZ!UirK6trImS+Zx=jBH(n8fimnUc8X z3xml^G_-cVjK;w30#WUzzs$QXmPn8Pr}Jw`3WZc&PW&FXU1NLH?}rJk$78zqgLztp z?#$1bexx#p?{(!F2tBo6w)taIxB?70|B*jTL}`zsc|(N?heuqa&1$}u93Sompmg)g zh&o36%!{BYOr>PhdOoZrFeZYYzsKS6PZEbo;`7<3gZEP*V1_E;BhQGYu3@BIfcZ{L zcwnwW_;f9H*(K)(OgpdE6-?%;JdmqWeQgbvS_)PrFswkWnA5+C*~vEm;eTf>63gML zL}_Fh+XRC}JIPhg4eco1)_-`9ef&Wer!NL{xxN{C> zpB%apP3b23O*v21`_1Vt!s#pNzTPF!f!->(llHQTE?E9!QWC!j$@Mk>+;G;UlK@4O z<|>V76HUor8s|$Ue!vDU3tExUMaxN(^PZm07v@GS4vZZe{bROj<^EXR1w8{65ZJQ* zt3MJPAsmW)Bx%P(cfLNX@YvrB{sQCwTC&_f=(_NW5D%FydRF}fbkSr9mH*j;mIPr9 z4`6ooe7t3N692Q=dvL8hy_@bls?Vk-#<>2x1QKb4y=z-P=Vno~j9~HQ6sOndw)|R> zJ3SK6KghUVZ6j-eNhHHa9+I0$uKNk1KFt25_+&P~N9;On72Oo|Da5E=St8?C-?U3Z z#GEEu)DH`$M^H?2e>iDS&QhnUi@96~Jpr_Y^_n%$vXubgM1T)+)?Q!E4wX-5Q?$_{ zV#$fz(Qa9G5ik83*T(k_D07#Ds6Dz=I-Cf{uZ zC&Ugbc_v0obzZN2eI+4H&<4JIIci{s|Mbx9K`AP+U$O<}Y_8dT)38?0uOFL%`e(we zlnSXuKq(P~Dlcofw+~a)YsU^~l`gKtv1OO=pind>B*?tN=r}n`a%VIVwH{Ba)N$|f zRDjfxar%IwpWvTiAV_)UV5Ok`Z%xAGFc_426%4}Zq|9ykpsK4MnHRis>w7cgCuug| zd-n=lU2R)RH-Qi{>S0%RC9M~rGHJu!UYX+z_4;=dj06DQVViW_#k5-#CJ-Ky42j1- z4sJFQTLY#vZ0BRR%%C|aDrcP#HFRm8$LKDTf8Qqf)~}_6(27BOx1C`NP$5kUHEk-E zn6X{>AL_Wz9PE9WCnNZquCNRe=;bXd(HYwj73p!>;e7*0bcQ}xnR5-p;xWDCoJSK>P8qg1WznnksAzq(1UU4+sPFWu`Q^%Y>6zX-g{wj%C3nLbqam$$Nk za9{YBhdx~Bd7To$R?^RnatSeGDNRUx>EAw?hVZP9IG2LbpTHxkJc^b%ny}^TZMfFt zZhivH=Dp}wjS=U98v6)j05I^`BUkiNoe{9fX`S;=O;*80K=6SL3QSS7`O@%&b@D?X zCM@vIX+|A5R9;_pzA#uHsARReWECL6?>zMgjNjvb{zNiwokbho)?^?z^o|uuI=l~R zMOj&NFw!~3Fq~U72)qL4uS%y>z{@HC9~eL}KSMXcU0wMHnrFyd1ob}rQ*R0_;w zMnvEX3;!u7HMi%}BhN<-u%e3(MpDKS&c}fa$bg3;6nhW&@T`FO%}NWVn{zNS_3s0( zRaKeEUTdZ!Jo8MCW-TYL1h+-=# z;YO7-Uz+#ZcxaY_OthQDfCwUdkfM4 zS|lWw6-+?&>%D>-2|8BWP7+6Jn zl3RP*vm=w;T7YFa{r=wVv$#RQIU@#>bo3{R`p)oDEgE?-RuGkX)}*7dLkS?u6h1}| zWr>0>mul^ZJ^W8kQPA^;5ZHO(T|i0Nl-U9GNL=`@f@js z6B+Y@^SL| z6FP6wEWZk_{p||j0s`yrKM+3(g>%*fh6oBU!M}|P`-b!OT9zp7-|p%}U`PKpVwV4Y w%fDm=fpB4Y=5PD|FF9X{|96(l4fr@u?VVYXFq_r)*soK3pmsk`*67Xu0qSuP&j0`b literal 0 HcmV?d00001 diff --git a/docs/docsgen/source/_static/onnx-horizontal-white.png b/docs/docsgen/source/_static/onnx-horizontal-white.png new file mode 100644 index 0000000000000000000000000000000000000000..13cc7a4b67b234a9fda167478bd6ef0836601746 GIT binary patch literal 46498 zcmeEugKyfp7{yAk=bfOyC>zN94l5 ze>jc`dM+RkH^c2e6wg8lcMymUr1(@y%QItp{z{utaT04O?PUAL>Y-#@`!jT2!grA% zW%M4x=W*@#f}GI9vF{L)KjeMN%Zl1oJf$o8cv-2WS$pD240*a;xo=j=;3t~`2p zK2DbM@~}AyyqF)QE-?wA4*78&*@pKAf9;ak1s%NnQgA5bM^2dT1=zpA^e<4`+piW! z?t<<}5|dsmXpnPt?_JSMAmo%JF#f!3%B!Uiay=7SpH?O68z#h7c$I`$L#Y0F3fd4S zLF8nosZD$JzJT|*Lw*=QpdXHi|33CsrNcij#tHFLWV5r|Pxgz_$^HB2&TC1EqXEWF zQzlQIq+3P4{qSU>Gn}7E>x{b zG3*8W{xIbq!htkI3L)M*y60ap5dY?a0CSD879Oc~?`76!{{33KK$>D4TJ`fpIhp@| z-xk@?x^AVGOBVV&(EBMJ{-%~s7#x;_Xj1#Tdy;K{l}Z08Qt0|E#F6^%*zXOr68^%Z zYT2!RP7ZlyGj5lb#9A#z=LUZV+*byO@rUy4Dx;5ahmHUJdXp{z(~V-2OjiN_&sRC7 zs8BT)leqw#(!Wx;FAcownyE4KYe8J$-#}9|bVQDd@)zqVp9Ug&fj_SYrr;DvWvXaE zuohbC{!8K|BQTH1Gd91Yg*xefB$mQ%NrL{yUf; zh`(e?5%Qz`Lihf^`AkOu62=#7g**#w%m1A(ggycDct(lM@Az6<@gIo|CV)h$q<-6f z=b9h}_KmtI>52D^xAMP%B~iauz`EmogDL;-Th~@+q@@s#?`s zeA>9pB*jW!cNzM3X99HqIU1#zPEJ9x4mh*+0R1yoQQ(sJHwPUhz)LS{r}Qj|*ot7M z)m=i30o8v?^(4G~3J5DF*mBl_e1~4z`s}~bQ3b#%vR$0isDYPrdn=j!EQ==e`! z4YWVI=nsf*jZ(I~e05(-lVpdrkl*NB@!S6n2}A(|_fsG@YhF6rg7~SZG8^?TR#(P? zlK9^VZeL32Vy1B3Hts*9;ntxj3>R!pG6eeKIxtJ z9cC6LsbHx8QN=H^bol*_&96!rBbWpEetB)+0~1w}vX|ODE|?4*DPk+!Da`&j$ElA= zTblLSPhzVrYXUE4!w z_RY0?sHX+Hg_jeM(;7i(Y+m83RPFo2c|_^f&_7I602WHH9H;yv_p10QL)KkDG?2Mp?qXzccOcK3&3b;Q}(nbI@EZx>CZ|E@tSLXyJO;eK+0 zpwXGtXhg|0<`xBymO8P<*4-V^KVtJF4dUNe06e7Ki{7i@B*YmaKOrJPU*hZXr;pU< zL4UYedVvZZf!%H-YW&%rG_&h71(=GZ6U`r`2=oCE=#i$AIfW5!U-|G3dOCA~PDGmq zq*z7*`wxjZ!1pzep$jj(yqx0JfTTW{37{!XB1okPdyHM0EWf4lXB^E@c7#9ZrQ17j z%p|%r^Lu@y5w|$%T4S657X8og-C}79xC2#gRYuTzuYr$L+Ed1| zOIX&-1(Ps6IQ^e}4m1J0-o=n6q~*a)a&1_@m2!fu&kqN&MP+oaocPf3nPU&gkAh*I zd;cLCI03NJl9uAW$9Iiai~Vh`C&1t2oU$A|F#e0!n!X?<{_wA(Rl2S{!cG`aS?AS{ z+V8cyJo!|ZUAa5gGT3g|n>`km%2n{|u|u96(vGr?>JM=dfVij}7yN^0D={6QJYu2a z#`b$dPJd)+x&){jwQ~8;)PsOS?aLuqQKbMI0p*Kfr9Fu`C50U?cl5t*^!d?jJgo%P zy)WfBWxj)#qBo7iG&bzjSG7kCIxcTY`zh{x8E3g!PiC+831|FXsCvcI~VrIX1f(=r5jCfEZF1v{S~gn!0hDi5f9v3#p> zo^)_B_z>>==_oP##fSMqdELYG6B@yy(xS%fvy2+HA*MesT*l)<6XZ3oHdqg9#C+*u zQXYqnR;JLPG(V04ZRIi${AZZtBCtU;O$KgR-FqL07sTUXkQSAV1!4XD-ID?%mOr#O z0XufA%%{a^z#qM-co+Dsd0SQhF0}ABE;t7q$7F4yhIjTEx!?=Ec z%#wh)^DT1ZhXh;UNvGBD+m(zziYqw`7_qUmoc`Xgy`@kL&lZ5P+=CfRV@`3@ZF#jr z1IQ24eR*>M@C^1J#Tem3h3>->zxHmpKz_LX80}rd_AiOUlo*_KHU|9l%SwM?#)$vR zS}y<=W&_v5&U^B4lb)$c@$J_`;j31&bhVA>PuRh&Y^TSDTv{qrvLiM&KOB;g%)$PDM7eK*3UzE# z*SbHTC-p-{%g`s%t*u1&{fxh~GBe@R0lhWeVm&x|mad%ve-xc{>9F6Z2vYFRyWVni z_`6e~PYHH`X)B1aZLR|rQ)r9^En&yxuMec}!NzgTh47@{y=A16M3D4ur-1FcKdif1s!Pc9pu!g<4DDx@^1!yEJ z6MMJqbKyO4fK0--`KG8XACVPM*OA(!`MooI62y{jK4~ZI-owzJ8pUjdLN-q#bTOYC zjWb?vlV1KDTBh1Z1j*tjBVIXmt%;$5#4M-+5Sci$-Mz>gpNq)F>CqKm|eeB+b_*2Rs!PmB}-#Ba6k_$iMTi{i}bC4|)-~~B))Yv(V!dCIo%hRx?xyeXz z?X`=-^I$%-UbVbZzRgFPIdX@No7a3~(Yt9}oHRydpZkMHJEd0_qOxJ%GPh!f?~gl7 zICdJp5lbe61*^R=y$tf*Ij?fg{eFy2 zWOSR7SqsIfP5O7_R-^;^Z8vv+hy`907Ji^*jvjd3(vt2&BetcvWJ_HuD*8>kC2(*y z{sL}zqE_386#Y3Lz8OPG8a-3;)ZvZS^vt^!v1pq(UjFg!<09G_&NxTnwpkx+w*m1* zUJt^R{>@=RKvBk2d~;pG4X^th^QZ5~J>e2>PIEH|?=0Hs4h)!|tg~x=-Itq4a5K2R zDDvFf)s5@{S%QmJ7&@rGrs3X`^oDv4H*-mR?p-*D|KQG!&AQA&{b`l&kl_PLE7jCw z#DR z^`jEKuJ#A2phOAlCyI}(5`#H{2L0mF-zxj{R`OA0OFR@y-x9uoI+1|N_^Q0?8}5m5 zc-@&~(Zxd%pcnX3zQa3{x6&9;mw>+Zp!>Kk7FQs?DXJ#gBHO)Qc>~HV0aMM7q=5CPkhczdQrVlKu=Ku^ ziL35&P>Ppq>cL}5P=d8SGO!wG`eC$>e8BwCe?TkNNd?G9Sr@bR^+vf#JE2I@58q|e zaF%(fX`A)t44~zhYtvf7rYfwkMZvQ(qR{XUw&&i7E=JB@1h1axPznd3Kqspt>^IhQ zyyEUDm~W(^%rt!MrPFoGd9`#njL0u1ypv1n^T;VXAH%|+k~G+=@VOAY*)#5o9vdQ- z7?foG@!pf!*`?3yJVI4@?(8|3!;hnU3YG1gl}Tg$7xz=A(vTP|O{MkT<_=@l#N zPYl>M+t)MQ4b``vOrlNcb~bj*;3nCOnYU?~TZ#qohuhvana-F1#hD`>MNqYPNqkme zW1V2Uw)IZ9c~^^_m;4+zU_SbE8C>3^ch!6(WW3Hy+J+Li8%xHXvs3CnsiccHF{Wh} zY|+H=y)Cp%0ayVNJeh2R#<#}0c33FFVw<(4EMfa$bJQpRTqWb=?4G7C@Almsn7IPy z-N(Vx^Qyc}Np5gfb_WX+MfZ}w#y;u~&bPbr81OY;g*e=J50BZ^Ih2pqXa z@L`pSdngUoPQC$TUNR%>w95VF9RvB#mifl}FR4T)PVn@o8i$p|ddF~EMZxaI^VINF zTXV7KlQ}Rp=;v(oqh!*IIX7bSZ^ZQEm}N;u11{3g>IZ|bz69Wk{Hpa-Ig){9SAlc+ zH;f~ZA-ciOI0l-y-zx1sWNjbV{w!jxxHSV7{Z+SnW9^lfFj^8uS$mi38FwY6TniC; z?!61AnroU*bxL>gGi8NssE`p9z#8o3c9}PL%%|)L3R{lIWZY=pK@hxYvC99AaB!Q8HQZ8OpER#m4rT+8b zQY0^~YuV?TVwd(!$S1kB6aP-ROw#dwk2jJO_qTancuw>5Vo|E;R7(W|fZu?O)#74Dc<-LRS`!_kD+LHh-nVF!I=^!)mQcZ>ZDrejP$ z%F!m#NP1DzqvZIJ+Cvk!@OhC>%4i|Rs`K?fXhnn~>vlDaCXU_!$Wm}VITg1#=&_$q zpGl7@f1;kMcxSSW^rPbpeC^Vlo_~sLi}{eYrLTFAYH5%f6EpA$;P47tNvJl97wcDQ zPsGcFlsT6*Ybc(yILy(ajG#R?`LHEfiManVLF3%-MQ*BzxDGv6ci`5mAET+@FIs1Wpz|gB z=tH7jjXPuZzRcRn`8sv*4WGs33?dMxu>(3eD9io56ITJ(%p;i#xdS5(zb;4ZZe4T) zd2nXqSwFO`+dW{$O`|Nc@!k#bVkr^N7v=YeHZuLAUv*rTx=0*t1c)x1LO(+$YIxmj zf}awPt}Q^<=qi8Y z@OVRst9D)y7cv_!x2biYhIk$P;?w$bAQ(J<+(OESLSAzAylj;ext}n9x@}y(fvz`5 zmZSv+S(lJ(2VN=mZq+HMD#v5xO`Y#byWmZ`!Iq; z_M6}4V6-zE$$t+)C~LQdZqKkfj)!wQvwy=W7`61_V$_?pop_%&xbzD!bWDxi()r4qjF4yb4vL_Jp*xI)`MR1werb~J0oK6+}YI7Y`}F+Sb+qM3m`H3X0wR* z$~C1JlR}lOb9y_?Ws($r{C!jv?`7{5;ej#lqKb-H)J#~El8B|D+_ZRBCI;27`F`D! z_y0x31I`)~rd@Zm`iU1vL*9LS&_YLX>EU}WKnm_9z^bq<4};u*9H;7a96<5SuoB$U zNH@~wO1r4?cHqh;nJi$Uv1h)t7XHd(ah<&RcjyzN3+0zge4L!FXj)hxAHN7@e{Ume z9gwjfI7^_y^W^~-8(E6A*_Xb>lwA8|{&4A&5Rv}@^R5sa8ZohB{RvGi%$f|ni}KJJ z6faU$;D)Tf!1AntGEv4jSl?vf6^sfsC01B+q0iC0%{Q{#5$b}Q6J^I1nRQ6z_BKMX z&U&S-;v($W6H=Ca-<3mVd?xrnB_kijk*U(hateeP?P(SoYqQ8-Equu@)h@H18 z>Q~K33R0~(WSZ4{N>D6*QWWI+-7RAN1<2Z!MwSH9P80*I_K zolUIb&E>6ACt+|$hG{;l3-!AE)hO!hl@@m4rYtu!LtUl#3VVv=6K7#WKTcrqPVcGg zJ(!U$-pS)syql|EZ~7}$W7#}trs;i84!WIQZ+h%}i_&T66)bG8{TO>_IOU=Vx-%eU zXk+LA+7*gz%0-}YI^>N+nNo;7D<8tYr;vX)?(to8d%*-@ZR*%m@@|YD)gurqwi)NF z&nHLTKZ-wo_QNYL|MDmh964+pCp$YYj3p9*qV`rj#)RM_hn^v4=)mr^r2)Acqg(I31{5pb7qedQ zmTxft&G5G^yh}%GT(p4ia(?>o_7E&)tKT;_FT1$jkB%t$RuYEDrxm=vtwA4)oQ|fO zNe_;nyp%5e>f-~!q0j2Q$SjAXIU_(xlF>uRrbwJ`ULe@y(l;e>*S$Kjpq?|@+@4;) zm=#wmVmz@8WyAD2bF`77`R64&%`SH&xz^J+T`D5rP@&x#yX)Y9I`yRz{MzRnJ0wB; z*6we?ecp8)l$CC1P(G&TK6VQ!wc{8ty=QT%egp=(`U2hSwP!%el=>dG0HylH7_NJZ zeSf#~{%Lop3ed2BR%zHCNsxMsYvX1@mxk4C8zN2$BT^^C+7hG@XvYx|4>Ysamo!_CQJvaw0nCs z9I($AWv_yM?wxY0gpG=5oB1q4L%m(DDiP8iprcvVU9@l-Ih!(B+Tx^Z2-_8@d0gp5 zjE-Rs`k24(^)=8bdVOO0CG8^1!OvD1^#0?ZD^O&O_p+;HZO@N0i>+|2s(TTBGxTYK z7dbnilj(vA9p7MdQ9zn?3F(2E#XCOI{yhtDf9GM4I+e8*_N5VE);el~|_gkfes zOX~`mi$EK}ptlO_Rc3Fp3;-$(XgZqZLyFnaKx0ll{neY7LMtv=vLfz@kP6SILUMRl zIT9Avw#diPQTSD)=PR|~tt_4y@0+;C3xU*SfHlT)H@hQPz&pbktP;>&=PtqgsxtMO zIo+9Zua9vdzb(&sez1Mv%}KD3W0hTyGDyVP%%5OMYKlag&-saEi8Gn7^?=@@5H?ofuRiaxxaw%^GhEStG&(glA&tyw0wWN%sqQPUBxj9P@v$>=kPcR zW~NCup-@3}3>bSc;Uukm;qnTKwuYN6TlZiW6a4}lKJa`sW&1({tH}xv41=_Poja9Rc z3u?k=igB;0Zj!`?(Q=cq_g#}nb+r;hLN3mXK_eJ2Q4|_or7abtOZdwQZA(2sfxG!# zL^Cyricz6gLMG_8>@j{^ja*14_}bjbdv=|;5B+v`ZJl@F>VWl_JExH|pc>o3;XImw z&Y53kX?RZ^rep!R%(G-PP;=e{WqX%>-EQdeIHK&(lXdF=fQu=QKGCWcWSL){4mKDi z|8Y(jRY3XPj%KbM*uf0NhH8QozQ$<~SI=pe&7+bxZ$sd)Jp-u1lZernAK^9&f__xm zW8B`}jG+kA_mi0*SE((Pz}`%`>H^mx64r*zJ8?WK4A?pSk6Pray}oFad{%+_NB(BFdAdCt5Du}2%iW9jxh_ac6b3#w;F&x zlaCtqbm93auR806TiEe<5eKjZI@lCT>SGZ0)V5e|#S!_HsLA^$5&ZES335N;9_yww z_fw@jk8sbtdK|f_X<}Ye@HCrDUpSNC3cp+}E;5@N6$%7tF53VA+tcKn?AUAcca*p= zM?(o4w;t~*8z1w&V@|?sLo^cyR+5LHh&|g33AUo zrtDtb1L~>GBzrTC#AFSND~DBKnxA@Xo*Z&}Y^)mIfAk%rT|@1h~QHC*09l` zo@C#AUUi5h5ds=sme1-OH^Tt?Y7|a5%%`(DU1{Y(2(kc|zI?n4jw91|>13~5SvVYS zl-~$B>516K!-D|++b{e~G4P3K#cJzC>}2MVLhY-Yn&4DE%1uL{vs)-KXsC0L*WR7( z=25*+;y%%Ig+1fT0k8(d>)`YJB1!%JjGx);4EXI)%oy=gI-_&FmxPIFg5=e4hmG={ zWQnGA7X|G1FXPJuK(5`seC;dn-BA09$Aa$(>JBp{DVX|=bZS0-{-F}yp6Q-BPu^|$ z0q4+{4c_0wIpf?KHQe0qbp^L)YOO?rPCZc4a_Bv|2g)+%>0}p3Z$QrR?XuM7Pj5I* zhM@UWLa;95Z!z~mw@^P&-l)gh=m&ha;BHZkRJkh?smf*yz|Mn)>T(wiQ>I5dA_GEk z9n)?}=bSZM$id9|4EFV7-MFxNWzJWsd^$_QaC59cOLEz|wKSurC6YcWGtQI~**9e$ z^=EW^7h?&(v-}p`ms30da8ddmxDPbTGpFBaH+R%;W)Lgn^WU~KNSf)G5_Bp*oqqL= z7gPmH#5ZNg8Vq?^sazCY4-hKMenzeeu9%&tutPy&GP@U%m42@eN@6ixzr(R65XN{8 zoi|qVIFe-bKxC5zH|fKyW=QTW9VyAf*Nc$JMv^GxGd2ydr(_5lHNr*-dh~^O&{L{G$f3m|(b#)i!*Yd(o<&F9? z9U~s;ov47h+31{omcEN}zS;a1Oal`wRp)|X-qV$gbJ1$%CBZXpY(X{qRN0g}gEHY-N!s*=~xN zs5ztpHrBP}KJfy`b@?e^9ptIIGgN=LyA|g?^dXGlnO`z+8v^DER>!jwh{vm>D>Oo` z?{woIo;+{ltbu9rDO`yLTglHj6D%}G+oT&m7mIYXAiyG}L}|>A1tdt-J3an9p2)BV zS<^$BC^Hf(r(q7vEC(a<`Y6$69TGY9ARTq-x2M&{MS2pr%uE^tlulI#1D!u_-d+yi0mY zl|D~Ms{v{2Me4RN@9KNd=Z3oUDE9muZ((i$Vb>)=O~zxcuK7phzY{R<>rqDM?ek1M z05#{hv9zzKu$MlqbYo$%VbDZ`Q!eJ4;1tv-+9hdUs(HS%yx&qkRr6!gr;fZAxs^Tr zz|hR>oSoEbJr`Uk>ZK5AXJPWtGMS$X1*AZtBafh3)snYa^ry<-!)RC&i!LJ@3*-e`H)2-O|+A(y%^xxnufx;oc}BE zNSd_fTHuCLVDG9}de(X2slm`$9uz-YvmepWA2s#Bp}8J^AB1V6udZbCg8w|HM=k9h z!R1RSoIvgp4ZMAL*K;AxOvpe|PQ6a%oD5y8@)skWgNm>~1A|G}oEyW)k%>=D;*w7i z8jsVX4^X93UG!0vqv%)oTf3Z`PNTfY)$#FNrNEAF9E}1OMx9niigvm1`DsPG=W3`q zl=>d1VwEYAnSIPWxQ_aL;j4Go{b_;BVXEl;EZZcajPJO*f0X&za_Dv4($9?Sd@iII zKh~DM$7s%HsS&q3BhJ%6+F$j`Be!U%Q@>T4GH5drrkU;q+6cTwsLiBzJna>w zg&Op|vdGsAHt0a|Uf&8jVgsSF{pBlELw&83N>n|9T!FGT@=_2m!N%v5=^n)2(-v3w zN~5U3Fwr2*P4G`vbGKL-Vk1yUjc(t|r&0Ag04JkR-&UI(xBHg^Cn$m3x1|6(3aJ!C zDf}yQ495eduUhK?X_6F1(EQ##Ckv7iXDtOvMDV1F@2l3{z~b@xjDSU)K`HyKa9$;G zhq+f3&ua?LG1^VX5!_25lbAx8q;9+b``olo~ za(Xpq%~1l~FE2C{(+RbKW|4j=$_NHf^DCY}G;6LYHCF?gc7t6oS}zEM6eZOf&lhEw zcm(2x<%yqOJh$qOP8YQUCodEzf!Y0&uiuOJ88Qtj6G}?#SX4I7rOXHpwKuQZ$-N4r zRRoCZk|zB}NMqKG#VYJ3h#(+NQTclFBwLcgPPC8eadVlgu5w&@qeOq%#l#^{O+}1{^5bf8v_2KPa2|IM3HO@jroH`OLT$mS^I@$uPTp07JsB!y*@=@9(Mrb8$&!2R^A=D*$Z) zg(Phnn^~_TgW(rbLrQ@Q-`HHY8%vtvsDT5ET~7BCotZS5@ck`rweA*3JFF`dis@Rd z1^9z=dl$HIwu5@fv4jfwTq~piTO2T9-hMTv%5bSKGChwHtb?txae)dCP-1>+po06s zo(@(t>^u}%IyXfN5-dLgMfPEzmWB&5w3h#u6{x<)W39T1A3Iwm^&$WdZ(DjD21&4- z=d32kKhH)FcbqhS2O?=hGH&nNDg3cCR~8h4Y0iMrv?rRGEQ-DtKdemzEo!oTynC}rv~s@Mp? z-7^!hNBp=v52~R9w}m4DNS%E`ZGHDa28kmaDC$KZLKt+Jzb}y$lW3hz))V^|CR0dW zrJbGL4Qq?yaybC0iu5-+wJBpISIujNe`2-PY^FT<%EOW%XRj=f9sfQ6S@5%8TL!xT zDDKJlvh(+eMW_F~tsbd;08S|s;z}T=2+0RK@^B?3l4z(1ZG$Wdu@p@#DQvo2_=%^g z9=01k1{7~o2)5*XC>MybqzDKModG5iAw?HE=bSD4q?ceT4MsWC%y!DyK)iQ$+xI5~ zu?=d&*D@m34S7ps17-}86Oe3Q%}8Pzcdb*ijn*(V5ohX~uWlGzS=#S0 zOBO9Kkj^NWfyJ3Rzm?E9572Vm-8WN!+_-(z={bsL4BU#cFc5Nt+g8l#Z3Rj0snvtpdIp}H< zfT!Ioc4P5#)le)S^3Zo%r%#Y0kbkdA<;Nb-2pl%CenojlcSlxndT%tP_A^GutKumW zp!z<;u_>ie7eE{LnA?=5KdGC;#Ea2AIP;H~b<3 z*qUKDZFVxkXgVeXFpXO?@nP2)=Ni|JEHbZch$lvwox0BJmZCd>fE0XpP>L2dOWJT( zo7qJrE1uN!A09o0Ib#!eDkXkU9212X?a8Q ztSfECEzVqWOw0RJ%h6z{D0yoSNJKU`k}3Q}`lh{Ir-E@M1t9iQUS%Q_6&0(SRHTyM zLT4dQd;n`Zt_sAP=M&C1IX{`=!EBzb9;3N;q>WLb6$~el`3`G$UC?P~2#`D77t=tG zzyu)Qwimjdvo~pXlJ2bmv{sgQ$m}dZn`%E9(`|4nb_OTG>bV-SE z$D4#aUdbT=>hoLxOqM;2XE+518WFeh=L~r6ob!^X?ZO6+OVpl8z1i%48o=5gC zbd?wNdE@xtpT7Df3P+vn+{T#PSFwZXvYj^*0DQs~!r&%Am7AAJG^{>eRZy{AA98(- zJzK3P@XXVu5c3~ttYjWNEv53INN&5GdkQN|Dty_(L(3k&6ys2&DtY_Fye?IN0)SJ{ z9?;!TNoFB$cJqi&deQ+e8z@P8*Hv#s0FvbpAa@??@*&+oxXoRpDhLn=D&D5Ejz%56 z?|{zYSlUH|2MHAmOjPmU0c(xqkWu#FMZG*-thNBLS`{BnU3ymi=2xdTpA5ZU3S1Cc zUl;@X5d9lGiZorSfa{nP4@hD~qJDWpoeJt^7a%v&K%LaxL|xG-!tLJIxJ6OeSg;WG zS3v3Z;S~ChfT^&mV&i=58VR#(Qza?Nrz3Fq$0Lq2Ne%mSN5YlIfGPktuqz{X1u}*k zZ1<`7VIpTdLHt=ZG*XNSbvvU&0|&wrRJ4_;(!uvc`>-FKw@9#6Qz1b3iA) z6<0uN&=H=M(Dj5d?YrEP(e%|0>f+qGh?6FkwomUyZfJIhAqmha)5vnH3Q@ZaFjzE$1a>*{9M9gSe(kv zbf#wB+wU=_s^3Dhk_%;CrR$7ruGXRIx+@Krrwe*!|yI1LQLyk77@nUF%+HmoX;vXsSGJ z2>MuZ6kJP&5h(Fr?_#0R!3PMg+wWQgLf&p}bVP*zsiqus2^BU6`u%3fJzk zj8fEYVSIoII|W428Ajr{4uX)wjbEl==w20~bi*X$?3qCFf2FY*I(~*=kThx2W$s~% z)Uo~@U**ukcV8yTC4X$-lRDW2j7WkL5TpO7?kp2$8Kp@K5MbF_0RZdphTwyNN3aDe z2`*5%z@OFzQUD;LHx;(WgI#OM7~q7j=mQ!k;%Ly(Spme&W7fjifKK?71JPICKCHio43@$v9XG~X3^~Z(N_f{s*Ybz+d zmw5@1MYHW;b#Yra1e(v{g>iaZ&KeZAd1$8{2Bm-zP*u_CI{R2-CtgHWbYPrfpSLu$ zj$PoSyNhz=6ovApmM#@)%QT8=r<2SC-gu; z{2H8~HJbK-5=TO$BOOe{Qr-JPJL9#DO)0Kly?xj_#B3w{2djB^po}ON3ed^$@(<#- zGAUi;TakpgQS8V;iriTKsD~0v6x!iGcapsO95e3@DzpGQ-_BBuNH@9Y{e|&|gIBp+ z$XwE=YK(=ECcA>tP_XT{WHesSp8k<1LHy6>fg%w|1gLvRN<8CWd+E4IlZR_}A8g$f zE70l;go2{oBIT|utN!b1#)@e{Jtj^r?s@!7_wD9MrDRkukQtn!yrmc;@8?nP*5N`& zS~d2Rm4h{JL)lvpY>5GNp<&IgB2wg$2v7!hF`>x=?wfT@%0;>9Lj{6b>xiu|NpJIP zW5$7?4{O*G4@}qQj}yw5DL}{S^!Ux{%PKo=vl}wAzpG3FaBskfE(7E1a^rg2>xB<>l}rpj?_S4 zM2yc!B|FhkXLB;Smdly;&{7>W z7_>90j}gA(yMYCoXGo8%8uKi8BdHU{0v#v^4eJ^hOzpA_}1`yW3T5clQdUmidZr$mzSP z{=XG9B*zS9jtSZ515tG3@#qJf0{q^RFvf$3EcjH(CkvL5yIZ@Dd6)FvsqG3~rri(G zMsSl~n@+>GL&b`27lZ;>8o1gXZpw)j@D|9Ehj+WWWGl-`04r?V?z1i=9Su(Oz4R4T(*`o!4;03 zF|v<$$sT>jro$-^FVTAk_TGr#n7OU~+|#l}&5`?UnTyiPNV8r|@d%pRIlnGGb^e+{Vq+lIoi{W3y9o zZ}J7_p68cG1J}~1<=8F_UnQqb(8|8sHMlEdI72Nt5A37P`JgmT++!VF;c7}OP|lQ= zL*6R^`0ltnR*wG^2`6?7P&WYj0k>M6Cn;})jySiO>l~CEG}-TJl=5ylO7t&zkpmL5 zoyXt_sY3R78NqMKw7Gw5(AupIj#Zl31^rGK{2MEPlOOUYTd?VJMwtVt@}YC=f@9rX zAhb_T8C#5w-T)4==kq(vX-Q2?&byg{eu=R>VCi$`)2au~uUjeM!$j?lP)f1^1-{~i z(Y?&b7^^J@>XOMinQlZ1TgA}zlq4Y%oh6FizC#0z6^${ag{#;v(y3~M$tLU0O>RW*x#`mO6~87wpre5*io>a=)yhl%<54Cv_v*aRJT>s?S5`B! zl!2Q6qU^5_T!5q!B@;sOZi{{Hw*~Yyu^EC?9B7n|U;o|)yc4Tb-8#-PPNDVmeg0dl9+vulr`et$+ ze+k-SPh)mpmbMI*PA83iyYLkSa6+UpjLBv!-sixqkyTGq3JzgH;ZHks+D zOAUC=d@eTv13D(H zesR{pH#0t4#?#zWD4%T%(+FewwVWwY98#FQZ@8CS7y5AR-O)XJjK_b^0^CzDP$gDq z={ss!&=&^oyx{73C!^ikuG{9Vzm;@PFUiNZ#74F>6LOV+2|Re8&w`HFOlc`FgMRWn zd1!T@?lZ(mkT+i%{V0i{7wGNUi8s9GNC(nQ5GZN8IrxGUAa+}jkPLXnXTR+bJ((F_ zo~wPpw%6*k8X9j$jNY+=kSToQCg1N*6#Sm zm=|Av95;Zr^NjgQxvSI+}T-6ssI4#R&c5aA3%A%ym((nP>hc zoLbhR@#D_3Trv_kD(yWLKG!~+3IhXimX+P_rZ7hV5o6p+7%yV9wCLrnX1B!riYlb^ z-Wp6Z)w9^UutV8*TJ^*dz}eu+E@T8^!VK? zd#G1UP|&e)X!#Sn65Rt%ebWW9bA8#NsHc{MPtzlgXb`fjJ9NhB33pr>iii0l*?Q?EgJEh0+0YrrCvpe(UX^>jMhRAh^i3 zWhL|EF`2LN{LU7ps(FPTa|qB;{>H+9!BEC!u~Sb@U?urbBd{vF7h$&TLu z3cg@av1Y;$;x;D7{TK&L* zzOO$V+7!nmM@yCS9eb3tL%M z<){>X zf?qZFtpUF6j9xh!DYkEvASn&f9*R4p(r^tIR9CvgpV7e#S4bhK4QCj%-G^6M;HM2y zudtBio-}q@UnXJBd;_PKX;bGXh6E)6iTaZ&pr@q#_^1@OK*zyBnIS{6WF>jq~R&PkL)+p1!Sz<3gj`Jc`%S8>aAX2=<);Rqyj6w!i{z+^V77rZy1# zsr39pd6dR2cyNEJpY=SrtN>#@A}$(F{az6K1AKJxL}nc{&M18^9WrAq>(k$d*+ssk zgKfiiw{u@q77wUN}xOcugwj&I4zPrH~CbW)((@GneAGG89jRSCO|&y#R$&xR3Z z*ThM``0MOXEuX9z45q?hhjE9rI_M*WlJcN9kO=B3HmAHP;qO02Ljn9L8q?P~^ubII zh4)H+0kJZ#C_jLfw6|=yW3i1Fdw7Sk{z5NxR`6r=sT9v;Ngd2ZXbRn`tu)Kwe>10 za3=<8#Jt=+tjmyR-ErDs`QiJ4h{Tu+xoMJ>`ZG8&S{JYYY>>v>0-!JloLfNIe8gee zC|!zY04dr0I7<5YDIFIjvgX6X;<14iwc2}k(ka2PH};rI$#=;o&pBTPYp1=@kD865QFK*964ot+KIFz_Y^3NH~klT zPaRriHKV;A9A~hhIaeJpWyx2PB|wbMoGFJt4G=elVE1iTUZLmE)P|xgzW=>edMzO9 zA)Ckqxv{?ZxmJ)9#?=zG9gxE1OP3&6%^j@k0BbPR@g+zuu`6o6BBd-zE8hq={5B@f zYRBUrABFB|9A#RL1MSjdq~iNXask{>3?wQ6{LkH6*3~;#2Q*0Qlhh`>ESv?(fkY_T zJKu?rc*uA~n5Zlpb0$zA|N7>6DbAqLEOAV3mwR+0YigPEE}MZ<&`((6d5$|9v)KgI z99M_lG8>dx33at_FUk+Z8o8+gI_Hg-!USv1<5a(22|dwjetXfGq6bf2dj{l?z(k4O z`3L*`Y#c3w6z4AO-a}*4)op@Lbw3zNdyK{D!tJ{)2UM^o3x)q5Q(qYnb<@4QgruN= z2+~M*cSv`4NGrXhl(Zr$t$=iQi*yKxG?LN_ER6z7h;+TP-1q(b-*0}vv-6ud=Q`J^ znQX7T0HesT_@Vl!Q_Q}YB_W8ot9Hj@Ipl^U+frGA&HkKn@rp5S-M&5UWUoR!NKcIh zN^0iJFL?FJwRi8@spA1cf5LMZcb>T}aKs}Obvle?5vg<2VUylkk@aO89v1Sv4>z%k zP}7r&7^?sfzs^S6k=;yW$!B**t!;krZ<{x^eujmDVAwl65VX=ADXglhiTU@x zD9H508j8N~VH3!)Nic=%iC;6CVLX*2HN(3ZKYrEt+x|?+Pg>2F1h5umPstZaXNUSg zLD~pW@9qXeoEcC0lj85b?m5I7Sl{{9rIkq+H`P8t9^HcH`2Cpf^RZX<-yNIs)cFN& z;iNJXr{hYS-nn&@4&_TKrstzUgi}UnUqwVUeWiE$994kDaR2;~Rk@`RA_~s9_tYO3+Ebst>S0g|kOZno0Isdr z9`VK`kB)6`R1%GR{r8^Jli$5&deRzfIGXv5zKMqzegsTvW>n}{1w7u+qC9o<@h_ek zZYp!3xWFQfKYt++ECPucN#wfUZVe}85Cf&O8%580xnRnwl&T7^wW-ZL6G zb1z`5Wwa6gV9(i1-TSPi(z)Ai)$r$4wgEuxmnEVX0`tsBE|>S8$nP>m5Ws2E))hO z0wsSJUdHcD>ZEH6e%92$NyhO375OU-oTNnR)-V^?4cb>dX0^Ls%ZBPf`%h3E7PhVP z4m|NTzGNOgL%??w`6>`6zUBhcCbf7U6OI@8zjdBaRtcXwlijC_e;ijR$*jqusZyQ~ z=DGiG)ry|@2XSX*XZOFqJ^$G3RfRhuaQ2NXw01XxsdBhAqHCYtzVBIs0R*51{3MB2VvIRHSMQn8zg zu+rh8)XODfux3ZVC>bvC?j78LM?*#$P>8-19fHv@T7LaXoJV_tI)ovfKK zM-$jzJQAYhaJ9qc3LkPNLPk&X58cqr!IqqPgyr<)s~`jt>LB9%G3B z=+bQ^#ykco90ryG6}*ryW(s*)!vGhZZ4IPkN40w=GfGTPn%&l)Mofd?zlZl*DnJ+7 z>qpi$6KZxBHU*3?oOQzTRjLHcy5Kj~z$fg;E%gl0IwoFxqeoM!LbrkZAGUd z1*i@>hq}@r{PFUeLJpD&APjY(3`oO7ZnislN?5v=_Ck|c-$eEAzPUpsIT|Qmi+hJ~ zbRV46a zclkj2?$7aJ=?F`w-;?b&47C3L%M8!@$Nha{jY-3cmwqDpqW_6RD&=xMPB60en#})BcF;5yd!9Uck|r^ zM7NI~l#&7FdN!}F$fh|EkIklq^l-Pwh}?MXPLVF@LFGXPJ+DkMYdKS!d->2_41~Wk ze^2pja(lc>pf-#~8TKKg$+XLHobw+3@#DCk$F3-{4*u0S#|Wk9N+NSCvCk2Nz;Y(s zH+0*5*ili7JNZkyJrb>R@jHyvaP9F^+uf4~%0rw7GCb*ZNIhwwBE*RAbDBLlvmx!V z_dh(_9-L!L=)F&G96ruD74&h~Ai(?$FY}oB;3j*VL7^bqvmo0D7C@zP2}>t{;w4mt zT@8J>Acv4QaapEaf68``*FI}Kwh!_M-ec)9LUS0rZvb zZ`wMZJLAWSjVsx+_US!N?N2XQ(g^AOT40Uue5RN8gx~Yb%Q&p5eNF_VX(0tl1$L78 zf8flCJ5xw~X+8JHn^q(-rJhlTT$KiT#z3vBzNwEswjp!3LWI`7C{}CEJl(ReSU!xi z7BAoH>?(&k(jq;ob?OrfGGhGpzs2wDehcgtuVl$uK#(A@@m9riylay0yD?jEVrp`Md$n4N! z9-C0MQ!xXQAu0v`c{xRF*x#>-S|%*|8M?1|ehrYKI@mNC`x8jl*3@OibxIH3^dRQw zR;P9*7JV-91N7|_ZF}?{4OL|@m0L>iFo28Mm}mG8=P1GMc_lNoQcd3h|4K`H-lm?_ z^)=05C)j%`C^df|1A#A$ z`I8LKEvLLe%*vdIWQ2fCWpgs{KhGc@T{*Et!8a?Dh%?v@%kNF*_F~Q-ZF(U6@uE&K zS#Iw9^D6LQmsG2jk8~nNvMim0LS|sJ^$F8gt5{ZO=v}UN=f}iI<_3|LR zNG-w91guF4slR`vkj~r#UbwkC@m29$%af*}eyrXwzJIM@rctHzn5jEDFtpa9CALGO z)InNvHWD6hy+Zl7Vvyz4-6HG(bNWlf5Z=cs@b^{y&;Hy5oo@;j$`9)gfwO+31I`+$ z;S;Mf61752HA~@bSo!N!#+0X==CPK#Hz1E*{Vwt)%lAi5U=fgak#GN2Anb;5-n=~W z>S!k+sgYD?jCZXbl4y_s{D3$2zw_?>YB^aZ@$mj|;bZXS_)}CYBW7wZnk$#;5=s%$ z?Rv-CYGRU)<|DYnDLDciTPS&mYs^p37TjhdQh~S8&jL%6{hu1Hmajgnocn>-nXOre zG45Eho`&=IZY9F5ALHsMm2&waCb-v-p zqtXurm&YBLUrPkyKIPiF+mpy(8FSCtlWuD3!`8ojXDW;OB^@!g z^O<}-WT|hjkAy%vv^=VGkk0THLhQc~8(wFQq&CC*?mOb;Pj)=xI3;)p+dIv7?4_IjUFpy13&0%>sbAEH zTJjC$e!$$DgntuveD?V>f z!O)$m*||KR4GU~J&4=U9970wWNVt#wuzy3f58=^TIHQZROXYjbM~jzPH%GPh%wYG&5?jNVNEP<@uUHX$-ZsK(t2XB8c^eNq<$3?1CGP7sZx)2f()vIsk=773rnmfO=01PG)YJT$H~?(nmPuA zJT3Y~$d3{XMh|T#nu>7lB&oUmx$wA4ycQqoc<@=tGm%DcHoSnw+m1qa6U* zj0rua`B=NjaRv0UkfDZ$yQ{*ESpt(U#6!~-T_P(Bf&p0^}p~;aTnKACuNW+jvOR2 zKuF2eP0vn#c`d+XxhHmAVUh8xI;Wm!{BO&2MU!;3?k~0%o?v%LRj_aQ`+$!{$I1&N zJP%6P&rhkVoraZ9P${vw=Xo&n=|y+Xz(Yd$Nv;;XF0XE!=>;V| z4K2IZbfIvIx-Cf1NQl zS#ZhHxa=7f}Ayux;v7}iq)6(1zW zzm1HW*Mjg5K8pFFt8jmknuGRACkVEY7NiwqRJWrfibg{0liD8Y5v%#kWuC$-B#?JI zi?6FCl0hVHZAE_L=!UB)V0km{4d_KIQi-;QanXka|NT)pQ89HE%*cuKCiOqF&+5p@kZ&w z5bww~n`ch|Xy@z8-0I>V4je2}5UD`Ea0AjtU!#@EV&bRbZl#n=9~>Aqe!SwpyXG1& zH+7A}J~<=_)=VRD>w%j#ZWva5=B_a;`7Z9XYcf3XQFsPuq;lA}WTnJh`^DW%v~3;d zjyK8!pWkVjB+*WO#mW-qCXIa%>=e4*Jb7P8pwsCa{l^nFk-=h#tO z3q*PEzafkgh?^2lQ)6uR5vP3bK5#}K<~6ADmeY_QD&N6;vV}FS)&5P@5sL%Y9LN#* z2Lfl{Anx>?3v#AY_1V}{-R+@JN-J065T&$VPe=PEPFWvdzz9z)QSNNt+fNb7sjZT6 z$g`9o<0`Sb+2aG6w5IZ(8_GloTQK{H&W&p`d=q6AJeHap_ekt;82MQ-hmqA6FJ;n{ z8ej876u&sgi_qw6eObC^k`rX6 zK_vAKEj8<}9D31!0!{R+uDWiVXwomy^)vgBuTxS>} zKE(Tf9@tMzlRJUYj+&5PSQ;eKT|uVj<@hHp69XN!GTwz^s>m5XpQcoAyXR>#XFATZ z6W3Fl!q=6G#I0Z0va|sqqq#+e+>nri`9LhoeA~qNEWiGOLmUq<`rysNWowie7yCfs zypVj-jx6!3nZv6R_#54_xRo&TnIfve68m6-+CLl4x9e)X1-k}HxT5{wH3n(k0>Py5 z+;BW)Xsx#kk-ElQ4J$1<3mDQZ5i#xmnh?;`dA^if1gpPK z|22O3)t)h5dsVNNL6-t(hEAf^f=GkW!)zRoW{`aJ5hY5_Ga!(=^T}}&_&(aWsYSnB zw}EO6IJ+>`Cd1IPuJ191lbLcelY%+o(0ltH<`qMO_%`I)KlN%0bgh)ZX1O6ZOLliM zgXn_X68)GiL`-3rtuI{4gp8$a)U=1mHU|n(N}Ls*?+H~)%4+~X95Epn$GERsxZ(d& zFh#8mkjo4l3wX5lHHdW>JgRoZNApp%klg?K?880}b7?B{@o{0*zwl~p(#6FYI7G1i zor97gW!rOq0xjOADhj|y8k}O|TGsVy?-F%b^ z)^x%nkzet(v0A+;Ou6Sdfpl~vzQvJ&no13nSr+!&D6&qmZ?<)%#6G^5KI{A`fSW|^ zf3f-R-scq*R97ppk!-LY<)i1rf!%ULI_N?3wbVr${ugcE4cmxwNMLVU3yhxAOqhgU zotg%go(f1-ss#gtC>~-zWECJ5gCL2?D7AjZ_Az1o)weS2`~puZAy=5_`)G#e%_lzO z6^B>_1vB{Kn`N)V%L`__aqBj!nxqcD{&64i(Y{9)1=!-me{6L$_Ar7RI(M4=J~9U@ zzaWv7*&1h@TZh7JLGYUWzg_^`{Aifp&M2rH1jg--7(_t(j^^mrtnccMF>0IEa8ZED zL6p?LX*s~iByF!%c3eZoDK#16>4LlMGRPyk?t?E@@#?UpxZs|138S8(vWjw{`!Tjo zx*LBgflu*gWly`xIP8v7X2jIa_eBv5ptcH(C%1WQFMAA=`Dd3Zz}OjJ#N~lwF(>b6 z!CEKPdYe`DE#`Ry-S0ycm(pT-#sdT1wzOt9k)S-9#5=QZSr|Th)n$a}Ioa_nSF0M6 zoxs~m9R&er6i7aS{63=5tbk)OL#kw$t?h06xVGBI*NZ@I6031(eN8ZwYo}PqX7aJd zBByA8;jsbZ7i~#cT9`Ig}XYU!(a5BInC=coYvTeA{7UKdc8_m9F37^?pj z3j(h!qdTjp;EMPTWQF6_DZe;|)|f-tc0;AP(8rstR7_I5d#*7?Q;gLfZ>f0h?@tQa zXt5X&CC3T2t{Rp2>MeP?cJ5O%#rL-Walt|7c4Dm0?c8&hYrw&O*7(;y#D$j*H&oBO z0blF|;&pMBPWox5*p4T|bj;)H?>7vIVlm=Z&=fam zbz0#kPO+nJzrQNVRev1C*a4_=oC46C!bgw^uxlMK${zmSbctuzm+am-;6l~~9tLQD zT7eZ&15f|2%I`kg%#yJ(4p+(qj>Jma)80={&kXjJHZgmF215GR%?k>V$sDk6=B_Ov zG?O-A8*SxR^jw|2;0o~yv84AOs}wo8|E{>9SA7UzU*Vfx$0`1b)h=SSqX2-&&jjS+q4|B*64qI zlrm{hQznn)EHcwBdX7w)7&Z{akAqlcS}@L5)Z~dlQSOmnrw4J?2UuNJ>q0cH3=o`g z3k$=1_t=s%(bm=)H>J^B-HBgEe)7l)7kMA}VTuEI+o_5gxf?GM5+LcU#mFCB@F&H) zfSvuXjs`5hG3v(kAp!Q}4j{`DASx*#mK)F2Xd7iA6(krj(wg{P}$HNmInN6x+++nY;1xSyUU z1%5#nW*lB$CWbz4Up8xSxp<&btf@gE_QBdK8QWoDCiX;d{n?i!@t%qAtBFLhwm6$( zxHrL7XM1U&vk)?R56Kypb%09L*kbQhzgQpC)X#Y0RG)LtDRx8Ni>V(w-)m2FtY^Ig z>sX^sT@N?^Lnjp7a91=yNYlDyE#z^=gS~ZKHb#3p!4AFJG^}q|;&>eEFKYR+KpiW@ z{6BUxg?M=XUEs`;$R}>RpXjz9K@~``lzhyf#fz@{4!Aa6=Qmw~H%-DPy@+_XqsZSo zJr`$9oA3*sww5P*_iV9c-P6*M*U-mJ1#V0~Ul2wumSvgD*kywRvNn|-+6{dz`&p-W z=s_PZqXNPzEm%s!gHZmTj!F&$-#$g+!zaHsW;@F3`fGF?@};;@tAi!pU-F z!jBU{>NIJO-^M@X4v>3VqCU*rJ8A7sGQ`hF_7P8@zeAL)tUyyi9Vz$iBS%iHM*d9p zBET60yP>StBNbT0>-^QatbMY#9Po;?Iz`_KL}H13U}b_xwAQBQMTlSLrY8|eKJr%- zuqQJfQlTbrom>sQMW60MG5FFbuvBaz_A#heU+1TOv@I}wN!YelF2d{w zUp{ogXT5&lO__iliJ<7MZKo~Fy>b*-a(SAr+xxL!*sDBg(QyW zls-uY(9A0pFE?8i6%)byge6{=h!>AkE-V=kWd%leY-Q%QIf`;Aftsgjowdwhw*M6p zQxmR6j0J(!)3AZnH@Ty0z4gnLCNTrn`eK>e{X&hzg>YTc<4IL=@;hau_-3_QZIR>d zk%1#XEM9t&lRVfHy?g`06=xg4_!ZQue#10aOd^l0nmjDH`&jb%=R-Anz&B~w3iL`) zv;-$JI%b?n+$b)R-CzM09kt3t$qw?M${C4OP*+Es!#}Ph(2a5KMqP#mU=$(0-{Ux! zloB_U^<3RHi_L~AkJBf<#9pC^Lc2;{wDE?j+9} zN&&?G$H@T|Bz?VV?>;cA-dWyT3KWA&y=#pFPB3av&W1#@ zT-ctYJB&Y+Fno%B9PTAPd{nk}5$~P#a-Z1w$rPfOe@qh;f;A2o&Pd}G;V)%Ac&Htc zWP!jx#Cq+2?k1FMh%vrjdZTK4EBf_ab0-t2nsMV-{&`XFq~^FEVhfq*6eEgZc@L-E zzf7;&{gG2VP0grNG*|rf3vP?1E)R2k#4M)a+xgdCfEEx~Z2;NGN`e&B8@G}A&&+DQ zp%>F6C}n7rDznxAJI4DNUEB{vcLHFiu1FB%sgq)RrvsT6uO-g{mlw95|HSgBU-(p% z9wWHtyTb?eMy~p; zUvK0%4EP;q>qSq~Vbg&3Ax3bdt#pN8Z$4}VllMj?E}~_KfR=jQOf{j$ijhh?7B5cc zy$u{?dRe z1RGzzwBV9Ymh%mn7r;c_BYI`)?EDxR(RX!bNli1ZwV|1X3cD6g;n}OW~L6=!LY{yQ|iYWU8N*9sG3sV zg9i8P>E|D>>4Qd$@I^sN-7lls`AI!pz*WU&i9HE{`0h$zCgbS^mjs!^T@X%ZZ2w^M z24{z)Us@3#OGQ1$oj0G@9;XM@+kC?vAy68WE#)d^aj3Ehptrg0%HeF@t-K1$!kd+V zCw)JiKWbVj!1lcwvueNbUWsHjQ*#AoZRy)&xPxqk0MMU}zDe+OVAf&&=fIG0E{dX= zD6jbYB2Ye97S~Q(1b3!ZdF(ld7qPFb?XA4eD)>Tz=`3I|q2`X7u__H!0l{!!rsdP_ zpnVTe_x?fHRw!1c093jZP`%r33lDD!k?{0|XOd&?smz+=46v6|K=${9+Uq*J!DM^c zl84n5r{l|q@CapxdgYNaFY+%>mh@^@W8l*ZfbD<$(+;wdQbGThewV5QXBEJ{6toeD z0voG{!Bo(q0S#d33@;fwsneX&M!#D4qYsRSSuOgUMx~*}RwcS5sVVuJcyeD#aN3`& zGs?IHQmhTHdS-O$hY5nyD^9x&H%3~=0VGBJeBOD%YL-Ic0l#72>_ktP|VCwO?rySZs0^Xlm>Z3BR&nyV>q8^Sa%(hCO7a?jv})U?5}#+JMDpw+CHSGUEdR}yNA-38o3IQLbh6f*gVmtRLn}; z8~3y=37IyKM;Y-Od`0pJef*C{ioN5owDX#udBllga!BgDLPyac&ilM#=JI96ssT7P zY;0o>kS6!RA_)z7(^SU0^AZp1=Ky4c@ZUlPD8-A#PE*}+)RocStR21ZO1h)xW3t}~ z2g2i7IZAPMrghNAA3DWK&F<3u`nle}wWxAh@eKN1xd3y#@W5xsbHc-O?D_-+-x7=+ z^w}N*-wnz8ucR2xzt^O{$x-=uM!(b{S6b2R{J}}aBc9OO%Y4z_hOjdy7W@~k(^R2< zI98UH0CE396kn$zf%WJxtQ#RLbt4MM<5g&VOHEkSQBOD0y=prRJ3Xy zCN|k5{q>mcT|y0EnP@P`)Njf`9s%j|Ok~xK3<<}oJtN<%)alSC5^c$kDn=0#p#0r? z&>A}xG@kJ`v@n+g!6@nOp$*m?GG8{K#tSJ1S3afDO=f6iAY?iVf2yQ9r!G@lOi>LD z#`xQ;GRA#k(jEVkbSN#JKTbi@x_7#8X|fcm6d`4zO!P2`>HaNK8VmufUB|%7ysICm zAEX$WEk72L%0hMsGPfd$see!dlB#3T#DX`TGC7I7uBnpIkJ-yELG=ZDxg|!^PqCGBCVl|@~3 zJ)e&Pz+wtSA!X3KR9syb()exRI5mQVz+q+dEEQo=2Xdvsz=BTtYkWKHvgVrOpSw@x z;zZvuCl4-_PE z3)hbwdY=EfhzdQ^i(EQ~^Shouft(6YXt^ph%VHO^Wia)Jq~RBSqOR+{U(YN*xt0Hw zU2#LcMRufxa( z{oFaJtmN3B)3wR*V%|e?35svjeiE?aMoioORP>^OD7HGwuz3Zk*vGeuJmU(?-jsIr>KROzq^2d=i8pAWhm>LfE{mY80V4fhJ1N&r;%*vI(z84%PgJyJON z-(nFzFJd$;+imNcA9we=ah7p=uhLu|&S^Ful1le{S8K>)dE!JuCx+C)CB!aYdr;Xw z(RrwzEzc~soe!NvI0q=##fPe_(!_mg*TXdLH6Xs!1Tk$yiI-O7Ar^_3pQY!=v4P?T zn2I1j_t*8|$ZB>kyfr=I6RUm3oK@$^s`h=r4MGe>NB)oRK@krL{|<9Ra=sCPi^k{I>(USx%bZ75s6rB z$su3XIh}z-AG=5DooJAyBM&1v)Tnq?0m!*!Fj}p)ivDKu2*T51hgW++1Rr#LD|yAw z*M6b~Z4zO3EzPKNYC#(!o=9gMLS>2eUQjIE{i#i?^^L=Qe-zAwnqq3@nb5UGRo z6ExZWum7W}t^#{aftwsXCIDjry;irXOqViwR7o@`w|gpcQ-LVOVJld6h@9!kPu@CB zr;u@g-sPcn2HiR%D22HQlFvi)8UfaiF9AK3B1c9q4j_6NjHz000C4!{=ltPW!Wn#h ztbR4v;~q7aS!#qa7wCrvj;+v9EEwG+#D)EyOWTXnYZrND=sSclOrJUsROJC388hz8 zUR0+L23jITep@PLACrozoPkT3t0X@arnVQBs{|r7)w;{mq)bGHiOj-;-S?;Fyp~6( z&qvey+V=KO_%WF*aPQ6@d);rPlrD3X(3HO*q;1q;RNn0k;vxPy224^3aFE+zz6ltK3 zur59y$M5oEe%V{e_8wc^gR-r^?|-9W2V@i8Wht%rHM~@DM}o5KIeagz-Hp zFhMG&VoS4^wXT3aWTPSd-R0HyCCOl%k4=^6S)=NHl#fDd zs%0>OAls?oU_rI|>gBLDg_@V24s!XEe?MROW<9C6$`^~}L8+<*6qThYFN64;3M@WO z2+2^v?(C=gjmX|#Tp19wraO$;n=Ev*gvfy@#)0@PjB_JJgL=HOwYCr8PZxUgYt!(v z;7C`A+vM^j6|HYi!so=IduCowe>_mt5A|qiMGQ!$Z2l3D8v1WVH^{rZw1%}n&nHOm zSuqu=!+Mh%f8mvVK4mLZ$eeMh1IDp}x-S2AZ&@Q<*vCElBkb5l$b9?0?pO@&B9#l1 zx!JND$gYM|-c~G*g*tSQeLN?776p%X+)X+T`41`yfly=mbS1u?v|rGBdS05c{y$^x zdDmN88*M?K!VtY6jByo;yjhg)yxCB7!O4oKfAKKu!vmMc53s;$&HhViscvHL6}(t4 zedTW;%be5f$z|n9g9bJ_er~wINsJi0`!sHk?+u!S%!Aa$vCE(gwh_c+FoKZt`PKJQ ze8dPFwFb7pL}kA9-Dilg20lt;K-c#F1#}Jp(;kJTH5m@M25p=o?ZqKdCMhmN1qMC} zzN7Uu$9>!2eJatRtq31V>iqzXT5DGMZg-fnNpCgYmOtxUkbK#FowgXRV`K{FFS0%w znEYH)qW2o3xyH0lBF9EQqhkIc@AAxX8<@CV=Vd?~={T2%-J_}4R%-C^<2j;NIIf?? z$j8e6I_WKb^8_B_2iXub)A$;0z9UzZZH=Y6gi772Vy=r?lp@h{ivKewI|EgLm&$ zl%xfMVE>@$3P~5xZWEKK+s8$Z+J9p5^y4YnO?j%0n2H-gKInl&QE4hJnBl0`a~Q3r z_*{@2=iPFQ71#Ng)4<&8)MXd+mYA18vybX9Jz)e`7gL8S2|s-+>Y&@U*sN(uRIIq$ z;5W3Wkos(u>F2c>ASWx7?R4=5{c9lx2ECVsQeNg^qzra`qB?YQJ3l!N`H%v|GZefM&f*C*~y{a4{lMtAg0P~ zwjG4{P&8!2YEtT8o_4D98^m;eo$V~m*L5T* z5olMUu~OR8Ikksb_~yrS_kF0#{8|KK-$Mf{C5<)Y%K}?IeR=iwFLC@XSU4zVfi9od zS9lY4EpdCp4&5A2M4F;5-ip+G<%*Irq0|Zd?U;4>m;-mcDH;^K%c?XU-#@-t$=c%~ z=6G&+{r5fbH*lonj;wRmh$62Pg&EN|y;v-Ol}Ne#U-YDvrBf=Zi%-2Yz2VNqW9KhP zv<@atWlPi6>SCLn$vHo?hFkq5P}!c-Xa!Ts$HIls0vt8yiX%NJCJ`K~KB6I)4FT=x zAs)ABtfA`-##L}x6_fKWcR1gghQ3|9O3Ci!Gg9qGU`*02>lehWY$f_-1l~t()Vp>B zn7t%-Zt%QwAJ?XI9^0*8JGYOkot)FjxY^9c3=y`q`}EwPLSFO?Jq~#Rl5eAL9iDm) zf4IVPE(E!!Sd5;#E6pKW!^(P5CVxMInV*;Wspz`f!mtPAd7!N?M3O`IjDZ4ueTCKK zk6Fwwfj!1bX@Lz&NVZrL#&{36K5236Z7C?rPagzgzO#OJwfkn(R~RYG?F>MJUd-0; zC79w}<<-vGGcQ=h(|x-35WM!sk14%(#S!$IXEBdQUEWm1@0G{8Ej>7`JO9)N^n7Hh zV{CGyhN<+uY;(fezhOq2{{i-wq?ae)^i*W@xPwoA=AP1~xFfSCC7xt{8rH*Dcm6Z0 z6(Sb4-;kc)qi{+Hb24&aAbmHW3y8~nb~5{#WT;<%Zc-Fj5W^(?T4N^{6gF*Bgn8B! zMcyOh@I+i8>w5WJgE7(t%^jmbUT8ob(0qk8O2++H_E{H+?XU%IZi4}6CK2sgD-g9V zxz#;4SGu1o1tS};)?bunaap~{xcH_u{YiJMCm>ETdhYiFs`P!xUxm*U5Sp*Zh=&J) zRCVUw^0-nBT#`SYPdfzW>1fwLKhdNj)K197DL`B6{Y7$l+~IrIBn?GLVkjX=wJt{A z#;ZQaSNC+h)laKKIn|kl;O32g_&iN;ONu7{)6{06_=0&RYffxhJ0+S5$t0fHf{L#L zrF`L*6Z0 z@RqhbKHUCOD)uOk@KrKNsc57}%9-2`hWKPY&^Jlk8Vp8a>ab2S?2Y4)$e8NHJSSoCWP*;i_#9=anH8hO!;4Z zgLGKjszuF^YJ$Z8G+9(vsWRH&gE>~0$YOuhoV7_1f^cgTWNFPEi$kCE>=arHP(UMDOGOy_-j23!+FSx#Ed+LHZWuzx!krmkUjntsrP-8$u0h}&+kP_k7%&aO>Wfc$ze zAVQ&|9?#;gViJL-Of;S?4tt~Nz2<;~l-V}4XLQqdXgF15eLTzpfaXNKN5)p8=)}l2 zDuiqCH~zvO8^eJuFBA701Ri-pP#sp?Qhu#;jwvZVN(C=2rx1T8Tw`F&zk-QN-#HwA ztuG8Q?mzN{wUIW*J1rf)L=c3$1bX+okU{Fq0}kJ5GEiIgCgR%Iizpp#Ug&`f`!L!Z zgF!orhXg9BIkk}xm&e~u`7sHzE*HwaWcW_LBEApt>-T#6!4vzJMk(3xgqjjs4ms{c*S$)<4KXRidjAN!1I>`&sosj=BR$?`^b3IM3Dq=NTj|LV-?9VWlxEkr3xS#!z`DAw7B69D!xorSZW|oe(Cc*tQHJ>Jahu#{16(&(L2yp59B` zx^GzyPDhOUHwAJuinijanEJ=U{7RK}6W24-^L{UQ5V-17wUoudrWvyIihm92vu=4T z`#hK}elbX?ZZEaV$d^eH;x};A2z!IA#5FDN64M^agRG`ND?#v%l$k+Tc)md2)+_sH z7^if1-rPR{Bw~8jdB@&$y?(_Bsg-bhS}uXQw!~fU5ZtZ2qVXn}7?wrNwxJW97F64b zN8<#%rAjJ39>YvbKbpuF8a&dL^fKhw$b_sHQ?1a*>HJG&{mn*75`Y?X+_Mtk;!9(C z@i^A!DkiWkNs@e8YFIU*Db{f~Z{SAI27c`fYNbPOE+Jn&Z9mJEU|XjtNtew8jmtLw zj^l-S0baxl%>ZqLMTYXQrL*i)vY)DBvBp>g)j&8$n&S_`uv zL_KKIoS1FVmtyAV?#AwilJX*lD6;QKF|#E#kpz z7+_5a-Zqdj7!Lr@svefR>rR zFz=vwi;0GM`X+sZ;iKv4TC+OKbmkYU-1wO{5yT2jCkrRPt=2+o z9Xc&n^4zGUK6qs4)Y|Fu*M2;dcK}I+Ok|m$t#~(VlhUrW!zr;;BZ%R|Gm-&z*#)$= z%=pSbereKG|EYzI3|W2+B?)#KbflpxF-jmM`ZOd^%ss^U#~y=JzlZsojc}$`vtc-O znFUIwPljJ&stF}>N;Rwq@sPJ*FS6U*ndTqCVL))j@tnp{Fn8XVTdBh5Q=$%=WRRdOCZy$K+h|cxx3HwjOy0-xko*eNOdr)Z>XI)!af~QRPlg#A`jZ&nV1>d<%bAf7-P_ zr$%%A=*jT>27mPhLObUv|tbJs$9vfl~mk8RSj!a)^4Ls{&pKl-mZ7d=iv%uI__ z^ys98Zb7FXm^&SJ8&@9E9d-^th7 zDlE=vJjcOfTP3It0#noTpbtTOT^L{ZCL*K8!9I;h;o{&tc8V%ctSlF000@1i$c$&s zu!;LC7S#b#_0iv}2r4Q?=~3mWyQ%;x7F(@+(h(0+M~2l=T(hA;s;Cb0Or>435n3vo zI^|m2<_|I(C^2z|1lLA>^kO{ek{(jeO&-tcn5o+6caZ@}>k(ocdq$Xq5I+DWoxqSHD!Ea6z*)D$JKY z^JrJ5i^JG|$)HzKGKsE|%- zl@lu$7@k(*;Sa;Uj7QV>?FknzYB59kLxcvusvzdw$)t#fd8<5Y+G*<$GQ+z1ULSRK zVdn)k=x6cTgi1CnlL@s?$8(@USgi0RMaymRwtE(Qnc;%(9qjiEG7k&C<-?M98_C&3 zo7)Zo9c#C5e%MSvafB8PLs-`X2wd1|u*c%afqiX#It&ZRka4Vp_!dAP-oU+s3=Lv36$CA+`HivP zuxs7B*H;Vp@u&9;qAy2Wq69R%FL1Pq=9waf%cEX1A?ze6q1M(sQcgpt?Ximr5K#k_ z4Dwp9fe~lP8Ms#3f=nfoSrdtD zY-Jt~n-4Vx)u`UxQ2wA#j=es-O9>E*pwnLOu+mhiiH1{`jZ$pVz^K>z4ve&w+j2}z zg`C%XDO`yuC&ZTg!#JeVmCfvNfZbQ8oKoj4UYWU8zhV(?epNbm7%0H-) zS7TJknV?J{emG&-^dG`fKOI^eO>YV8`zqaF7s;Wq>1vhY&;{Y zz8F1)3Hl2nnNE-i86-|9%{rpWIzF7Iey~rHEE;PIDRAEQnU@Vpks{MnKI%@XkQl#5 zG)`LMpmuc5R+zKi+?=aipQXVV4>!p4tkglz>dB0QM<3Lc+^q{nu@q)dQ-%>H+vaoP z8c$YOqi37mX3(GOVab#L0R3apG~69hxhe@BXtHgUqKPizx7crCq6)6akZBJQnLzJ! z+UfbW{TTdCYiVv1_gU|NXPd=^(4QJ5H)*e6#0PifCYhy9Z!*gSm6VV5cF!s3JU^ZB zZ!iGsJkLQE$-zSugT@yq?8GOg!x{Yr{#MA@sx?p=mHZ0T*}M8b-&0TchvQOV~4 z5Ew;c5Ws-9Zz_xX{9c*v#t?!bYk=79w=Brz)R7845ed9lPFkhJ*gE5-rzlpe= zJ1j^g`BJ<+bLGLS-C1w$)WZeNB7IO!$UnxKP$z3H+(RXKrpp)F`{wJ0=eb=#ug7Hm z6#V9Iy6U9&C4?8;3i@^Q2>GGZQc3J(C50-^O`8{f>N&a4orKm|NGPqQxs&vB* zGd=O<7CQz4FFg!5u%DFOJAk8E&V#KIe1BL{J?*Vqw9qJQ*2z<8qZDHGZ>coFx+M-< zWjLo_`!J=IWr>T=hr<^Unp0uKJBTzoS%-w2Pl z{&gIps|=-_9b!pHd!TcSZR&n^Sq5Fe*j$e5`L{Q?zaiiMwLWI~f$j^l)KqSbLKQ3; zSAG6shDX3S)&*Hl=H}j4{+0x9q`h9i3R_fM&>ar^Go-ZR!!|M-x^-Beb)}p!Ot@Iw zI@XTg5pgKD$r5N}{*Dyp4)L0ciF-c0$yDa@6bqd+;h%Dq>@Geq;kT%nMWq@Z-WGki z-z6gHZq#X_lM)0ODxfq&_7>s%(XYR@uPpITm!RDno4zs)Q}6W%ne(P_j_7b^IVRHl zTQh(cl9k9W;>i`06gJLm>|j;<9=io%eETtl?r>BOWNInnKHeQ+HgGiYJyF3x1wIbS zk>JI=#IdRU+OKOmL~{m$tO zPAw1RJ#YujcYBL;INCF5q)~our#0MsbT7*YOx5d9kbGHMA75Fc2G=V0Kepx zP-p43+KIZ3$bV6}a4@8C1x$uAru9!D0-7Aw?b(^9pL8tThrd+sa2G{h;PF8cV8cC8hhUlqce_f_)HFYbhoF&cQIJyS*& zwvx+s>4{q2-1XI(RDQ2wy#I&xRY@LWTo2Qs5DFhFuL;ZAd)u`v2-(tG>56{Nte$}-!X^-P+%3ZgI{*GD^snXS zmufI;v*4N!_nI%TfX9Y(3K~7()lF{dDnCg*F9LZ}(w|)p3x3Or1v4H`4$=efWljmk z8O#l}PS2@79omt@3QUV$W*=MCh#84um#c{cCs@TsrGW@Yk?s2*HT zq`%5yYIX2qAjR%rOYlHhVaZzA*eNy~_>g!g9Lq9R?MW80Gla~DfLX=6&l}?2=m?QH z7E7_9tmX0CTLyPW3+NOG>1vEBs}rU-h{jODEyl1jw)sODjb4hTs09J2Z#s6}Pk#V= zdsU~n<(BM=V>x_*4f&(@2j^Fw@Nn^wa47Fi&nEX~|8TpkbtS>(ZX_oFk~ek$ITFHI`bXXsj1UOl`rf|DQlbMRM!t_}T zFVoKhQ1=zov(G9=g8eD$=ENhXNvlF!V5#F%LJi9sZ#2M{HO zl9KNCjKY2Hy??{|`DvcnYwxqqI&0V3D`H8T5eWC2YW7C17JdRLl2K`0rdn0i@hei} z#V&R40S~e=X&W@$H@>mU?ic)vASO#_2vqGxcM=crkBG!%!Gu&ckrsRV;f0|PqagrD+9-wUwewXX)=&G~TcuF_x ziHZd+Ug}+!+dbIu*qHg2KRwUEkCS7!JL3UP7-jCswtHw3LQ^E*3JUAMaGf}_PEHpe zrEUvXKT%imrB$eQO<&QdCNCWmWkV!wNjcj@uz=pcRK!~?jhugbOCcj~x;cuq++}|T z7!Mix;b{v0b>2((Dv?kGXzdG5!gWd$K!5T#UIH&gQ_ym+K>xJ1q{f*oMfcm+=d%{{ z&{8pb<~hq%K+em?&&eSVo@}!;zEROk!NrqTTcR8jm@#3D{m5Yu)Lw@`b?mfKjg?$q z8F_v!h#Py!1~_4?s67n(sf4q+p-Yw){6XYkPmX5mc)v?0jVBIpvTa!QO+Cz2o8ppz z{QBKRVMa-f=VCH6$S8eIIpDpqGnInebOlpuK$v-riM4SDgW|RiPK1|S^j6(6s z%CW=l&33%p#FZDp$~)}|1B9obaCj-b!gj;>$IUIMoX$ky-o0SqA9J!p(P<@N#-m&z zi2K>mvo3HaD<+MQ>WW^ zUs|fkL19^ebRJ2H=A9$L{Fr~20`VjL)MQUrLJKvztmzM!UBx1lflb;E?pk#Jhg^!G z*mVQl+3AdX8xHTn9!`C`fS zJ!^HY>zAPD#zizSb^VVF?oj(|(={}pmHW&KBr)~?yOxw32kcSy zz@{wUvMKh=jeOW(a@u9P#~(xLh`ynWM@RTiF&~cnkjZC6&=grHd0T2U4N%EvMeD3= z#*?*JD({ZL_b}=k+PEKpL)P52x#ZA};+h$g1iyH%q^(D5T?=v`9}kAiBzh%y{Dev% zFzy38b@Yo`)@oochELO4Q4;RNwC%HsbmV3*tgyWDeyKgOqT(96kP&on9BD0hcmjnG zkE_($m5u>v0HYs=*+Og^g~VaL3+}M#(x+TvHq6i2`<-|Ol25VE{opbiUBFO~FgQyg z@9)S{vUxhGz}s!m@tlx4XYIx5a7YRje@HD`)HOj3#l64097pplv~-1@2yDr{d#U5e zJOZr%vS=C=++F`EZAi^GF@#925|@36Dvi?kQZ*XA{WEk%VZ{O}!| zEi!&d#H1E@tCF=xra#k?wENKk66^23xib<52X^)$Vb`y*SUlQr+BPin*u@Y<3CVlC z7hE{9b*BK`Ht0APyUPHrwfe~a)qfIZ;5rSCS#~yE+zzJMWy?@ZvXZN z_5nx5x|!?I6;wru5N1%?5i=<6jTL7}SjBn)Upu_7V)r7lld4MGIw$aC``ol^)EZw-{b=+4*8hi(4Vg|%x@a#YOWXH2--!~|JE}DO=xCtFH9RlDg|#A0Twc`37a#OT zFuDMFO7uw#HEo(sl&|Y5l&vyVQEts#Bos)I5$#NZ9{3QEDo3SZN|?dicfQj#OC1yg z*7f{Drkkc2yZB{+3UxUF|G2!7)vib z#6hhZb9e>kFh1&~{&BDQ4IU?`d(>OK&YkjjRu9i^cHZDXGKM+^tYzd%w^Xy_{PAek zz+i`~aEf5?e4YQtL)06Pw2CU$P-SC*tFM0b6$kS`U$iy-gaMcKdr!L~lZ)9Jm7fU5 zc}170!9k7b5wE^ZTQT|72ho8P4+3?sJIAlU7I~b?(>W)`a%MU>i~tRjX2{e8p$Tmz zt_P8kbMg7k_A$6A@L*fciN`d*0!W2)mUuB^gK;mNn;r;Z1_dn2I(GS;cedhz5PU^a zu}$$ebw-ak)rrfHFw2w^_onO4(GAOwQ-QPyZZt)hsShoGB1b*G$5m{$Per$$6bHS* z5#v!1?kYlnYQu=(TypcY_qA{};~wKj^^+J9r$pu{+Qss))#ROjoiUeg9lfaM^GfPa zxs?WL`eHaO+~beK1F}l#@-F&1mWsNOINe9y7bKRfO#Nu1|E`MG@8=qT>r+EiBp%}W zNX%Sh%?gDZV78=IQC#)azw@hwUklSj#fS$g{lpTLK{i!Be*(&L&$ZO1w`V(bh7SR!<04wmB@XtogHAx$bdTt+)M(~ z3)2$+{oL62LJhiDNUk6dWtO+HVdnynV)T<+&%OFIMppM8{6;BE2hM-d>Nl^>HQAc! zarI~qORqUoUw}s= zq{&^9!^`+j)Jf}q_wDUIqVw@AHF#jHM=1((`b^p; ze&;X|NqAg;4PHw?IP8nfV-H=rY}(Tq z_y=XS>kQxZ3h?BCgolcSX+Q5Pgtr4uCIckvyB-!<6Am4%*0P-&+_}cQ=~V-GDf#Y; zUk$&E%85Bj;{xHCs;fOKGH|@055Ye@wTD2<@?2ukAkHboT<6q6Sy=h2LJif;;y8kC zj*8_fPF%R#FX^Ge;4C6y_deBOx1{*a^N4^m7z!<#J}?a24jX+<`$opy39h*rj7|WK zk3wz`eEGgoi&^j0GO~+5bn$ry;JbtwN`G&kz19751BPukduB#H2`oS2(iVzLc?+bs z)Q5nd_`>%TB5vl(L!}xkxSb`)*IDP6(6D?TmYOivKwoJ7HU`=g`d*;L6cp8VCx$Cf zd!9J(r1F6`J*dH>7&y`~_ddrJcIPoaUQNVd7)*Z57Ai{dfYIv@pig#VA0=3R*K#;z zAKMfYN#o>6M=0ATT>b2J8Kdoa7p)epYrcM3)Pk_yp1Aq@T`l)(pu4WEka;+S(8BH3 zXZ5{Q&@S9ChXQ0w9a1_KfHOSAT>V{s3DEJl!A6jP8mDxcqv3Yd-j{cUhzzBd-D?gg za5!f3gwIabb??UTpvU~f1t+i_qShIKS5x4?@$vaD<$E9tcP3KSADtDPBOCW+H&MCs zxPo!^w^l*BKTU`l?AT8cy6%c{xdgCrnXk*tJCVSrh7>;0X(h?7IjfxXcCwZ`|JbK+ zu+*zh0_y<(H{;cR-nGE}ju>7QHoGFFO@#AvlOe0ILR?WWnw~Gz_SLpGVGvv2Xdd2R z=$iOIo^1X2tiSUHc{5i|L)CjUFbOoceO$iQt1vz2-Qx5RJnAEVH;X^SDIXIo?2a6a zUAz~nvMF}gDiH518S18&mbxiYv>ImKD`J`rhjC|*~m#JmtXV?k{PoQeRaZA)qL zbs@;t)=t+J*ojyM3qoL*%>S(`UHo-zZ+>wy(CG^%R9S21b}>Wl!tV>G&5bg29a0v= zVl#80#VuIcR);%Jq7M9E%?^<#_rsVE|5RH|Wwt_^K&X z&L(;3Vfqu0!KCr;feqAoN&G}YcEGf|9ZT$gToo=s12=Z<^|r`e3sW3osh0K)E%28i zUlS$P153*~tUQWzImkBdmj));3cEJT)B7(MszGB|J5+xifmr4C zN_^U;2^8o$9)}e@eLzcx=4LN_x+$=1T97EwN4UmQLOm>9xP{X$#kXrcH{=!vWY4~? zpUhwE=STIUn{r_yM{0OAvGcrlx=^5Y_3V$4PP8}@51iR`+>&b9QibU-ZUGGf5q8i5 z&f4J~+HP>?odJRE7VVcG5Q03s6GhDwWO>z^Xj^`|inrg9Vh8*4W`b4hox3HP&PERA*!4X-=r z!Bdi4hl9Vc`b+drCzbIWRK)wSZ4VIDq`m6*`Qh9xLM!0sGA;+&#pLf#R81HTNo!b} zN)+_Pynm$oI@77EF?L?S%xm;$XcMm`!X8nJ95cq5N;%%6GBZ$ea_S{q*G2WF6P{iO zkQUVup5~+TE}Sf?<#xJ}QH4~wB=Frue$i~i`Y6TEscR)jp;MhQ-u>n+L3H&PGe(Cl znrYONQ*_B#_T3j-&|s0zV)ZkvkImV73JLR~j(&$3RP<8InQ@Wg9MP0!oJx9XDUIb# zqcJ4;3C?%?a69Q59xPQHWz@{$G};qqDIA3tzn16`gi@(y*+@)zVCsq-QgAg@goyPL zV*~i=o_~?%Pi>BTGW?jb%TYbJxIN1Apckb_MFT?if*EPoQ7wB;TK=IkZY(InQF1xC zyabBprU*BV0@nBx@h8$csy~Z%m^4!(iR=y}1bF%Bi3uIbCfn9i`m2{h=U^S=mi_~z z`*J&8ujf}73G~K*Q_#H7qr=01V)N>3D-OJF-7cRGnaAT!1bg_i9rySZ0ec1dQ<2(= zVkXuqL?NkvYwckgX4`N>sV3`>?C`~~XOR}j%w5I#?VR8|GQD!Tt7Qf(t&3>!DRLis|-}MoW38%Pwi) zFx^*yrI;aq*o5;d0Z@Vygfx#hKYD!k38*iCu6XGlv$2VKLsW_f{v&fniupNz+MO>s zq7hPS;z$CdSlbtK*xj``13&sHhVj{khXqVe7dF&uY}ZOEnWDxOdbf?I6Uvu$L@;w| z7eL;|7F~wTKoY~%O=Y*AqHlga9Ve`om>w_8l-T})ETZ7Na5$3kkf8<7EiFl!9d{~F zfui(eGCT$?ZV~EvvSd~n@`YVnaAu5!xz75hp48nS3mTG)i0nlyy_}k`PEWaB-i(tZ z9<>ghihxJXcaowantPbk8)MR~G+wR8?c|1eGis+gJ?&Ryjb?-0gv{8J1y_eoa7nDP z`lnC1tH*eROtR}fId3&OZ!y-lcbL+vcBCg90JZHKgbmYN*Bv=_Wk6+DL5cdnGK0-e zN}9JNcYZYK6>KSskW6r&JVSqM^jmpY*<j8!nWG2R^kumZtmsjdn z!?#>#b%`S5Vh>8Ji8yFN0vAq}Ln7iQ;IJ}qx5*jps&>Yiq*6Mls*j;p(qRforf1B9_=4+k@yH1Ogq0dr=6rJRf%PUa*TaJd28r9_sUp)N8-2U;ZFi zSnb8f`cKc^VYYo}%}d+!`S-o?%TBnVM>8TH3D9*gnoYoF z9Sbu)TzG<+9NOsZEFuh6ivKi2TX(VJfbTrCiYN-{{H{XLA|?oM)%pvmWjQ@bg1X3` zIw^vLat%@3juKE1Zkav9p|ovO@sh9w-MStf;F5tw5ctgUtJZ5Co=-I=`F6XmYAXhG z9bO_u?`_v1Of4z&0vmq!#vEJ z6|`LNl@7!XC6=Q9>q{f90}S7>7)+9<&#@gYb)dF(C-duz`Y6!dBCA8L6LRcbevML z(M}rAp~@W>p2w9_3w$p18XUmG5DQjXpe@aswdtvC5usT5?xPQUCqXXiM#w>!NF0*o zR-cQ+9`j`%|I)rFZ@2%29802}Konj3uJJJ3OF&yqEE*f+28)ToXEcli26^BaWWFLR z%bh5WsB2T1h;sYH`f@v`(;VP~*RR_CqXjMGD$=vR|J@C(E7Mi-wUs9Q-B8POdM<%y z2?g?q{Y4ybVah(SpOkj>7V(DRvsM=`<0hQdRxB*s9!3TM0jB=Kl0EL)%zG6gFu|P( zCJAm9Sz6T;%hDK+PmsHp6vj7HLmhkXNZ;AQX}qidvf4^R&tauLq97vrlECdTYcy5w z17s%8M|=6(!$<5+54IdjlQ{es08`zxwH8gq~KAG{7VCuTnS@{;!l5VjS4eXn*Z;p-!ik^>JV<^3b)#(0J3F^;}Qzq6~Gn}oAux}MO;)?umL0#BHe%-Nkp)AgR915dyE)RpKOK*#Rv=58~U8@cGPrgWrB6lF1qVZG(C$N-G< z66RJDS#Mh96ONnMJ@sn6mwmski@#e7yv3U1_8x+^m##`K7`Pgy{zH}V5S?4W=M|zJ z8Zvdr;>XLp+MQg>PjB5n0fMIP8L6!RQzques4UjO4vRr9Kqg)OgB(APF7NPrqFTZX zYoxBrWMxMfwB{se&}aQ40}WCFbd~9l<0<;g9Jt>}B~50C{cdIg`IT$g_hfdq*FUg)d+%qo<0 zV=j4=9_~vqKM?BG9DnE|q^CG~PqpSJ)F=1xa&%60B?PZ(Q%B`;fSCaScB4KIevbWS zi#izevUUfAH$WP~*Q2sEU4@n_M<9CWS2Nd^$kjZlg#Vo`zEE+p zOs3_BH~tVm{eYXjVNQ+D>vfBPGQy7!19KiP7hA5MM@DN{5B5{h5;HWaWObD+5U ze$2h58m_idw>9SZ%qUc2&R%z%vkyryZcTt(g$-5$@yoCXQXtM2($h8gT^-ssDjcwZ zthoo%b1#wAv#SfAfm|=;v#0dYm!Bvc(Muz>5qfpmyEay!ca2#@dH#+-&`-AaMfpv@ zV%=!8TB5L)I7VZk`x-Hj7mlkw&%S#LmeRoptfzdPejXWWSDjR%RJAZtza#HXXCCz~ zMEihptgb-DV89hy8T}Fxf&5%^M%3~#YFWnrwgNT+DHhAXeASpi@` zaUqP^qLeGv4HFBc8>)DKBJ?}@dGVj24y`$QtepE-3$%vp1qJRQ(6J+Vz5j=#>8ll<~>(ULdC{|*YT0*34Gs=e>n zbv{2Zm$5<6W!2q9xg_B>vyC~edhcJ4dF?4beAF@ZojT+{6x+ILz8QE)&RHeC@_on) zKAB;xXs`RvW1Zjr;sK>9-UNRc!59rq;#iWdDK%`0nLGRPeFk{|Bg7fuqO&o50p4JM zk*gLg46_PP?vGtATp(=cf3kaStCj%|F8vPmcPQI7c0*VbUkdghN3bx+#Vem+%!H&g z$A-IP6!k;2<+2b&M*x%_??cTHzv|WOBP$M~H-8qq>bmN>+j8IheAXgq+j~AmyOrcs zdlpZx^#JHy&=o-maW+ml&njEZQeoLTVzTnkFcFA={TIjg;;w<@3Pa!=*b!sXwTPs_ zX6wWv$mG;g2(RtBHCN8Ot)~Rb_NSr3D*SQlH6W~X+&3Rj-W8M(hOK_#Xk7k8ZfJW5&}Z7{e{dhP-Dz96>@CS?a~-Ee$`T_ zHqfgvPc79PVe>2QMce_nmr|ezZ_Q4?;PJJV0ZM>>NKj@a2Z?$nLL^nWz`n5XTCSYy zl&`Ni`-vzUM_$U4E2ll*V9W|gQ)!d_800wS1>{@%3z99&_V$nu5TW%xS5CZb&XMTw zKyire(zX|{@3^EP03feyq zRvlh(S6Kz<*(HK_11jk*g%||tR`|b5fB|)dZpHsC0LBCF|FQ&-i2p|z_91E)cqiif V96l}f7Fcn4@>t_hsiIlP{{XRCD6s$l literal 0 HcmV?d00001 diff --git a/docs/docsgen/source/_static/readme.txt b/docs/docsgen/source/_static/readme.txt new file mode 100644 index 0000000..c0a8962 --- /dev/null +++ b/docs/docsgen/source/_static/readme.txt @@ -0,0 +1,16 @@ +diff2html: https://github.com/rtfpessoa/diff2html + +Copyright 2014-2016 Rodrigo Fernandes https://rtfpessoa.github.io/ + +Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated +documentation files (the "Software"), to deal in the Software without restriction, including without limitation the +rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit +persons to whom the Software is furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all copies or substantial portions of the +Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE +WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR +COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR +OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. diff --git a/docs/docsgen/source/api/backend.md b/docs/docsgen/source/api/backend.md new file mode 100644 index 0000000..6bd284e --- /dev/null +++ b/docs/docsgen/source/api/backend.md @@ -0,0 +1,35 @@ +# onnx.backend + +## Backend + +```{eval-rst} +.. autoclass:: onnx.backend.base.Backend + :members: +``` + +## BackendRep + +```{eval-rst} +.. autoclass:: onnx.backend.base.BackendRep + :members: +``` + +## Device + +```{eval-rst} +.. autoclass:: onnx.backend.base.Device + :members: +``` + +## DeviceType + +```{eval-rst} +.. autoclass:: onnx.backend.base.DeviceType + :members: +``` + +## load_model_tests + +```{eval-rst} +.. autofunction:: onnx.backend.test.loader.load_model_tests +``` diff --git a/docs/docsgen/source/api/checker.md b/docs/docsgen/source/api/checker.md new file mode 100644 index 0000000..dc10b83 --- /dev/null +++ b/docs/docsgen/source/api/checker.md @@ -0,0 +1,17 @@ +# onnx.checker + +## CheckerContext + +```{eval-rst} +.. autoclass:: onnx.checker.DEFAULT_CONTEXT + :members: +``` + +## The `onnx.checker` module + +```{eval-rst} +.. automodule:: onnx.checker + :members: + :undoc-members: + :show-inheritance: +``` diff --git a/docs/docsgen/source/api/classes.md b/docs/docsgen/source/api/classes.md new file mode 100644 index 0000000..3e5bb54 --- /dev/null +++ b/docs/docsgen/source/api/classes.md @@ -0,0 +1,294 @@ +(l-onnx-classes)= + +# Protos + +This structures are defined with protobuf in files `onnx/*.proto`. +It is recommended to use function in module {ref}`l-mod-onnx-helper` +to create them instead of directly instantiated them. +Every structure can be printed with function `print` and is rendered +as a json string. + +## AttributeProto + +This class is used to define an attribute of an operator +defined itself by a NodeProto. It is +a named attribute containing either singular float, integer, string, graph, +and tensor values, or repeated float, integer, string, graph, and tensor values. +An AttributeProto MUST contain the name field, and *only one* of the +following content fields, effectively enforcing a C/C++ union equivalent. + +```{eval-rst} +.. autoclass:: onnx.AttributeProto + :members: +``` + +(l-onnx-function-proto)= + +## FunctionProto + +This defines a function. It is not a model but can +be used to define custom operators used in a model. + +```{eval-rst} +.. autoclass:: onnx.FunctionProto + :members: +``` + +(l-onnx-graph-proto)= + +## GraphProto + +This defines a graph or a set of nodes called from a loop or a test +for example. +A graph defines the computational logic of a model and is comprised of a parameterized +list of nodes that form a directed acyclic graph based on their inputs and outputs. +This is the equivalent of the *network* or *graph* in many deep learning +frameworks. + +```{eval-rst} +.. autoclass:: onnx.GraphProto + :members: +``` + +(l-onnx-map-proto)= + +## MapProto + +This defines a map or a dictionary. It +specifies an associative table, defined by keys and values. +MapProto is formed with a repeated field of keys (of type INT8, INT16, INT32, +INT64, UINT8, UINT16, UINT32, UINT64, or STRING) and values (of type TENSOR, +SPARSE_TENSOR, SEQUENCE, or MAP). Key types and value types have to remain +the same throughout the instantiation of the MapProto. + +```{eval-rst} +.. autoclass:: onnx.MapProto + :members: +``` + +(l-modelproto)= + +## ModelProto + +This defines a model. That is the type every converting library +returns after converting a machine learned model. +ModelProto is a top-level file/container format for bundling a ML model and +associating its computation graph with metadata. +The semantics of the model are described by the associated GraphProto's. + +```{eval-rst} +.. autoclass:: onnx.ModelProto + :members: +``` + +(l-nodeproto)= + +## NodeProto + +This defines an operator. A model is a combination of +mathematical functions, each of them represented as an onnx operator, +stored in a NodeProto. +Computation graphs are made up of a DAG of nodes, which represent what is +commonly called a *layer* or *pipeline stage* in machine learning frameworks. +For example, it can be a node of type *Conv* that takes in an image, a filter +tensor and a bias tensor, and produces the convolved output. + +```{eval-rst} +.. autoclass:: onnx.NodeProto + :members: +``` + +(l-operatorproto)= + +## OperatorProto + +This class is rarely used by users. +An OperatorProto represents the immutable specification of the signature +and semantics of an operator. +Operators are declared as part of an OperatorSet, which also defines the +domain name for the set. +Operators are uniquely identified by a three part identifier +(domain, op_type, since_version) where + +- *domain* is the domain of an operator set that contains this operator specification. +- *op_type* is the name of the operator as referenced by a NodeProto.op_type +- *since_version* is the version of the operator set that this operator was initially declared in. + +```{eval-rst} +.. autoclass:: onnx.OperatorProto + :members: +``` + +(l-operatorsetidproto)= + +## OperatorSetIdProto + +This is the type of attribute `opset_import` of class ModelProto. +This attribute specifies the versions of operators used in the model. +Every operator or node belongs to a domain. All operators for the same +domain share the same version. + +```{eval-rst} +.. autoclass:: onnx.OperatorSetIdProto + :members: +``` + +(l-operatorsetproto)= + +## OperatorSetProto + +An OperatorSetProto represents an immutable set of immutable operator specifications. +The domain of the set (OperatorSetProto.domain) is a reverse-DNS name +that disambiguates operator sets defined by independent entities. +The version of the set (opset_version) is a monotonically increasing +integer that indicates changes to the membership of the operator set. +Operator sets are uniquely identified by a two part identifier (domain, opset_version) +Like ModelProto, OperatorSetProto is intended as a top-level file/wire format, +and thus has the standard format headers in addition to the operator set information. + +```{eval-rst} +.. autoclass:: onnx.OperatorSetProto + :members: +``` + +(l-optionalproto)= + +## OptionalProto + +Some input or output of a model are optional. This class must +be used in this case. An instance of class OptionalProto +may contain or not an instance of type TensorProto, SparseTensorProto, +SequenceProto, MapProto and OptionalProto. + +```{eval-rst} +.. autoclass:: onnx.OptionalProto + :members: +``` + +(l-onnx-sequence-proto)= + +## SequenceProto + +This defines a dense, ordered, collection of elements that are of homogeneous types. +Sequences can be made out of tensors, maps, or sequences. +If a sequence is made out of tensors, the tensors must have the same element +type (i.e. int32). In some cases, the tensors in a sequence can have different +shapes. Whether the tensors can have different shapes or not depends on the +type/shape associated with the corresponding `ValueInfo`. For example, +`Sequence` means that all tensors have same shape. However, +`Sequence` means they can have different +shapes (all of rank 2), where *omitted* means the corresponding dimension has +no symbolic/constant value. Finally, `Sequence>` means +that the different tensors can have different ranks, when the *shape* itself +is omitted from the tensor-type. For a more complete description, refer to +[Static tensor shapes](https://github.com/onnx/onnx/blob/main/docs/IR.md#static-tensor-shapes). + +```{eval-rst} +.. autoclass:: onnx.SequenceProto + :members: +``` + +(l-onnx-sparsetensor-proto)= + +## SparseTensorProto + +This defines a sparse tensor. +The sequence of non-default values are encoded as a tensor of shape `[NNZ]`. +The default-value is zero for numeric tensors, and empty-string for string tensors. +values must have a non-empty name present which serves as a name for SparseTensorProto +when used in sparse_initializer list. + +```{eval-rst} +.. autoclass:: onnx.SparseTensorProto + :members: +``` + +(l-onnx-stringstringentry-proto)= + +## StringStringEntryProto + +This is equivalent to a pair of strings. +This is used to store metadata in ModelProto. + +```{eval-rst} +.. autoclass:: onnx.StringStringEntryProto + :members: +``` + +(l-tensorproto)= + +## TensorProto + +This defines a tensor. A tensor is fully described with a shape +(see ShapeProto), the element type (see TypeProto), and the +elements themselves. All available types are listed in +{ref}`l-mod-onnx-mapping`. + +```{eval-rst} +.. autoclass:: onnx.TensorProto + :members: +``` + +(l-tensorshapeproto)= + +## TensorShapeProto + +This defines the shape of a tensor or a sparse tensor. +It is a list of dimensions. A dimension can be either an integer value +or a symbolic variable. A symbolic variable represents an unknown +dimension. + +```{eval-rst} +.. autoclass:: onnx.TensorShapeProto + :members: +``` + +(l-traininginfoproto)= + +## TrainingInfoProto + +TrainingInfoProto stores information for training a model. +In particular, this defines two functionalities: an initialization-step +and a training-algorithm-step. Initialization resets the model +back to its original state as if no training has been performed. +Training algorithm improves the model based on input data. +The semantics of the initialization-step is that the initializers +in ModelProto.graph and in TrainingInfoProto.algorithm are first +initialized as specified by the initializers in the graph, and then +updated by the *initialization_binding* in every instance in +ModelProto.training_info. +The field *algorithm* defines a computation graph which represents a +training algorithm's step. After the execution of a +TrainingInfoProto.algorithm, the initializers specified by *update_binding* +may be immediately updated. If the targeted training algorithm contains +consecutive update steps (such as block coordinate descent methods), +the user needs to create a TrainingInfoProto for each step. + +```{eval-rst} +.. autoclass:: onnx.TrainingInfoProto + :members: +``` + +(l-typeproto)= + +## TypeProto + +This defines a type of a tensor which consists in an element type +and a shape (ShapeProto). + +```{eval-rst} +.. autoclass:: onnx.TypeProto + :members: +``` + +(l-valueinfoproto)= + +## ValueInfoProto + +This defines a input or output type of a GraphProto. +It contains a name, a type (TypeProto), and a documentation string. + +```{eval-rst} +.. autoclass:: onnx.ValueInfoProto + :members: +``` diff --git a/docs/docsgen/source/api/compose.md b/docs/docsgen/source/api/compose.md new file mode 100644 index 0000000..b83dff0 --- /dev/null +++ b/docs/docsgen/source/api/compose.md @@ -0,0 +1,44 @@ +# onnx.compose + +```{eval-rst} +.. currentmodule:: onnx.compose +``` + +```{eval-rst} +.. autosummary:: + + merge_graphs + merge_models +``` + +## merge_graphs + +```{eval-rst} +.. autofunction:: onnx.compose.merge_graphs +``` + +## merge_models + +```{eval-rst} +.. autofunction:: onnx.compose.merge_models +``` + +## prefix + +```{eval-rst} +.. autofunction:: onnx.compose.add_prefix_graph +``` + +```{eval-rst} +.. autofunction:: onnx.compose.add_prefix +``` + +## dimension + +```{eval-rst} +.. autofunction:: onnx.compose.expand_out_dim +``` + +```{eval-rst} +.. autofunction:: onnx.compose.expand_out_dim_graph +``` diff --git a/docs/docsgen/source/api/defs.md b/docs/docsgen/source/api/defs.md new file mode 100644 index 0000000..8291a30 --- /dev/null +++ b/docs/docsgen/source/api/defs.md @@ -0,0 +1,62 @@ +(l-mod-onnx-defs)= + +# onnx.defs + +(l-api-opset-version)= + +## Opset Version + +```{eval-rst} +.. autofunction:: onnx.defs.onnx_opset_version +``` + +## Operators and Functions Schemas + +```{eval-rst} +.. autofunction:: onnx.defs.has + +.. autofunction:: onnx.defs.get_schema + +.. autofunction:: onnx.defs.get_all_schemas + +.. autofunction:: onnx.defs.get_all_schemas_with_history + +.. autofunction:: onnx.defs.get_function_ops + +.. autofunction:: onnx.defs.register_schema + +.. autofunction:: onnx.defs.deregister_schema +``` + +## class `OpSchema` + +```{eval-rst} +.. autoclass:: onnx.defs.OpSchema + :members: + :undoc-members: +``` + +## Exceptions + +```{eval-rst} +.. autoclass:: onnx.defs.SchemaError +``` + +## Constants + +Domains officially supported in onnx package. + +```{eval-rst} +.. exec_code:: + + from onnx.defs import ( + ONNX_DOMAIN, + ONNX_ML_DOMAIN, + AI_ONNX_PREVIEW_DOMAIN, + AI_ONNX_PREVIEW_TRAINING_DOMAIN, + ) + print(f"ONNX_DOMAIN={ONNX_DOMAIN!r}") + print(f"ONNX_ML_DOMAIN={ONNX_ML_DOMAIN!r}") + print(f"AI_ONNX_PREVIEW_DOMAIN={AI_ONNX_PREVIEW_DOMAIN!r}") + print(f"AI_ONNX_PREVIEW_TRAINING_DOMAIN={AI_ONNX_PREVIEW_TRAINING_DOMAIN!r}") +``` diff --git a/docs/docsgen/source/api/external_data_helper.md b/docs/docsgen/source/api/external_data_helper.md new file mode 100644 index 0000000..cb3cf58 --- /dev/null +++ b/docs/docsgen/source/api/external_data_helper.md @@ -0,0 +1,61 @@ +# onnx.external_data_helper + +## convert_model_from_external_data + +```{eval-rst} +.. autofunction:: onnx.external_data_helper.convert_model_from_external_data +``` + +## convert_model_to_external_data + +```{eval-rst} +.. autofunction:: onnx.external_data_helper.convert_model_to_external_data +``` + +## ExternalDataInfo + +```{eval-rst} +.. autoclass:: onnx.external_data_helper.ExternalDataInfo +``` + +## load_external_data_for_model + +```{eval-rst} +.. autofunction:: onnx.external_data_helper.load_external_data_for_model +``` + +## load_external_data_for_tensor + +```{eval-rst} +.. autofunction:: onnx.external_data_helper.load_external_data_for_tensor +``` + +## remove_external_data_field + +```{eval-rst} +.. autofunction:: onnx.external_data_helper.remove_external_data_field +``` + +## save_external_data + +```{eval-rst} +.. autofunction:: onnx.external_data_helper.save_external_data +``` + +## set_external_data + +```{eval-rst} +.. autofunction:: onnx.external_data_helper.set_external_data +``` + +## uses_external_data + +```{eval-rst} +.. autofunction:: onnx.external_data_helper.uses_external_data +``` + +## write_external_data_tensors + +```{eval-rst} +.. autofunction:: onnx.external_data_helper.write_external_data_tensors +``` diff --git a/docs/docsgen/source/api/helper.md b/docs/docsgen/source/api/helper.md new file mode 100644 index 0000000..90ea20b --- /dev/null +++ b/docs/docsgen/source/api/helper.md @@ -0,0 +1,163 @@ +(l-mod-onnx-helper)= + +# onnx.helper + +```{eval-rst} +.. currentmodule:: onnx.helper +``` + +(l-onnx-make-function)= + +## Helper functions to make ONNX graph components + +All functions used to create an ONNX graph. + +```{eval-rst} +.. autofunction:: onnx.helper.make_attribute +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_attribute_ref +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_empty_tensor_value_info +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_function +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_graph +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_map +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_map_type_proto +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_model +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_node +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_operatorsetid +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_opsetid +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_model_gen_version +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_optional +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_optional_type_proto +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_sequence +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_sequence_type_proto +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_sparse_tensor +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_sparse_tensor_type_proto +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_sparse_tensor_value_info +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_tensor +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_tensor_sequence_value_info +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_tensor_type_proto +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_training_info +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_tensor_value_info +``` + +```{eval-rst} +.. autofunction:: onnx.helper.make_value_info +``` + +## Type Mappings + +```{eval-rst} +.. autofunction:: onnx.helper.get_all_tensor_dtypes +``` + +```{eval-rst} +.. autofunction:: onnx.helper.np_dtype_to_tensor_dtype +``` + +```{eval-rst} +.. autofunction:: onnx.helper.tensor_dtype_to_field +``` + +```{eval-rst} +.. autofunction:: onnx.helper.tensor_dtype_to_np_dtype +``` + +```{eval-rst} +.. autofunction:: onnx.helper.tensor_dtype_to_storage_tensor_dtype +``` + +```{eval-rst} +.. autofunction:: onnx.helper.tensor_dtype_to_string +``` + +## Tools + +```{eval-rst} +.. autofunction:: onnx.helper.find_min_ir_version_for +``` + +## Other functions + +```{eval-rst} +.. autosummary:: + + get_attribute_value + get_node_attr_value + set_metadata_props + set_model_props + printable_attribute + printable_dim + printable_graph + printable_node + printable_tensor_proto + printable_type + printable_value_info +``` diff --git a/docs/docsgen/source/api/index.md b/docs/docsgen/source/api/index.md new file mode 100644 index 0000000..e36404e --- /dev/null +++ b/docs/docsgen/source/api/index.md @@ -0,0 +1,71 @@ +(l-python-onnx-api)= + +# API Reference + +```{tip} +The [ir-py project](https://github.com/onnx/ir-py) provides alternative Pythonic APIs for creating and manipulating ONNX models without interaction with Protobuf. +``` + +## Versioning + +The following example shows how to retrieve onnx version, +the onnx opset, the IR version. Every new major release increments the opset version +(see {ref}`l-api-opset-version`). + +```{eval-rst} +.. exec_code:: + + from onnx import __version__, IR_VERSION + from onnx.defs import onnx_opset_version + print(f"onnx.__version__={__version__!r}, opset={onnx_opset_version()}, IR_VERSION={IR_VERSION}") +``` + +The intermediate representation (IR) specification is the abstract model for +graphs and operators and the concrete format that represents them. +Adding a structure or modifying one of them increases the IR version. + +The opset version increases when an operator is added or removed or modified. +A higher opset means a longer list of operators and more options to +implement an ONNX functions. An operator is usually modified because it +supports more input and output type, or an attribute becomes an input. + +## Data Structures + +Every ONNX object is defined based on a [protobuf message](https://googleapis.dev/python/protobuf/latest/google/protobuf/message.html) +and has a name ended with suffix `Proto`. For example, {ref}`l-nodeproto` defines +an operator, {ref}`l-tensorproto` defines a tensor. Next page lists all of them. + +```{toctree} +:maxdepth: 1 + +classes +serialization +``` + +## Functions + +An ONNX model can be created directly from the classes described +in the previous section, but it is faster to create and +verify a model with the following helpers. + +```{toctree} +:maxdepth: 1 + +backend +checker +compose +defs +external_data_helper +helper +inliner +mapping +model_container +numpy_helper +parser +printer +reference +shape_inference +tools +utils +version_converter +``` diff --git a/docs/docsgen/source/api/inliner.md b/docs/docsgen/source/api/inliner.md new file mode 100644 index 0000000..6ae1050 --- /dev/null +++ b/docs/docsgen/source/api/inliner.md @@ -0,0 +1,13 @@ +# onnx.inliner + +## inline_local_functions + +```{eval-rst} +.. autofunction:: onnx.inliner.inline_local_functions +``` + +## inline_selected_functions + +```{eval-rst} +.. autofunction:: onnx.inliner.inline_selected_functions +``` diff --git a/docs/docsgen/source/api/model_container.md b/docs/docsgen/source/api/model_container.md new file mode 100644 index 0000000..48641cd --- /dev/null +++ b/docs/docsgen/source/api/model_container.md @@ -0,0 +1,20 @@ +# onnx.model_container + +## ModelContainer + +```{eval-rst} +.. autoclass:: onnx.model_container.ModelContainer + :members: +``` + +## make_large_model + +```{eval-rst} +.. autofunction:: onnx.model_container.make_large_model +``` + +## make_large_tensor_proto + +```{eval-rst} +.. autofunction:: onnx.model_container.make_large_tensor_proto +``` diff --git a/docs/docsgen/source/api/numpy_helper.md b/docs/docsgen/source/api/numpy_helper.md new file mode 100644 index 0000000..db521f9 --- /dev/null +++ b/docs/docsgen/source/api/numpy_helper.md @@ -0,0 +1,63 @@ +# onnx.numpy_helper + +```{eval-rst} +.. currentmodule:: onnx.numpy_helper +``` + +```{eval-rst} +.. autosummary:: + + from_array + from_dict + from_list + from_optional + to_array + to_dict + to_list + to_optional + +``` + +(l-numpy-helper-onnx-array)= + +## array + +```{eval-rst} +.. autofunction:: onnx.numpy_helper.from_array +``` + +```{eval-rst} +.. autofunction:: onnx.numpy_helper.to_array +``` + +Arrays with data types not supported natively by NumPy will be return with ``ml_dtypes`` dtypes. + +## sequence + +```{eval-rst} +.. autofunction:: onnx.numpy_helper.to_list +``` + +```{eval-rst} +.. autofunction:: onnx.numpy_helper.from_list +``` + +## dictionary + +```{eval-rst} +.. autofunction:: onnx.numpy_helper.to_dict +``` + +```{eval-rst} +.. autofunction:: onnx.numpy_helper.from_dict +``` + +## optional + +```{eval-rst} +.. autofunction:: onnx.numpy_helper.to_optional +``` + +```{eval-rst} +.. autofunction:: onnx.numpy_helper.from_optional +``` diff --git a/docs/docsgen/source/api/parser.md b/docs/docsgen/source/api/parser.md new file mode 100644 index 0000000..8660c7e --- /dev/null +++ b/docs/docsgen/source/api/parser.md @@ -0,0 +1,25 @@ +# onnx.parser + +## parse_node + +```{eval-rst} +.. autofunction:: onnx.parser.parse_node +``` + +## parse_function + +```{eval-rst} +.. autofunction:: onnx.parser.parse_function +``` + +## parse_graph + +```{eval-rst} +.. autofunction:: onnx.parser.parse_graph +``` + +## parse_model + +```{eval-rst} +.. autofunction:: onnx.parser.parse_model +``` diff --git a/docs/docsgen/source/api/printer.md b/docs/docsgen/source/api/printer.md new file mode 100644 index 0000000..76a7367 --- /dev/null +++ b/docs/docsgen/source/api/printer.md @@ -0,0 +1,7 @@ +# onnx.printer + +## to_text + +```{eval-rst} +.. autofunction:: onnx.printer.to_text +``` diff --git a/docs/docsgen/source/api/reference.md b/docs/docsgen/source/api/reference.md new file mode 100644 index 0000000..2989c61 --- /dev/null +++ b/docs/docsgen/source/api/reference.md @@ -0,0 +1,45 @@ +(l-reference-implementation)= + +# onnx.reference + +## DefaultNone + +```{eval-rst} +.. autoclass:: onnx.reference.op_run.DefaultNone + :members: +``` + +## ReferenceEvaluator + +```{eval-rst} +.. autoclass:: onnx.reference.ReferenceEvaluator + :members: input_names, output_names, opsets, run +``` + +## OpFunction + +```{eval-rst} +.. autoclass:: onnx.reference.op_run.OpFunction + :members: create, eval, input, output, implicit_inputs, domain, need_context, run, make_node +``` + +## OpRun + +```{eval-rst} +.. autoclass:: onnx.reference.op_run.OpRun + :members: create, eval, input, output, implicit_inputs, domain, need_context, run, make_node +``` + +## RuntimeTypeError + +```{eval-rst} +.. autoclass:: onnx.reference.op_run.RuntimeTypeError + :members: +``` + +## SparseTensor + +```{eval-rst} +.. autoclass:: onnx.reference.op_run.SparseTensor + :members: +``` diff --git a/docs/docsgen/source/api/serialization.md b/docs/docsgen/source/api/serialization.md new file mode 100644 index 0000000..2359047 --- /dev/null +++ b/docs/docsgen/source/api/serialization.md @@ -0,0 +1,119 @@ +(l-serialization)= + +# Serialization + +## Save a model and any Proto class + +This ONNX graph needs to be serialized into one contiguous +memory buffer. Method `SerializeToString` is available +in every ONNX objects. + +``` +with open("model.onnx", "wb") as f: + f.write(onnx_model.SerializeToString()) +``` + +This method has the following signature. + +```{eval-rst} +.. autoclass:: onnx.ModelProto + :members: SerializeToString +``` + +Every Proto class implements method `SerializeToString`. +Therefore the following code works with any class described +in page {ref}`l-onnx-classes`. + +``` +with open("proto.pb", "wb") as f: + f.write(proto.SerializeToString()) +``` + +Next example shows how to save a {ref}`l-nodeproto`. + +```{eval-rst} +.. exec_code:: + + from onnx import NodeProto + + node = NodeProto() + node.name = "example-type-proto" + node.op_type = "Add" + node.input.extend(["X", "Y"]) + node.output.extend(["Z"]) + + with open("node.pb", "wb") as f: + f.write(node.SerializeToString()) +``` + +## Load a model + +Following function only automates the loading of a class +{ref}`l-modelproto`. Next sections shows how to restore +any other proto class. + +```{eval-rst} +.. autofunction:: onnx.load +``` + +``` +from onnx import load + +onnx_model = load("model.onnx") +``` + +Or: + +``` +from onnx import load + +with open("model.onnx", "rb") as f: + onnx_model = load(f) +``` + +Next function does the same from a bytes array. + +```{eval-rst} +.. autofunction:: onnx.load_model_from_string + +``` + +(l-onnx-load-data)= + +## Load a Proto + +Proto means here any type containing data including a model, a tensor, +a sparse tensor, any class listed in page {ref}`l-onnx-classes`. +The user must know the type of the data he needs to restore +and then call method `ParseFromString`. +[protobuf](https://developers.google.com/protocol-buffers) +does not store any information about the class +of the saved data. Therefore, this class must be known before +restoring an object. + +```{eval-rst} +.. autoclass:: onnx.ModelProto + :members: ParseFromString +``` + +Next example shows how to restore a {ref}`l-nodeproto`. + +```{eval-rst} +.. exec_code:: + + from onnx import NodeProto + + tp2 = NodeProto() + with open("node.pb", "rb") as f: + content = f.read() + + tp2.ParseFromString(content) + + print(tp2) +``` + +A shortcut exists for {ref}`l-tensorproto`: + +```{eval-rst} +.. autofunction:: onnx.load_tensor_from_string +``` diff --git a/docs/docsgen/source/api/shape_inference.md b/docs/docsgen/source/api/shape_inference.md new file mode 100644 index 0000000..41f9091 --- /dev/null +++ b/docs/docsgen/source/api/shape_inference.md @@ -0,0 +1,25 @@ +# onnx.shape_inference + +## infer_shapes + +```{eval-rst} +.. autofunction:: onnx.shape_inference.infer_shapes +``` + +## infer_shapes_path + +```{eval-rst} +.. autofunction:: onnx.shape_inference.infer_shapes_path +``` + +## infer_node_outputs + +```{eval-rst} +.. autofunction:: onnx.shape_inference.infer_node_outputs +``` + +## infer_function_output_types + +```{eval-rst} +.. autofunction:: onnx.shape_inference.infer_function_output_types +``` diff --git a/docs/docsgen/source/api/tools.md b/docs/docsgen/source/api/tools.md new file mode 100644 index 0000000..0b23db2 --- /dev/null +++ b/docs/docsgen/source/api/tools.md @@ -0,0 +1,14 @@ +# onnx.tools + + +## update_inputs_outputs_dims + +```{eval-rst} +.. autofunction:: onnx.tools.update_model_dims.update_inputs_outputs_dims +``` + +## replace_initializer_by_constant_of_shape + +```{eval-rst} +.. autofunction:: onnx.tools.replace_constants.replace_initializer_by_constant_of_shape +``` diff --git a/docs/docsgen/source/api/utils.md b/docs/docsgen/source/api/utils.md new file mode 100644 index 0000000..b08d1ab --- /dev/null +++ b/docs/docsgen/source/api/utils.md @@ -0,0 +1,14 @@ +# onnx.utils + +## Extractor + +```{eval-rst} +.. autoclass:: onnx.utils.Extractor + :members: +``` + +## extract_model + +```{eval-rst} +.. autofunction:: onnx.utils.extract_model +``` diff --git a/docs/docsgen/source/api/version_converter.md b/docs/docsgen/source/api/version_converter.md new file mode 100644 index 0000000..8b7754d --- /dev/null +++ b/docs/docsgen/source/api/version_converter.md @@ -0,0 +1,7 @@ +# onnx.version_converter + +## convert_version + +```{eval-rst} +.. autofunction:: onnx.version_converter.convert_version +``` diff --git a/docs/docsgen/source/conf.py b/docs/docsgen/source/conf.py new file mode 100644 index 0000000..518d0cc --- /dev/null +++ b/docs/docsgen/source/conf.py @@ -0,0 +1,107 @@ +# Copyright (c) ONNX Project Contributors +# +# SPDX-License-Identifier: Apache-2.0 + + +from __future__ import annotations + +import os +import sys +import warnings + +import onnx + +sys.path.append(os.path.abspath(os.path.dirname(__file__))) + + +# -- Project information ----------------------------------------------------- + +author = "ONNX" +copyright = "2024" # noqa: A001 +project = "ONNX" +release = onnx.__version__ +version = onnx.__version__ + +# define the latest opset to document, +# this is meant to avoid documenting opset not released yet +max_opset = onnx.helper.VERSION_TABLE[-1][2] + +# define the latest opset to document for every opset +_opsets = [t for t in onnx.helper.VERSION_TABLE if t[2] == max_opset][-1] +max_opsets = { + "": max_opset, + "ai.onnx.ml": _opsets[3], + "ai.onnx.training": _opsets[4], +} + +# -- General configuration --------------------------------------------------- + +extensions = [ + "myst_parser", + "onnx_sphinx", + "sphinx_copybutton", + "sphinx_exec_code", + "sphinx_tabs.tabs", + "sphinx.ext.autodoc", + "sphinx.ext.autosummary", + "sphinx.ext.coverage", + "sphinx.ext.doctest", + "sphinx.ext.githubpages", + "sphinx.ext.graphviz", + "sphinx.ext.ifconfig", + "sphinx.ext.intersphinx", + "sphinx.ext.mathjax", + "sphinx.ext.napoleon", + "sphinx.ext.viewcode", +] + +myst_enable_extensions = [ + "amsmath", + "attrs_inline", + "colon_fence", + "deflist", + "dollarmath", + "fieldlist", + "html_admonition", + "html_image", + "linkify", + "replacements", + "smartquotes", + "strikethrough", + "substitution", + "tasklist", +] + +coverage_show_missing_items = True +exclude_patterns = [] +graphviz_output_format = "svg" +html_css_files = ["css/custom.css"] +html_favicon = "onnx-favicon.png" +html_sidebars = {} +html_static_path = ["_static"] +html_theme = "furo" +language = "en" +mathdef_link_only = True +master_doc = "index" +onnx_doc_folder = os.path.join(os.path.abspath(os.path.dirname(__file__)), "operators") +pygments_style = "default" +source_suffix = [".rst", ".md"] +templates_path = ["_templates"] + +html_context = { + "default_mode": "auto", # auto: the documentation theme will follow the system default that you have set (light or dark) +} + +html_theme_options = { + "light_logo": "onnx-horizontal-color.png", + "dark_logo": "onnx-horizontal-white.png", +} + +intersphinx_mapping = { + "numpy": ("https://numpy.org/doc/stable/", None), + "python": (f"https://docs.python.org/{sys.version_info.major}/", None), + "scipy": ("https://docs.scipy.org/doc/scipy/", None), + "torch": ("https://pytorch.org/docs/stable/", None), +} + +warnings.filterwarnings("ignore", category=FutureWarning) diff --git a/docs/docsgen/source/expect_onnxruntime.md b/docs/docsgen/source/expect_onnxruntime.md new file mode 100644 index 0000000..4875050 --- /dev/null +++ b/docs/docsgen/source/expect_onnxruntime.md @@ -0,0 +1,75 @@ + + +(l-function-expect)= + +# Sample operator test code + +Many examples from the documentation end by calling +function `expect` to check a runtime returns the expected +outputs for the given example. Here is one implementation +based on [onnxruntime](https://onnxruntime.ai/). + +``` +from typing import Any, Sequence +import numpy as np +import onnx +import onnxruntime + + +def expect( + node: onnx.NodeProto, + inputs: Sequence[np.ndarray], + outputs: Sequence[np.ndarray], + name: str, + **kwargs: Any, +) -> None: + # Builds the model + present_inputs = [x for x in node.input if (x != "")] + present_outputs = [x for x in node.output if (x != "")] + input_type_protos = [None] * len(inputs) + if "input_type_protos" in kwargs: + input_type_protos = kwargs["input_type_protos"] + del kwargs["input_type_protos"] + output_type_protos = [None] * len(outputs) + if "output_type_protos" in kwargs: + output_type_protos = kwargs["output_type_protos"] + del kwargs["output_type_protos"] + inputs_vi = [ + _extract_value_info(arr, arr_name, input_type) + for arr, arr_name, input_type in zip(inputs, present_inputs, input_type_protos) + ] + outputs_vi = [ + _extract_value_info(arr, arr_name, output_type) + for arr, arr_name, output_type in zip( + outputs, present_outputs, output_type_protos + ) + ] + graph = onnx.helper.make_graph( + nodes=[node], name=name, inputs=inputs_vi, outputs=outputs_vi + ) + kwargs["producer_name"] = "backend-test" + + if "opset_imports" not in kwargs: + # To make sure the model will be produced with the same opset_version after opset changes + # By default, it uses since_version as opset_version for produced models + produce_opset_version = onnx.defs.get_schema( + node.op_type, domain=node.domain + ).since_version + kwargs["opset_imports"] = [ + onnx.helper.make_operatorsetid(node.domain, produce_opset_version) + ] + + model = onnx.helper.make_model_gen_version(graph, **kwargs) + + # Checking the produces are the expected ones. + sess = onnxruntime.InferenceSession(model.SerializeToString(), + providers=["CPUExecutionProvider"]) + feeds = {name: value for name, value in zip(node.input, inputs)} + results = sess.run(None, feeds) + for expected, output in zip(outputs, results): + assert_allclose(expected, output) +``` diff --git a/docs/docsgen/source/index.md b/docs/docsgen/source/index.md new file mode 100644 index 0000000..80168d8 --- /dev/null +++ b/docs/docsgen/source/index.md @@ -0,0 +1,19 @@ + + +(l-main-doc-page)= + +# ONNX documentation + +```{toctree} +:maxdepth: 2 + +intro/index +api/index +operators/index +technical/index +repo-docs/index +``` diff --git a/docs/docsgen/source/intro/concepts.md b/docs/docsgen/source/intro/concepts.md new file mode 100644 index 0000000..53766e7 --- /dev/null +++ b/docs/docsgen/source/intro/concepts.md @@ -0,0 +1,364 @@ +# ONNX Concepts + +ONNX can be compared to a programming language specialized +in mathematical functions. It defines all the necessary operations +a machine learning model needs to implement its inference function +with this language. A linear regression could be represented +in the following way: + +``` +def onnx_linear_regressor(X): + "ONNX code for a linear regression" + return onnx.Add(onnx.MatMul(X, coefficients), bias) +``` + +```{index} ONNX graph +``` + +This example is very similar to an expression a developer could +write in Python. It can be also represented as a graph that shows +step-by-step how to transform the features to get a prediction. +That's why a machine-learning model implemented with ONNX is often +referenced as an **ONNX graph**. + +```{image} images/linreg1.png +``` + +ONNX aims at providing a common language any machine learning framework +can use to describe its models. The first scenario is to make it easier +to deploy a machine learning model in production. An ONNX interpreter +(or **runtime**) can be specifically implemented and optimized for this task +in the environment where it is deployed. With ONNX, it is possible +to build a unique process to deploy a model in production and independent +from the learning framework used to build the model. +*onnx* implements a python runtime that can be used to evaluate +ONNX models and to evaluate ONNX ops. This is intended to clarify the +semantics of ONNX and to help understand and debug ONNX tools +and converters. It is not intended to be used for production and +performance is not a goal (see {ref}`l-reference-implementation`). + +## Input, Output, Node, Initializer, Attributes + +Building an ONNX graph means implementing a function +with the ONNX language or more precisely the {ref}`l-onnx-operators`. +A linear regression would be written this way. +The following lines do not follow python syntax. +It is just a kind of pseudo-code to illustrate the model. + +``` +Input: float[M,K] x, float[K,N] a, float[N] c +Output: float[M, N] y + +r = onnx.MatMul(x, a) +y = onnx.Add(r, c) +``` + +This code implements a function `f(x, a, c) -> y = x @ a + c`. +And *x*, *a*, *c* are the **inputs**, *y* is the **output**. +*r* is an intermediate result. +*MatMul* and *Add* are the **nodes**. They also have inputs and outputs. +A node has also a type, one of the operators in +{ref}`l-onnx-operators`. This graph was built with the example +in Section {ref}`l-onnx-linear-regression-onnx-api`. + +The graph could also have an **initializer**. When an input +never changes such as the coefficients of the linear regression, +it is most efficient to turn it into a constant stored in the graph. + +``` +Input: float[M,K] x +Initializer: float[K,N] a, float[N] c +Output: float[M, N] xac + +xa = onnx.MatMul(x, a) +xac = onnx.Add(xa, c) +``` + +Visually, this graph would look like the following image. +The right side describes operator *Add* where the second input +is defined as an initializer. This graph was obtained with this +code {ref}`l-onnx-linear-regression-onnx-api-init`. + +```{image} images/linreg2.png +:alt: Snapshot of Netron +``` + +An **attribute** is a fixed parameter of an operator. Operator {ref}`l-onnx-doc-Gemm` +has four attributes, *alpha*, *beta*, *transA*, *transB*. Unless the runtime +allows it through its API, once it has loaded the ONNX graph, these values +cannot be changed and remain frozen for all the predictions. + +## Serialization with protobuf + +The deployment of a machine-learned model into production +usually requires replicating the entire ecosystem used to +train the model, most of the time with a *docker*. +Once a model is converted into ONNX, the production environment +only needs a runtime to execute the graph defined with ONNX +operators. This runtime can be developed in any language +suitable for the production application, C, java, python, javascript, +C#, Webassembly, ARM... + +But to make that happen, the ONNX graph needs to be saved. +ONNX uses *protobuf* to serialize the graph into +one single block +(see [Parsing and Serialization](https://developers.google.com/protocol-buffers/docs/pythontutorial#parsing-and-serialization)). It aims at optimizing the model size +as much as possible. + +## Metadata + +Machine learned models are continuously refreshed. It is important +to keep track of the model version, the author of the model and +how it was trained. ONNX offers the possibility to store additional data +in the model itself. + +- **doc_string**: Human-readable documentation for this model. + : Markdown is allowed. +- **domain**: A reverse-DNS name to indicate the model namespace or domain, + : for example, 'org.onnx' +- **metadata_props**: Named metadata as dictionary `map`, + : `(values, keys)` should be distinct. +- **model_author**: A comma-separated list of names, + : The personal name of the author(s) of the model, and/or their organizations. +- **model_license**: The well-known name or URL of the license + : under which the model is made available. +- **model_version**: The version of the model itself, encoded in an integer. +- **producer_name**: The name of the tool used to generate the model. +- **producer_version**: The version of the generating tool. +- **training_info**: An optional extension that contains + : information for training (see {ref}`l-traininginfoproto`) + +## List of available operators and domains + +The main list is described here: {ref}`l-onnx-operators`. +It merges standard matrix operators (Add, Sub, MatMul, Transpose, +Greater, IsNaN, Shape, Reshape...), +reductions (ReduceSum, ReduceMin, ...) +image transformations (Conv, MaxPool, ...), +deep neural networks layer (RNN, DropOut, ...), +activations functions (Relu, Softmax, ...). +It covers most of the operations needed to implement +inference functions from standard and deep machine learning. +ONNX does not implement every existing machine learning operator, +the list of operator would be infinite. + +The main list of operators is identified with a domain **ai.onnx**. +A **domain** can be defined as a set of operators. +A few operators in this list are dedicated to text but they hardly cover +the needs. The main list is also missing tree based models very +popular in standard machine learning. +These are part of another domain **ai.onnx.ml**, +it includes tree bases models (TreeEnsemble Regressor, ...), +preprocessing (OneHotEncoder, LabelEncoder, ...), SVM models +(SVMRegressor, ...), imputer (Imputer). + +ONNX only defines these two domains. But the library onnx +supports any custom domains and operators +(see {ref}`l-onnx-extensibility`). + +## Supported Types + +ONNX specifications are optimized for numerical computation with +tensors. A *tensor* is a multidimensional array. It is defined +by: + +- a type: the element type, the same for all elements in the tensor +- a shape: an array with all dimensions, this array can be empty, + a dimension can be null +- a contiguous array: it represents all the values + +This definition does not include *strides* or the possibility to define +a view of a tensor based on an existing tensor. An ONNX tensor is a dense +full array with no stride. + +### Element Type + +ONNX was initially developed to help deploying deep learning model. +That's why the specifications were initially designed for floats (32 bits). +The current version supports all common types. Dictionary +{ref}`l-onnx-types-mapping` gives the correspondence between *ONNX* +and {mod}`numpy`. + +```{eval-rst} +.. exec_code:: + + import re + from onnx import TensorProto + + reg = re.compile('^[0-9A-Z_]+$') + + values = {} + for att in sorted(dir(TensorProto)): + if att in {'DESCRIPTOR'}: + continue + if reg.match(att): + values[getattr(TensorProto, att)] = att + for i, att in sorted(values.items()): + si = str(i) + if len(si) == 1: + si = " " + si + print("%s: onnx.TensorProto.%s" % (si, att)) +``` + +ONNX is strongly typed and its definition does not support +implicit cast. ONNX does not allow addition of two tensors +or matrices with different types, even if other languages do. +That's why an explicit cast must be inserted in a graph. + +### Sparse Tensor + +Sparse tensors are useful to represent arrays having many null coefficients. +ONNX supports 2D sparse tensor. Class {ref}`l-onnx-sparsetensor-proto` +defines attributes `dims`, `indices` (int64) and `values`. + +### Other types + +In addition to tensors and sparse tensors, ONNX supports sequences of tensors, +map of tensors, sequences of map of tensors through types +{ref}`l-onnx-sequence-proto`, {ref}`l-onnx-map-proto`. They are rarely used. + +## What is an opset version? + +The opset is mapped to the version of the *onnx* package. +It is incremented every time the minor version increases. +Every version brings updated or new operators. + +```{eval-rst} +.. exec_code:: + + import onnx + print(onnx.__version__, " opset=", onnx.defs.onnx_opset_version()) +``` + +An opset version is also attached to every ONNX graph. +It defines the version of all operators inside the graph. +Operator *Add* was updated in version 6, 7, 13 and 14. If the +graph opset is 15, it means operator *Add* follows specifications +version 14. If the graph opset is 12, then operator *Add* follows +specifications version 7. An operator in a graph follows its most +recent definition below (or equal) the global graph opset. + +A graph may include operators from several domains, `ai.onnx` and +`ai.onnx.ml` for example. In that case, the graph must define a +global opset for every domain. The rule is applied to every +operators within the same domain. + +## Subgraphs, tests and loops + +ONNX implements tests and loops. They all take another ONNX +graphs as an attribute. These structures are usually slow and complex. +It is better to avoid them if possible. + +### If + +Operator {ref}`l-onnx-doc-If` executes +one of the two graphs depending on the condition evaluation. + +``` +If(condition) then + execute this ONNX graph (`then_branch`) +else + execute this ONNX graph (`else_branch`) +``` + +Those two graphs can use any result already computed in the +graph and must produce the exact same number of outputs. +These outputs will be the output of the operator `If`. + +```{image} images/dot_if.png +``` + +(l-operator-scan-onnx-tutorial)= + +### Scan + +Operator {ref}`l-onnx-doc-Scan` implements a loop with a fixed number of iterations. +It loops over the rows (or any other dimension) of the inputs and concatenates +the outputs along the same axis. Let's see an example which implements +pairwise distances: $M(i,j) = \lVert X_i - X_j \rVert^2$. + +```{image} images/dot_scan.png +``` + +This loop is efficient even if it is still slower than a custom implementation +of pairwise distances. It assumes inputs and outputs are tensors and +automatically concatenate the outputs of every iteration into single +tensors. The previous example only has one but it could have several. + +### Loop + +Operator {ref}`l-onnx-doc-Loop` implements a for and a while loop. It can do a fixed +number of iterators and/or ends when a condition is not met anymore. +Outputs are processed in two different ways. First one is similar to +loop {ref}`l-onnx-doc-Scan`, outputs are concatenated into tensors (along the first +dimension). This also means that these outputs must have compatible shapes. +Second mechanism concatenates tensors into a sequence of tensors. + +(l-onnx-extensibility)= + +## Extensibility + +ONNX defines a list of operators as the standard: {ref}`l-onnx-operators`. +However, it is very possible +to define your own operators under this domain or a new one. +*onnxruntime* defines custom operators to improve inference. +Every node has a type, a name, +named inputs and outputs, and attributes. As long as a node is described +under these constraints, a node can be added to any ONNX graph. + +Pairwise distances can be implemented with operator Scan. +However, a dedicated operator called CDist is proved significantly +faster, significantly enough to make the effort to implement a dedicated runtime +for it. + +## Functions + +Functions are one way to extend ONNX specifications. Some model requires +the same combination of operators. This can be avoided by creating a function +itself defined with existing ONNX operators. Once defined, a function behaves +like any other operators. It has inputs, outputs and attributes. + +There are two advantages of using functions. The first one is to have a +shorter code and easier to read. The second one is that any onnxruntime +can leverage that information to run predictions faster. The runtime +could have a specific implementation for a function not relying on the +implementation of the existing operators. + +## Shape (and Type) Inference + +Knowing the shapes of results is not necessary to execute an ONNX graph +but this information can be used to make it faster. If you have the following +graph: + +``` +Add(x, y) -> z +Abs(z) -> w +``` + +If *x* and *y* have the same shape, then *z* and *w* also have the same +shape. Knowing that, it is possible to reuse the buffer allocated for *z*, +to compute the absolute value *w* inplace. Shape inference helps the +runtime to manage the memory and therefore to be more efficient. + +ONNX package can compute in most of the cases the output shape +knowing the input shape for every standard operator. It cannot +obviously do that for any custom operator outside of the official +list. + +## Tools + +[netron](https://netron.app/) +is very useful to help visualize ONNX graphs. +That's the only one without programming. The first screenshot was +made with this tool. + +```{image} images/linreg1.png +``` + +[onnx2py.py](https://github.com/microsoft/onnxconverter-common/blob/master/onnxconverter_common/onnx2py.py) +creates a python file from an ONNX graph. This script can create +the same graph. It may be modified by a user to change the graph. + +[zetane](https://github.com/zetane/viewer) +can load onnx model and show intermediate results +when the model is executed. diff --git a/docs/docsgen/source/intro/converters.md b/docs/docsgen/source/intro/converters.md new file mode 100644 index 0000000..810aa10 --- /dev/null +++ b/docs/docsgen/source/intro/converters.md @@ -0,0 +1,317 @@ +# Converters + +Using ONNX in production means the prediction function +of a model can be implemented with ONNX operators. +A runtime must be chosen, one available on the platform +the model is deployed. Discrepancies are checked +and finally, the latency is measured. +The first step of the model conversion can be easy +if there exists a converting library for this framework +supporting all the pieces of the model. If it is not the +case, the missing parts must be implemented in ONNX. +That may be very time consuming. + +## What is a converting library? + +[sklearn-onnx](https://onnx.ai/sklearn-onnx/) converts +[scikit-learn](https://scikit-learn.org/stable/) models +into ONNX. It rewrites the prediction function of a model, +whatever it is, with ONNX operators using the API introduced +above. It ensures that the predictions are equal or at least +very close to the expected predictions computed with the +original model. + +Machine learning libraries usually have their own design. +That's why there exists a specific converting library for +each of them. Many of them are listed there: +[Converting to ONNX format](https://github.com/onnx/tutorials#converting-to-onnx-format). +Here is a short list: + +- [sklearn-onnx](https://onnx.ai/sklearn-onnx/): + converts models from [scikit-learn](https://scikit-learn.org/stable/), +- [tensorflow-onnx](https://github.com/onnx/tensorflow-onnx): + converts models from [tensorflow](https://www.tensorflow.org/), +- [jax2onnx](https://github.com/enpasos/jax2onnx): + converts models from [JAX](https://docs.jax.dev/), +- [onnxmltools](https://github.com/onnx/onnxmltools): + converts models from [lightgbm](https://lightgbm.readthedocs.io/), + [xgboost](https://xgboost.readthedocs.io/en/stable/), + [pyspark](https://spark.apache.org/docs/latest/api/python/), + [libsvm](https://github.com/cjlin1/libsvm) +- [torch.onnx](https://pytorch.org/docs/master/onnx.html): + converts model from [pytorch](https://pytorch.org/). + +The main challenge for all these libraries is to keep up the rhythm. +They must be updated everytime ONNX or the library they support +have a new released version. That means three to five new releases +per year. + +Converting libraries are not compatible among each others. +[tensorflow-onnx](https://github.com/onnx/tensorflow-onnx) +is dedicated to tensorflow and only tensorflow. +The same goes for sklearn-onnx specialized into scikit-learn. + +One challenge is customization. It is difficult to support +custom pieces in a machine learned model. +They have to write the specific converter for this piece. +Somehow, it is like implementing +twice the prediction function. There is one easy case: +deep learning frameworks have their own primitives to ensure +the same code can be executed on different environments. +As long as a custom layer or a subpart is using pieces of +pytorch or tensorflow, there is not much to do. +It is a different story for scikit-learn. This package +does not have its own addition or multiplication, it relies +on numpy or scipy. The user must implement +its transformer or predictor with ONNX primitives, whether or +not it was implemented with numpy. + +## Alternatives + +One alternative for implementing ONNX export capability is to leverage standard protocols such as the [Array API standard](https://data-apis.org/array-api/latest/), which standardizes a common set of array operations. It enables code reuse across libraries like NumPy, JAX, PyTorch, CuPy and more. [ndonnx](https://github.com/Quantco/ndonnx) enables execution with an ONNX backend and instant ONNX export for Array API compliant code. This diminishes the need for dedicated converter library code since the same code used to implement most of a library can reused in ONNX conversion. It also provides a convenient primitive for converter authors looking for a NumPy-like experience when constructing ONNX graphs. + +## Opsets + +ONNX releases packages with version numbers like +`major.minor.fix`. Every minor update means the list of operators +is different or the signature has changed. It is also associated to +an opset, version `1.10` is opset 15, `1.11` will be opset 16. +Every ONNX graph should define the opset it follows. Changing this +version without updating the operators could make the graph invalid. +If the opset is left unspecified, ONNX will consider that the graph +is valid for the latest opset. + +New opsets usually introduce new operators. A same inference function +could be implemented differently, usually in a more efficient way. +However, the runtime the model is running on may not +support newest opsets or at least not in the installed version. +That's why every converting library offers the +possibility to create an ONNX graph for a specific opset usually called +`target_opset`. ONNX language describes simple and complex operators. +Changing the opset is similar to upgrading a library. onnx +and onnx runtimes must support backward compatibility. + +## Other API + +Examples in previous sections show that onnx API is +very verbose. It is also difficult to get a whole picture of +a graph by reading the code unless it is a small one. Almost +every converting library has implemented a different API +to create a graph, usually more simple, less verbose +than the API of onnx package. +All API automate the addition of initializers, hide the creation +of a name of every intermediate result, deal with different +version for different opset. + +### A class Graph with a method add_node + +`tensorflow-onnx` implements a class graph. +It rewrites tensorflow function with ONNX operator when +ONNX does not have a similar function (see [Erf](https://github.com/onnx/tensorflow-onnx/blob/master/tf2onnx/onnx_opset/math.py#L414). + +sklearn-onnx defines two different API. The first one +introduced in that example [Implement a converter](https://onnx.ai/sklearn-onnx/auto_tutorial/plot_jcustom_syntax.html) +follows a similar design that tensorflow-onnx follows. +The following lines are extracted from the converter of a linear +classifier. + +``` +# initializer + +coef = scope.get_unique_variable_name('coef') +model_coef = np.array( + classifier_attrs['coefficients'], dtype=np.float64) +model_coef = model_coef.reshape((number_of_classes, -1)).T +container.add_initializer( + coef, proto_dtype, model_coef.shape, model_coef.ravel().tolist()) + +intercept = scope.get_unique_variable_name('intercept') +model_intercept = np.array( + classifier_attrs['intercepts'], dtype=np.float64) +model_intercept = model_intercept.reshape((number_of_classes, -1)).T +container.add_initializer( + intercept, proto_dtype, model_intercept.shape, + model_intercept.ravel().tolist()) + +# add nodes + +multiplied = scope.get_unique_variable_name('multiplied') +container.add_node( + 'MatMul', [operator.inputs[0].full_name, coef], multiplied, + name=scope.get_unique_operator_name('MatMul')) + +# [...] + +argmax_output_name = scope.get_unique_variable_name('label') +container.add_node('ArgMax', raw_score_name, argmax_output_name, + name=scope.get_unique_operator_name('ArgMax'), + axis=1) +``` + +### Operator as function + +The second API shown in [Implement a new converter](https://onnx.ai/sklearn-onnx/auto_tutorial/plot_icustom_converter.html) +is more compact and defines +every ONNX operator as composable functions. +The syntax looks like this for [KMeans](https://scikit-learn.org/stable/modules/generated/sklearn.cluster.KMeans.html), +less verbose and easier to read. + +``` +rs = OnnxReduceSumSquare( + input_name, axes=[1], keepdims=1, op_version=opv) + +gemm_out = OnnxMatMul( + input_name, (C.T * (-2)).astype(dtype), op_version=opv) + +z = OnnxAdd(rs, gemm_out, op_version=opv) +y2 = OnnxAdd(C2, z, op_version=opv) +ll = OnnxArgMin(y2, axis=1, keepdims=0, output_names=out[:1], + op_version=opv) +y2s = OnnxSqrt(y2, output_names=out[1:], op_version=opv) +``` + +## Tricks learned from experience + +### Discrepancies + +ONNX is strongly typed and optimizes for float32, the most +common type in deep learning. Libraries in standard +machine learning use both float32 and float64. numpy +usually cast to the most generic type, float64. It has no significant +impact when the prediction function is contiguous. +When it is not, the right type must be used. Example +[Issues when switching to float](https://onnx.ai/sklearn-onnx/auto_tutorial/plot_ebegin_float_double.html) +gives more insights on that topic. + +Parallelization changes the order of computation. It is usually +not significant but it may explain some weird discrepancies. +`1 + 1e17 - 1e17 = 0` but `1e17 - 1e17 + 1 = 1`. High order of +magnitude are rare but not so rare when a model uses the inverse +of a matrix. + +### IsolationForest Trick + +ONNX only implements a {ref}`TreeEnsembleRegressor +` but +it does not offer the possibility to retrieve any information +about the path the decision followed or statistics to the graph. +The trick is to used one forest to predict the leaf index and map +this leaf index one or multiple times with the information needed. + +```{image} images/iff.png +``` + +### Discretization + +Looking in which interval a feature falls into. That's easy to do +with numpy but not so easy to do efficiently with ONNX. +The fastest way is to use a TreeEnsembleRegressor, a binary search, +which outputs the interval index. That's what this example +implements: [Converter for WOE](https://onnx.ai/sklearn-onnx/auto_tutorial/plot_woe_transformer.html). + +### Contribute + +[onnx repository](https://github.com/onnx/onnx) must be forked and cloned. + +### Build + +The windows build requires conda. The following steps might not be up to date. +Folder [onnx/.github/workflows](https://github.com/onnx/onnx/tree/main/.github/workflows) +contains the latest instructions. + +**Windows** + +The build is easier with Anaconda. First: create an environment. +It must be done only once. + +``` +conda create --yes --quiet --name py3.9 python=3.9 +conda install -n py3.9 -y -c conda-forge numpy libprotobuf=3.16.0 +``` + +Then build the package: + +```sh +git submodule update --init --recursive +set ONNX_BUILD_TESTS=1 +set ONNX_ML=$(onnx_ml) +set CMAKE_ARGS=-DONNX_USE_PROTOBUF_SHARED_LIBS=ON -DONNX_USE_LITE_PROTO=ON -DONNX_WERROR=ON + +python -m build --wheel +``` + +The package can now be installed. + +**Linux** + +After cloning the repository, the following instructions can be run: + +```sh +python -m build --wheel +``` + +### Build the markdown documentation + +The package must be built first (see previous section). + +``` +set ONNX_BUILD_TESTS=1 +set ONNX_ML=$(onnx_ml) +set CMAKE_ARGS=-DONNX_USE_PROTOBUF_SHARED_LIBS=ON -DONNX_USE_LITE_PROTO=ON -DONNX_WERROR=ON + +python onnx\gen_proto.py -l +python onnx\gen_proto.py -l --ml +pip install -e . +python onnx\backend\test\cmd_tools.py generate-data +python onnx\backend\test\stat_coverage.py +python onnx\defs\gen_doc.py +set ONNX_ML=0 +python onnx\defs\gen_doc.py +set ONNX_ML=1 +``` + +### Update an existing operator + +All operators are defined in folder +[onnx/onnx/defs](https://github.com/onnx/onnx/tree/main/onnx/defs). +There are two files in every subfolder, one called `defs.cc` and another one +called `old.cc`. + +- `defs.cc`: contains the most recent definition for every operator +- `old.cc`: contains the deprecated version of the operators in previous opset + +Updating an operator means copying the definition from `defs.cc` to `old.cc` +and updating the existing one in `defs.cc`. + +One file following the pattern `onnx/defs/operator_sets*.h` +must be modified. These headers registers the list +of existing operators. + +File [onnx/defs/schema.h](https://github.com/onnx/onnx/blob/main/onnx/defs/schema.h) +contains the latest opset version. It must be updated too if one opset +was upgraded. + +File [onnx/version_converter/convert.h](https://github.com/onnx/onnx/blob/main/onnx/version_converter/convert.h) +contains rules to apply when converter a node from an opset to the next one. +This file may be updated too. + +The package must be compiled and the documentation must be generated +again to automatically update the markdown documentation and it must +be included in the PR. + +Then unit test must be updated. + +**Summary** + +- Modify files `defs.cc`, `old.cc`, `onnx/defs/operator_sets*.h`, + `onnx/defs/schema.h` +- Optional: modify file `onnx/version_converter/convert.h` +- Build onnx. +- Build the documentation. +- Update unit test. + +The PR should include the modified files and the modified markdown documentation, +usually a subset of +`docs/docs/Changelog-ml.md`, `docs/Changelog.md`, +`docs/Operators-ml.md`, `docs/Operators.md`, +`docs/TestCoverage-ml.md`, `docs/TestCoverage.md`. diff --git a/docs/docsgen/source/intro/images/dot_att.png b/docs/docsgen/source/intro/images/dot_att.png new file mode 100644 index 0000000000000000000000000000000000000000..3fe69c898a8daca46094b1c762261cbc2677f19f GIT binary patch literal 35875 zcmdqJcT|(X_cw?Y6)`GE$A~B>Edl~k14t7@s&t5e^xmZ3mrw#p_QCRf%a-$=vuA&2cmLp!%*=D2nLBr8?%dDkCRqK6BJ~BP3ltO-)Q^

z_Lblw|PW1Au=g+%y$sDT=yTmVhs(tz=YXC@6}9$E9tvY zP_PCa|D6cOZdy=Kgj79}lhJ-*vPuNSu;}`yP=Oy_z4|!q{0X$krb1G^vhY?MYgl8E zP*Ziu{8%#5*MUlP4xe3J5h(DjUKz<-Msx7P)84!<)_A=1)7MtNn%$B9!RRQ5*r?af zZt$0#c~C|nE_u1}!HvZe7a67Bwg~sOlo(f%5^zgJMQck-Yk6W%@eMue-Wj=hKtY^$ zPcG(w^F9T|JH{L^#SPCCIK|76SA-K3W|TNeig&F4cM37pb5~zmi+p2NhwgWSi4h#9 zWq4Q-Tbx^7XMLz~lo%PJT>LA7y7nM>?(^yC>p2*6pz)KZMCksjiIeIb!wqi< z^=E+wDR66Q+}?FGs3CxmDwKoq1)5#L1r4RZ*JX&m#B;y~0fc(W|4%pc(o7na67W|h zQRHAKe-F>g*;eds(=S%}1NML8M14>;-An0x6uJQW7(a9ApGxdR6PX2@oId)c+!ASF zL9a{Zfc+@(!|Faa?(%_IjqPd>tG6D3PyMUV&7B(vmLn9u*Pvjy)dDX3Y3y>YbvZ&ZNv*iAy#D|I)075tqQa*`>*bUMynUsloBz% z$^QlYzrmC%!*yltJ0h{+!U{p)0U|$mD)8Ury3M~GYV!Hlmc|!Pzs_Jk7Q7w1@(xRS z@E^%$MeuYgL`bUP?iZ;a47l5s&D4L!&N+pHOvW3TP3V-*fXAne|5(zKgh7N4N91L5 z#l+QG{1>YV!?p0`x=Vj6jRs+{kjKVWJHHsm5@FVTuNK^0aJDMMnxoILt#oGDh`*wv zaIgnFr4K?@_C6A^lyG1 z+WnU{+#3bIHrX~;r9~sf37q3Nbw?EOFkw~ zL?(WQo_>5#oZFV}rm92By#SSn_07Y(nkT;gGfyuvNCy)m2kE0)EuVC-*;%bR@pcJE zyX_c$vWGE!GlEOasleJHi`$Alhy-t% z1x_t5EPeOS)u-*@R-X*B-mf|7@jbq!xNcig%KTSJ5H#$r2)H7e3`L}El z42elXgZ?!S{Ecr;dwLTrK!qw@ogLuiTv|^M_2j*QS?sLFpXmdkVNzmQ%l>m4uaBzF zR?nrvaF8kfCDN1a(Y~hMxq_oMnWYVDv&rWgB<)7QYK{65#DROEVTjko3;bV1N4q}N zN6s{TbQ|3p5GATSHi~xGd9gH=p*LL6%cTnAdpmKVu=}rh_#jEj@aWXNUxQ8~10JlC zGZ`fH67Kc2*3n1x1Y;}HN8XmT5m-iXgJdmHp+t0LQI~s>;Vdg&@WF+6WfHMvyxKpk zxvjjg2+>O?H7!QnJXSOV#a7Ebe#N6H%*dWxQ5-3SB+6}<_rE$ z9vp_s1JO+43G%YFO3fkhiR=2_?h_+xJXbF&gBFJcM&>jckIqff%3u034~a^79q(_y!Vb>goX?#YIa`f9n`@uXDr=wYBJL(?RUw=GL1fM7r?J(FE56p4D^%ps z_p-U*P>(`Bc%@3$o$jRB9~`e&9OI!}++q726%RS6U2PB-EKLj~C82 z&^Aj3%&QoOP#Y2*!(D|A?6Ae%s;_>3C+OE8#J4*G91Ycti1mt4g6ec-&}IwY zYQk-fY0Ab8v&rb6kzi`rb+Zmf$}*~({J5j@NQf1lpRYA9eoExj|D#FswnwfpT?6VqG__Eq9JcLWTmP zpEIi|3j|-}!NmjnZ>~;dG%jA1;KQ%EKJtvRnfs_R$%TeAMpp*IPu`BcoSIKd`*Tp3v((;iGcSV1rNdqnrU>e20N^ zh(Vos)|+-3Mz^8Brp0MObj-l~MO9k*T2BL<8rpr0r^f=DK&_BhFCflNBU1v-Uft9h zEr<}dF=!UP$Judz>`S}{x;RM`m!+0N>o^2y6Lqs%+TVAUz10m)jC07AzQgAy!;S%wsw_`VprDC?YB{ppv`bENTSGGxnCH(weSewx>z{Oxh>8H;yfDLKdcq1CM|G4@E(uD4Q_b^n%;*x@( zQpR}v=E7ZCN&`&gFUl`u(OdL%1FVA zyfrbsnL+FmP#reAUo1wxAKTDH$ct^@l>>Z-#3@cc?biF|tLLT-I^Vwc%*Ux{#x<0E z?k1_2N!8wu|AO)J1iZs|DSjTBzkUfEClcyceR%f^(L>4mB_95{=Q8OMRMA=KFI!Ltn4<`_4qqGq}P8= z-NXm^m=vRZXUnWdw++ONW8z|#TyEEBwR2{3hTmV_)#95FkM(VjcfIguFAc6GnBM)n zoBzMe-I5{xUwV3Zxa4HcbJ5a&eB?YF=ewc5x1l8T! z{_-VoZ-VB4Z#)Bb8Nl=ZudJqUicL-?ORMlT?3<|&qhavs3G`{A_ErYPByjbt%QWIc z>J|U^c$N`g2mPs?OF(PxQ>RX8;vBxZ%qkoRyi5W*;b*ZvRo}nt zAYDNzaT0%OzrDm&>bdVXc3^Cn0vO69ul}}SrDG(7V)1xvPM267+*qCZWZwFrjhGe6s4{sMnnFsb*>7 zDd=-kE*dRpvbg8!`2>Xi)ohBKuf6!fY;rRhCVbHV7f#0Za2If8I)x?M;EJ$1JEZ z8Y^Ch2Mn;wDb4=8uu8Sw4?i&XMzB3lhvu}0C!?hfgtNupoomV-V(t>Yd?yF|Wp3xu zs*|p0q>LPg)UH$6me!WZ!^?oI@O75H>jdg-o_Fu$*LcT%WY4i4M_p1FV6~D)<}ShA zv(qBh*(4`#%y$?dyb0=mL-|zsyUf|?(lE!`|RV;?$h>sVfKjKi91>-g2Z% zFJXY}fb|~hR(Q|dGdDS0R$w;Ss2d2HZ}jTWJKxkA?7Q0ro}IoawONdKh(JJPh?nm} zB$57GC$rDJV`#|^3e&Ip4m3U8o{zK+H=fs~xH7&!QyaU60W9it`+#pE;z+J@){qjr ze)h$AT+pu<(}rV;H^mt_`%F78TBG*q513CTy&|yPFcD3cbP3q*NxHGKzjcU(oPTi^ z7xZ#J8YS{}`vGa*m%iB!n-}Epseau1?9Hza;(DLZ&#j+-0nF2l7n}O#yPmCJKehC) z`tKsmT(j>?S{1xDYg+wbK=zw6-iUn?X*p`n~H*18`vo!YXC>_`F9R zJ-tV@nnrSN4nU<5w~hZs&k){ti&YxCM8q7a=@SlmOSv+*b6no6bkqERj=!Yep1WxN z^%{-fa7e`9y#5=d1ojr;km|#gdL0(p=ER#q&rlt3evjbA;qWz@bVIr7T_wRCuSdqu z$^@4wFCys&#N`PB#p;s3KE~AAY}#YTo2xuFl=&;#(WU zu-W-(;&3~A;hpdhx#XA%!x(x;nsk3zqdSApQvJ{bPFPN|!*o`hiol};&I@}T!v?1H z=37j~&rCG2juF-09+V*JWAdT>i$^sEoqEK$os*W?-maychN1p9J(85xX)Q`a^_7@8 zhZ!3bv~5A#e8AdSrN>00M0qbUnFa|-r~ze&SD?}mft^ry>O3vjVAZ)dTLoc$?OsngarZX zq9bliajGbi4VqC>w-IhBfX}w^S+^Xu>P2!mq;tr@Fi@wIw(G|arTcx3sAQbdspL1m zzP4e$D{DTxf%jl8moyuZsPd$EwYf!=C>fE3%+o@7RD9>g1v>hSnzeL4=rS@pifPE4 zzbD9Lv$X6kq-zep77&l)UZ<3=;h5I)3E%uQ_PYI@Z&QvpmrEe*8d(h_m>Lh9iW?F1 zh`FV6Q@7Ln@#0(8yBal7vu~mb64%X8zJ-KBCStEensd*Lf?`#&D&)N<)3`^*@?zmT7>9h9}W*lA7 z+||pBD;-)#u)qs-4pNVZuclrRzAM7_N52jexV_>fJ$2rr@n()FrT8GP0yz8H6)onK zc?&iIg!T+~s*|D?ei-V;qc^zEfs0PgH^E}Z<+HMEp?=<~T3?BJA1lYXUWb&HEui@R z_4lArA(?ssQvvHMCv3-GRLs^4vDxFJkqU7MPf-4uqBdvyL0^-oqtf^zY##K(U*W<* z_0-tCANmlA-FI!aC)5WF+r+A`1|-C`cbx+4uQ?xL=T>P4YUn-)S~YafhTBEYmP=}( z7=45$7ANjDJyQkuX5lKi=eOR_nDrjH>>J~QVoi5Oq7Ab!wxo&P3l8!_ z4xz7gH|uA&bkF$!{yXXgHhS!{h!{DK80v_2^7X`Z*Q^@nAzV9V$b}!nDur0@V@E3> z*V$)FgPpdO!>5Ksq3BVrqVZd8_hx z^G{z~#S3OkLKI52SNwPs9T3}O%*;{#uHcGiA6>3rE?jZ=g=C(wnXQ>z+TD@|My(zAexg6iO-Gruefb0dQ4ps!_pXm%t)1GY&WVYafdatH6 zg$2Epil~0{rHh^0D`dxZfx-BIPWGbH)!TN5M6-D=bCs-ADum5M_jtzK<%z8~{O9x$ z4#sV!kq`5mcO25pz`|oz7YT(b>nk+o++72M{dz{;OhK@8e<5~6;BQM0`Gtm+X9tW4 zKi0<}vy1Fzi8p-%6T�~2bjOtXvSAr}iAMfsdEpr#{uqCr8K+Wg6ghUYGEK~16x zt!#>(&0^W4XBt%|w|e-)1+ypGM3^r)hsQL?Ues?FUO3Y(Bz!c*tGV2~@567OsQ!*L zc(b&0T3<2zFmwrtHRYb1=h+fdM~MJyF65aVF_k)(G`~u71-Nif=YWq+;=O2dfB@YI zK$Fr$)-yOoq6u(_LDUh7uLv5bY2bTxCXd%? zmpYnc@h=~mtz|WrMmo{Y4t?it9-%9C{wY-;jDxr?%JO@B_9b^zCz^GN*=I6T%RTpc z-)EQ5x=8=gE1}REGx|8$<#D*7>fd<+p`F}k`>HeQi?FTbIwW9diXCq;sOVZKkGUJ=D!&LM8Tx_?dh zMGC3y>07(2VIEJ1t=XqPqkc%t2t-sr{^HJ-)#RMaP}?_hoAQEvp6HZRD*JZ_FyRR<4g@P?q%X->tGb?)h-`}P7QqnRtk%0gC!C=2aDOb1`mpXDZh3k|CtUnwrt z%qXIFJCq-jm~i>AJLuFTX=WQRS;r$s4orJ?duP=;k3_Kg(9BdBPrxqZ>3Sb{M&#y> zfnX*BwDwTCxz&pIo+t4^Kx6@ah4e-Z{&{*)+m?TI%HoF=|5}9BdB&Y!qb`1xJdCZ4 zURB>w>LZNuLG=XFvku>;=j9YRONGx9{q(OQ^AKaJo0*M&cSfHgc&#NaoKR@|VR-j3 z(QOc=rU#f7O9BP*!h*OVYlo<&Mr&glfe5#o$ZCUI@a7J zj^?y^xa?S6e_$V#?-_&*YQ-0pOy*#ASI~(b0w zIt00HLv8QL`YXymzSyc*0A9(TZR3Hwrn5I-yZH>i(Z^WRF6voOW*l0V?HF>c*{+km zO!P-l*WX@CsH0YRVB_ZX@*}HhyzejMl3Mu92Ko4f62+4COKr0#@;;6 zLZ%9K^0x4|F%=u&kWcF-96eW&{@&Ptl6fiIi)G21xy*H*EKzWNY8nRvV?lcHms5~_ zOUhdTk!DAdH zk>3B3FQEHsl>q0}8RK+&r35Q^(DiVJC4KfId#zcb7UJt2Bgm(PnATXWwYhleGgXR@ zVQg=CEaJCMVwsHnDwPW9XP2A5ddRX7bVEXyHAD6o2VfnIIyGJT+SJSD0{(_d$7k=< zvAeCYiR}6)$CE*buT;BpRfjv!{Fxi*@9DjvK=fgtS>LI!(a1a(Lcq5|7ItJL|Jwm) zDEOs%=T)%W8(CsYnIlgx0~cU)fnK;_b5x)7ScgnTyWBZ6N-Z>r-BmK7GF$lX8K5?Z zZ1p$8t(w&|VEjkYOIqtuA7zWxLY{-J1dHTmKS*{-Nx+Wrp=-=hyZ0m(I{JkXm=jBsL&cICt@UV`b`k~X& z3z^^$n3w1gOMHy4>(T2Faf!GS_u0k3ZuinK-_q3OmJ5wJQ2XDG7{;F`6F!L!^iT84 ze*BZikVx4TxO6DjVA*%g4%mTA^b|6yRHYna2TBnV``p>sb-ZB|;Tw8QoSDmXSpxoq07i87}6Jy6n6-e$d zcSBc9ri71=eTouZ{wnncN&0uxU%$f($(Kv{SdD28- zrFFS>%+%~eX3zqTJiRct4mu9qN^Ldv*n*N*j~dtbxM<9dZszJH{*dAKcy)w`EM{JY zd7xtIq(hA%be)5a9W)BbbcrN<+VjNiC;V{JF41)nW~hTynQExf4MHKu9}f{`pV%lL6Psupg^Mv-Vo9UX&Cvc(tvb^?N4nRk+)&d zDCqAC(@!_p@P{MIMvN!u2QDz}bG1T-)78j_EiHlZO&=$WII||?ZyppigTqGHKh2w# z7ZvDuw2FjGNNF?}hbC9*N$4#(ZR$jLN9*0gbiea2vFjS4ix}0}W^l=tAkV$2&n(^0 zB!141OP+6Q>5rE#ns8TY z_qRl?G&GwR{M7R}s#GpzHcc)q9@PTH7JD0ZZCWm3-+#}l3G>l2Qw zw0p77d9eO|xi&e6-NW*2C62s^8UmuOe^y8{k3Uq`TmcU~reUUR&@IE3Bk->ifEpLHmJ<=vlZtMthGWVjd{L5aouueN^E{0a zP4mi_T-Z3IrE|baYH>oyEx1_nWnW$Lq2Ph|so%^uiTcVtbDwy-u}{y+bu$MW6**lZ z`0+0Idy?&y@22x@vvNY?3HfgOBlW4zS!jp0r3I^fGf9-NY)QArr0(C*H*qL2&vOMP zO75<`XgKiU!!~eNDIx+k{n(QNDYG+VDJ3ynkGi0>wv)Z7| z>T1#Kf72~VqAsdtjXOwg>Qfb8uS43zi?eoqNE3qztx|uKlA-@2-4H(yG0OVWW9AS$ zkJLM{58sK0?gIYS!2YkHh@rB>j(7eoc3}~A^Bx9`Gbdw^WB?e*cKnRuJ=W`MY2-d) z2jNL7_o5kOA|&1{HR&$-B_+Nkl!<*NCDxEBB88@8XBb^TAF*weX8NjYR1yYhAr)`<#0T7ZypOOUVr zW@TihL~9^lls4jWpD!M|BesgJ-`YD}cLM)V`Z(6x7m|S3W~Ib|_yFEck5R4-vK>Qn z95wz9#9Act_tQzg(pqtd47J1W=l_PVADH;dA0FjwLr|sC#P9OqgXAVeR{Q#?9Pr7^ zk`z{AasuM0BkbW#CpFoRa9HEioj?H`GSQgN!9?2$O%5&}i(U8|B~RfWa`Fz)I^hts zq6kxiYUWXV9Sc8+p5I&JWJWY5FdS{)HJ@^R3Dd zRV}&aZ<=jpu8zefw~1c@Se-JD09E*IDX8ge-aZ)~1b=uqh8Q+E$gCT_d^!hAJHJFW zy}0f2F5sL83o*H@pDoMw901GNZcnC;0oH1Ee{PLfEn4!azy zipj$!emYYe>etX9wuQmya{7i`UWvRet8%284SDCpDwrIy8|cNjj373WMQB}?E4B;o zu0waGs90*he1J449=5|6S)J8ZMH-I45?RC<-jbT{>2LbtMrv~6K*~JiEX&Sd~6m+XHQSZ z=acS0EOH~^8n=!fp5UlgZW}4VDm@I~tv#c~=0Q7N^edCKx=EM*z`>o}-O32B69CxD zb`16&Bf8h7t^la(mK;Sr1%R*0M2mt;Bzp}3d==~-yUjL!s{Ra6-1P0+OWFg0(JGg8 zi8IpS0OBcC?6q#Pv}NJ?>)Q$M3oow&5VW5e9BxwM7yG5d;%~`OdcZNVym9h>LW1u^ zcl0l=7MnF+I8goz^Uk(}0Wj}Zk7b>I$S?NWt5ZOc^U4VSzoJayZ*j*^v53QO92N+1 zQ=mjWS7OS1RV=Nnuv8Ss!AbzsrFSwlrQ4tdCXM14ApYR=1HirKQYirF=)}oqZNKFl z*IwlP4JiK~K)(Nhf1g$HjdTdq{Q|a^8~2QUqs0Kz_`gSj|3}C-r1I-Ms{cT~JdISO zU#|${R`QK^p6Tbw>`#u?)=5QJ;&u-d=J=j+%ktr6AmpJMf1mA~+0u(yX6*XoQym8F z7k>OqCLbhs#8b<>Bn1CDiEm{{zQR3H+`oP4#v2r^sNTao;GzY3z zr%eu~aix;xMK+8z5TJ*nJ?*Ly^MD@t_1{*CS=_~e=zep-ncu0kUPmL({pLFZa{qty z38l;P99aR+WNxFK?!A1V7fO$(tAlTzlhU4jmD^OZ3iwTHlAPB9e?)Rwt{DG zP#w-AX}zL)alMmHO?Pf+|36KAl@(hEg&0TvC~;nqq0Jri*F~_n<|k+iH5zQLj=Grw zSG&V{y&OjhUKQ-NB~t}<)&LB#Djyi-@oWKF#jq=nE@vDq^E%ms$Bf^>qO#Bad$Q+6 z>PHhh(g#k3((R!vfS;~DQ)3^>eZ-LiZApR4OB1haipDaQz`KT3=?ZyzYaUOZsTOHS zAu$9HuARX9sflKMWZpzX1T8-Hpm-mVfVAk4=ththzeuT_P#= zPqc{^-^z8Su^X-4w~IP7wiSs0k)8V_ZjZKBr6<hl@qy+sk^V)%Lx$NNpz-jILQ zCwMKFa%lyZ3ou75ui1qEjb<;-m93PGtpaPrVJDm`5fw!|GLPR;B=Z06);cTX%LPzq zhe?92+BM^ByCJU3hpw7x0dfM{dbzF&Yavws#w9uPyoC^0WYNU$H|!D27Z9-`xb+d3 zc|*+!!2~!@Nt^UPQ?E?JGhWgJ}P_q=Z_>Hp^~tZ2;7+@%bLgoH?F zY}XHti1;PYAK@~@CtQJ(0r)879LCAw{nabXKvDo;KF6sclnSg+ev)ao!Q^aWH>fBmCk&yqWcJTv!w{$TV5-#@Jln{`d}e!4>qq$2$8nL2_Ro32p8=~Otg zl%o0b=lA+P#{ay4Bp=Fm#QWB9RExmyuzoJ<<8Q+~esg?#`fRY}Q+SANGnA{b78t}E z6@W{h9$QRp7mBtv>)NPji#=JrCZlB1w! zItQwEg0`tt>E7Rk3JZC=cYkiU?V?4dVHAvYLbf||#YYnNLOw48xggX)f&%T{&m^h7 znBfz6ZrOKeSi6Ivpij|9i?Sq~m8 zf*+@@4yHC=if{ci?ny%-exh&meAy}66A@; z40k28lt8rWy2)ofN!Jco9D;!>2O%Zm2_L|Ea&rncG7AK@C>|cfqQ!{oh%tlSVhM4{ zIHikfv^DT5NY_LP!aRMTG^<*TJ{zfWYXWjZyhcn^beZ(_4>f4X~ zK=BW+0Y)^fj#}fGjlS)S7cq8jjPcDsaOYM~7Afw?j4Y~ZeKF@{>1zQ`32C2P$#5}Q zxWgtY^3tvOS%9mfVaSqMsL7q8VGP@!wSF#$V7Nc2_*+hnP%Zc1w;Z7mx%rGEmn-6D z65vLTUvuZY7{{aSDePZM5<9xOxg&4qbbQlri}*+f{vM;`de5yyaqHg>;}ot#<*0XR zO>u&Tqkr9I_++C*s>*7E`_ChF4+mxkt+tMBVIMy!?w)0xON>a{p@0qa&ew)Y-BP8r z-}|wQ3!a`Su91h%V^o=l?e8?sy;2$ya+&z>59MC91(MKdSwS72sIbazS36o};@lW# z*-^+bMD_<``tfWd$a#=?F}dA+d1*{I7$wMaCCA>+qZ-t zjssPZ#k7^dV+hXh{6*xKHt>PQ8{a5m>L8skH7aOU_uvi6U zUzdZB8T8vXdXgrF%0J!2?l0t#0w;2+2PgY@T_L=@6JdnR^k^h}b+fFxM>3y19xT_6 zU$XT6bCj^X$01s5JMCwrR9e;!7=tCU8@GyZm1EHO2V+l0Yx6MeNsv@{WozK=ukn`+ z@=&|g+3$&2Ve(lYh>7=Yx=iT)Sl7E0LPu*MPe}eDz-M!_x;Z<=b?7ctJ07PclTD4U zEJzx!aes}L{>FZoP zAlg+tFX2>i1L;|9XcG;lEvrj48biRT0bHITW)G!6!4VG~8GX?pAph$cFbt?!CT7xY zD6A_oGCR9fC*(J+{(_+!!0iWU_4uWx+rE79w(;%^#wTkXMlE#CS5Me|mWrz&B!)D_ zjbI)r5>;9M>7*L&D`{6bwX8`Dspsu#kT!G3@@y!x0N{!OAVolT#zd=vgX3y&Cyi@B zSf{eb*ysL)$nS*%5j^&Bk$1LE)zDwu^^z5YxSzr4zPXc!s!1vjnN=QDC46Z1D6$|D z-o}o-UlTwhQNx+@mcuLFA+O3gTrYtYk%$AXy*Ez{@5eW`CVe7v@&LsD5^0~T`CbB=_brj z3emQ-dVVvf?KN|=c&A|1Zmf;TiD+$Kj3bpJgPwk(rSaQ>V3rB0vowOU zXcVtqk&Kr9zMPLzRp&6&E4uD6G12zy?XqvjEPB1&FWysDmi@~++)n&^C0~3?gIPLH zSgSIf2htCIdtbD=s7cV|tEYnH@we@AdRlnqvb{zng~HCT3jX8JGj&)qq=Ek@3VoCZ zUlm*NmWt{a)Z*2&l$=f))M`aJPVgKav_!8Dd#kUV6!W-}wTGGRp>Igf2%BrRUD)a+uW(^Qxk~OQs+{qVUOj*~|NUfGFZ;2tLz;3C zOEwWf+uzM^9~el_kH>jDQdZQ=l@wgdr*;oEv36i^tT`ENVx0($T82+08T?s|FA2WN z)ep)o>yKXaiH%IHAY#?3p_nJTU)&9Ts=~pIUxPa~4p0y#- z?uM(L7ag85BpS!*7wXy=by6FLrZTr1w6|{~2m{GchN*UQU2kt}0B$A3uo5Q(4+w!Y=sCVl z()2&wM`jl}@dmi1__SlU-e2S95dpa5j~xTAyUaA5U#=YlxbD%0CuA?-!cSIC`2f)l zw;oAd$oV;)#WO&-V#q5W(dJ4cy)Q$&Vmg@jlp${c2y+Y*bnpvY?b`3I=0I<1E4J_8 z@e$FB=WCYnkO6`Y2;hcSo|HR@2D1(!VNTw*%R_~(H16MlxIn?W+p}lVjwjz;$bpIh zF%pBke$nqdOp~cl(Kh5K2u<2KTsW=!&dRkx#XDl|vbSW25os0aoJ&g|QXg{4T}Ok3 z78ge)-J@Us3?Q(do5!668nH+duVlm_Q_>e)fksWI(O^}e5u^L!o0yk`^z;5S;x8f8 zGpBIjJcAc_SS82bDBgYt%7K>j_POb6Gb`;t!p>fOCy@hf*A~wQ22wttf4iI2V3F=j z@kzX5f^b-t3i%Xx{~H0njP1E1OIBa^U-vbPwjC7EB-hLsk!1XU_h%=*%B2KdFW9yI z5`VL)dLbzv!ACr`HeRXbn5XC^$tuR?m_l0Aa;W!f6`x`g8C^Wn2!s;~>*;`4lPm02 zZ#f?{CfMZSeb33>{D=I$P zpGAuj{C7u(5R1L%T~6U!1GTvI>yl6;H&WNtZ^J-;8o1#JTE-2OX+6>-eLle^Wz zz2GZz+>PN!N)8(zyy^npLz^hkV2gq>c=>`^yj)iKKsms`BMe{5`~*s21k}nR@uC;k z-(Ylepx+k=Ssg?~1(;~!V2MCDOAat1IpnH|-)3|F^A1jd;B{M6)+N1KPhNjXwjU`Z zB@Mpl=*S^2`N|{-n6Dtf%6gTpK3YV3UVq4B`$B)@mq8$!%&VjkCH$J71a`?r? z$LopgAD4UWmbV(+-;;zC>p&5d zfGOZ^-s+HVkScQ*&SE5uun3tLrd%Wyqt_HHABfOYpvGdd>u;ISGR6~5@Geg2Y znNC57UNOv}ND<0`@-6PdbSjh8)fT=M+^0``!1fk!8-NjyS0Zu@ea4+IS%RSBS%tWF zmFG6At4(0$?-PAC-|n#`rH~%<&(~Q_52hVa@CJACcu=qcp%Z2CNw}<7VxxVR@^aTm ztrt>)=|)rvX+JH#-lH=4Bca$MHw*Dx%;jVXNpGNLz)DH$CU2H=iXiCa8C_`Ma3eG9XMR`Qbsit-fr$0|?4MPs2+k2tv4HU;2Id{vsg zG16Q9PWZ|6w@`bEuoTkB#zewyV~;geazEX5hbxr3MC^Nvn4y8z-h8cy=p}UtM^@Am zlbT?X3k&7_rtDB8HXAljri-ONbNQ4yg4UbFJXQ&^gAwgkgTL$^kc|&l#LLu(xw~!9 zAqkOBvV}qG3c|PR{jiEh5saP1_gZ)u!Y|4LOHh2UscR*=Tl0n#xCjd&VI_Pb zc4N6ik;}oT-iEloc~Cs|g!FyD>SU3_J>;-mkYQOk*>Dv|E70$A>0GBdg>pJqwvO*M ze&1T{B+TbK7~iz=#%*|OW_s8Td?+#oHA?qiK2dmd?ST4}+!-_&ZsKEdND9>`Cyy{4 zV{_MrY`(02-W?7((aIP3P!VNMoM9ao2NBzT0Vx%NWR1f$jn&NL_H~vUb19^I+wCDh zl0*4X=2!pasTbSgcRu5wig5_h$x~qrDXcJQ45$S|16%ND; zjTO>703hlC_4*Hvf}B!s!Js>eI@X++T*<>T-t-hd&f`FAJ<0tIuBE$``DH8mPx!Bp zn))z*bz=4dlCma50SEazRTLrrhTrS0IPa(*o^uqe&+I7bp*qH z^IrK6yaI?O$o<^Y8J^WrT^q=kIpRrmAnqwqqb^k&K{t81DI1bCP5!P;yaS07c$w!c#03W(^}yiMhPJ z*WG^|Q$YM=VJdB-j*6?Z?W`TGJ@0)E_7L(lxlY3uA6genfqO-W%y3uo-SBvldYSD) z{RoL*yybW@hZ3h7jyUjLoi0t}Vw0r@Qe<{J`MuRn(4N9`w?ODZmrk%;Ws_x6M5rB( zR$>#+sGmn88DQ0csfK;QjmSO#Qkl}H03a5w&6VgfcGMD@@(S+7qDW{t;8oreW8u0w8SyUy1+}O6u1zO*TwJ`amkE@~Vq zMKF1-d9=VIN;-Hqg*9APj4hvc4(#WV0Ibb-mujow)aeU2P#gq>S!kyBz=s#m2W0Jk zT6bOZXxe$HbSWmt^NT^Qf_L`i_(W#>Jv3RE-%?WVB9G~Izi+7dr}^R;+2UU3ZzVS= z@XI2EU^36TxOB>K-wrE9&ixp_hfWqqaO~yWqD-XJPIrw5vJqb-dp{Qe2(}9NF&{xZ z?9NurfNDRl>jO%8%Se}(nJdnDFkmsMO?kbPmLbY)SEAO-0$NRTQJ6X;}n3Pr$F zj)LLKdc;FH+wud#Jn-`yclvcWva5T;7S3)L-tKA z7c5Z(Jo&MKg5yww7~9)7KUbY{Ne>n1+lseWc3(!C&0UQ^S#7@T4>PNZOGIjFomIkH z$!{7|ySbXVI5{lpidPKmtECgD;tN$dgvRL&h1{7gBt830Z4iX~tOZ?W$$_fF4`^1| zom|Jh&*tP|aunnBls)7~T(wl{f_N~k7Cpnbg&K1`L)8$^DS}*4!(O(wTZjRQgsANZ z(}T3*TNh0w(nsXK_!g2s{xq2@sV{+L6uv$E6ipr<>p^j@OG!JYkfgn*4HblXO)10E z^WfaRj=dudHLcdFiBskzR4}1qc7BMEB+OX)Vf{e~FMe2i3VS&8rf=-%`T@cVZ@Z%*zMsNzVtnS za0)9P9#am-9O8$S>cpffa6YpdDZaw2&^u(dmL_oE_;#>JX+!qF)2_3RL zjB(C~g(FEqiGDYn)e$KSIE49EoL2=hj>xn8X2gJaDDPORSE?E6NEhgDR#R($y6(e@ z`O#nBH&$28N2Y<}t$+lyOE?hiHMA3MP!_qbTiXM<-fKj3Tj$R^XtQhj!B(Bw@h1HG z;0a2FPpv#rEQJj@;1plrT{)?w=G{T=(4vpC>-m^*Ib!CQX=}yEI$!*<2CGfN=yM}Y z(hA3_cw7p?_*USU%}n~uwe^?DHVFqsvgX~X{0gHSti97LRu)# zkn)GLxXK0L7()%HG&NvYCpF096Z5~y#&wO~dOaP4k-1ceK81B-H;fu4lS)lLZ<@Op zja9!0tL}^PusJkaN+V-DEd6YLz2E3qA&h1A(4K(sm2 zF&+GhAaep3$$k;Zbu1=Hrqa3Cw8k?1$JB%PaT$KCL`@XDfrLX-5l`K$m(Sq4JS#3g zmG_H^zs#pt)>st-a=N@kZMLFtcLTV1_g41PiXZq%nMjyG-~9yJg`= z{PNj6B$xdayH3*+50|lM$w*M#Y~5M8-6!6m*P55Z*l89Rngfb6O^bOGYu@zBL=+*1 zb*a}<)}HQ#tX#9C9FR(H?VBKO1=>yVl~nRpF#asirVeAg2LgV43fMrU&)=MWc$x*E z8BpLUFW~;S?%p#hs%Go=#fXRq2m*?Lpkx8bNkD?2`|j>$Pkra7cFRPay=>2KFGB$CskBA=Q=M-_IkrzIQcqwv35!zF9n8 zt%NU*vLP}Y@oxTchlj1cm5Jm4XG{uR0~@nx0%n zrhn>bpu4Vk-HRGfT zLSvw%RG#;RUf$oL6w#vLJf$?(TAc)+AatAt&8hk7cPyWgOqo!Q7T-Jw9J1I&bPvN` zL+o)=$$Rlk$7uJ4MoP)W3Cr{5NM()e(Rx4oBG~)QwD{E(y=_^wS*)Rh>I=m zSE{0^(}NbrT%Fv1NVD4ow#}BhJYSP!*X{1n$YH$l4OyDzQZO~1AE*YKC{ApF-XDgH zy|v=V8E+WNO&wJHU8W4v-_SGKD4!{Az3AvJDPEn<-qu#R>Z-$RxPf#ZJ0Zr-mn)m8oxYx{@Hf4$Oml%mGH*_}! zgclx8rz-T%ef!GEW(V@~J|*2d3{CFS%jI-4Y+@D6g;$oWTn~|pa}ORFS`%7E$7jRd zE=nW_S#g+9%-S01z>9L~Iu8nP(#wy8cFmu5S#G`uqeZOY z#3RpoRW!EoRz__9o+i++Fm3tO-=IB3b4O#G!Y6JRNgr$^k6v;_7_7${Ry8ar#=|Q+ z$P=3@D-0UnT2alar?-GqocoA5dB_F|-($h6tqJs> zO^LeQela&lisvK}#{0O+m23XVX&A!!cW`T+{{I}@$dZR=Tw1l+ZihtMR8GWI!~m=eIhcPX?1A2t<`0K^$<<Zw*t^Bhgl~?=kkZRfe zUmjBP1*YD@+P=PMRuvzf>6qc_S+01>8pM#;a)c}xCv`w-kXzn;+=Uz6`#>ze`uVp_ zWl=PxkwzNxl*V!irMKI&@FuYGQBdeOz750zM9lP0!uP{xZ3WWezh{ZrztjZ*J~?_e zGNgL}LIxu@wP4mM$lOMf;uephRD>NosRO&)R{u&5Ep^Z+xfERw`zLHenI;1r0Y@BK zl*-WPdfs(s#e>}A7yCC$z0H&R7tDHeGbN#Oag{+ERzX(RIhfVN4X<~RMNjRnO@^M3 zOuIh^EXHMzWf1|2em$-{4x&xJ*i&sgc5ma>bW%V95^!ZhE;TqlcmF+_5Vy)2EYG1Y zD?}3Z)UCGA+VW?c@zrrk%+e7n>Kd3|JH@4SKGl#~`}6oJd!P`_TJz-)DP&o4cjtLI zc*4B%TrWnIss}aSOf_+vJi2!f)a6slTwcq8>iE~;i^$=s3a5)**`2Jd;v?m|vHL^$9H17fomT0i`LZ zCg#H+?3L7SjQd++V55M>w(wPaM(I_*pD49R zaTS1xMLe@!iteij`D%iW7V)CN27Z=vhr-g)mQ5a{ra0{XvYih$Xzz!Jk$d8O(wS5{7=Ym!y%<%IQRSSACOHgop%cy zULKF#Pj6UzN;s#p3OW5VJi#lRpXe$&L;7A3e5HLiD>&zU*h#d?D<6E|`zRleF9R5C zJ=P7d17B8$EZJXx4#`^&QX(C*;%|guE^u?yve!%0eTnwM_ruSoQAXu8Vs?r;JSxL- zzbomsf?w8J#4|D~w99$aeZ}YA5Y^ozwBP<_)5JnN-YcO)OLp3cXAq&De@d5TU6C-sp#SjOEW4bljVH58B>s+~f;QOz%!8@`OYC zh3%t?rB8XY%9Fj+yz;R6fxkQvtk#ZC>C0?{Qalkw<@DB1=iAA>LuzemG`~*m6r%qb zyZfAMwjIjTkYF=ybF0;JUnaQCmk;dd*laZ%Tj!kWSck`B{Q!C&hZwvNAhcc>Q6M0l zI*Sp40a^{b01kZjzxK3v7}z9%z`Rh+-ACRchgW1!7?1hXkk9zIma8(>Dq}X4TW_vf z-jizQCa}Y@xQ2z%##j{+blUiQy7jR@r?O-45(ahebH^z0z-r&W#eWZg%HKX8K3vNA z9eT!+5@KOQa}+Rs+%gv_;O`bsx`Fi!3*k+Fj)(pOu2Pq_fZ-{zFq##2F#oMuI1?vKvsP9_&~^egqqT?cs`e1@;dBR=T_d z^OF)GEMd_4Iu+K_Oc^uR$$70K*{)!*feWwz_jrW9jrSGo+ifwI-p7FAF!1Zka<&)M zy3s|K&c4KdMu%b@f|7g9InPSw1rO)vmoql=eKPow_c$M-4&;IN4EKj;FW8|PtnjC-XN6%fu zK*FbUaTshCbJ(Lf9P^O?!&x>LyWrxMVqFES8RT3b29+WpZ#bd^GE&=gJW?p>_sB>1 zaC$coj%1)ZoZPu|%5(h|K0d@{&UeQ@+i-fkVhCLcl_A*rfPD0-0J)U;(jGj(uTQ#m z4j007oGVxmu#*yuV6dr35HkJM)KC#ogt3c~$R1e5sTpLSf{m(FY_Og25t)7y?hnXW z@Lw1c=v7#l|3q>@XS#&>5A%)xUSG8N**HhI-f1p|?#pO0tj)A9#of79d<`LW7u>lb z`b(r~*2bNOLc1TQvQ@Oc##dXu<$JU7^6kqNh0aaes~2B`l4DTQhHtW>);Vv41uZ5Z z`JKcTH~m~bTr?C6PdgU^eZt3t7Wjf6xtLk??-S9Auzgu|6Q9D)X0G+He|>*K9X+fy z#8&Bm2iE`_m1>IM8X2b%_n0S>$2J>8kbfVBjrRzgZQMO#Fl#Xr2v46s7Z_eHyl8zY zr(TgYF@bXSHDc?5yoh%6q#@Y7JbPTHJlVN09P$CqUe(3oKmXHhj36zEgrn?oTpTRA zauMspr-|eINJRQ+z**7YV%4X1Zw@|}>wp_v&^nevDV;Ud?Dc}-B9#PPONxh*w_8ix z>Hl>dO=Xm}ZEuRBPk<>}{_H+fMNwAY%aAhu8~0`up@1Lk2y zjyDtiyJ=HT2j1{IXH|I!WbRg`o@U>XL{w^;KC)M)RN2+glNnrbYhQ~U$_zLOPd}w) zOiqy1o((69_i{_s-TdjAS@PS#%Por=EQodawSqTOSV1Kkwb!AksxGf}PtD_kJ*qT? zA33us_hfZb-VelE7$4vHGQA5VTA9u;O27L}i&!%vUv~464&(tPT!7hD#Qw-B z>?90Xt5H0E!D#xrbgY8xHD|^ef-?|6G&X3_F!;u+J9{%=mBBugYT$ch$ONi8 zx|k^4KjQ@Of8v4t8+C}m0aZ#F`a5eL%?wywoFOaX!dAHFZ>aR+p(1lOVvhCvUUnm{ zf%{r1sa@^+ocodoR(Y9Ju}hj4HKa-NN*sU1KRi+s9_|Pj;)OYW zwAN}7D#Fboh4gW~MSoFH3RXPK!Q9*V@Ms=vto{Cu!m# zo4t+v$`jmcmhhyEUsm})BkM-%U9#nO?+gIZ>T|m_rvYVj*E2A4{uv(4z9TY%ZP}2U z+wTi&?qhM6rh0P=@nur%>P3NV!?Om_Nz%4r9kAlpPuWea&6Jlav(-YHc~o2B7-L@3FY0H0(_k;YY^ci3xsyij;wUjLyuj73oGfPj51})h!|WTe^L}C z+g($XuEcMw05|{qQhre;x*;S+A-2|!BaB7$zUfgVW3RkgmZO4hY`X|iW}U(hI9u`N#9^!5te)9lYE9;Snyg**F}2!yI4qJ( zE(qmv?_gl;RFLcBzU$?_wEY#$v_qku)DvpcP8_F7zI;0@OulLpNS`{>j*`)8qx&jk zdwudq#hioj02XV|9hH;Qswc60lg(DI8f;lz~ZN zcHRUzt!Nq=UMp;b7pd|b6%`RXhu2NFS4d3f=RZB+&r}-G?nCB=4z7NuVLV5)M%^#s z^ljZK=F?-H#s1P+L2N2wo1U4_xF~R;;`K@Yv6C@;`c|De`+VVar_0{uj^d=*ZBP z$Gd_KR8h(`yo_;$VsG|z%?5H(0r&%~nG)?ogvL&7FfX1i%1>x|Hk&U;qc|RKfM@xg zVmt%jtf0Qur$)U41&!J#ur+lPoUJoZ|HMfj*fx&SEkmth{4w zcZMAKQu(iXes53rJ0h%M_>^!yKuCLlhbW`9zR%SVu4*-bxViXCj<=SCtZDh4p0sac z_cEoaly^n;GXd*{u?#1OfRPicrSk228iPjKTIZ0a{0Fzr*`Jc zg+pyMUb_LOsd*|Y$qd!1-+G@A-!^UG`KtHUK^;HjZF0xx%Ss0$S@~)0AXKQjpwF1_ zg8GY?!Sc-ir<}C&)5Ti8H+p$Nvzy9q#;oYR7__+PMVK2WKXu->NTog>FL0C!W;gf^ zZ`D0@?=g!q-HBz!^F-LhC*@i7<`ZWHZ;HpZ$_geklI}kaS-4NPYP}=5GjaW(N~>S9 z;^)}fvPx}MkxuLz{3ECC&dQdpj~lt>wG57(a_wnto=!QwfG92H-3>`Uo7_m74cRO& z5#7$42x6fx>{L@w6}Ap7m{Aq9k@}fGFEHDK3fU+gPnXgyluObLW^I5(!Be*U#ALDr zCZ!_wy-CRMDy|o{)99cI8ro$5-&&G#-RS|8lIout;l%osyU?GLQI3qP9V$9*4k5}r-R{I zOX$DY|A^uKUUAIt@jx7JF@wiS%H>gkyPk z$uE$p+uYudE|;0PlatD0>Z)x1cAF!StgSV1o?!w5I`$hu5}Be7;Sp_;Dya1gOABB zF$-XIz`RivN8;a#(SjtfCo6*^H!{l?I1$5J0Bto#rw+{-jLXi)m^xyGWYAM*?^`^_fG!vy(F4#Fad_=j#0<%yEqDokW#=_)@Z zY9ex7py`baI)EFCE_k5qRn*PD4K9B?HAx+SvFLEzu86+$F1cv7h3Q&+@MyW63WjmT zbOq4phXto}_J^9|Q5Ip5FD`_l^Nn*I3=8kgM7LMx*J`m&v0D^<<~ASx&8#4Mf1!?* zs#jPquS5C|>q6VFBgsDxE7#iACZ|>=!>>t}9(`s3BcOWNz^^zX)Sf6$)O?Z=Y*oRU zN~U6V9I6!r)$*o}lbtp39G0EEE(2zu($8_IhS|F{ImN1OmoBAr=OPCpGubv%O3?#f zds(~bmUxr(;Db_n|C(dtm^AHFB)Hd4ynVg;y*AW)ynQZduh*UI6_cT&+YO3xXSh`z8shtUP@}ZmEu^nmwqJ5HZW{3f(Zf6CTit zXRmh5Yhm67+RY6=BNYJwVeMHMgUcSSmvd&{$;Vc@oAt2Qe=B9B$m+qzIr!~PK_aWH zT-RRJooU9Uh`|V{<+!i)x>e40f%2jh06F__yVx7bMbiclbgH0b*a&@OK;tN@iAY3M z3@H#4TtP_P8lrNg$vk!Cd_a!Vi}_m*W%EJ*_cAZ+Xrr`uMD2JTcDimCalcF%tuu3{ zMiZR}n{Xb{qT)N=*omLr7<=IYb~1lN0Vhjdbb&pfRjcp1eL8K{p8XA+EWnYnq<-UW zHdR{7NP!LuyB>fF?TGdw)$-J}W`7q{&M@(cj)9P9?vAAree&`t2f!kN-}@B01LhSM z3zZkGLk%=Y0v-bYN%@|`R`p6+7G)YV*3;8-#95~JzJ;QQ1ksl}`igRy1wj57go1uV zQpc&r(;&2&CvOV$YSpbjhNf&jHUL%N0Kh8_I8ONE?Be&KNr|HVJ-{bGP14t2dc{J}VLgZqfT&mYtX);yC`r{2yVae5&N{0~>sos|+{1fJ@BVjpCtIvQp zX6h#4WrTPg85!C4;1Pj9z*V~cX(p8_5nKSADk5y81}jC}bP-79vPEyK=8W#m9`!hD zW3O^Us=389spMAbw>^v24dpg9zL9>gW0t@L)iVT<#?hnqHQt^4z!p}NbI>e{pu{08 zN9{T8;n&*C3Rm~~-A=Rx5$zj&R@OOkSrh%cW5tGNo;7XpUv3}Obp$Q?MO+Yu>iFQ= zW#HOUxzidB>sscX>#dSBBSpm%P37Uvbqf@bqYjPSB`OMmbPzhvmo{peMpMVA$JP`@1|taIjw zGsWQ_ZrNPg`GD{7enyz^3POBuywU~|&u^`9rVLJOmwV;Dz;pCve9C7G3RmYqts=Fo zwP&N+WKH@|8n&R#2eAn7K;MfUA0Outw5^a#V(kKP-bZh0UITnpcJu6H!qL$YJkpAS z=+f+3xg$VZU!d=|A4&l!hkhwi^_cyzAHSM0enG0tM}T4tI7kQUtSZICfT}ArPxC~1 zsghl%(y-zFWc{|=EjP;cD)@cHALW~o0Kw^BxT87DG2hruhz>dxN*<`^YyP~W2POdG z8+^9&64(GZOgjzG5jD~x&qbXdmD z&3z(x^o%jL9gq-D>UoBF3LgE}M!jG$7{1kuoGsDS;4-M&U?ys`P>&fxAU2WE$Z*+H zGgS@*e%I!_b9ERqvj_00ZOhFe{MR5vy_mKZw- z9U#_SMEroK_=R>af@C1@SVwlS2&yTt3&2`@MgPJ^6Vw}k%!-$qBP*?^KOGq2!m;0_ zYE@EI3#zus6w!0Ygo&DhNOxylQ~ul2WE!J9`ha@{JdCtqR&Sr2=`zC)^{0=>zk-|N zjiNFW@lrni8EC?OymqX9GBU25AU1`J(V@U*pwvMR8t)7^Zlg!Wk*&556iG!GE!a+9 zdT|n%gGLU-popE@qrxJ&o^XZouzgFde)%r>fDf}V`?|1}YK=dAY-3{R8%xIt?(r>B zXw%I>Qp;qZ5Z=)}fiLn1CKNybY)KC@Y^cFmR}kophw}mF?LN?K>^XP{Md4kP#BojW?dAT*RZE#atoe>!3R5IR69JjqxRDyFeABEroz5haeo zL%F^H7;Ec(&qBCx!MntLR9Gk5JpiHlGjO2MypNaq)Y%ML(RA`BQTFL2wr{h>+!O5$ zZD?Zp8bGzIz?KD7T61r5Kjkr+PiT1X23T6`%AiiN_vg)zy2Z1BJ*IB34cq^{p6EvN z3=g#?4Q=@d_DQJk)aioV1m8hv9~+xKwutl%5T3;jW5F< z@DuE)iNgT6>Av*B6fQ$&2&*a>tA^uY>GPs>w7nTAB1F=R8^aLX@jfAK_ zRWoRs>t27@x;ja=4%_0(W0GQ=i2&{+PIh>btwJlYIuEoq>f!+dLKV3(&c;tG=EDvo z$1R}Bo*wn!dj zhD2;zj14;(b#a!ZpZZ3+ngtA@hjCnEs2>6~`1U+)>x@CtIUFs#(KVtY@1A-q$;e@m z-R(tBzxRM}dCV4>P;x`*0r8qJyIXjUA$5xr&x1aF2`!Wxko7-TZHZiT_-X#>x5S!| z*sXvsT3BSOXdAvvU?!1>0SZi0eDUC2;dU>B9YrZFhI<}iJ!5C~TpjdX?zS+WwS zR0fnCX_F$T?i{T*w^`U9010xfuD6_MDpo}DY2Dw8kRR~L5q`m_f^)-QQjg{6gAi%D z(65hfjt2M;lBJ)BZ?2)2rUWAFz)pk0D3Q zS&U+L2Co>r>zog(MksdA#l6ja6$s+e9yNX)Bz_anVP)`rW+$qQ%O!36H)|gGDwC4K+uoYT9yjAdSVA^i zGB%e}=akVggBL>mUwW+_;%IKAj{YIi;JYhQ-LYpEF{jsT+3gvhFo}4_vpYkdKQeVD zw{^pqbDMOQnqU2SwsZ30uc8%53z{NN(P-lrOXekWe?AdhV8-lndAC(gLBAfPZ zWFQr+BM$C=1RGvofbVRi3$NfE`Hb;VJq1DrmhQEB4Rrj+4shPAse4dJlM`qDgL-+x zxS1VyL7e1VS150|)Hw*3fH2 zfTaQ|Jg=%ii;-Pc#*aohbY|KrV;yrSC8y z$dOKY6{F2Q{o8{-sJBow#vwCsq6jM8Dg6dLVTN0yT(>nk15-OC+8>-4|DtBM_nn&g zqiTEU>O}!0fUX5Pai3YIg~RJhhZ~>ZE>s}pY)LHz+|=nvTDJ-TAW<3cr;@z0PIDKr zehb2oxe<((b==sJW_R-aI`#O=N4y6VQ-=!us+#8^SX%^0|CjBRj+* z4VKg_8%lW`a+kzk{r+SKjz`zG&t5x=vC;n?Ck%STj{N z_@FIP2Hs{`934)!;tfDJVj5Brx(KLbkhlYQ&cg9Av>_ERd{V;a7riq?b@adVixRd# z6qc@tz&+1;l9_{*r1=7|gCc|QvL0%yyRFaR%QkDr~H1z*2SR$Ikyh?;#6M(sb=mjZ>xYb_(H2dvU!(Nm|)!Jf@aKm-iaqtm`3?bq?# zGg}fT^$ahJ+@@KjM{Qev423)3Uo))<&3DULK>~-Jvs4XRQ5d!5aT5%}O=`_`P-v5` z@g@19nhY6UwcSp3X>T;dFxAWdhY&04o%dZCGnOpT=Vd^Bh~iFWIBv@ zNCQr$m%{*?!{KT)nBRllEfsBDL+(m=(WKJwQ7h|kOT*7Yd+*7nN)FQzRVjmMsl%&$ zj}uVuf@ywvyh%iElisT@foT(O6^&<$m})9ubr0E)>RDAN9TWen7$B$kN#)6Oi%zxN ztq~o4o;x?f&3NiMcXpU!gwsQCk{eA7WejU=&1+rFb}Mw(RlPw#0yYR`!x)HmQ(gf% zLS4=7Zvry8kfH_p!gowwc2OSHmJL9Lvbwk6()f6nd-8W}`_f~Qu;MgF&LYZ@oQb!Y z`-}}w=`PG*f1b~6m8+!VxGc7;8D5;FvtyL6f#8QN#SAw}-Eg$cX~jwC0Y)_YWPFoj zLJhim<#NDy{9?V|!+#n6k3l@cHb4{PX62k-bmXKGY|q_j2;EEd<-%loETF+i^{!vN zu{FUiOG_{}Dt**kc2I_~GW>T=E+&#)oP(+Da7*fMsl6UU=AZr%%UsaOon395HCT*cQH^!Xy6I*c@k=BJ}Q;8aj+h}{i!;!B{-?3s?@on z;~QJ>B%HTdFMn-SD^~MCZ7ooxpP^})wSKZ|5XCFdMEZ$Qa6v?cqPA8!7^8_9xX4dq zl_z2wN>T|+8zXmkQ$LVlyLb6A!aEnlnhlcv^nDOJ+$YNvuVDEuM@xeUg?M6O zw>M^!|DCUFV9Z+fA459dy$E_XLQdxqAVa-I$L^HmDzto6*#6)i4wW7+eK*K8a|@C_@RXIAM407bisyPn3Z}nv#8Q1q z6O59)j1?P!I>beH7Z?$B$?mw*0%~Byp6Hrcbnw4f43X7KcbFEb)LNX>)O+~e#VxF^ z6q5pIXt}GmJgja!2cPEKpMGdTJ+kEdMC6JNcH(sqcx7}H1!M}z4r_4}K^zuoa=rA7 z9H83p-~2;gDj!BL_X!)Sc6{W#G<_8i_yl+48wSmZXe8K=gS%9P%M+wT~EGrJlQjmH$W524mS98=nC#)|F7C=Ebmm&F9h(zS3M* zZy6hi4Fu?2frQI_e3&SZAT&(!>jTx+KG~R@SmEJENVwcFHTPZ5+3*MUU1<@U9t1p2 zURNWm4}u-zgZx&JvphZoz)i5U?mT!hdE7=DT+Ic76TJesit4gG(F+r#X#CBwZT`EG zPnd;~0N}D2BB31CcYWnmNniq2u71hH#i|1(qp|0)u?bZ%_q8}DynqUPzZtzrc7 z`18~YzU*v$1T2@U!23cVSC~|4*m`?yoAoRv_d+9Z?NyIINYw_~{vT{i^0hgNV&I^1 zrFpLpbn;GKW0J4UQ`llO{{9{U8h_e#kAVB~8R2s_j1XYI?O73+Y;QyqDWP^wFjn0ZZ&-_MegL$S)|561!gS(yq!NaTKLD$^1{7CCSq2O}@ ze&_FxfSr5)+U$(z!W4YD$?;D#Vtur88Y>BiA7cI_Au;)PHaPGLOp>^Vwu>H`exxUt zKtX|zFd*64$5)&4EFdS{!^tbB?~$S`)_vAwL)Q)OGV^pCdCXKHLPGVz!*sZp^KJm= znMG|0+gF(e)QZ6)CaH_XPM7A1z6G*fV^pgaBdJ2 z@PE@M33U`bXJb?zpu27a`ECr`_0sehis8fYF&e0ZOWNo)1E-aDVsKqjr0SotsfCo2 zp`ifRm-8L6UJW#rk~fB^_&~GKz;TkL=axkW;bI#dyfJRJAcwMU3`;vzI+)X+m4wp8 zX#WXxq8^RAR(Fdo{n`6x!eQ>(k>MVg0iU%pV0~O^IE~uTJLWNN^-4UqGO5B|CuM=q z(X#?T1Pk%O%Om=2$2OSqm=KqXJ$1*t6bi&>h_V=@dC4)$L9E-MW_Ie<133v@w+cEX z_=XKH9@{TTR2)xodCi{y$JjKrb|{b)6pZA*w|O6S>7w*Uh;?psQ@T2&H+9Uvqb}Q>w2p*cL1 znJX4Fk!DjRkvHMg8F}o)T>5N2*7qdo&&J>mJw5JGyWTkek+?=AReant0gY%8euncT z@a;XoKA%00XR2rq>y4BCjhs)XvpXt-hI!qNO6LI1O>W?r5D;vjRe7gjZM^mSaMfqB zXGhmb!!ATA4S{dfSyr@hu0NM{Jh#*+VYgRECs+#>%b522E&ZRgO_|(Xga+Vcc#Uqd zwPEtL1Kh@$H27m)Qpbi72esdfNa^d1yubP0BlGKJ(qLC&fuXuClKWk8=~y z7>dw;z?2R*>~D7W@0|nwzS9l4FqhVE$AaCaQbghrr91T48BOVdju9`8RPcHigEKtY zk_*f2|AdNC`*F~d2jMtNAJ8?|n7oiTN7}J3sxLZRIq!#aA(&3%H{-2=UYq*bR#mTJ zs|;lrk8dddtpKxPmfwM zU#>yD($0sZr4yyI#tNd$LmRcIzP`f!Uk7WX4^*w+dQWM2t5yu(rmpC%QtR*JFTE?^ zSZBK;wk2Q~lZSz6>R#Q+3O8t}`XPGI;z!?=XgPhKTmY=>F#Jya+b8H`N(2VhTX5`E1XM>N#AUB4Y&@1U7z38y3NnDcE;7XB4RfJ zst3EI?{>_i7Fk|@FD<3Svxph zWMlWa$i{B`io{-mwW)SPLG|F>;Dzn-y$P=zd{PJ>rMl6T7#J+0d?M?*Rs7au^7An3 zR47tMz1M24CKndB?yLgS^R*Efo$Zc#LViA`h+RXljXRT-Gf{A1@(}+6Jd0nuevAUR7t|Ll|!$gGzXIZfZW_?gF8d`I&UWH(|4PF z1{9FZAoH%mbhIXKXE(^g<_hrz=n|Xn1Z% zWjt~mHT%po!)1zcjQQAhf>I=^`#ugWzdkSfoQaxY4X>f;V5q?Ip*IqTjLA11;eTnR zD=}kXp$R9P59(ksQvUoiKg|KH|Rk=fFphpMQwvLS4H6NTd z*gTZ=EU$HVW;_99aIDhgi8x%+nqCW>i{CAG+?c1dG@4{)oH2&u!miusC5n1_W~g81AzF`{oFcv|bO zv?I4{Iu}jI?WdO-Dbn|EYDK8y!FGOqWA9nt0pp8fG)SBtl2kNWKZ;46ULsee{kW>P z@)348cf8WL0kw!Zf?>FKDXPkSdS+FuhG*Icui1&6 zBz=}c>n-1_rT#w>gV*i@dyhk6hQSVoXd2+7EXR9$t8k@CMh6Eb`-9MZ$mkzpfs84< zoL8gG*Y{{vPKpzb%=&y~X#`=irth&C2i@DSXyQ5fT6d_n literal 0 HcmV?d00001 diff --git a/docs/docsgen/source/intro/images/dot_if.png b/docs/docsgen/source/intro/images/dot_if.png new file mode 100644 index 0000000000000000000000000000000000000000..fb223c58cd0bc5ccc14429d3303fed051e1402d1 GIT binary patch literal 94683 zcmeFZXHZn#^Dc@36aW^j59b3+v1jkKx_kBN-Otm#faz+h5)nKkz`?;GQdd*b z$HBQdhl7Lr=k875ic?B}0q}9-g}$l+PQ?H%8u;V3qr8?p4h}N@-o;Z~;BWkAYGyBR zaOk43KR4o6G4?n(*9GcI@`nCa+w(*jq@7=7e-FFt#8Q+Q+qlL~T7^X9KD(L?bk4V(D@-1v9VPen1|f; zdKmtC(?8kMAKgTn17_`#wprWv9A$#PJ0!XJ9{YxV%*+Ji15)9T%0X@b9|r%+KOpU4 z7mqNq*RO4Q_UHFv-kI8_a2tdfkbQ!udw^(=6ijd&I0F$lBFD6P>={Ha1o(z50)zvB z(LyC8^|%by2JF+z>J?%Ag;x_02Od?rNZp$_;3P$LK1kuq({XjkjeGKds<#;7rX&jZ zABYt2aj0+*sF(-PFK!Q@IA3t?f{i0|Z#ujN9>bG^AekdTd^ErWd^id)awho8X+}8C z%Nq}%Wp@#`f8Zf*;}E{owNbtcej;)gjPnF10>r5Rxl!}KVu~DX1$Ci~j}D%UaQCT? z0~OqQ{PVdFpe{;MGh04dPS1{Gx+$WtS`|m@3jB1fAsKPHZ!Me@`GVw6c?o0?Zz; z2;90qI!>s<*{>c?l9vOKNhWP!vuKZs>wYB3;m|ss-qqAd|6M1ADm6dA6yRnNr35nD zJF@WBHHiQT&Q2~0_fV@@J)>XTT_f`UrMXVd@1GWlb%X3qMRhkC>7|BK-`MCMDfa{6 zY$DM`CI@sdH}4tu4)d;t{A3alfv>LPDWnmg(HTmNRvpon;r%2EjJ8d4{-sx#QnJR8 za!6`bk-3RQI}tHPhae?{{BMY`xH-cF<_HbZP87 zn+$5{NxxNuI=j?pVd`JmhaG%wovq8J#P|hvL~-FMs59k$t@s%Cbc9EIThnp!b=dtO z%UWdhacf@R5PEXqrdU%!-@HMOTWf&jZ(Cf@pi~3MKzyzLj?K{VrO>Vxp{j1;1X$6d7b5agjcTtWpaz%WSufQ zl0J(ToL!x~__gsjTJbT%QGqjpv|B;*+rORsA+S5GTSlrDF@?DlI@$j36K?SnO9$HbJTCC5EMW|O6v-zj{m=jT7| z-5P%Y-3P)IiNJJfTb@!#5;{si2CifrR2L*Z-*4`KCq%*s?4jo#hFXO|mHi^)oVc^5 z*X8B;NJ~FxPP4>3Db0>|HvAirz3Y}S?k?B)!h)`9&E036ir6;{6$w<~51OP(G2}6i z4g^-_UF-WvQQaRlFTz_aJ`p9|UmwSfPCCAwAu^Kxoj^a=Wh5E&EtYm**KkzL+aP)j zl7CvJgp&MPSn>)4v$Trtr=EI5Pksx*60dt_p7D>-xmAEd&q7Y$n(^MMo9GLFKbUIu zluD6%xJfeE#zK0V;nP8gdu}iHS~mJwQJRX)C8wDZlTr<#*-`6>Lu$Dh-QwZj*A&$Q z=n2y5#bX{tVDCGOw@bi>mbsiqM3RP16jVy$Yknzf2}yqr>QrQ<_a0syGbEpmY)<(U zzc4bJSP0%kb77_=gr37L;|Afm1W~$)vmVy2Q6ZIt2s_VqKaTQCBgrr+v2K;Z({X2b zC*{aOl1xB+c9f0LZCIIG5Bzye>{;{Jw22ERvr-bQW+#XXU)-3|ha>5)3E^bW6Fr&C z$E0@m!N@Cmzn~iLrRL#<1@z=W=iMOXb)}MyZI{!7Kq0!dySU1rKDzP_OX(z_e8uKhITq@cd(eotAeYO^9%M+1bIA3R?)m{b^vv zeAmN3=r#-ej)UvBIx{-i>YbpnRzby=JXF}sA}5y{yfhbJ>LML?7)X1dwm3%ur%76V z>L-ZoKTtgegU)TfYWo_q?4%YPWI0Elo4%wQ6K2~xMOs}7>W!JsEY`WXbc8iX#?-B|5SYqH?aUQU#INvZ&}uHL{q%9}*6fC( zi=rmtZxAIX1aAiuJjXkVi6)aw(&BtolDitt1{=GP#*mh56w3-9^2;$8lNC}PsA_D8 z?Y?QvqNVixG5^NbnOQ-3xN3;5pj%|Vit z>9(&vI`bxImwngbh5N3@`*)ABmri_M8OWF$Pde!n)?<7TI1zb?4>{K^u$w(Tv5szf zr2X};Y{~b)Oh>o0LUh;G4H9KJ&!r9=T{qbZ>s^K;KvwEb`?o8pgoqA<6NJ!+FLkXAbNQeRp328*8cedUgR62ONJ%*4ptERhb z&|P)wZ%?7O_ zJbvTvnHprf%lBGWpXF~dW(xV0+T+A?=l!Oej9$}claEvccfn_8@FBx*cx3AB0^z(L zd<|hOco!AQr_tf72&LH~C3DFQgGBX$Gq_r6&-+4UG?C+oYo`+Wo0kAvUGjc3C?o`$ zxD3zV3pJ})>;aOY=jDU5Es6Gv%bY|Z&}tx(Kx8`cM7wKC z@H?C^24$D&;exg&@_{;-8ZS>FDDle3oBK1i(tD-3GDbQ(M-9^>n@lCPdBJtK=QPDtR&IFwPT^<`VNgeNsoiHKPI@b~YaaO1%)_e`UiG}G zH-{n?eQt-`8HP5q0xXH&0~^}Egv2-4bAUd`V6n!_5k>rVG#;{X87@b-bn zezT*7s?k-4q2=WphQdvQ!^H1rM|f!p^;ODmc7E4SyTyTHGHUZ;mDxDXsVXrt@f{OS z7#6?nhCE7Li_)~!v8T@r_d6et!?6HDX$JEaisz$9Rv_0DljY_(lY1afQ%{cytN17# z{0M%+(_YhvkT$CO^QprvE}-oTYj%)%^Oje=hDbl=l)?v-$vW2K_0AR{TQ#$qGSiVi zA=$sW(Q>FgpRHakx$V8i!DURZ{Ziwo*ABHPCNT)fiw}R z&?En@jx1z-6n-4V4g?8u7ApTC;C=IbXtIl{cXA-;iP9t&lmL7&7m+yT^mTT}E`}m@ z>VCV$+~bC?GW|P?3C5h=eu1;oyU^@So4k8_M-8GQ^Ie$_F`kS+AG$uOr!etre?u6o zWTJKK$e-)lrO|!N>M7w)Gf_yrCN?v=@PuaV!;{I3Lb8orCF`_=Bk=T5>u-fKZoXVs z9_*HO{DMw*>GR#AF~8%K4j~}NXbl|X8M7U2jjj=tM#ny(%$tevU@HP3gcd>Xj1^R0 z*Ld-~@Y$?{K>Y<8?!IX2t`L_SO;z|$PW=G7al;2D{pWf8wDC&SbkIe`+q|cnq=#Aq z-l;$5+l=SHVt4wYK?@dOMc3OzB{R*pDIqQ2OLmL&Be>43S|;Zsns2WIxEer6Bv8|i zt})F#WulWAvJhhe3QRxGl93ChzaoXOhm}4f$@a;oZvZr--^!FQk{nW#8vE#4-0C*< zOzh0MAi;ex)}PIuF7Yz}Wq^GE;3z{DzS-noUnCwp>pv=xm|d&m{Asnc$19z%eb6NF zI!z_fUv}%hncJ;4LIjFnDJ1JCl4(cgfz?RXi)Vt#rv3%Z8r>Juz$*rL8>jj?_VkDU z9e4&GtZDZm)jEiMuTAZ@aab1r03rGO$m;mqjco1v3!AGZ0dN8U-P$}|ET^syaBJ3? z?`UeN-VVZqUYBo%)+qy20M4y3V@L_*{|*Gfe?QH!qAL(66q5{?V~Qwh6RW`M0M$IDVA;goK+{29oTToGT-ctroY0KkuOn15&(E+PQ;#YF$5VQid% zwa}INhlbGrg89p(e`pxfhrozr073+47}N;hqB;KmSUClY8?ZSzK*Jz~u;dW^U#16E z0FX!j6Q?+}^d43&vJXw>BZY&@z7);Qfdk223Ktld|3Hh?e&uYdlCt|S9C7ny^#~1o*xNA_YpwUM zUfp=^#7hX=E!rwit>w1{*b*o5p}1GRSQ8}5zcnr`8rH;rHCfR&QVhLN(xJPM&G z#>q(}!0+{@=NJ7zgB?BT7^`D()tvlC8P2s^qkt4RAx>*lLQe0+7F=^~zU=Lz!>Wnp zIC5r{DJ@yzjvoS}?0em@7n(N%I2)Xe`llUu&s!$taB++hxFMrxoBLHav0k9k#ZU;` zYF>?NFJbxPhA~#|&hICGHB1nJjIu9@SKYlR*98hn-b>!ka{28K9K@OpDJuXzi@eXz92b}J=m2!qN;Hw=iW5MSdJUiJ^6RFqJ2#0Wd-@4rBaXfxNecova;=7l?A?XZFs!cd!&9==@Q7D+Dr_+I88*Qr=wIF z1OQuITA3x^jWb}N+^l#pSE zR<%Pz9tm~nf;NKrGjE0}Rv?%B7Tm%%ENXJu9l!JlDLH1KYdZb=5Vhz*A*s{3-6o4X zjot*ymkcT#-)@pM1$bV$8&<87m@n7G*K2yZ`zb$9P|aZ^L3(B(`+7_0>=Icz+{j(> z>j3L*7OZEyA6#r23tNXkA}O!mqqahkyvaS+I}Rpx#WIw*k*U$`-GQqXw}DZ$B_+ng zenBA0a1&#gbni3dy4kK?2rWTOb=kH4Qh9XeeOnK}O{U$YEiW71h3k7yH!5wL;Kj(4 zUXhW(-B!sfr`$CO5w{s*4_}XkjX)sa6v|NW5s%`KpFyiE0d&`9o8A-;7~x}%G;M+A z>LEQcKG|s?RJ>WLjV=;y z?=|Z6qTY35ZjND&Tf63VM8MAD(ok*!C^~Z2!~x(`->bEpxE?Za5I{Sj-ZfLiFW$6g zp4h%Vi-bT5D4VQ@Tl<6n5katb?>1!($>1@`Y+)Qn9kf>F6`|w=F?t2GqH+)l1^lsA zQPGlXrLCee##i@Xws$e}WdQ^QHmg~KY)jV6UJPrw@uMHAKGbH>>oQqG z`3eHXkylL(^CefL!s?A5YZEo2-lvel97SLyYm*ipOkeLWJ?vem9zodQVzT776{l21 zg+P;pAWIGAQFS$8i%WPauqL)D4BPsH)&(GmRCj^$ryc6&)GYq7fuXJ7z(@E4ELsP2 zob*uU7wYHFZ1NbSKZtdaTq`7+Dz}&vJARtpUx>AF{3H?qDh}Er*1HZ#D?xd(+(&=r zUB^2!sJ0bLx1rHuMhjBvDI|?7w1DiXooO`9S&@w2=qs2J)cN*&7H`jp!13?M0Wl;! z0WmtO<8BW;p^42%&!|p`;9(&e5jwuvnbDbY|4b#B2;ilOn5-TuS)( zCM9g8`65pS5W%xKXe=yajnl)|UvrR%Q>|HOm|u|}S^O>R+*km9()~a{pRQ=>-ka7e zMSO1YfuC*$D0y`igf{H~%%cmJm! zYQWh2E?9T*n@P#FSicDnWCBs}(4K_n1zD()V8~Nzwo#iq+Mo7x_q$%(WE)O5m`FR; zX>{o(nlfs!ppx7?Ax(y2@ZXh5Pmz?QCYKQ)^NlQro7fQ*493Ggu5mUh^iZ=@eE6=> zo>VJR&E~F(%qtCyvM^Qs_H+ad>ZZ=$5;|hITDy4t!7k*HOr-v}oKXK`HCk159U`V3 z12u2UnHuqwuiPQE=S^c|R=Fl`beT7$U?&qj1o6=FJNCTumaMm_H+_N_`?y58`zdN?!WL}khaixx_!+fr}DsUnjQ=yN{qE4%E(o7Y0Q z_uib}L9uEf!fsv=M_eUieC5xCXY<)RLgl`F(ytWIKm?`XGs5LUX7?W*S{I%|qjam4 zyoHAA#m~DRbmOw95LddY|Kt+UN#CF1K3sQ7&l0%>ZV#}FIxu~n@ykeA?@QbBti@q>pAs1_C;85!RyHsOm z+c%y(7@dU8{I4u(Var!9{WDk?2lL9=`;DzG>ZaI?j-(2*YG%N;5d8f()-=)=-k#5i z*#@Td`7FO4(L$#WBX)>iD~H9M^ye$~R`;eP=8WyAQ_G+56lH*#)25O<6=Wm#q)NKt z=lbNbR~aNC{Ds1Fy?V}zNYb=g^iga{YEA^tt<=I_z%(|mRzJvpqkLa4_@@1}_q(SC zZjSaR;(DtJ$3obBenuf?b(!kimP-i6Nb1ikSnwJ$Oxo0p&bCe_-gnl%CgATJRdYkX z&tsWj?@{|-R=3=%BSD!6s1w;%BB2@HX>k7Dz0gwSF)$NrN;_JbOamTNj)KT{dn!E|oU-cRF34-1OZ>b4QbHjIUy|Qz(vg#?&XX^TIrv6X*ak_@^LzUmO4iGK^|Gv&qcYWk={`1z zswWuM^YPdcJqO9_J38gS7<>7|u{ZOD581tLdmbpcE0E!WK)KY@F5uhEidy$|QkICNx7SuX) zbnXTG1!lAq9AK_PVwHc(CT=E`fgssPj2kaHF}UwN@<6ha*U+9uQUUg4S64y9uJE8L zA)+=@v>-VfvAZ6*XTEwL+?dK_+XxX3dPUf3o_0Ia{Sx@LJ+dnzUG``wt&@#{$d^2pj+GO1n5xx1orc(m+0X{?I#e3^q z;^TlcMhBf(iNbC`6V=I)=r-ZcKq6+Cy*W4%3~p08+ffcf_N!epK`imqr=tLP2-2}K z>gymFA5t@qv$Z^e1k!e`RbZc=(qw~o}KkXnA@S7G|{ zNS)=2#dl9uZJCN&A-|sIM_$RcX4~%C|D^t8e-j^U4YF35V=LywBSCEB4uG$yQ2{&{ zjZ@ef-ymU$jar<2E^S%lVoE1JcU^^&ngoBxkm(wSF+O+;^@T}j9&dEE!#^(Hyy$)C zDSH6f19V#m_lVY>ehL3+*`-juEGi|TLNF1(L;a>yB5Ngw-vHy+X_Ro>?SfweQ_2>U z@82KDUp#!$>q|Df8bzeNN0{}3TB+-yg47H3w|64zShnqA5zs+kqYG@i#dd zo|m%&T!Gvn*7J@04W3EZKdPT2eyx#t^1e&FKp-xQomi+xrSCbCj>eIF4g%F4jylEpI#Q&2#1F{Ob6SMB*d-&9E*)1kd2?5;`amR^gAzbCn+BI= zKrZ~rbp3z`>Q3=rt}w!L3>*ISA`otYIWFRMK@;#R=M2~)5Xd;kT`**v8vqOAMQ1fi z@+HRd5cv`hfVL@?8P5h2K zjd=RI=*^>Kdwj{9sTEu3(oPRkEYzxAe~zj?F7t+%$+tK@IDc?7O!RJv9+QA}KJMiW zGE+TUuRdRMZI_(gAXQU(eteM#Jylfy{dlBK_bjaOm|Bbdm_ga@2L5BlR$HL8!H&n7 zLEG&h?+$6V*iwa>l)ePk~l8rtxn`_)Uo70S}i7HrOi^1wLPxGEx8kqxCz5QM7_pYuml_}!q- zF%b^$Q@*QQIqm><^**bgBOTL5ii15kDGk~0fk8bj)~Q-Qe1X?GRO zE?<89L-iaNU4jw$k-`Xz))-IL!nby9up+*NNg6r1wcIvP4s1DTOW$g~Gst*!Z;UjS zeU%W*73Or@B(NtqqrV_aaOJ~64bAr{u+(cC1`2ew{;H+R4hjEp&od1>2`0tSboKm1 z%n6N8HO%nK?<6s5PAZ1y8l>sa3SmwL=u~2S-;fZUSosKog^y?5FV%&057h77S`dt| zBY#fBx-(P3V!z;6ThJ0lMotV4?vd_sNHjFld<+POoGvS2|4u`u!T#?ej9E-1#(n*G zKCvou)=I&mx;f2sO?}=Jzv2AN;Clc7 zj8ZSs*x2{Z?7HgUbhygW+>b2^By=)X-cVQ)g9f_Xbjoa3u#jv(VP_9}=Jp1@rv!X#fnQX(csmT;}CSA{^gN;qTb0IYkCWT}|7( z9=Yrm6eNcd%!&HLcv*M(mssYjNZ;&r)jGc5x;0PEKD$A98FSGi1IE2XmVuerwi;Hf zzJT%pAhI0wfYS5&y|HAvz=vAP56+@V0~X>APK8_l2^nK%Haxy=JyX7En3JvKO<8wb zipFp{qEifyL~17DiE7h9z&>1)T=T+Z*jeseazQpqjE9jwkk3-%rXw<71|G}tAKbvu zA0S=ymJxZsWZIe8L&{P8kDd7I_oDC<)f#H(Wr=yc?h@h? z#K1PIEyRKRFJ2g_XbyKbi!sXW4cI5Q3?swf=G%*5G-63CzWb_;s42u$4J?!=+|J)& zcF0+g=38w0sy^FMYhf8`KaRdA6c-^bO}#LrH(>aS!RU_yyx9B2^#;km{A>t5%c$r<+yuVF*=@7$eN zp9UTTe~n6fc`?v4xcON^96Vhke=6hELjye_q&u^?KLY)z&;uG#p!BG|kymT_(6 zt`+_tf~ZZ+SlezYw?|r!_kPp35yoZHuc=Y~lw?Vj^PyINAH(W7%RSE!_PQFr|Coki z@A(LlF-WU$Ou)G5v92JSAGn}I|0K|;_Ci}(bj6BxmZukbdev*ISa?NA^7aDlzL#M37vL{pg-sMh4 z+LDhRrPe7)py4;H&D*3(c?)FP)};qrh)a4CJ*f@BX&35R)7on1Mwtyuh)DNBuP6`+ zv*0*|L~qhM3_yyjvtutd*&9>Je}_bgHrVBE5#F~>ioDRPx>`1?Dp>8yJJU}K5!h=( z%oD-Eg9639bUnVT>gM$W=iP~Onw`<0&KT~tA?g0xSbTvSXM&X)Hwtv5d&ZuGEV?Tx zu-VH%9a|lah`e_BVd|*%VofX*^Qdl9qa=PWAikziQN7LnyXjfx(sA*=l)^$xG=*zK zqGQDa0Pt-|Z;Btn6|lU|5w)``cA$HIW&15&(}~+AwaHsB%~nUFEe!U9^wbbi_i>j* zwT`W?UE1=iL;hQOB1_*Nl5mc4R+X?Z$+$leusup)gsT{~j)}{id0LV`N-eIkyr$}6 zB&&Lek=}crVnTzaGtdVw3Jfjvc&ocyNT6$*mRw6IxkkwV@hHQ%kJOW;bxPL947?)} zY0uw|r6a~9cpF*oo#y?-;-f!rgHj!`e>8jA&5f@nI#TERq+Fu`fyul-oZpi?V{b8L*t~sM4dYZc7?CkUQvkFpg<(BaT$@Og# z68~!%0inx%jfG*O3bZGa`vK;IpU5;lCh$L&~krwCoz14g-_Bz{IEn~tYEzCe!D?4n_ z^gB;`k$za50t{`rS4u%?K$@rPRia$>x`+-S>EM%jDyK5gdLcIqiDF z&&NyDsx2>+i6&)Ww@zQ+{tHY*EhtFZk`#jsJq*nllOPC13eF zRJq8##pj3LCa){eD0R~xSMjuxT@ zi?s2FUriuxVW)fvxNwZbvj4d>A8?Yv$Ftes^0=g^>QN}HT~C+GlCbf(zpvp4uPP(`eQB zEuB5+vl}%lA%Dps4gO;puIG|_qWvSd=pNnba8joSmOICRp}JHKtxGDZ-_{Zh4lNQD z9QA=lN{@MzXZBZgIzvBnX!MO&j+i<5_e$|Sj7iqGa2d&M3{>;O*18<9G~@dxib};- z(8CmalrU3Azk;4iV;SQv^X_^X_i(?B><^|Z-=%n>P0CHKU-ty&e5gB{{vy2b(Q7VS z#5wtVvnEjGVoaDizOfZtab_`YF_G|72&CO!mNS-EqL>SvQGN|c=BSm~Xv$qVdzrG6 z3d`iH%+wgK9JjU+eP6MiNO$GQ;{~XJ6d-@{aLvB zq@y#dds+o9Wq+9p@|5QCrFsEW>A$agT5h&K14@qfv#ppq0`CqHQ z_P`54dfOa@KqG?gOOriqso%1jvOZ++7O+^>J1+ej*}-uCGsjIGRe3~EXV?OzB#a#3 zImp(QrD1K2o3j2usT5m4jAyoPnm=X}xz=rEfgGA`WXT=h`2<*ZXqiEW^~yWtES>>Z z*{I&Sa9KMj>U`@R?EJn7F9>=`)^v!Ld7=dtfX>CQzD>njP|M^y zaavNMyqF_nr0s{^W}O-YJH3ov>kmksp-8;;b_+YEQU*C(4itC4++}F}c+yP@Yp+z7 zl9k06S}YA4Y&LE`zPxtnEgqMLl7L+ zyppJavaR9*YW95el`M~WQvC^awS&XbWy9M12~)7Aq`F*PUGtJr5%{Di+m~Kv`>E}k z#|;$~JQymjdq6+V+jZY&XE49_hAxIG$L;EJFRaEFq_Eq(C&Q2b?+wT20n(U-pgov6 z&MIccwGxxL)5Dfcmh`0>V3@x6OZx4u94bM`KuI{g%e-Ny_ZFvpcV@t>udrPJa%RCt zHQ5D+w|Tz2@|ykp{1GuY=y;d>pJ#KpRNg&Q-ok(U^46+vH=W;*^v=)M#1!#gdx3#- zxSrj(Z`eg7IwyBr+tsClp(+DRj?;US@~52rW8Z0eZ=LgB#9aJ zE2e9_o)yG1-#;L}PW#_Hl;J(1?k#%W$s5Y_??a8h?>%d=sswC?w|I5bfcBV}o@W0q z-&M5iLjki8Zpf(2ho0{&@taZuttfDfu6xhwJ?5JLHLWiY-E(B|^`CJuZv zHZCpB5$zYc{U`==g^ieb;k;n)%q626(q;R8g*U>=@?opJfJKsrtnh8yrB$fvFQ2o<)uH7DYmVQ z3XN@p8duO3TO?M7R9WCNSNZ@e{W8I+4(qOZ;&#zD)(~Qm-ke_->@9&9@T5wiOJ8wI zf{YCinBTVRjoH`mzaAPuqk#&>kyr&JbpUB}Aa3BiO%37_zDSA8UEMVff4*9NAM07n z-CzFHIP`-AZ)~LIS>SgiH%I{1fW*?i18b!6kYg{GW0i8j`RfGr$gyV=4A2DNT7_Wc z4L|^0o@u8thU zXd<}wZ29d+i(`1$xp<9i@wG1$(T0DY@5er0Z4g%w6)S@9F7*IAv6fI{=ke|R`zVzH zAU$b#9q_u#ak)C1Y#%75QKb)HsZ>DOs<9p6XJUOjr2e~E9%!LX`L z(j=hS%Tff;Z*GJbAP-Rn5rLPua|*E=X}Skt_c%n%keOZ%G=Bq12#%)pdgEUi=+(ycNsdm}736c2q+ySKwnap_D{1rzr8d;=85s}Lef=q6LTWj&8JS4i z_h9zdi%jt!T@!Y|c{3$VZtS);)JAiIF?5#qkMK&;gzHBa`BUhLVeX}s`5jL`My?fc z`x@DqM%}PZ@1LmGkMTxcRrKUPO_OyOuBOgk5pC;S_--t1IqE*&S&1%ffTKj3rgjzP znE09KcXnJ-+tOME$tX#+AY<~B$Wm@Fbt==sRy}D@H>Rnart&Sr!Z|hYggKyNgWhDb8V1f`sb+CgTF_u5Uw-L8k!@Y z&Wb#hlG16REF+Xf0iSGRoo!*Y>n4)$;?$=)Uz{Il#qd#XA;My{b{CjU9{tt2Q9+$~ zGo7n-@5f7Louv_HJP}M@DsYqIj-=(VIiRX7hBj_`9=rqdzrU zPx6Wl*Hj9^r)KwB?LNE{sy${Q{tooVEAMbWGVPF8e9`QfUYHn%hjyL*=sWWB!}3;B z##J2Y1%$nt7a`MmsD})c_|AW|y{MRfDsnN7@C0Q=@5R%GU3ymhwu*Y*dXk#$%Yn zQT5(cf_G!sF_{Y@G?EsB3!?Ox>Yp|SLVAD=0#cZ4TQro6>#It(sw=?k@Y-`>VW(F= zsG&>g0ji|^&Lvm_I2CvAW%ks&Y&FSEf_WzGQi%L*HGwvHfeG|t=K{wB&_>48zdK}w z_DNI|oef-&sc&3vzD3vXc1re&ks^b)wT3zTcP3OET;JG{6-=5IBVig<6?B%rUJRIH zP{1zE5A%~&q|P?ozYN=`Z5a0pI1(HiVceWwH8j*DRaVM5g~nGl5R^9etr9;pEWW66 zb^SwBrHZut^~T$unAb%=OO>+33C=csFnH!NHIn9j;t5s0N%elgY?|krtVR}yxH%oB zi0P27@97D3TAbmfUpw_ZO_3ji@E9$FJn!m>PUTl|uE>ghufaP9hB<1M6|0!3NGJ)JpSZoN&}lL5rBS zs&GLQ4cHiMfy@rLuLf~1@7;YNJ?Q#(fnOS6_s2f=ygR273i!k&n>*===d{~ZB`qllKyvH+P5(JBDl6GPt+{KI<{j_Ih_)EavoQ2hIK_1O;%Jp{|SG7qqf zOnnayb;bue(oF>(hmh{=c)2z`Ci@yeXKCiAeN>=4h=5DcH1T~@fO*>}(@`6ofJ)C@ z$S9%J**AV%?hu$qRi&4UjZ-tDOZRCPdD#92Jg*24N40*Fii4qQp!FS@l_0$M`LYzD6)qP*xS=Iv%>^$&)RpmArD8rkW}>nn&lE@izuhXC@5|Y@CR>#flvJpVA!g!#n4P zZmt%*?T<)jQU zoAT8;_d?#m47#UP{vg$78&!{)s?Dr3$mf|tXw<9CfLHSSc(M_RwqFd+b6=g7_lR@? zlUCV3p*(@4zGAa%wx9=&k^Nnr5k>XjRFhIYwJjItXSg4|pXIyRS}rW8Wmx*C!>Sa7 zg&Jz#Hi+77ha_fFNut&aeOlgc@tw_8A-TD2qo+p(7&;X*^?io6^Y4+yDl7C^-nUxs z|AQ-n(MX+M6+Bpti@f^Dn(bKF*Ic4;=5lm&YI&oiE#AXxAECTrF_^!&$ObH~9d-V& zfLndN&q(jv>Q|5v*HtCEyipGt6w^iiQc8=q~|1+qm)lxU;cChDAM>YKU^D_Tw zHCvS+YrTCIQD!#PdxvmUpHCet&Z~k;qOnDu^>@o2+^g{P%&DCD z)@J1v-O%swf=kaj!&>aG-4YdGC$G{~rhVxayEtjIJu>1Wr!LQ?Sz0@u?_1Fc7~KoD z@oaCe(R#S4{}A$`XjAM*m$^^XD}l}$!~kCR4<7Sb*m}vUm#zx8(x1ji@-c>dRf%@> zQG3X#r8^fk{_8RQ2v^9ir^y$Ap~HHE=dqIq>D;Q5fAUjb1-g@!r6H7DYIXIFvS4Kf z+csWFMt}`oZ|t)h9F9G;E!QUT+y)W^9m&)0^;w>ALJuqbMOHlW&*gHghX*M4wp?sy zQ$37%?@70MG?7TBpG`jtP#@;b@^$`U-4s2v^U3NAaF&<$IX1J24E%_(bhSqS%2=g( zeuh_84x~|EQ2#$O`_TxH4>3x!nm7_4T%4>MZduIr_sJpsn@;Ti2Osel`+u`(JXo~UC5)usyQDA!r zMQwd@LA@c#zP?yT6UeT8utl-(kpVf$%Shnp5&RmJe@&8<_APkGUlW=0NgjF5PApxKphI{wn2HGS^Uz{u&JXT>5 zDsuezIY8eaj&99FnOVmt%?uR#ormXm#PpgbdqW6CTjYli1iIKGLB;l@Y}zZ0`IVJA znl~Q}kqNkrL7Hst5Iv0Wxu*hP%$MdWTiT(Z=!qN~5Y>%L3&dY++2uY}^~xlB^8aG* zJ)@%Po;OhxML~izNkKq?hDMS~QnG*uO%x)-M z!`;XDn|uFj&CFdhcij*3>GeH*I6Kv@UAyY3r?lLvA}*u^b^0dQFkOQ@WTwgZD{Gu^|nD()M^72*i zaVbbhHM{PAvBAbe=)x9WgEyhn5jnP*ko%BBNQQp_ZvQ-`TU6@&9Vq#&j}VN^f9#W; ze*PZ*1q5`_?h9Yet%)7XrQkBZ8gi52lg)Dc!?#^uE*L=bW#_dwkW6CXf4d|%mq0`I zeswv*_PM7ubOM#r#=XCPv{KlRY%Z!Y&)WS~F))GFC?`noBRpPtk59*)To52o0D`?TG5tOi<-+yVzGXn+1EaNxA9x zxA;?M59gKA??8*J0K`IZ=VnW~k{5fBI;!9Rat0b+Oow=*(s}b$R>EtJQ_`KbGWzqi zW{Ef+G;2s<+y>zWjHNc<-U<*nG=8#7s@<3VWGa+XUt3$qxysprjA-MY`#ospcxWaO zDr5&y-=l)|#{A?2ehsyouTBg^Zm&7UYpY#lvuAkPJlu$hto7mAgWD>j4g zc9bSGUYS`TRcPj15)=z2UcRX<=A3YTF`t({LWRWt3cMH?OZX?B*d~o~go^tzbQii3 zrqoHyL3UyBU_O=R?U#XEI>TLhjjl*;%O% z_n%ZYPT9s(ButXWh{yk2&j}J0S>I1AIiDL75=6X!qf)UFD_$O~G{%s(=_|8k^7LeG z^P!v7QZhi=`N_rOAton`Yif7_*V)CMwng@RcG!AVvDd@t{-P(|2l`Fk8aGb~bGm)r z7*E(s=?--Dbmwq!aT#WfD&CgxyD|TqUgm>0lWQ$5^md=|e*8{RdOis=f1hKc4L65t zcOK0gMmorJ!EpZy@}Umc%?n*PsQ{gb5g2MJb(j<&pvI@e3HdoE z9}3qH^Ax?oh9;6_q?CKZt%7KMU8?>i-NY-JjU|4R1`|<*67dt=PS#K9{j9fbgt4g3 zicPU3(Wq(BB68d2`P`ag-2TE(t&@ZF6Am}?8R`dSE>*xb>|tjoDp~p%Nzj+^4(MX{ zzZrxHo^ssH--z%y^RBPbk{WZ-E^sbcD5$mj9M6`AbXoL_d%|yU3RB-vWeg0wTenN0 zqFL`4s3|FPQ;Kr7JQXsnBUEj1$6XX6J%tJhV{iF=;K1{Y4&UK^VcK35eo-rG_ zFOT2f#%ybDwhTm_$?}gM8?<*~OFyMB{Ft`Py65aqQtr6u<6`BYTfx2&V8ML({59a+ z!(jG5)pTPIynKVNh} z=-ObQAu(fK?sY-gpM4nG)MNBsNz37Ua$Bxku&$LFU)m!Vx#H)&dR`8SALON;Ini#c zuo3N1(e(R^)Ah3)g3qG2~7Q}s^i-dUJ~RYZBFNWR?Sgy2TFLM=Z%{LsE+wMI_io1tgmhL!|Ksj@+V-V# zyoSV-BX-n$E{Qtn%_~q_A7dnkq>UBrD17XEfpGD z+3tQmVIx4*3ckHS8d9$|WoK$A)}pj_g!8=b(oZ_P79hQRFEfXuB)*UZV+pI#2X@_q z%@YjLWK{#q2-rf7WlNtnIcI_1b+)WLsf0AmrFC;f?&D;2)ck&d6RI(1u; zSoAhQ%3Bil17-FqT=>11q5+{0sl8KFQz!8#`Ate_?!HDxOL2Ob)wgsE4t0og1sb=( zrXrFfjcrjH(ksba!YxLy+84U&zlIWm%97l#Nz7@<}rV{)IK|&cmO{SS9}O zcNl_jx>DyX_?nn5#pA^7VLLhpR;Ln|mz%0;qh{Tt#?z(k;DGFE4zThueS*2+H9OEJ+OOf-r(qk53`0&>+vRE0W z<^4!Ki6>Djk7%sM9)6K$k~HVK#c-Sa=|C_(WA1*6IH67cS=zlOfE}b}s>n1JX~F4{ zVQ*>&px1TZ8wSGi1^IDI#Z6hXIw5p0wNoo&-95lca*p`t`kwOD@eN{bIRt&8{)mS4 z&>2nV26yKDn4-+67We9$vy~tqD|qSuRI;%>S#jfCma(veKGv;CX|pect2`~28$?i+ zs-{>(p&Cr}b69eaIEN%8%&cMd+o^>5T!y!j`fq;;eqH}D{fONoxLDK_r$uPo-i~tl zN)PJ}Zwn>zjgk14F+5z!1NKP42i2B<#4&|{&ypI%?*gu~U#XuL4?TfHvyN>+g zgSsw^>f21{pw0y8k!gWCj|oddL`ml@8Fk`<`B}Qb!4n3B#OU=>LAs%MluEnIGi9^b zDNnE9ZL7dr=&^?%JcjPYVoe#4EH9V}N-ixRsY$I4-|g66&k~SG2H~zC)_M8Tfv2hsIsOS}-(8+6tlst9W{m7e^=4)CfqjUaNCV0} z3);3)jrs{5r8;uEz4tV(ec(s`Xmg*Xag5^#{>;*XTK0B(DAXA85Ru^po<>3kRm)9upU6YRYsfhN ztbY}`l<@a+m*G2df2YnP*b#9V9taYeF(Pndvedj_ip zM|JYwP5|aW0HMB4mnX#@6Q-Vgbb?Q|cT!6EXk_t&uY@kk1`j8DxXRki*VmIZs!M-d zU2ObvMSSdWWiKatblC+4!T{SNUA@0!)ssVP_|eCDE}@D%*z^zL0m_U;uf<*ZnIz2> zZ?pStxY_KzhOGIbh9P0K0Ky*TFtwbFS_Vk4jSq;6XbBVc@LUI@^fY2+X9ig5lXeiE z`!($Mgb8%c=kjAhzuxeH@%ehbE_vUgN3!8hKl7}8TUYYTb^xX)7+%jI+x;bMMikaB z#a&sSOV5>5-3ObxCUu?( zL3BWkqZRAzba##LZf`UsTjT{vv!S(3c9(zg#(nW>$+9X)sv5j16dtV&mMx5VBsE# zkcDRh7jhoq<7S^V#`$VG@SPsUOirx=x<(`Rq{^{6AQD|L*yHvILd$9Y0*{S0b7 zP0MfYjMU?;(5fNV6Mg?kJem|arCjLP_cGSeQ7F^wiZn^(4|10~8xr9^>caxXb;29o zu>}g4A+XJ2Tc0?~pVyWq-u_NW_f?%%De&c^E0UQ z+)ehqoWRWQ9L`*t=a$zEU{lh}*N^fy_nsm<8pGx61{m* )Yq5XdU`JYY;?+R2I* z2xUjlyb@7ME)JT$os4g?d!FKTL6~d6&=>~b8Uwnl_-x};Y0oMR~|d2g}hIob9l!Q7dGcBItQD_?lDRFkU)1vu_+hq zQb`)-)HAB)!w8U4e@>@Yr{(f8lloWSQ)r}r+9Tp2=>WeeZ)nx83TH9>FKDiZjHYbL#8lbD1%xyi{_gBJkD3 zql>q4K61)#d=Zi8Ag}78*%^($G0fvZFM_)FMjscR#q@p`N%LsK&v5+dQOI}t^~LB) zOBIw{frqd#6iLSwE~}o*EfS z!9IGclMB?ls-^lqF7SkDkgXtmEIFJ6N3rR$ZYX(#uOx`R680i7xp_ZR@42b2KG$NP z{6$MrBpy){n-^8q1I*AHY^;tpvE2HP)Cf$%&oQ;JB>M?y{0Qy8jx$~s;7|D16a@D( zf;biDQkbjiit(coxiH0g?$FgbZj^xn%r6>BN{0~~VJrp5Ps z2A*Y-KkhKH%9SLEw_Tfxkxtyaesw~eR-t4-9R8B>g+DqLI^U9u$1ZdlOung31I(%4 z8SA+siaN#pE>OS7PZ%=3yv0JQVBE6%uCiDxwdzE38;7eZnNZ`*%6O{3q9GMa9P&-g zZH-i+0Ov@k>UVfsm8Be#?e&WXmQ?7{uB_|#1qj{L&5hk{q5UGTUV|Ru$V7J|hcw;o6MlokY{bmvvVVGVj8Y^@juenvVbTwLjlMpF1L|tn9yf!Q3@`jJD z&!(u^RPUM3pS#*k(`I%UAnH;97W%y+6>ce+bq2FWDYi8>)*qEk zVc|fggwl^O&Xo5l@8nYow7yvCS!iH*LPiDC{&q+`(Y)~zA2AUbISscn|3>TW{EO#s zp#4a?u~C6y5tJWvM$of>>3V?JI4;&5Jx2`r$+D&=&yakssePX=oJf9swyU?NaJ+0e zym}sE8?uKb54LMiMuzaU13oN8cFSR_r!)mJ=4=@JkiqiiFK-Lqs8o8v=y~S@7I!`Hw76+5Uirhm_)6ygpb5IVOP1V|h&GIQ;oJ+?$~5@-YZ zYK-(p)K#^Ou+^i<`jB}z)q|k^q#}SbrDsrF{R`H%0F`LUwjZQh?xx@d2<`_M7*7FS&m96M z#z%oyp&?`RgTwFUfB#b)LBJY?*bFfYxbNKg;XulVmIB;(d;?c4&&gea+HK9r%?_EV(cwyt_t1pu7KyT+>NGE`rCQ01G?SOE1a3PnUQ*l+p$jWLBYw0R913z7i z9{$e%9zdjE;DK&_gae+?_|Lz%LCz_<&};%h4KHHsK!$jHBF=-iz^hMwms^M*+A8Vk z!X4kfjp6k{-WPWy1>pV?r;daku>5>bv5@F1iqWE9_YO~2z?}!Wk;$2Gy!6IMyldQb z^DdFl$(w=j6yi@AE2Y{qMg~UX>#We`bvuf?=dUeQldgm=Q8N{Awg1 zLU!toiGMX98|-{s2+9!FB+{}u+t#2bdHve8Yki;zHUrtmg zpsE(vF8=6JS@qiWcE7P@%;Lt5*fh%9-a!1TUWAH%$RKBS!vw-5%tED&gZn7UCI2(h z?djLLhQkUCe?l&er>k3*mnq!*N|aWVa!S7c;g+UNq9u z7;*haMdOIL-p2}F;^8)i20hGM>*nQ`(drvI8**te9QGHIKfuNpObd$3oThse41i4i zTEve}5vaXf^CXkc6W4uEjL7_Ne7un>l9pA7TK51Kl|tTWuSWyj3{L0vlSh zw9l`Og!zX@^IBj{)AEf+T0KjnL;V7y{jH9fT(1u6w;nvIBpUMKgm7~Y9#yCw-;7xt zSDkt|@Kl(mCyg8^>0et#`89yleIci9M#w(Ady5e;XX_Q(BQ$=%u^n0m}nrEP#4w^$afhXw_TKWiM{QC{3v@uiWM>m&i>@CxSAwc{mSZEb)er zUt(7ACV(_q#StPvoIIOP)AraUIS=7iT7 z8s(8zZ}kk7^RiqdKp5?}h6Ecw@bB*WA5xJ{9rr>z|rGpbv{*T`M z-*iCD$ZcMJ3XaBK2V4L^FF*$w1U81wN{}?*{^aLbsjJkuyz?=q9LZNgpAW}C{Oz`ziK zYzaWrZ~J{2e9#>-Oy1r?FIXRMH1+;Bzk_1{vWtJd`#=6~p~y#X0Ybgb+;%YUC&ZkE zK2>B@7|{*v8_4p853QQbs%s<^^WKh4!E(0nv#36|JE;=a$vI$1G1||@kI4xQ3JeFB zg6P{NNxqLm2Ks}_A{rZxdoM?>D0mR3U-<>s1Qva~?Je%nP3c&Hjr)v!4;eZMP>el2009DE$q)YyGX=E4@gLm;Zbhtr z!}9lnwR`7)Tf&ZB#DU{~v_LXBYw-PT;DGv3Xeow&mRZOe-U`kNp+daOzplf6lmyxW z#rSFz6+AbC-v=zp-~GkVhSvPoQ2zh$@A#{DFF|>WsfRK@N!^<+FLHgRz0omyYxd-| zFKl`JMJYFTp36cst<2$zrn%rrLSO{{S^CTQ?a${otv;M6I|XP7^N933dsfd~oL~`y z*4*`QDO5Kv3Cp4EVORCwX&6P6k*r@Mup3#8kuN|!e(S47(fmX($~Mx?UGNI&zuAp& z$gMCiyqRr5LvH1#OHpLvdZ$EIYFd|n)uZST)I{qPu=SleX-!6cNZAz_!nI zpj*MS<0w4{lw`jAw{vtbr}Hw=brq6)NS{OcLcFE1bb$a^lq`ciLR|)jc=fr#dOA|j zc3$;yTB+&QW9P!Qffa1u$J!w62f`Tx!M#cKL65fgIEwWK4LY6TPQC4HOSmvfF8C$A zcaY4QzPTf|lNe#s(i_K5?yvqXeTGv-C4SW9@}wypA1<^vqX`E#?IjCDOV^-&f$S$l zmSW=oA-cTW9he`#1DI4I7uk!@bWjlyo1AuWxYC}bXKGlLQ?>~3)|Dr8q$JwUYl-mZP0KyJXe@pa7Cz{^Xh zaj^ZKY-Q$_Fv?I=nGroSHri0fF*?4I>Sl9C9tCWsze5ZPWr@VBjn%$Xv)BGX=;^EV zXpnbeYpES>xPQR6+wf-Vpvv;uIfHnV=UBUW!SsNu#iqA~ZBJT}$oShxp!$BPgbXLZ z)ps!$C%Boq zEJNi)} z_dy}1whAH_31;j+lZB0_|5B~~_KJDZn@G53lEUgLVrzLcBr&|bsU!c zQ3srHzgHFA+aaf|5N?}v1}y&tt{S(LLX_)>{+1`Kp<|xFQ(8B(^S6W$?^md+ksv%*hz-ZSfY!U1_Wlxkka^ZQU|3wBpVF|vK=UheI z4S>S?Pn#6NsITV%Uzx=-#G>7(cYBovUnbMP58@?=DrqHn)P&mT@lSlY^k4reXuZrC`VI5pTxypw~iZ}BjQ7Y10}p9ht-WIv|bRI6)RgWHvT#SOphhi{jp ze`jUIB-Bs1uEXohlM|{Q>qj)x+dHuz=^x%c;;`3tmkmQyWZ+*IQ&_knQFp=Kioi{c z^h)=zSe`3Ka`!|v;@UVvtlYcceVMprsYOaq)?&%NnQez8+2FYz_29P!T+^j96*t*U zif+%vueKpIDgZ}3-k}7_2@(fb263V57oyBVCp@0BDcidxXec=ir{=NX`g#E`&Tow> zkwLRO>J`&sHsAm%Sj7y8D1^FDa3<)N=JpTFjlvZzoW4wCaLr2FZdIvk@}X@HcqsiN zDPs>f#KhH8hA9oHul-ENp)^OP3Es&ax0ATyOTdm?CZP&9+O`yvOWfly&OJ%GUXw6#9*3Q>9ua!>b8&2l(eE zD~s&ENniO?{IVu&q!kiRzRLhqEeZHV%=%mfv_D8htgvj*z~vOdwWJc7!H zp;W%Pj`EC%V{XT(qb|*Nd9`|e*Gf*&^R2uCvcGs5B5^4XT;!PW~e^^|!+m)wjbfA-j>nX_2@t|9j)zCX zYZLT?v3K19($QBON-OW}FfMN8EqgREm}zMybt=QpDTy#e^9#q=(pw-TmI;HoWSPGu zqx4d&Hx%zx)j9X45x;ZuEE2oiPe$;`#8G^f!2eP%y=>gJxEf3LGU(aKQna`I8&j-P z5}aj57(qEGoKvNhP5)O3GhVcmX{J-;%@n~FG!SyjFMZZSGM2p(C)Zpt8_J0OfwCrV zmP<-Qt?Ke~sRR&cB#15vcaOyANo-Y#eBXQri>>3Be9|hUdyAGyb<`(QmPz)tE{!ya z`7W-tt8`1XaeVLWc*a<@-9Kq&OjhHP)$Q_tnQC#Ar7JGEnow}1FfqH@k!?PIWq=DO zoTTI7bVKO4m^0uenK0(DlRhogLGz=D2hv0x&niKN$TexRz{@np(#ZiGfEiD=B~Fk- zH+p2e&iuXmso@YTyZKnD-@Moadp8Ar}ZN~)CU+w6;WEHVH6e9wu zI2K&$(e_^6;!#}h-w0zz5`CSxAGl756V8n`H=0zjV|_mrSyvp%0+Q#C&kBk_&8@%! z1El3aei+P1nOY4A>fM2>3oyCSqKC8D5~SXUBZx_OYWM+GZT_AY9+MRNwRT4%*xE$)0$qfQ-~kHht#TAX21*;Iru8 z*)<{{Solv#iebQSF-!r(umNQ!E=7ZJq^s;oC;N&RyXr0Cw5IXzx=QluNX*umOvMDnJvE;d)E^X z-F8iK#&X!bt0&GcQCCpDyr_+?GY9?=ADWQv;Y`S6+%MJ2wy3B5N8}Dfvkpj7F&V>AA<}Ys>D2^F zP2OYjOJDQvG+(-+^EPJa235l=kD)O90qnjxsR|BH4L_;j(eR_U{2Co5tbn)FGx zN>NSIIG2(dw?f4+|1Fc9yjqzB;>?6kDF&4>F7WpX!32NJ1I+poyA49Ml)F+Q88~6~ zg@F<^(@Ep2SzqS#VmI%q_nOUQtt-X5tehSVzq=c%MlX-9x~~^KZl26pB7WLjE$Up6 z)?i5S5*6-;_3@CVD6wCLaJ^?enpWeJ$fS{>vT4!4m&sey{mNE*h7mL?zE~l3Z3VS9 z;Y?oPB#G*L&Ke-8$El>jSn4`g!|HM9E0%Ui`CM}%ZhijCfJj_SkpPSSH%{o9!g ziGeHdAiHea*A+{fGIxvr+6LIy*s2?rCs|IlB0zTU=&L&j3yWq#qzuminjA6J2H|_5 z>+L&G{ix`0+cQ`+J!F-San31H)vzRg?23)ti)Akh1^!BDiLw zNrWX>d>)I8uayw`9L@PMS}-%!f0>B%#}YQD9jW3aiIGXXrDZglmr;!vkK$)-x=e8k zZ`VZQd&Kt2b82b4XPb1A;?)xI5e2;8$R)K>x~>P$B}=3<3=A!eGxE0@SB#4B42}5^ z=nnCMVww#}GZb$!{@K0{oP`F(|%J_VrW&5p2ix=`1DMz6UeGkk@4LSpwO+g`Og=C_;00enQ z`o`1I!#;DvkfB-QP$uYcbDD@}c=%n5WXOycAB(OcpP`uBKUXnidg zwWUd?Ls&?oF3W2p7#rPfshZ}tnN7?VadqlR{#ohr(eH8KD2Fb1RfutYx;`R-SJ9rr zS2U^Ou;D@;V1D>5<;5*|ax{*GN(^W^*oj8cV#Tnvo!B|RGxXcomygE9hbM`;4Yj@t zc1SPmAD^}vQDLs_{{l)B|MoQ`bCFOZXO3$d&dZQv&s}^J?r)x;uaQZX334)f#wS?q zp8M^U4|@uSgbdEPFaBFcX}};jg~Yl_KhP`V&2Q(4PXTU!EWtLC%$3cxn#pBWVu$gh z>#VzH`)i1rDSCmcIJ;ujnl2&Fjg*(a3R#~zINb7u?|SMO*eka?HPTkqcCpTS8K()s z|80;*M>LuyMfpkx@)Is68bm|H^J1Ug`CpvN(@gePpVds}Ltj0bJ@tNi&~#V{PQ;Po zvW<9KPXF;wpakPaxncR+<>5zP>gTLjHs0;#sX7egYpvmJmucvptG^HTk?l13JZMOf zD)kg~j?GEo@DsmtH=^AM`5Gk?SSa<5Q4Un~`6Dd;rlNhoG`OWO#PljII7g_o?G|42 zK-bvXaPUN(M`Lhx4i^9>Hc7EjtP~GgP;}1?_fd~B7>cjp*-Jlp>vVM$eCL8b07`skLM^@)Tl>^P3*$!o@>;|iIXjjL8++En!KmSX3z0! znCR1-kLM>(X3#BdXUF*tAJ*P#1hB-}RljxF&M=DOUeEca9y#9zVEge|SCQ`D!>^_x z*8MKUhtwv_WFKiv%S2kWugca`F8BYSc6F&`WYDnc^)<$T3XhbQ=+!1;%OJd1ZC0C0Cwr*l8V)q1OKN{416bA6s6ZfYcX z;b?oaeLjxw^pN3{kt!ASyxhliDtXcq1V>O+Y^-sj;X0d})ik(#=RC75AIUx z$7`^u;YJb=F?hOPs343`&GlA3O5ZM~0{;xEZ&V!4# zcS!|LEH}!4e-qPwm6fH+MdGQAOMEuw%c zKvl~rfj`l{BO=l=!Qg7wDnXWhh<>ki@q^^ z?!iIsb6G##U~aU&^KB4xcP|h|+?4}m<3VZafdP%#*CW#vBGnxwIttPcw2yx=hICs@ zyq5cYF=tujYJiUr?XtXZDisYWy`S^qofpin>icY(Z@$<*f?j#-SY^LcCxa5si6t>1 zH#MBPm#O``86|SR(P6xM7A}vj96Sk)G`~3|y=KGy8 zM3;q8GCL&5ef0<2jrG%qa*c8{8U$I_p2T_vZ+{m{?p6z#HxO_zCp{^P``APqH1{NS zn-SC(hIRw7f&FLBfySEL&+FRZ;93W+wNA9RepZewchcinh#a{0zUtvymnep!$5N@k zod`V@kGS&J!fdG?8_5~P(0c_txRtg&Bzs%kazm@ed5>WB5>yaVI6MXNF%Tizsox9Ex)9AQkBb-JLR%`37}R=jKE-|IVXpdj5_wk+s-92;BTiah-jC0YC~4;8>fg zQ4?mch(36VvZcOZLQK@sdk#j3JYN;N8>w;1a&D6El9TTY?$Qy7(bO7?v&JK51&fl5i9BLQ z-72bGop#EH23R8$*4#w-pO19EhptFU#?i#4Sx&`qea)vMPU^l*g_lQqPhH(!M?*R3 zl&d6A-Y}cO00+J5>WW@$i!{Z#;Z;cHEt|g# z|F-2@PC*RThiAiEO6HQHqtF(5rV^H@u=o^FM78|c zO^1{FtEg-AbpYMBe){SpH&q6m*is+YUe5IF7$yCCTWJo2!3^Sp&t*(Woyz%JCiJU1 zt}<284U-3D8TY!Xkaa(#>y(J~V5*5#}S)8;HnR212w++x-I<+n)mQEux> zqvH@fQNTgFX?Tut@F#CxYI>_2Vq6`p1E7p~Ro&_6DfxJ2n|8=7uHU!T;<(U6Lb(<&a` zzZ+B*TUgJR-V2=(?dargwVZDK&UY;M7QE@u2_8 z?EoUGU>Aa*AS2ZozeUX$#lnk{DpHwJ^_ww*!G(``qjQ>U6p$0aV*om1dyF2#M@e1ZS(md6- z6uZ9E*J+slfK?Q#w~8O>CcypQo*FZcOin+iV9Jip&^V0^EOH#u zX0Qc?YsPx+O>|Vl8W~&g|6JQ8uy@~8&~tY=X4*YL(34~6Z`;?r^l{Jt>`nZS%5xgH zq#kr-_Gt%Bt+n4tkvkeX*(=1;%XCzqJxCog7g!#y=cvovC{rZ2yI4kKdsKeUC`wc&KLP*vxUM~(82>T;yw zMo;bTAc2wJm>${~KfUGM@=8MMDs6e|Rb3{RXf4&6% z?fE;1pG@CS=c{I3LfW~@bDa$HD-`qm9m5Zf(yx+(sSYB^%070i5h{EH*Veyb`ss#J zhm%9BFu#Me{FpQlKc{yGF1m#;Kam7Lf+5$#0@Z44K>TFBG~WeKj`weq7v7Ux=_n+7 zaV6H+p2?}--QhB*!}}LU@XznS`~S}G(9A+lm$rgm*4U%WX!|btfr-3HK4*!eJu=|l z$XZDdK3D0~+iSb{PbT#@5C`wHO@TK114Zcjz<#f3eRF607B)C@`4-GS z*5~xS+RE62x~3>85kZfiY2jkZRh(X9A-{Qgx&dMc+i%#bbXx()3V1zRPp$7jj;TKg ziT}O6hd+y$A+0Jbp{5jA{sJ){B;2C^@=%O)vZpqRxsgGgnS(I-vz+|P^MrLr z7xMsbEYOD1lkH(2*>NHuXX77yNO8+IxqHZMuWWM!%8N&RoL?Z#uL1Uq#T-=c{kJ-) zQzEDjf@Lb7C1`6a;}?2@SkPILJHni|(W2#}qbVKP8=Bk|a>mx1BRB4VLe&3eQ(+^) zqb&HG2WqE?NLN-Pp-znGmbCB880-q0bsiZb>xUqL^o9!G6TxcU*4#1_fybMwpbyHE z$)m&lJnVrV1gSKebx|%lD#PQ7Td^DBmUoII4{nVx%0c`3uk0kc5ax1>sC;w=HNa1E zu+4pJlAeh)owre8lW)qlW|OBvdY@RkiDn+_xlC|Hv#hF(ImnfhH?2Jb7O8^<`*`in z3XP7gO<9@=-|`UK9+u-6Yd(hV2UtVeJjXY)LOccf%zF3mD#nvBs%m5O9DmAZ zQHEw!`nq2pfxpu8|GhKfPt!t;BB;zHCU?L_@^@w+_4poAc7KpU&B<~!1QR$AMZn)R zx+wvoK|8vlR}m13<&Ru^{vqkDfZl7w?1JLY|7);CKt;i+F#GQQ#wZN{MvQ%)ZnLiV zhf^#VOArq$05Z|hlC$lbcFzJbTv+a0JA6rbCM7P ztQF41#LpqZjo+AffQ$;p!@KU>PHYUgD5gSxA?5!)O$S6Wm9TgL2y6-9PX*m}SeHOq zX#Af}i>pX$#4#>@$G8yXZ$Q*c9srD2TW^wIyLL7ATsp>@S>)9n1H1wz{c_F9eK!z1<Ul=k>HFoq@; zlj&%?S9W@%tn$UA+t~?$;PH>iv(r+C@-@&K?=^N={vBd1Wv#Qr!NsMKQp-38{CHOI z%k*mL;)(qegZLRbH@z@k{_*>Qjc$YQM)>bh}5M&Now?`bOPkl&m0$11~u%* zElS5_I`$F~!TE|ml++$XCOZ^|IomTQSDqf!K)tr&U zos+Wxalm~ruW_&DBJGw3+M~Gxs3?u!$3G@Im2O+|)jFm*VUDs}m_I$;t4X`=eab=B zb^$A)U(C{SJyY@H+}B9<7gP1?&hxwv&U?KEi`m!Lw;d99tbclZAP~3yX+HL7CPcRW zUcUzqE_^#$V5d)k!;vSZBP=W|*Jahr84r!wUVXMDcxE%8YNX+^({Fn=oUiAlav5V; z1o0%PACtd~S-r%3sV{?8?EN>b+tInsvrOBAUo+BT9VcGRcWbe4d+qpfPl&r+%hssa z1(I@=F@Ty7iHa$ldr1@T(~!?RlC%- z-dtf-DYsn=krhB0PS7|5u4BLwq*3W`S7UxVk5$QxROB7*^#|57`y10~Qjylf2H_e& zTzpAc5{vQMy_W}bm7kA!Q8_NuUZNh$)2ZX%8v{}sm6MYL(SvhHYl0=^c=eu=H|%*C zU^;J`a$!&C{|c7e6_A{F$E0EDy2N+{$MD+sh&r{v!@EvA+$P@x(Ytdvc#=eY;(fYt z3Z&*)p7uHZ$}X}Am%VY@xCAHs@rB1XUR)UWrE>Uou4d8=5*olZP8Hwn7gw(f@>)$>fbl`vea#k8(wVZk7(b$p(MoB20zj)1%^bSEv! zrbtI~b(bquQgMJ4r~;2CXRTu#rck=$DM^>MiDd4-rVlvSn(r0g0HW3#^XW!pz(5yU z)ppNZd$dWY$Z`O*rSgLF@y=}Gj&8JPKompcL|H>hT6MtdTso1MvHj%=+i@dzb!|;j zPk8lq3ia`~OTB5bX|>^v17w;GI_ncvQ=tz??iU^Nu2*gWjd0{FP;S{pEzGg(O&~Ma z*JnE$s@y{>;#d6nA%FTZNE5W_9Cw)Zn{M=*mdp`#E0c=2^Sbd3=zVETjaFs6c?GNcpq&0F}HIqoy1@ z&h?NnyWE(Q-KgV6ITMhJ1YO1ut!Dn~(r9~U-3BO1owBeQ#7>N4bIpV_I4twJk z*5~y6AG~ABGvEL59q6ri!M8x-YEY`}K2)a*)~Cnwf@y-q4LuAlN28Om{7zheJIXMd zJ%Desnay?M%4)94;poODXz*yQ;u02CeU{d1e>biGkEx#&&gG zrNYcexlN!mPPl6h+OZive6I0d0aR6*V9WbOstdTEK

-ULXm0z+KiYM8FxidlaK^pUK@%y@Yqo9vm@>;31! zz+9NHs>35)a(GfChnTVcgicGx7uf%%T4)%V_PyRG!Q-^0sI+veVUAuZ;@j3ieJ|&C zphL?x8Yw~r)%^K3#Y;y|Kp_I(2N;>zTp}%)1bp%4j{HA9W4b{qi#lp*(Xqgwjnpyy zWW`93Y2f0mw3wG8Cwr4@u_f_NYo8tuyM&$w+VR8`#gfVnhFqd{wrWK)>OK-zQ&$Ja z&)p5%9Tg_DN8R_WPQf^~JDh;i4*={<=R0Di7S*u*X(O}Pxq%A9B(Cm`g8!44~eXLJs_(?YkjXL zl705c8s2CIB+#<2sX+&vucuR=EIcl&2ps_9bsI5BL)nR5#DQSJ#IhA5{sE$rgjc02o%fPzI4R6hXB_fi2>R}=x!AyDt$Q~QR{mPdz zKS~vMDvtwkb()Jb@+lT+mgXLZ3_ap2Ucr!vRAcCX3{%${ExvP(pi)IbMg}UkG0C4N zv!X(s?}=jRr0Wr26+Dplqv9uLF!L#VKF;$gx8=w8Ium(9S#;zBkrqhasP8rgP-{7n z`BBVv@UEjR5#ny8U3wJ`=i`r=uOlabJ!sOKx)H$p%^kDr!ADzPGPbbXu1_Fobb$c5 zA=*%NdbpgOo$V}9eLTxH`Et}^3L7zSe6V$Jn4D8K0 zYPr~xqI+9^1Dy5EtLW|d*RwngZ|zjs(Ek@rUjY^6+I=sG3Zis~BB2aYil9;s4FW@m zq>>VXAR*l#Fvh!oumEGYTHc_iyu0GQPhl$i1W67@S#C8a-)F0UF&G90;{H?fH;$Y;q>O&hK z%@p?~&UG2L1TI`Hq+#>-c4rNf0gCJe(i>1bln@h3d1&(q->0Q?hT}$PU^Ieb57ifX z+z(&h!j9NB93l16unn-Ce}WFB!(8W7%0XWGrqD5oz$3Q zubm;22+=s+&RA^&&P=H<2&jYK&+$A{URD;IbLyE1B>Baxl=kbc{VDf-AjdURKP?4l zk_-nqcxn!ORXwxfeca=7j1<`U6j#7(b$fG((D<;KhHvwyOi3el;6_pJ5m}>-6e(Ep^Y|TgeIh6q9@+yTO{k?<-&|O9)sG1 z!XAJcQ`D39|nE9%1KsjsoKhI-Wucs8aQ8OE-yK8A{W#xIiH{YxwExO)B4OK$u z1Xr_CtqA{tL(Q0Hna|0=gy13FkS00m-Skw3rg-zdH3=5%*Vw{)}r~ zRq9?&V`C)S3>_d7ntYX<%*02_H48aYd?D3;VZUO?>NPp_#=?h)jbm=WX-zOE4o{c? zQ_!!EnlXZ1nLZ81xs z_Psl8&1S~(k0eRREGj;ZaDTyucpx$3p*fm~6E9+M31*|8W>g%gO(u94eOl3|u)B1g{jM!c}HdeizFE;}JjF4`ox<#&d(iS1Q? z!Y$m|eMd$(eFboKJzP>M@2#!ZA6>^QWu>J-3TLhe+p~F10VwlYzocaL^yw>dUy*0+ zG<-Q6v}7n~Ag`PVy}7LFw(+#E#`!Pol#6)^T93<8J-l?-beDNgL`}bfxq9mLmEAmu zDiU~P^X3p*&pWYdU$psgo&Z}sCtq!L-+!p&citV~(e5r5(pOOekt!i|U^HaMEdW6x zUgsF~ItxF34716(4$xBkWG^BCHvsKcGw*oSbO<;66u0;3^2|*3^~VjE*HnAA{2`zz zVmjF`Z#b1#y!lj=VO3h;vVUMQ{nkEMmarq~V^{y!cd*3YsquTXt|ZL!q&UEfg0)BV z-+TCJ9%_h;@$0 z%wxsBsvNRPX=VVrhQX~t@>4KBo~CbjTvX%s#QJx;ag=g$Ii+>|QrPqn^e3C!>W=0! zaHq$LnRQ_0zg3uR%YS0VQ0i=+y=RhmU+3~jwfprVWU$m?zGQgV^R)5nTB)SNdv6`h zDwk3910&k^X>VJR;JRgCl)lEv(6#Hkocz|JN!S$MIX}P2)lhm$}9-)S2vlpD{W7h^v`Nc;nSzufI^3zoU+!N^WhUz-syQPJK8#@oUpeYdFO^3YN zKn+_)8^@_mzZ6<{k>`9T(7>+sJI^*9EompsV4#=68J`#hn9;%yEoztjYmu_4y2$d! zI9}UV*BZ9{bu6dC>TxGm2^L^n6=F(RG|5mJ>eRZ?)8#tPq}q^}7k{?BG-#}Lrj0Vq zBrzeqykn#>S`JSZ_HaXapO`W;vD2cHiaUwVxlKB(oRQvSuXSKaQmH@kT(a9j8W!H&o$VG) zMJ@CoIjZ64ukE5eqy0D5=_00B64chqFXAY=JgJ{t!Z=gKQ*G?k`~lHQN!kfw zUVqrrcYOmoH~bw&=xwL%X5YH~woJHz+xquUQBevuV5a2Dc?!QipW-8S6jWr=8qyjY zD~P*P&$)?>o+&Vh{qO~H#KYsov9U3zKi=^8umJ{%SK62jeBZ)IckVrIyiOMt11ds2 zck@TA1F#c)5h()J?_Bfk_7}4T82&q?t#X5>%&vV0Yth<<(chr%8L zNF%sO(`5bxP^+Y-c<;@|hqj*De|rtLCU#ESz%9d7&+W{2-Kg076CyYi#}ixho!9=1 zw|>e}4&%LSsSTmo^IG#%@0&)e92Q$6F7ZjBf9$qv*GrkNxuW55H7$R)JH4hVWYT!} z?4w4zYD}!3pe#+_KptV5X-~4~;q?XcV1?M&d$?~0x@S>IjGdP;{7jkwAuIBS^Qfb= zYfPsX%tKV|KNFnmr7zI0(Lxv0)^7iJ#p@RMK)toRf9b@l=%tH&C83l~+`rI84W7^4 zQX@Jd0+6=)2?P$Vy-bQmD1a{73D~^0n2sc=7m0w_?q(pNWXyfk!iWBM;<3f-^QM#|;u{4}&th zRZ%_{a%%UN^7<=S?j{8(ocDqLX0K$h*5$>l;7anVJ5|ArT& z@UAj<_5*C*NacvuCz{5Iib288-UcJv&7>B)nM+~+n%neOn7y_h@hSKkqM)HX7t*Tv zC@d`T@&kur-bNL2+Tl}XxvCpLhscvk?;p_KU?5fXRc(BD^D01qX}e}t-1D}dC#$1X z>Pm#YIs0j*&<7s|BAKlT-yZYlBhMIFdZNe-rG~Y-`VBn{F1LEC!P@Hj(LEa%+d^Y* zlFdS&qflO-7K7`D#;g8hBJ${hkR>a#B?W$(fiGv{e0IfxHg3g~mG9-~L+|E`CjYjAE!=4O z4)9jV9xKB2TMRx*9rJU!!rx4`1#Fhe?{_5=#)RicDX>sP(!H+zH}3I&Pxib~(<%To zYe-?(UBW|z&51eXnCs242cUhI^aQvDBsf*;P!sV1{7VwoolsSys9VjP+5W^{3ep{o zmd-YsaN3OzucVB*zWg~>5Dke zz?YI<`pUEzZ`d%uSJuMDhAx6s%U6~E_U+H&(?Idk$8bT8?3gtcBx1a0iGw>x;go6v zsIO?~3TZa96?yvD>b?sMn$KT&lI6$0$h8Ql7cQ)}>JA&aMPTH0!{S;_V^?c&qLAG+ z{RL?Cbs@#VeyRbYJQ_Qc)u{Zv&UXDKS0rmfK z(s#hZ7;$6htEcUrL4+qkKe{ylouAbDB?)o$04g!EN$6g9n6wR8x3!kyYMY9xXO0>3 zt;M~1@9dqQ>O8F@Z_O5euGs6PRpQ8?dSE5L>)O-mZ^93vQtn^}cYs%U0|}pYqB4Js z+pE%jk3X+g`ybRRq}}e1rg88ogTjH3hFjMc9$sFrESB2b)6Uo9h`O?=bOj@Jvez^4 zaCK>kDuQ6^jx2}UdNcjr3qk8)Hg-*|TxsKS_ZOd6X|&ZY*$rmn#7pBxnkt-vyf20r z8K97DbQu0@Sd*MB%V-_3Q-A@GM|){=g0|zG@&DjGQ^(vr|28Y1tJW^#Fs*Kso#`EK zx%Em%e)PS!FhQ!66Ib@)Cq_}Nvuq!#>0$09*VrqO>BOc6;~zm3OEZ5>lmnfg_Q-`x zPTM@WY_E>l`SF9_9mI>htkqv>m34%#RG_lbG~6%@Pwa|&2!QkYlN(Wwtdo_!fAjHKxp*;bp)T(sm3(KYFIRIj6#e6m#Z-#4{MWm-kMC~F5$25K zm_F;Uf6vA3ee#jf<@fJ!p=n}~k$XS&t}|(?1`6I9^CPnuaUV=$qtn3+m;cTCTBO8G zGeN!bj>oLHhg2q4glC+={JkXohoGAENAN&EVIHajJI>0NBQ@Qpd^M+`Z7%%ttNnA5 z`8|oXQ2{td#p5rjp3=rIHg|GtEn~{_1Z({FRA?@*0Lh%1qSqzxD4WC#UW{B1!8t~W zKj-;!4riF@s*uxyl$@ROcd!oQn2#*E2X0)^GA)4?e~)!1?e|o1{zeTg=PO)R$9q83 z+bTgwj!=asoj^Tw?`#+P=%WjQ+HLg)Z*aZq%_wpi?AP6%bx^Ww(f2H91*W`H+#A7n z(FnZyj_4%fv%oRXMF6*vNh1@Z@I(3SQK_Pe{e0{N{_&@JIerbARszm=OH0dYXr%Il z7ib>8e(q%nk>VSMTR-@bC9iFxn%B~$MF(3Hr^veiK>rzt_ z5=eDyVg^+2$-MXpC%fdmwJVr!;h<2KmzTq#yl(`_-QS{csvtPvdBvd-7dzfHfeSAo zG}4lG1ExaTH-1p*T~qhxfeF~NB;k_s+tPPgBYzN}6yuotm?O%kvvSM^8tN&0j7qw0 zT2awDFq8rl*G6G+x3@sjmq^)tZckBh!h8I^s-Hr|)30H@Wd7Zb#97RNo*x}k&Sn8`;;zyhl z62-s6?|0AIi-O)@EOzRQLm-tq@%^Rj4-1JxjAZcA=0Q`%k1PNQ-6>S+Dd0#hjPMnt z+`n#8I^Ks`wkW`5r`iq8VaUH%X}$hw8`ag-yY1vegvQYi&`cA+Wn{z~w71s8Tx1sW z1QB%bGS8{bHa&PIxQD%iQIMRBqn|XpU11`BHEe4*QOGL@qY-Fn?SyluqZ`x&!pwT zu@;Jnq2uG{P?Z4xP+ zv9L803M?o4;y*Zl+6|pNXXZ94IQL!sVcqr*-3hn(t9;bosqa9;yf>dRPqObZagbaP zPO8dY>O6HW93eTUUMjfc$JmO98#NNa;t0{Y5Ed5Z7z*3?P5i1GdBr@ZNrEbR{AR-T z5Qw$rPpDaTLpzi2w?~nX-&!2lt1GR9L!Ci|JE>|L>G<-J{UQqIsrMnrZK9qE>x_-| zgq0<}-xskvH>rIT`y3L>;TZH_X!xxeq#&If_?$%8gq|i7`kdWyFOLe7{ZY4KqvvQn z!iR@lk^v3mz2Sa;{DM0ogS=S&vIip0<8rGZkYx9qXBK+WnnURu;8~5?vn}S3-%d{W z5p*daRp%I}thJFMHG@KJrJHX@jdoyd#gF)kCbjcDU%{`f-|<) zBI+mofReiX9U{nOTzPx#EBWf`D!qCbfSCw=5V)~doi9zi`wp!}S$B8$H5+o0AI3|n zQXd}|`8C|^>IR~^JmGw-6gh!O%bTgh$v9mbO7)8y@NwTYY=X=o`ym4UcmozYVwCWt z1|3us;YTlFvai@w`uezgc+?~iopt+L_VdA;OJB5YZ(pKO?|99$SyNp-BTaG7aiKSm zl>Qh&aSjq)d>$NW3xy8H-QbMiY2hls!`a>|_b2wq4EHj&o^?pp`-YF%Fke!g(4lWP z2iXic{@8F2Qc6mKW#rOpDolO(9^wm%Nj1IOOvWtszVNZM%5OSQj!MfjZv`EIUL;F? zb;0Iw3=ah+Tl{c}{LqXMInWByngJhedvJlno~5XX&x5>CkmL{5k&sST*M0VFUWiO| ztJgk%GZ#x`xiVVi$%tO)q=w>>OOBdRIK7NGgLD4A1+{y?^;m6l; zs*uZP`sA3q#KHR^4P2@qHXNB2Ms6OnOYPLzqyer z3EmG-NUnby3(1){oD(#*bOV5=3IbgNdjE{ga2Dy2vOS0u>zabj%fCQ_(?=U!=7n^G z=pYp($&h&Rv2XytMS`6G$p8nv>Kon2v_*K`;W^4D9%4N)n zeoS{4Q?p#$(WIe{C`AQTbCAWHs^h8P))IFTfwR_>tCR-r=s>9RBHYGRglY45Mo!( zE#W5aW&6<~ko9E==P!fo(gI1ZNBcW{!>(*z>68BEIKlOKv42%(Na<2P1k*mbhkNNO zfgE&o=&;!HVk_$VpUgy?XrBvSpfq?w=h{Tg{t_0<=A*AUa9UKs<02_E!I@O-Z{%qE zjkZbFg?nzrh88rUEf7_g6ad~+ZQo*j@qr2(W!=(!7a|+;zh~=%hzu9@m z&~27>JwO%>Y@cfX}lZ?rzcR_SOU_iaqcknV`!`a>bj#R7yWrYTCw2S$0@+jQ>=h#@+ zx!A0CiHRN_9?jIR`E@zc=Nhv>J+`TOuGwSBic^Ic?*?&amrxb_NU;9I_ZP0O-(~ZHZO#P89AieWW2+B0 zIu>m|0495aW)INPas2z%{7}zb!ot>;p7chNS0%)JvqS0{AeCE>KTUEQ?wvP$gcxi9 zVsB1inM1N-&YnFxL)J_q>KS_DjR_D1u;AY?3P&+~QtVeaVV8GTo@x{@n}cC@9f&DzwQB<^_q%?H>uv^6U9OsW{W+Ak{;^YA_Rl}D za^mT?V5K0W!*tdD+N3{uyXYwAW##sfXXvtB}>aV?)Tu=^T>ujY0 z%@LnzZo%zXnY>>J5CR8T?qm7@Mp#!!Hqo72;D@b`!TIu=7k_2=_j=F`jM;0|V)GuR z>OULJsrK*RJT846qTtG(N&?hWCgc?nf#9rk^A_eZ0-%iotcuk z?+^nX1ImIH4}Z**f&Hwd<-svKIFq!580jA3QkT+Zpj(Nr>F%|S|5tbn1T6X3)FLyn z4~Y{91&4qi4v^krE}M2s``pLaSqD#=qk)1SpVnh#{weG=y6(r?UAZ@qJEVH82GY@0 zvU7$4D@coQRFYq$g<%jq?tXh`9yCPC2`zr3#(?P+LqxHkZJYbzf3BjUB60rR8J)Ct z(Z5vyzgGZ`-5XJPPV-Lo`6sO(Nhb|;cYJNr7Y=s0vQz8*Zb)|9XA^&8we0wCO;d)8 zScBr@H967GenGTn4(uNja%f28qk||ZlqhdEa|b<_kW_c;)oRWPlIUQ4AR8RQE15fY zwAy`FSwJE2zQ^nEec~p-R%>;bzYa zBQ!iveM0x@6&#m;@rjH30CDVDCC}>X0n4m5G0A;vfe)^*@RO0+d1Z7DBEj7~oPm>* z)1)>0=B-;|&dZnJT8v_zyHLT>+6M=+T2j>|#8;%3b4bqcqkB$;f^~P}xeVJQ%=3;R zIcji|T!?rb++~J*y*(W^V9?)BeZ>~)*9iEPXR}GgKeLYlv+CMdj0(d$hB{M+fnC^f$1>y))05KJRJA&B*y_Y+Y-gi^M}{Hv3F^QWNZ7BBBA zm|rwIO1c+aZeATEp=Ni@oLZNl+%KcgBKW59XnS=c#Y?rp`>3{65-VGb5AfbE=|R?$61MX*0M4N{_KCS|Q1}pVoElRI#sV1d^ZxU1X)x5fT=43BXGURrQzN6P8!4XA`6_lk;mqPT z6(bgT55g{^UThfk4m#{C50NBD)zsE{LL-|j=8+B#NB|C}j^G4OAw!OwURiTcij!}d z+Ad&}6Ze56P?HpO+hFsNF5Uy{@waT~oRJFJQ9XWX%j6%}gq1r#nwripOXpZdKdHN@*#kJq{t*OO0!TQ#)@>B%dshN4I5Xoiobr;YbHrH=oKN;2BD zFqo^>DT?3tKG;HSyC9cj{zIqw{Zj4~#i<2t3|c5jCA68}`_Q$=^ejJ;bzyaiLv(R# z<+CU&bP_R%AMal0{03%Esg%UGZxu#9SG2@~9m;sL(rznHgi}C5{9Cr)Kc<)ND^>Gk z;ML3Z4D8&OxRIBg9VD=U7(5aeNJ$m|!}CFThvv`eVCQeJ#f|%(yK7)!G6X56mD$Jh z^DBbl@Ehc!?pxowx-tw8g9S~7%TM-FPK**MzYH8u@u1+JEdLP|<*1)kMlXGq@5S<7-1xM-T%{9OCXnlt4H4RP8 z^$G%Uw6qcgz&}`en;i07kyOmy@ctlJV$iR=6V#5{cfO?neR+UbAt9pW!Wk9lCHH+5 zHHTs0FHAo+Fz{ZRl+Fxma`k6<&}r3ax>~B}QH6 zx^XFvJtOy$Jli)&dB!(gh*RDWprOE#3Q4oH4x&oxO43r^1a zUT!D5zsY^Dgu-r_0a&mIc`ui4+kyks80{Yt;2`*^y zbV|Kdi%mf6QLcI_{l2*GE#O4up(7o-t524lo}R}v(%Q>PkmZsDvAe~4)b)W?dew3; zm$v3YU*Qsf#W>ry*GSB6B#Ah=0{~t`>Ad32JtcoWCwdpikqHQ3mxbOh9>$3j!5cHl zU}7JucIs$v=f50=Pl%0;_3WTIoy6PY+Vi5I>h;~w232~Jm}oyg!W~JrKML;)UbC6} z9WL84nfsw@yUj$-bUsA81Yzi$%6+H5Th0s^{XAz%KAj$&6@GQU7)8u zs8ZL`YFScrkG1ySfX5>8+EdujDl|}Tpr@YTesq5()zQwd+L7~ZJ6OXGwicz?hasyH z2Yfv}CUxa@h~G+14_J3Bq0`q16dlgGOJ zEYNp#biRJGHC6-NjEe6|O*xBNHYHnJwH3mc9_^b@k(LBXay zQuI*|nBL;LDGiGwt+YSlieUIv-N7zJMZ;|{z23ZWqt$~+z;?k>(MkO8WUbrg0!Lre z5V$i4S9`%ZHuC{V5s-9?Ee6xV3m)`Jxo!qNpg z{;d}Qw{jI$n+Ir-_WHrh-P0AI#ZYeQPjn{Y>(RxU9f10P;lYNUX4v4p zrm-4i zBUj~vrCh=1`HR6fp6!f$ut@6HkxDxgky_-NlUqnX zQqR2&m{q9*lm`9wI^szJvJ?4NaK0WAw=fSgm3Ni8p=aa{@pu?|u{cMk!|R{kv?+28rwp{wZUFy{0#BQq0d%~)+cJ6SB*)nO2D z!2Afk&MtgeHhmTpl6JpV3t#*4z0eRWyFIq>Q!vpqFfeeAoCys_ju*dWxMP<8xM&(W z43U?Ur^Le_{bn$_oG;F&q9)bSq6PR%m8qY|DvJdb;&+R4x%RUn)djrFJ%EAjgXZ?y^?GT3_0Uw51^C5gHCSm)gsh5ms*OwFZUNtVOnH_pho(zRwy}#^y z080%tEg8RO0*Udr^sAqK@dLAXlzx=jD5Q=jPoDUny|@J97CJgQ;L-+2rMw~ex=Bog z2f?`1 zT@|kR;8jDAz0a3k@Gb){*#C?DLu_U3|N>2peQ-6kOIF2}&LrX~dFk({^ zhLsq7h0A^ez|CJ&Hm>yOSCI+&d^8dFCRc3qD^K9n%7oeht&0NHKeUx|j*RhXA>3Jl ztT74&Z8N)W#E{h->&?a;O0{%Tz!0L&@7NzEcS^H}NnpOW0t{&-bCQ%KAopeaq7o2? zmpLok4oGRF&`^A+cNn_!FaXL>aXqV+r2EQVXz$-=SL^>EM_x|VC3wDg3~rUNr`Gb_ zTOQ0?xSpJ*CT(I_9jtL9(Hhz&My-p?^dQnLq+O@KXGQz%7oR3h7aXdzHK$Z=miJ}Sa zle$Mz*)DX{L!tnwE$OeA9po9XlDmV<%;Pa)T~Q={EhZZ3#Jjq>?!DzQf{`ck!5c8+ zB@g}oqUjh+Pux&>KlzA4odsnXs9Zh<%NT?cU#vu<(K7>EwQ1k_Wr!H7emu#2vVi5b zP)hkWM5r21jE!xU*1`Ve#mmrI4VMNd6@q{kgq2W&Xs6xJX(!tS&7)bGW$Kf9sxGHd zBpvVLPoFpCFWWZ8Mng&~65Ig{iFG_R@=QKhXKj>l4-&^3MbOnqDq061zxxSnhf5a$ z^tchT(P_;&-9j<(&?!Qy{KaP!aF6Z)Dzf?NQs0MGV0~RR&V8zTjmr*#fv6ajArN0# zYp5^EkmF^V^9}0WUEPF2!9*Awc-2QfSP-1?J1e~;MX?@G>;Y(em|d)cS6Z}0;+&eD z{M#+nRV)ATmrUm6@wx*rem0Tv1)Yhca&mx`-^bSg;=t{}r)p21F{xeq4b{#_?lm8X z@W)aVU;;lUa0yjxzJDrG=0P}PTgU(YqPm`5C)okiQCDI%3ggNpmcwKt z>n-6326nVphs#2Fr<$P;93n&tAb5`#BcQou9H6%Phw!t>=R)Rvj^}+AaCOjwBUZ`< z&~Iy&p*5K~b6MsbUS|7o-U0rGQo?6xmPenilXFH*iMkCVn(ZuVClL_>mT+pMey_IUlLgrfweRKm=R{puc6Gf%Lhp^2yn zxi=%J7>XJ6MPt(V&XbZ38Dq(rSH9;^%Zj+H9KttLCsHHS@Rh?Xtd5TAIiH2DR(Xdj z=&XD&-KQa|<;e(`CB#ZdJ#~2|jWOMQc&Fz<*n2SO^BhA<{%EbBW+lDh541b5s7jhy zP!hDRifsVWhBFlD+jkcFuKNuU`d*L^e(?0?KL|!A_~2-wX>vRp%dy%9_|%>M{0O@S z1JI_tZgbv94%L1A3U^9#@+PVYib%K0wPaRC(MYPWH6WBqe)~WzT&D;UkBok`t4LA= zg%a>em|wE4J#jI`aWN5|D4XytjskIw-;WSTgnT6fZv>fm{l$s!w64TW4Dt#Rgcy>d z=Y7}c#DBlJ^NSe*@BqrzZbu|lD%D}e47}G*j7%RWg?^`;Orv(a=*N&CK$2#l><~89l7tCFYb8=n5{pRyt={9b{2lp zGAep;Agd%EriFH(6qhsv!_aS#!C)5d)6_j!Z5G@|0{d~F-1jze4X6kq zN2i{Zh`n+>%}FAEn{&PC$-l1NH0hkCVq$PRlF5^(p80~G0jmXArB0+UuNvX8WQ>f` z9^=kTAtroQ3bIJ5k3}B;Rv-%}s>ncKk^5E>zb1W$ROeB#>n#ZlW5!$^n}OQ& z#5MbGJOXJm(wKCV45u(mc7ymJt?0b;eKFq9d+L!YcO=!#P4Jf&#}wPoVwec|&cM5g zz)>L6$A&&f&4$F?$oSNiTw9XyDu){#&I`SgBL>mLoJJzkp+x}=#Eo!WRv6sA>RieU zPGKg(GD6Ni{?-U)1cWc*5g*<(D^hg{*5jgqWk`|xV+?fB@=ZsLC-~A*oUDlRa<&n# zq+;+Au|9^lCm>EMFBU8Dk_C0S^-dkjic;t~SX!>HqY;}OnuZW?B7=u~Be>2xrEUHp zbflK70mA7s4bnwY4N}J85JFbN-C_?0*8wfZ$VpPbwdSE2Tlw+vN>V(?AR$s#PY%7$-?%;s}^(8UmO%)LLbV_ozZre4WjyqXJY^f8qbPcL64KQxTK^x z<{EK|=LQM6uY?2~G+=4TmGll=hUx3K%Oxp;aoTz6%Nm6XB!Noa<}wCnbaeV7!30wbzk06%`HtAYk}7&S@H~ zIu^~1uLw@N${#f;Vh+t6pQ-VWUW=!tV5d|J9<-!=7{-Elg|M%9BdM-_Nv4^AH}`Ti z_|%tIxZB=G+uF`C+R(@NKsjE>2slPYqjiJxspkRx8?Phj6RV+;pV6dQq~u60`Hg3E zag%5=B=$*viLb*)IaeZ?2Tf{HpQOfNr9!Xr9{Y+Oer0_w(YDy(E*!7PMA%2bY1B*k zRQ3<_C|OoY_<&S#jz1H@(y_j!?qQ>0GF7) z3!{6-AmOh?S_VUh8g!>mo@5>TOuZu{=D4b^mFgi_p-3lIKt(QJ{u7K0dH@P8rtJ)_2;NQD8bl6-hP5 z2i>TM@>%~WprFu1*UqSI0Hc7J@Gt@^3z|lqlhvkn{lWQtm+9q+h7+K`GvmLDt)Irr zILUCL0-BA}?oli##Y4`z3R^rFnJ2TI_M+LYOETjuODVq z`rHUNFDQqt0@fqX1D+0tZ+A(|Q|RLi*hOyg^3Efv`^Iv|OK7ZC$7-O)$t*m<0A1Rf zseEPl2-Yc+$jlux{&F=bWSrY7@Hj~*Sx)7_6KOU|QsARGXf|jf_y$cj z6(upWKcw}1#ogngL24*bgO(G+l*>M+CN&Ce!j>AW0qBukm~vl-i`T*5yB<4kLc;mz zDqj>;qshot>=f(RY0f@QVDp)ic=j#RW^Z41j7l{_5mFp1duGyD+6a9iIG5JMl^S4ig*fk>mpvJVP;} zAVI{5OV{DE@WA66D(>zgJ$XdP6T^3&sILQxu2iaL%ikTq{tJ5theccxAX{Vr5}=O= zj%w~=(|>5iyXhe&9O>sW;`{jtuo12zc7Rk`q%BWDih*{#3Df|1UZe%XZk@-={DkJd z0uZ_54@j+U!~y=}_y8H1{Wwwtn3A8l=B*}$jG28rgzg32=HnY=Y3ZT&AyS+4Fy0VP zut?tga05sCW~2e6t0Y7iN~#1(Dgcnwf-Mes40 z#J%5RYo8_w0Xq@iumr| z3)JE$U346~>6Vyo@WCvo3P-^nqk#VI_sMnLn5iEXNmb_n9y)gAcfG+^UM)t+x z5JOnMwjQ#v1o;`@9ygFAx((7|BDQwTaXS0i<>aSo)1#)HPc)XOPPTkn3Rz}GGBPgF z&P{PX;YN7~t63%|*xbg`uiy+w{6tdHLO7)H}xqcVC;Md-GQ`nQ^tp6@I8fS!GkE8PI!gx*v z(yt7aVE?-vL@-C=pAuq%^O_RGek=_LOnsUQQP&|J%N%~MiEepGXHF`WIb^OgMv-m* z_=Xep&tLa?oC8{go}Jn9D+-twYF&2Ad1B_OFp8f6n;p$VEDPctLF2_)P@j)rC!R78 z$hyaXIdc8)cA7RPVTT8g2&`YsDNfFtH$@@%pHIDi&&ae9=rj%U{{S|n1ItR>@Vh=$v|fr zPP3t!U;^@5GNt4%U=Ya{FUXTnxPj6%v`CuKPQc`N4Fww@J#97tH4Ux74a=y};Va=$ z{R}HFI!WkDzO3-BLU4yb8_S#G-HTI`YS(N(Pui_^$E!}MO%G7BU2|F?vU2CVxxi)0 zGjQN|c|9f?nWWat(<N<*&0tmpD8_Hn zWzm;7_n>tfq9!vwg6vR1KtGeh{aTh%C7Ew0IYRnR3c^;1VueMWWxx$j@rDNbOzAc+A5;oJEZNKK!m zouV#rvLaKlJCd@pvgjGoKTWJtaiXn{AAQ5)9lZr57qnDc?8iCMGMfHAPLwU2wdpU5 zZPLU9A#%-+hZ*tn+t5M2RCzDd2TZ((KWGp0tg7_y;*r(~c-4d*g!u)WM-oFd4R+=a z6?(9Zil%10zc>osDaKo*wh*V0u&;WngKBbzR8~H?My8tjxg}9b=GEz&DA*As#5MAa z5u%vjudtv}UxQ<^B<3-Y6Vj<#cX?4Dw?C6L-kj@#p9scLapf(S_=yI!4g@2^;7n?l z#MGvXgG+eM30!z4uaNP?^C5Fga%*gpE{6EWx1YeHHyrPWVz&tJM7zDfiird#Xx^47 zy@>*L8CO5h2*J`Qofaj}$%+`t+Km3 zDUe8>M7Y#7K{|m>-b4cPiX1+w$^1HWK7hg)trgplosc|!LZy3|WsmuUWYi<1atfY} z$pYC=KJhxw<4>=!fMsQz*Z@sHw(ZLTqwp~}M$!1wz!h|Z4~JJIVxQ53_SWUNXixza z44+RSLz`)l3ZHNk%YD%@pR!P&vc7`+yaDa;Z`Nj@->xERQ)e^KEiM3r-_*@##NL z)$jq88|~bSf4|JBjWqa(*VOGHbCaykCb>~7AsLIlfs;WuBSa6(q6eqE4jeLs72O41 z7EEkyI;qZr4I8kbkvmJu5ZZCn^>t5+tZqhpu&zKp0Y?D>m7v#;x8}(Z z?;j%`e?K2dMUeoalSm@S<&$m{#LAd9uyIM>AB5cbT2XAPX^k?$F3c*=bt_Yz@Bi0rj-EeYT^Fq6b57}=4J<|cz92S~( zAED#4t31IsSiZ%|B)iL3P%QA|oXK7>LJ-t7t77{?jW!WaCnMWm;a;U)VB&GMae+h& zsCkBUGIR$YkhBE^$P73#XVcac;&2pzX9y3TCw8?+*VE9_7M-EHsnaCg*76YJCyEl< ztHit+TfdFIK0v{BiZxjiHDpBf(;cQ+Ph|yM zG{S_490M~p%_^SYnt{j5jEd9h63Pc_9A%LUa_7sYH{KQM&}+Nn&i+MB>J77avRG8_ zq5mi=DplZ@-rK7pjzVC$0a$*Q9RV%RJ>toZX(0+Fi_ZzZZg1+C@4k7X#lO@F{#>!~l&{56%=9iV!<~g=ONrrUzQBeM3#E`3w6xW)VuRxF zcvw!Qynsm1%xBtSsiuKkGP$Cvb}i-k8ziG5z=9C7fGE&>|hpbW5MFLn&XQi0plhI|O+Ikq6=yaMBhmm!l8u}|S-uUV zj|^U$-uYj$ucYuZfSyh2@7atBF>RQDpP)bPJXm!m(*jOI%0c=W=ZJv>o_YE0FGnsnu_+0R zhMau{J~uZ+R3!~4Tts<^uskDALQE^t4E&Q=(lodW-(VdhhBHhzX4?8Iw0_mP-63x# zoRtqAShO0*y!dewwji8JyFU9vR+!;0e;3yE^PTlt5SP+^O^KKa&YL@3vMG305QSGD z5vQ4U*rf^SFyiGmbK<{rdSd7*r5vVfGc28kRo|$F<0xR?n=~i2v@zhbJmSZy(^TDieZp`Q2({0F zOHS_I|T2Py?p$dxqIv?>R)QT^v zZNM`ZGZ6AN9NLG8CSFh@Y(&&>Q0e%Ro@qv;2eIbHQ3>R6Qd+$(3VMY5k3xdAmN}^M z?ykalp-yAb`Rge7%f*#dc5k@{Z1CG$su-DHX{s+3Nd|eiW$i-3aFv*_oFWTHg=!py z6{BF0X&V*Dx)5hfR(dl?8xP^N5gGod5C zI)ejX{iGk5%H#-!$mr~sG|eg)%$u`co?bv(E4xBi0!lPR6qEEMM{$%gyt zYHABI5e#=;V;$ePXP83US~|-XUp9K{3LhpW8di4>l>Sn)#qpxgYREuxkS-4~_|4hL zn}Isl0BZGDuKjq(m+PK57E}{ZywC3hrKewGQWJ+q5CEd=*}Mqk&3(y?|ESO6J3_$Z zFokd`#8b_yIIQ&E?Mu3Nm7o1>M?}@;<|d~Lbck2cyHyRi%627_{SbY$%pb*JbKZ@d z%%h)lWK5iPvGlWMK`tCw-?aMI)$GQ?Ji;8Axdq22zOdj;=^d~+zn*rg&>CjAWQ6hS za6ft6RMnv)Bye5Uy_D%@++Z6^f$+qf_N>U?P<#F+?K5fmIy-uaA3_>H(pj(8!sy=YcOQRuCnZ=lqcw2kduph}$k8i+}mcA`Pc}VYUVa8PFYG1`A=Z ziTG~`SZ~JG&RBwcGbdqznT$uGcSjGIu^-^mmTp)KFvTjiu{vb3cGR}cF5>B5_ssL2 zd7>+H9rYGR^KSDFf6E-LQdd;1424=)hOzz<*Vt3*L7Dui00ZfK^?VP!@YjnL z>iPX2*$YCR8K1F@)VlR0qe(|CQnb9cc*rt2nzbO^ouG!p-sve1iJQqN^o?h~zEsa) z6YY*BT>q-JX*Evf*lD9gyL(n!STP>sAMJ(5H}dC7w|LgXd~cuqq(b>nUhC2G8%;)S zRD%3xFG`<}qM8Qda!j}?&pZPLDdQx`LmpuCpoS*b5=WIc*xo$`GY+mCKiJfvEi3lY zk<=^aG_H%&UpBQKyQwSSVL@@!V8&3SX}Dw{4AktOcX#y+j%a>NzWD819fc3Tw+?)> z%db_{AYe6ND8v*lu!wcZ2-Q&Qwp3-6O6hr7_zKYmX4y7Bqn;<(aN<0_wUhh(UqU~p z+Hb~!{vit~adqV|ucmhcDz~?8<=4A}D=4^+oAs<9Jp~)@vU|-+n4LBc%C4=*{T^kN zz--7I|EStn#cXN#ypr0$ARk+bKZ7m>4*&|!AH*CSldJHoz z$>4h{vR3CU;lQU zCZy>hI6M5I?p~k3i{T}t2cr!&wFB;4R`(|S`G7j zOHST)0A0_~mTyJk8YV`}&l(8?#-N1Gz`hz|Y(5)$S~^Iq}x>y4#fq)C-MAJb@*FFg0 zj~;wf+!|Q@`|=#bK~dho9GkSHe#&LtWvmJ5l+FAmbN-)PA2nwr7Uo+;Stl7cN>Jb7 zs4>>FnMfp^Wp}O*keC37Jt{!cf~mJC;~7ygQEe0(Q*Z&E?rNiAZU5lL)++Uov0e^1w}$JV(9#m2GKxf_+$Hu*B-{siTS0Gr{xB-J!ovbzX(K@othG<+GK7K2Ff;UHEdI;{d+Sae+05yRgn9=HT;1P| z@pC9#Affhc1&X^q7N1ym;ba+s26UcG5lNeq3_6s=V~P!k0EgEbjrUp-$!?u%C+!4pmhQ$wsK9h=ygz#>@ z_IBW;QN|KmN{5VPCoU8a7DX+$o|BOq!I0FI%O`FIH?+hlrxJdJM@7J#(dWkv-UNTS z@`&gc{z+^~-|~}T#K0FxcI~>Q`14JG=Ecgd0o#rj>+K$VnlbREJ@9rdP-CdD(2R%KDkMj`&x6tv95oUOLhFd1fW4 zw+=u2{NO=eo;vgjb26r40&#eO<#mJXp7!sw%!PVqH4eWe$NX47Q zzTzuErlt50{Jy*#CQvBoc=n*yZ;N&GpCK@B(7vA@^6*Qk{1Pt}(_fgbHmsLq!Xt!_ zzxh+w@u9B2rFpWk_K|ar_{OE0mGSLv=xV|Pd&ptEDRntrM?vANm)u9skH;g{XYt<< z!HO0NuT`pW!=q8ILi`iaMr1#`yLX9MrE5tB^){6TiEkCH-PkCh?=P~; zzRL`Hc#hs?AXKnyqApRZqT*A-$>u~SbfSRp+%U(ehQK;na8os)?2Tq_;mHcIHi^L8 zD-$fHT&O(oF4%^2v^A?6CA+Vwc*ECqDtsc;j+onx3nqT`6u44CIS z(hqnTomx#Ki0g`)g3O6Po7B3%-Y1zKg4`q zBe;JOZV)mm+k>*nUD+Rg)+G$g%*EOoPku!@YK9UJr$l;aef)XRwGq@!EZJ;XUE<^3 zuL13)eL4m9FZ=`BAL3+Y`C6C+?8Rr%3B|nz7ynuT9U|Aq#HBm2w&|;Or-c;f{)@d~ ze%ONHt|9&9W}Mcv+fP2_%Pf+^PW)GSp2QV$U-O$3ysUv=8czEiCY1X!PuGp_bsFHeK#ckUmS(nDttkPIaf(Px$}#H(<>npD2ugT3EbXz5Fce#MQ9vLxa3X9qV+ye!O%t@|ilLHhP*-! zj#kF_(4^2A%V-(7^66n?g-!Tl3{Z|nz!_e%F9&jOesS!c!tcFdvIF94hs=mWO~p zL~&REal{3gdnnhq0w3u5l%MD)L&I_LV)X1}1V3>K93+-(6&`C=?5n=NS3wFLbTT-c zJ`(TueGoPJh@o9#`XgLs5Cl<&HW<=?Z{w3f&Jo-Z^G5if8p*Zl0QSQzad-&##f-K3 zjTYxopgDm=S37&Nb$kr$0jh)4W+1VU)(N*4yvre9UC4yPL-!Fs%cho4eWN`f9G`(8 z8I~O}xfzU_&9Dex5FA+GC{J{7V9E2<48Jx)*u5t2LJ1IXGr^Ogg8r&rU+2| zbzXqhabjBnfzz(U_wQ&8=i;f?P&M4mCPg6=#kO`z%F zLV;2Z30#?@xATzx4nj4eN$})3eNI%4!gn0bz`z5fZ(U3v^NbDbIZ8ZMj|XP2!2BPP zaET(0&Q05H;>OuGN?5Xugd*5(K0haUY|Iv{Qt394$g<*xOeD~3aqM{2nPJWC6dcQOgeI3=b3VAILZnqB-DcAT9!}U??cI(g~ zJ{On_kQ8_+#5@D|Je6BF%mA*CCR~F<3knuXqg}wvp&JH`YdVQz1}eMR$#Rvh4?SIFT6ob$`&bc9NB4#icCi5;|ORJ z_&{=ReUNp~FM1!X%X^TxH8g&}Ge64C$uZxYSKKjnzd1kb>dRSu`^M7&QNUCb7=hD- zpo)?tkjAwm`atCsc>j$1fmVj~Rhj0z3dCPY(1i@dL!ObBj$H={>3CXwx+I$O{ljJT zvuc6QXa8{6q>@NJSg2q7o+gXV)_;j%GiC|pZLp?>z2g1CW`8NqTwK&a2tBWl~^o6@i_=Yh7p2VJ37R8wo~BxhUNHS|l- zJERUcczy8w&Y9QI+aReCE+0R#vu~~jKQGeeF8EfkZuroP|0X420CWe*IL-1#YOv_ zb!23~?4R2T8ttG5ZfU`?hmbhO(q%bd4dt|bj<#O<>~f>jH-tC^(vtVU-`m*T zfP4Z4I4q7#1bzCBxand-4jAh)&3FO?(I%E@IjFrBk?d_)*91KR3L7R|BdOCbDI%{u_(2T-xe4SDxNOU$J<{P^_Hkkube? zU&*Qb{xxdn{5L&5IA`*}ye1uys)X`-Y^8^e;Wwg>59Z}W@hzyt#!MZA)rWL!Sl>pS zWo%XS|Mgt?jA&6T9(RF?%g*5!=Q%2sZ{rJDdzJiOK4n}!A+pj$t5Oyj`;K#KNG$w4 zaSDVRbR9_?h;BHL;RI}QabQudfXz%tghA{VmmYv7NQjCzN%Ym5c)&XV!zwvEQu%P! z<7&;{9Xp$jV+iD&RvTd5Q$qNqjNK?&Rj=B$DsrIv^TFuZY(N~H)?DRDF~c{zn#QQ> zTE>1SzOu7BJGH*5>bR#fZvu7skgiR!Bj0Se@H4cS=R2R%#cbL(G9#;HQM46{+w%rW z!y2wZ?}0i17_=d}FORK5fJK=+ND9vD3?C$2z#Bw%(nLlE)IsX7gDP;R9 z>G(4qpH@(>-OF8%|8`0Se&>TQ?SVcNq^jtca0v2_3@q_+AtI7G zdnOZR6?}3vBSx%b(#)v3>^N^~xjbdS&czzS=~IpV+x8bWm|7oo}H_nj5Th zVedpkmzVW*AlKmH;7o3=AUh6>c9CS#8fFLLNgIF)$=NDlCrz+WdodH9li{bPZ!W?z zC)uu010r%F94{qbz;EZycMP=Yf-eNo&Xui{4O~vI9YjK?G?NH45DiD>a35e$FsZPy zOJlXDHr^G@6FCi!lqBI08nT-m=1yop020}uE>-3aEBRG2Cpar)~*AcVd zZ?5(K0>k^@Y^Bzp2?4qw|J zsRCT_Elm^DPl%!mpniZN5hJkmMN+EMO>2)~Qb`yS*_-3M=IHDdYVuj5D;CE*65iZ! z1yOK}*elIDC={d8z(Du7p6sZb_3!E{I||{TB5`RIDjA59<>!=xjmc*Vim!o5CBj%X zt0YAf9--*0>K|ox1!+|hM4HdFv>uH}+JZ+nK$hI;b&o5S&J5pXr<{FhSL%avZ z;+%LN{k_=sHkz^N2a{pza`TG;XZkpg{~);N4mzM1OGmvYTl4F_6n`Fk8xA&P!-Fs9_Wi&w0sa* zPRyC!f&htnMu3`O4?}dKUt_-@a)Z6w+UP({8AsLJZ-_H?LqL@c`C!iyMRbO8V`g9%0lm zshm>jMtdj}O0Fr(NEESZ1;(E#QhpV?-eI}uL@K16NYF3u|LpY~FL|y(J7{Sn&&rG| zF75Tcc6i1ebvjn7TTbvhwn|h*00HtGW(;9pL+alj)-7igp@arC8(XMDjwqSGAu*`lNG(D@4M>?Mp zXb>_cJUheJ3Iv&@=V`c*8p#yRiLfASPoVN!t*QgWLQx^0Jp}9QEJ~dnOH?6$(}6@w zC*LF;jHKImkpPaovr=whfsf2C6a1w@WZ`50Ttd*<$hu;8`%-X(tDRAc`7&)Iqr!R-%2wyf;S|=t!_dYFwFSFyv@qPrQw* zW6$F~A*SnoJ@Alasx5kDeH_VMM`-V3+|NKkSJ=G@5d|km<2L4`>f>hXARA zGP#_Q{E&8QxxL#(H-e1BGh%%^XxPT@&r&Lj1L*Jbdz|k+zlt)sGS#B(D6mh-H+g}F z>DUW90LRg!ZORxkGfoE$ga!JuVZ^>_iI{`0!B@#uIz7d~Z@7T{v~Mjisar_qQxDGjp|y0{By;&jmeeR)i_yG(wZNDU(OVpxQTk!CI5NISCmq=_Kj% zwMemC3dOSLJp`OK6@@G5b_ao)!Ak>#|H=-xz+k}jF6lj zA#g$n(+trBI9&a<4)iTe3N?!mIWpMStnjPdNrn;9h~I)fvr zG7|X;ZucMS1%B_ov$3a~Pmp+MK>68<_$zP$jsK4-%Q<_Th5o-T@=Vkpy6ymiyXQS5 zy`LR-2$tay1YXIZ0v}uRFB}$5i*!f7l=isbJpnq*hP4nyR@{W($Zau=h3f)ge^*3- zrzt6V@p94biR;60Ie z<-WUe6ogFh$G&`I+s7jF)e#K;-Bd`L(hTRV)wQ+x-fS`vUIdps^kU>d=<#iVp>y0L1AiKQMbGh+BaNRs zhTM)a+tO+Yw8~van6$oT_CQBH?sDS(m|er)iIa&d8S zmxUJNDb5|DWnb@Ivgw{eX5kqs?ZVm@N&M|GTn@=y zhRtFEpQXWyf%WK`5b|EJdsp^pDth^Lz<1&5Ks5b z_bZ4F05b>-#Dm;{&mS3c0mfz>zf9GabqXQ8QMe1B;Moz}F?jWQck>9IxWpFd!B;J% zsnU$Cw6rfPY_b&){-U07%}|c@Nz%O-maks3`Wrzp@x@;VN{nVXUBU}KpG@sV!(#$k zuy)oS&C~0f)6C6%AKk%)JLiw24t@XSL}P}-S+{pIu_W9S5xcQUx8jR6;YOL$w?>-c z_{PkW@JU`z-s_)&@%Y-)y*`C4Zv3MX8*i^VZDwO2@;KI02lxMIS zBzlIOrEI;~(l_FYEX~^n9UOn3w|OhZ5-mK#NFbl+o6i~^wM4MQk`d5 zkvLw_x_BuqB|EcG*1XMzm&KDgw%EQvO!ef4!F(^E3RUKR&FEKqew+&nLMXC47l8SE zpmcpR?R&;AzS@oH?(VvOg}p}!1?SK|j?905o{~}sgZ!p|;ZD2_>NL4WP!22({YB=g zL>6IQ5RhSg-F+&2=-{QYXMU;Wv2@F#sF>tY3fexaoc7hdG2F%+i9F>DfB&d*#OEZ(H0HfgDo~(Wcpr|41>XJnsH%Y6oIaYph*UQO~9K ziaw1*KksFR3tflcp@o7_KM|`1&=3-9w;JgK*Ipn_L_$JlW5Uxj>fcS|>hu4_Q!D=V zeAM@G@J;=*6VE^$l?hQS_*pxI2IFHH@=(E~_FhQzs;6b^774vatluYVdOgG~pV_V} zO1(C8qxnkz^&K%1%+_m`f7I8I+Vz-E@t{CrT-b`f4JRH_Fj9@fKG+IVy>Lee4Up>7 zudp3+@l+KP-8qy_={wur5ZO6@X47#_%U4{$d`XSHYx-sK35@q>FP=Y_KW)CNs)vei zp3x^nl-l`--WuafEkhnSpQPF<{zwwXZds2^mo!@~27Ms-iDxPrUMdM+S!OY}<#XVc z3NFP4ucIIA2A*wKYg}B(?Ebi^?jP7lL}K$xr0S6djn4yi&IzzIu|bzG^1U;Xi90Bk~S7Od@ONCrZ4#2PfP$;mFfN`7kzYgOm9GgZ@=hu%4LtR)`m zebmsRs#m+mkXEwR$NF#I{_s z>(#LJndp_jqa_0wrs0g!{@#9$+ipxz0qm6WebV#!_Ecx*4851>DiyHGZIh--I=cEi zGtFWskqMsfOKhyk$)Ax}ra(N@?R-&4ZY@CbI*CgVvI*Tk$fw1F(xAwceoKnyMy-3r z=Uh~1v=4n?kgv5!nL4PFi}K%XORlI(WUFV>^ndI>-Aj8jII7K;lSPuaiP2cLc%7W3 zNPJYNvmgG}**}6gv?GE$*6`>TzI_lb7Tr9+=IGEuG*0h{4#hx;on~9@22Koj^+<~J z&t$s>xzvEuDMs4Q_3OQ~{7M#;n2emw+FuVFJAX6eykC=XFTb2)EJ;?Za&lCc&#{p6 zomD8c!TtJO{~*Gp!P*{%<%H;@gf(KqQwrNmm9<+6)_x_T`Q7r)tN4;cuI8wm#EorM zbhRCogpgTBoVv2@^K4?>d~U({mbS>Us}3Op@1G5XNel$Dq?=&XG?t@zmgr{C!9ED= zg?K2mfyF*ejUop#Dqrpe9>|J5$Btu%HFJxb2b9di^dz?mb~xtS@Myro6(fUB!GWr4 zocm-?QtaXy+~Zttg{Jx5P48XEWTyUWXU+SAo)gkBkgQTq2k2*{>tDO{haSp;kt3g6c7VxA>rRX*!J>+p(Z}=p9@vZ z8-`iBk5azRRPj2GqyvLZAQE4XGmNw)t2W~2h$qmcBRJVDM0mMv0kwI^45HkeMs8|` z69u*t(U+`cjUbRTd7Xbq*A0@a)h2ke8K1 zEZ^Msh&|F{)`Nm0l#H^5F?dLn2-x@mDE7?3#bIa1_WxcMFk}k#4`fgP{XuSS*w5c( z4u>-@-jwi+r1L3@E^`AYR(iPsnoN!JV)p%!vvwrVQkL+e;7z{rlerKPkYFWXyLE-$ z@2s2AB1qhu6hb&3;}A!%;o8Z|!yngjYA z#4X_F0E2|3Q=rwnMpDqls0i^c)M(!=3I{J6W;|eX@K4SrQ7;(h=0 zQwY*whDP48pM!`%K3M~~A}Y^~MNCF0_tV)JIj}Y9E+E`XnnIUM`Z>&;B9=@Xg}Igj z<6{!H#Ds*xXMGXT=BZe^LG++l-E@e3m0AfbwxmAHs1&%3buTXq1ltvyUpA$WsUK*c z+=;?Nq;WUtEznuR0wKKK9gfq8m&gZU2tk|=V2w(HxDEA?58+TRSkW$P9;Vbk%pK~k z+zisl0sF%Tm_+u;Kgx&;t97I22^J49U2^q02-bgTg3mQa;S)|C^#92l!sDl<$)VTP zK}JLyuo{7+oEnwjet>4=Mf39~-{jnhLg0^MiFhKoBG$Vh639cAY$D2s@R|ag2MW|O zie7QKfltA;9$yaDp6sv(p!7zBxM@T>YP4C*FvtYmP5tMvScs=&v`>J1g=^HOf4XlQ82Rb`jbTU(s*10k<%oq1t_c^PB`uw(f}upBd{AKKeM49)wAj*Wxl={#*&xrfWG4Xi$fs#z)Xvx zfgO<*G1@T}BYC026zAE&;?*A(k~UnFX5orLDIz#w@lBXxx7G_)&ke~F4~!WHE;z@~ z^;t!us@N(LW3?vqESavjA97D~Ky$W(40Es+@|(t_Q?9{_W$DfjnM*S(nycX2Lh!_r zx|bL4u2xBQB@j+PmbADR;w@YzsVs-6&lj^tIoY$fun1!{QasXHFdJ$VlU1|rNB?Xl zgLw0;X^Af^_Vo6#Bk02mcR*^R(~ZO7!1&gpfbogh1_vncy8RT;%-(0m@p?wnh*_RKL^J~8EYXPNk{lP2%34xnxB>O|d43esn zUo5*QM+v%+Y>YPg+@2#i4C-E-tJg9YAMNFQ)AT(*r$Du3?lz5|8%Rg6NsH48Rz;-O zAMl2bG#%}YQh)20@E~lmKIgA~ zBMj6Dzr(ka(^Xw4k&NrhdXabB?(pYj8I6ix;EI#ED)WY~t2f4_vx-n|EIcST0bA;L znj-Qen}(qDv{koS5@w)Zp9a!V39N}m;W5$mr?LThk$!9C%7;RBm7!^n2nPq|h_cj+ z2iyQn?H{6|E6ECpjcV|vuupcZSF2N7>co|djdr?iIe*>axr7e(!)BlT9H}jGRiHYv z^m+uWP7soPu>t9)DYNv*>cltYTxzKQPLBsTS6xFN9jrl13eLJ&du)3>**%j(J1{vjMgq}G& ze=Avo0d{Qj>)~{>xKzsegAORzKW2L_V+!&}&E5fegOBPFBVf zaVnk8yO9x7cy)(BC%!TQ=yv^DFXDvguB8c%A-E=-KN*k2d1~_05|F`MWgHWgU%DgtOIp_x}C+9yO}3yvjUc`nO7_M+3Squ6#4Bu6pBI_8MhV ztWy0%aX0U)u9kW6^2%SDyW3Vq**jj3q1rgc#sia?aiExir{}3Zc06h;SaV?ZIvNFA zTV@R87)H$YMmT2popf1!GP4^ww2S$y(DXliQ-jN>3-)a!LE0*b`#D#ezOhUQ0TUb< zt8L8r?T>DIcZLC}!H3}1Sl$}6(A}54diE`6>`L+&o#>cOuvS3i&bbM$FAusK6LK3G zjcX1+N~YU!MH?~4%m^wRcY5V=Grm5?vrMhtDG4v4^>OmT^HKxz)HiWD+4M=!2!@fG z-1d;=gcV(Y5W!h6bYU#tZZua-9u`>ZQbr7M$mCK(3@+K5k=IoK1zvAUDWLT}tkF0( z36APdo4p?3OgN`IapCV1a>IbpCP{N9qqgV^M4PX;?(2;Gd-1Hsmi`7wuQxQBS?uWU zdI^c1lC)&B=Xg~C9P$C7lescc5il{9{I9@l@NSI zeBikgOl43sM2>zXn5+h3HkyAoy)rEsWTkjXrQZ9x7c&(ebl!}=K#i@VX^ z*48K4OQ+BtAndR?2~K~No%@f>gB&xapd_&W94BPd2rr?B2_Ui?-TOONcN?Enk!g}j zM!__L*iJ_zr|G}11Ooo+<-m|J@UwkVAb%FKZB0aG5E8{sfIn(h>=``Qxq)KdpFWFz zj{*4UULlhLUHQSUkeeE-sXr zH&B4D!ZQ{Rv3i}tiC`juUpBPM*cEVku2nHlL{SK%#G>CFX#k__Hl!aBi_le|T$5&a z0%!R-O>&q+63)!0+&%kCBQTH+xDwYYn#Zv~*BSIe{r$ z_z%RYgf%C9%KbtQS&A^x1@^x8_rA;Tx6B|l&_gj$7|xcG5D(-e=Rma0-=UHL0SIyo z77?~cTJ0E2`+*lvQ;R5gXM1(}kxi;StW2`i2gEa|nfQ!w=;SJrrqjIz`Kzmu$F?ch z-YNP0RW{wx@=~BV(fh0F{Njpz7_@Up0OsOmVe9mJ0K(8Ol9JkF<-q|gU98R;-UxTb z&najmg*Zf&J~P+gQd!{79P<(!gJRx8wjDMhQsmv%-}L~b&A@M-Ea7(y5)ZtA<3m)E zAB0Z#3;n&&q>jXOY-~aYy>;B065c|OS$IDo9%NQayi4TX__Z3_R2C?a7Y&gp`s9d< zuqmq{4o!+F=-1)2`|48F(V2iJA>D8Uat+m`aJ%H#gORXHQyM|)lW zAxT_J%*s1|(q3I!WpZ7oQ*-rQLoe`6o|_AP6i4?DrHa5}+W##Tz53|qH8Z$V>9s5vk@--XK?Am_-x5jfkeXO9 z!OUbf!l>#u>O1L7nM@t-7w!~sS@Ixe?8U>0)aC1i|9uic77jj185yT|ug|gk6`8_f z?TX`r*)$rCw-d`+S0QS{yw-=r{_UM3ury(Cd$-rJ3gRu|Qv0ykUBf}Q=o^jYK@wf%me_8d6rTi%*2>p@rw zWw=<+4?V1Tf3Wt?FwDt-}kdxN|Fg$a5ZSN%?ZvJ(%q>>>714 zhu9O^X!S&7fqD-s{rfAHmX<4zq>*0}-A+(X!4id4RYiznDmK93E+yLT27Xu1yW z|2rgg^zYe#TvMjwGd=v|d0OMn-z6z>tE(1x@%m3zL2hr-4c|`4a8WDk{$Y$=uvdRW zzdpiux*9{Ca5WF=(|^Fh>KFX*HdFp?yuI?F)*zI?nenyvl1U4SI>5wze$T}@1 zUp96o5DK=xwo6Q&__u793+r^tQkL?({i+6k^Oj1LF&1a@DBzXiyKGBh0n7b}@me-q z;-Z_BKq9c?<7p7>cKB45+xg(}{M%gpl7Eawd@e2#!LUB}z?0b4f!l%-Z zQQ{OJ9Jl?~5Gy~!WmxfD&EGVO*V`cIcg??!^ES2mN4?FbxOAA%$IgKYBa?LQeE5n2 zRvjdwR?}-pfp_pR=AH@EqhOJhiEJtU$3@rV5cTS4#0E$k)fgOB*`E3o=Diz<9d@-&op!mYv{QM|eh zhS(KN$Sgf{hPG3MD_O1AOSjf=#1fbO^Wefpo2=Thr#e8NqrJO*^@D5Y+bmYN z!D9mHQKbTb&|=I{<#1P3b5HzbSO%TQ&$FWnXMSHaK7`}?R7oHBcf*2B%y!C~mwCsr z$A>C|2vV_j4`Tv^DdYXO!Vc|L zsIorBaLcy=$)}QZQ@)85%#sw9^QxU~e?s3^q{?k`|JJ)drVj@umrT}1Vyreb&aT}_ zKheXF+JD&TJ$G_5nR!Gi#by6)uGf#-o5wFVMLZ(f)Qwv$zOYBn$s6!LSidabmCF!b zUJ-ovy9Ce9^UlluHRt!e4fmEhF9+4!;82C;3^D8eT!k~eea#sT-k!!W-r?gyOAZgq z`x58HDQS6URpjGh&4V2tW$tBke)Th|7ZZG<$-YUPe>rg^PnhPa`UWm6dYWtV3A@Yw zR@BfR>Z_W~3YNHuovuG^_m?(3-?wEoMc6LVF&^S4J=W^DeQU*upM8BZ=tupBK8>mK zD9NtPxk^p=k`)jM|+i<|8i~a_O(*q9;UN7{=Io);p2nYiSTrh$rq>{ ztFFz^8p~8+xNBs7^v_jz|C|2Z^~dI0BT|i(p$o0&Ull5^OT|1cYw=0TB7fI(ZZw!^ zQG3rrgx1F4T;bi>6#mDD=tC)pQ+;no*g$+*vc*vq@)we9a_|=zI^ONOy_LH8cs|S~B-1HlW0R*Oy#8u+Z~LD=m3tyQG)gMXDIdz; zUFC!2Q|}^tEa%>7?AoL{CKrgq$Hrc-l)l@3IynOQ`P^3DTYS7PxjjvX9$rrgq#L$K z2lqdY%gw3zIW?|8aZGlH>c~~N$@%&Dl2+|3_349!#7@sKFa2(tzKJ0=b$ewm(O+FT z%JzhTA2n7pZt5Ms*Tm7AJSU|*rZ^r758(+|(OV;AHds17!;eUld-vA;x3UV4y%j@v z7%xBAUuC^o<$T>4x4N0)xO;E=upEVG*X5|1h>w(z-~`c@HgOPq%=zznMgDFZRe317 z87PKe487l1nxn3^uP7SeTqbxoD{r>vI@YNtQ7i z))+37?vwJdC62L&MCO+&d8j3`Jxz^%Y%@}^bd`U* zHbN{U5|7tfHMt&w;g{6iD_dZL^#D7nbsQ9$R{z%H&|`szZ?C94{;3f1(DP08?v&FW zLp@to#%P${`!3_Jxz8493TW0VF5Ev?A-v&W{jL>lyA|$v1`?WFO9&}J{ZSTZ%=pC( z*Cbcza>YQL3xxB36-FZWXOTPNj*0&8dACAb`2legw_Oo*rkQ|OaB6x;gH{i&saTo+ zt1SzwfD~TB8c3hV>#nCC1UW*i3A zg%k!{`pb+L;>l)a|Cr{eKhHXLN269mkzK=OhcA|1jOPaS@s56twylPM|LSM|L}8y5 z^oKY?bV$O&FY(iG2@rjHt+EQNGyg7uyGPUiMLy%Az*KGOJV|XV?^GGx2MBH@PM~W& zBi3cK;t9F3BY*mwAKRw$T*0kwD{iZ!#44C5NUHs)!W$+#(<(JZwC?3I*l&_keLx+} ziX*YhzrDy+W736_53XsIR0za%Lon#zG(@_@{aan~X=>*rAy<08#o)&)hSA;yTSBih zGU}@}n}{foZV=2b4~t#Fr4s`#_i4+YVoW}+NK0F+a_onRPye2`S`&^7b>q1^ON=mT zhixh6G8$%a5cO$DzE4I>3a5fpjGJNX;hlLDx_}3Dn+87+*c9H$_IOh7IobeRUpA${P{vjFjmd2QkG+F~GrEetPvs ze0TRF&R3z4sNW4!)7#ajq_g$0qt?wG>OI_YSU2XDhrBQ=v|(gvB{adm?yn(%4!a)y zjUEaG5Tpi?Xwg94#b*WNG{V=OPiZ}GOQBLWt*{?2PtMggd zU->kNPm#vh7G8>hpY`Qw6`CS?*|?&eOOM^sdB}0A4biJ37;{g#rt}eO!57RexA?tZ zsE|kB3@`~1*}q*7#LI_mrb|oBx_w*8`qUF$IW*PU5dl`HBB10GVp5bt5)$3{U@N+I zL1ORuUV7$h#SgAV(ZhvrSbh_mn?jk8q2>Z&HyPK2Pp}n@m*2t~Q3? z61?If1w6>3e&73T#DJe;%j1P*M&cE8mi#6^-w{78i%4%zYzW5O>LC!Y%Q>uL*eCRV zCy(m$Tl2E)>{jWFdu@hayqT+{x@O08Po05T^EE-@!vrs%B%TA6u-meU8fYGjp!Dlh z6j47h+;6qSSz~$*#Zb}6TSw$}P;MJaiU^&2>)+@jw7y$M7rC*J;b{dW=IE1s@4+#k zOxBRX`Kd+08hTqv?TSrsf|Hx||Eou3FWSX8~@=rs#s@@|~Kyka#=hmnjs9ZA` zUU`^AbD$_{UcQ-d<}4DPpmQ$!{JMq)ie3z%XT!gVJ7^ zbf5Lyhu3IsU1~1!YjD}1c#u@ZV zOmGviUPkQ%zY4Z&tV6?37Akl_U}-is0dAe$b_W)w`C2v(d~xnuL6~@b?QcGH3)IUi zTm-NOFQ?KgV}z^~vB}I7E`NNi(#&&RHLEiDgH^GosYB%~AB;gra@)lYX@y-cWKP4s zg-_K%qpeYKCt{7GGvq=SV-pwZ-#b}c)dw)<(ndCfUzMh^qMH}ltSv4{{nk>!WUmvC zw9=*A#?uoOzase12AJ$>4%!k9Rz4iIUdqxU@04lqh~A{kk1BzPgpiRpMuF6>dC93h zVucpZ^9-I+yPjEO)Kd2$)N^B2IhXIJam$o*{2YxBnl@bF>z!+m=ls06ly?ZXTI4=- z&oK4n0RANhDt-n*=Kgpi!rFh2P&1iOJ`;Q~$dye{;^?SNg*@M)V9au(PqSS+o$MJU zsoHhI_#;fgn1F?HG%;g@NB?Zx-u3tI{oDlUn!`4wNV+ zSIN&}T{x}EP$IFoq4NESzMT`Umc?tV?$1X-D6^{1D8*G`tkXO%R;odn-r{Sw(LgBDHcO|qFN$t>joTy$Iwx@?N1YTQY_^d zQJ=LZKD{n>nR3<$@ot8UcLKN@yOK>2;{7bZz#Odg>R5XPKlu%vO^#;QyxY{<9#zlp z9vjk6V~d%5ol);*hqtv=ZwZkx=} z=eip=)fjvBJKb5K?PTJ2D(U|u+gi_G!4%R#*pQ|#znwmAjP75ee06CMBjHFlmdt95 zBZZB^vACPut3)rf=q&+;ElJ@R(X`;(IaHFfFqGuHsmU5kvVX=XpRDua_*S(5s?eiK z^z5Z^Ptztl-k8at3=99;_9*?>peG6MgX|~ck5-?kGiyBZa=#kIg=iUZ*EN zuZ;}~q4{C6bJBZWKN>C>oi=PHgdT=SZva$q0E`9DJa&*%_r?c2TyS3v=HjM zJ=^YsrTueKRm7WUXEmmG-$Yc|nZL?&Zn~`Ka!sGAT@4#MLU_UFE)dkiOrKBCEZT@R zYzX3+2wys!u=z!%zOS0?LH|G;Civme2s`$htY7yQEOWH&v7jCbg<03n-PUgH?=W}p zLI9^SX%~k4jl6B;#CckmL5~q~P(LYbJ=`paDj4m1i0vBoR>=AO2mR$@*yCtZkz36P zGxUTJZ~AX#g_}Dps>d*IaX9Cc^3u{1b$s5}l;Z}6$Z+&Slr@xF0?I_3eDjs!>%K?I zI-3}|g9uG-f!Q8A-m-Goe+$+PHl9H>9Q^?w8hOGkEc#YY9j5B6%ogK^iNhY3D^YioISSvN#iPg8Z|I* zh?8Li=@tHafc;mrA8S$a_Hv?neU*=JRNjc|JzW?LDot*VBCGqOR;Hl@zY2%KY{Qzd zrhk7}+%;xdMFJi6YV$}+mf!0u-p-M9qInGVy{D$C9cZz^PtHaY%(Y$IWE+rL%$Kk@ zSgIn(>Z1uG^c2X!;E%7Zv0M!`l*o2NJFB0Ne&omMt3gTe=E~mnZOF4P_IG?15K(ZQ zc7UhQP|=G2Fp+!8Ayln*^+(^{F{K-HM2ptERc~@$5-Cre?%eauzJ&hyHwLmixYa66 z8f#V82K|-dDh*FR+y(8^d333}JUpnoPV!Gw5nUYyh^n8H=&Z!LwP27T;%hgTo6p;R zCk}pwUmk@>d<0#!%Hs&(A6zSTg+Z!V!D^o`0bFWN~lFKvN*T13m78TXtMOao@9U&JebVGkg_~LQHj6Q9AWkJ4Fp0ACz!)KATs_Oo| z7q`8P`N}aOGQu7gO}-wd{6_P`>l>H+1?Ma`Sxlt zMujuY58GCSpDx5_lD@swF?~h;`hC%FX_dE=d2O+cYCaE=`-T*T#pmNZOV(Us!^X3C zZ1Ef@t+Hy}cPbv6|EIk(52xyT|Nc>g%2*Uh#&isckU2x;IObyr5t%YnB8p5EGG#vI zSqEjFDO1Kmk~u0QV;Ks`?04^@&+z^JuIIX*|DJ!I{`p+j$Ju+Ywf0(T-|HS;?*Pqo z1xo`CnWLe2ElF+CzAX|fK7&)^*sI-XS<7M9Bp1{!7~l4h_1;(-JNruGaIR4a3N6JLMd_vci`o<}^Vhh#jw=_dm3<~UaUT<}p zZ1!GIb0m!CFfAtSKO3MyNiEB!9Esk??HBQGAD$NOWYg1hO%LF%qr?O|D6UQ^?=C!O zFi=Y!X;&O3%k?e0)Z#GWp925SMg`pEd|UyQNUzRQv>3ONm&ZAu$qWl9hAs65g*;;R zADgSYsIlk61LwO6t~_0^n0txMcU71VV{0U9zzWr5oRz>S=u5T!=G(NGJMzohv^NaQ z4kyt3QCz}xSHBSVu?3OeKQ~HTRj#Jey&HVsJRGon>r+OVbr(;te)roMXQWgKk*+@+ zdE9G5xMJgkC(ca1n?Nwja3A2nGDMQ893+o9RdBdDcRY9N1mZ~_{Z0u9nn(mbZ~{uk zUMW(7yi^2s&1lV%8vS!7D|BtfpH4ZF-t4RBVvS;;39m+jpbdBGO(g36bomzpqKe3h zWdoyAXImD%qTHVEgb7gnI6peYaTlTTMYke8#hDJVvEP?ttgm2!o22Ll!iRjx8vJn4 zjas)Wo<)jSeM67O8>_!JFEoW%U@5hPakUAUpBTT9FXTpEa21h`SOw^kC3iCj<&l|w z=d2rM=RWBn>Yef^Ho^iO(525!7>by?nJ#yrSl}(=QaaR{8&vrj9#0v|LwOmQ5AYvZD@!2eEIsNMqD=x=TI5b z*)icE{=@LlYqRl!b<`ZqVjerJ=X)I#fvLOc{Ao+xp6E)San_^vf*fij54yJU(dS!J z2r`zW?Q3dt&AQVX86A4v;>8<}Qv8W=mvlDZ5X#P!kMNy8jJd|WF3YOl9?J7R9}ikG z-JV^;Ki!EXnTtzV^8g@+?#1t%IE&W1kyX6X=l%C_V<{cpk9p=jBnB#mZKt;D zyf;ygY(>!AP&E7-NtfFA1qi`KP2}lhiodJ}Ej~pY(lxx7?A96aLDs%N9bz3#MK$v@ z-4lKkXw&osIScI&~hFFNbybzj6g`h#dG6Q;-t#m<_pOtC6MkAL0ksB|%_l`ldALCdF1oh3;idC;ms3bB z!hVUh9Jc~x78XsX?9GPk&Sk^Bzsu9+6~BM>+G(ZH!f6HLT~<9pH*#j?D~8%>awiXY zc*{-;cqU9skttkqBdV^>@bMw~3o>fk#2R-`$dC*JJG;G}A#UPV0v<2*_(B0o7x+izD zK>y=)C{LXf4Q`K|4k-0J3EYAV5f@fDEQcjX&dpYCEomt5RPFteE`(+hv8J6G?yUY5BxZ^Y6lNbwvy#B7isG7bWT2`8Y zW&sPHjTd1%HE##-B#*kHpgaipD&2JYpjy_s_hLE|0=1{Bd-muDl6|CE47)BZ!|c5l zfa$a$Eg=J8?<`Itlt@~WD?S`PmziB5NOd{H2`18ZACO%0xj5O3}lklk{;wwl?sL&unuhC@xoDET?^gB%a zbt*GHa0`-LUK87*`1oKNr`!&5=b@Fw>GAIX37?z({BQGdCJY2rn6HzCI383`@ zUQst4^1$8u&w!ZMTr*N9j!*$m{(-Pb;fgxj+Qx|Eg2=fXmrVYhUR=Bg0CIwtD=QZb z6{Z!ua|X(gW=;jj33!OJqGZ%JL%&j;0M!`28&qn*8bB#zfSO2dn4iDw#w`kr@~I2&S1wt-9XS{gIFvzv&~xe5 zvh!D0ZrIt_bQ*3t4qP9h0_{Ox;IQ4rDNMue+GKsK~m$v8HyUluA9UrB)uo+y*4B=k$M$9SZZ^KPo0-ns5s^D6xr0${$b9F3lJt_JByjO7yvR9F($2R(3fl_vOf}&%B!$-P zY$p&2I}Zwxa!$TEl6rUgu!!5ww}wX{t?S3c17XeQc%)26%ngx_DZn5NP@6k{vug_& zA2CBG%u=VMUf=_O7YxA1Tyem7J@h*aZsP}-XO<20(nhQ}dXlnbO0&DJ8EKbxnR*EY zFTX)m53*NruzyHOoJL*`oo4%Cad*npqdGiDT2MZMp<99iI#mdgQSz25wU8w~mHI z_=l?z_I2o<{1!p4_Y8l&DGHg0juvxs(^c1C55IKR{s%C)*AB=jKf@rpU$+tRVLz{z zKOY1i<;FovyvR&rR4I=fqb%Ywax-LbA}RpTZ9qB8P}6Ouy^Os6XFqhlO#Ubb(+ z&ASg5K261-tZZn=d^4N(aiuVvS{y!nDtG(*+m_$BY?N1T(XC)Xz)ksVKuTShay7ns z##6hzBU}AC_$@!qm>3wj1*DuWOJhW)W#p=iPAl;#T+|Wt^#axcMaNDuuQiuN+;SXg z>NI%+kQ0l9ym=90N%upWV?)@zpKXr|J!M2A1pqcueZs^J2Q+eW@BOs56n}6pmB6La zFtW1nX^ZD7Sc{JZ-A+MEjFQxs^XjmS1Gw}^GIRhRA_NWzx*8v!N3-X{dcd%-q9w*; z8x-SBVRgk8OrTlAx7Q;K0xP=Z~0xmv9~vCqW0CgBKlD`^c7g3=#flB0+0`L>?^<8~4vi z!1iSd=nVY1-#?e|fmV(kL?r-4P z86zcdsUuo~V!4q12TN6o<~gL$U|9Zn77aq!hcwvtk3l1T36ZAzj~f}3&Q9gdE{K49&BR)3xp7WV; zzGs#4ShF|L0#^w32}FN+SUl3}a7`em*m|^^&GJMmTlO(k&Gu7~#dY76$67p7#vZ=% z%Kdiqt-|;$LckSDbinbzhzQWB?ruSJ^rQ>Gjm_r3RlCum#nr#vT5#4#tjCai=StHx zR^D4jK;-$`OM5)-m2FnjspNwg!+i4aS!^M5%35 zCOcC)5y46r86yC=+cGC~)w4y*C}urLw9HW7euPm=a%j4GJEjt!rma^mws7&&JT^i6 zU|6(WOtj2%wtImWd;JuxekQRt9gsXrNjbCrPH(xDO7A0>s1heT*pFW3^A z%Gwj9c*A=O8oleR`=K2n>k+%G(vmXcIm$ru2YO*r`oyF3>XSAes(V z+KV<-iLF)Tw{h_?H5lGK&PUD^?|q{g)hKpOswT2B3LwLa^kCxh-dHo} zR?i>=w+Y0?UQc^8p0-lfya%JK>H?xk)@K3rJ>>3wT*M>p0_ zC5YyXZs8wJV|~x$mu&qwZh>&~Up2QrX6*-!Hy71x`aR`%eU<&iuNe)+u5@S%TgyBy z8qRrM_~cud(8xT&*!Ew@{)YBM@{>I5u z_L^WJxFnKNc_Z>vcV6m~si)*PdJmFAM}EkLf=1i003&p;KefAH$nfalW`d*z7jTsgSMY%W!6P+&_6e z!_6dD2G5|$%~cl?THL!?N;Qld)0GMc<}+(2FF3}ly5*xW{p<(U!fX{h0Fjir^_O{hg4QZvJtPQbq8@&tW4-bLkcd$h_6IhTG= zoIN?e*O>FPX|+I8n5S3ToaX4woTm?j4rSZiLJMH>ZcTCOe^q0%ErzLXEAd-RQX zoyT+zwGEj;SOJSrvEQ|WbZCp>Zi|cv7g+EyeQDa6MQ^SawR2sq{hodH)5pbA)0UKD zE;un((F~LPj0mNgQ-ameSC^SjjRx73d?;%;rZQcTdBao*>gxIwYwg34hJ_Gc{XiIc z=1E$kv9Ii2ONp{0XJT#!>euRNoR6C6) zu5tV)C`2hGfsKs@3U`pNS5XZ8%ysLb!#U#A;S|rElggf5zBl6XqnG6|e~J`cQ?lo} zyv@6`1C{LUEOG`EMHRm2M&wV)(LmAQ3P0Gljd>Nr*c4gQ zSE%FxM&gDjz@2Mb@PfU$rbKg=?!;*|5e~Z4cVVaNZCK}0<6o)BPv2J?SYSp&Wuw8J zgnTG$)VCvatZb#VUkm1bG%C{+#jZ;^qmq$59^1%X`?kD#^6D{#lS~ZlO(QJ|wpTj} zKW6W5Z;}qpl*3jBgyNf0A4miiS2Ot7sHt%+?`dIHv9_zeg^4z; zB!gCZ=^->=q@!7>SY0R;SkkO=!!{(2$}DQq7<0@%Z!HgfwSdYXXqMW+!Dg6FU;?~P z$zb+--dViQlAkIhzT%as(JL{jDtLh)VaCF}AM`jpU!tbsag=YS_IP z*xvk;a3eEP=jm4LvY`vvTbtv2(FO5V$7tO71+GoMmy)TuNAA7wmV=?4P)04PJ93LN zzT%oonlZD>_aViEFS@1Q46mrd3w+ygZbbEj3}|2`4}hr_CO&p1lkmx@PTW`*YkT(# zbv6U=Hqi=I2fv=w(4D(GU&Oe`Btb>gpI@23M3-@)da<~Qw`ow|z>QgFD|(K;JUUT2 z${p5q^ZvwFJrb4JBLPo!R}=GQujh~(hI;vYEfs8QrLEfZ;OWQTY}G7jJl}96ct1vY zne43=KB=E4sUrkia9ve+iP8VZfml1m5ckRmKl2GjSw+5^iuo_B%uM<}XnT*#-8JXQ ze8FMX72$z?lrLZfry6tcOm!jVRZMdSf?HHe;VqND$)GmwWaaSZsPVD5N^kb8j%08vlYg{bj+g32Ay|iU}yG{t(-d{A}Q~fYeS| z9{iqqT?aOP}J9-O`;$OdAI2FcA$@%f* zfv|77<`?~vGcH74Y)=-vW((OB)GC)qK`KJwp zApW}1ep&fqMuH(Wr*k zX5;%a34%$@L%zIxg_dGN-;)-Q)fCTf7_qP(7oB+FumGf!thP!k)Tl^ zAvle`A|a{=hlxiF;BePiZ=E@KE(&|%J3smjt(WHH&-{CIFN=@)c#`3qk+Z|af_n+y ze|c*gW_5NNWQqDInAha*(7&r_y8Z4-abcBHIlJ{&ztxftVY=gMRM@(E-}5C1`!8Wk zcy}hgfwb_kuqyOEU2;RLnP#a`MvG4Unf@`h*LjXZi_O<-td6)bgkE42STr8)(&qaq z2<d=`o>Fjemdg)%-$lN;(6v&$y)jJ)GD47qJEvGdjvnsFRg4o zgO8O-X{>TDK2avqX+)H(+&YTc;|VXtZYxw1#l!nQ0Z;}in3uxdA^S==7rH2@ZbN#z zo=)nG-D3xqqRFENuhz+Dc&;0QNZ51z7pt4E7{U$Px6Yx(^Xbi74kv8Bw#R%}nttkI zs%43$G@+lF(Z14d8^xbq`yo7f!|^fiMzvFmjsPv0%3e|n-mGZ>UR zp>$(Ru$dGS&}g<&TXPmXuAO*ml-=C#sR-WNbreF6Y+TE4)0#BZtMlO*_qm!YEU0%Ph7|37J&(4A-GTs>OC%g?wvN;8{@%^q)Q$v(|hjE97{uZVfWE|aWt&qH7=zx)KZn4l&yLqp${#N@4 z>B-yTpXYCO>&q<+U3?VG6Q-fTN~#psQC_kV?$1CpW!^Rj*S*wBe8;6pRqt7=5pnqx zU8);8pw_zRMSFFw<@GdcmSJ;qX9MWpu;!f28GpYO@WoBC9|uot0A~ws?C`OcfG5op z!`T^+`&IABnK>CNrvKznMv!RLO?}TPvbvPBe6OCWZkjf7`SP#vRu3AS z(z}1*d>ymGeYc9W;$uqQ@2u!Z)}tamFywW=7*iHI_~xvAVf<1_3D1OuW-X|JKTEys zS9;5sQ2+`(A=;Ex79OLc*DrcKo%pS#gYvx3E3*D|{DU+-%(k9-dJ;E4>rpZwpZ#6- zi#IJBs}dl1bt!NecT31P-8kb;57lGp!FFlqh#V)alLBGFrLW#qB$d70=x{u#R>jn9 zME!cUWq63N2J9gnN*9ooxRv}5RB{1 z5mashEBELg5rPWMsh!i!=BVF66eOCTFTiB_xPI^wDG>nHf+_bm17 zF{#z25{e5>xYHEz{j6YC@8pReTKU{9w_LD=C*R+Nm3#({rQ9A`K1dF_pA#@AxTxop z16CF9xV$WCDu{R{a^01R-#=vk(uKinqjKekvc1OFm=krUA~zMtAuwBN1T6rj2M-cN z{QhFD{G$s6%v_GHbUZCXf4~cp`~d(62294&g@+ftwgoM)-+5wfF%$QLF+WC2C#K@> zHWyjT~ z7Nd1X3Pf?2eH5yP|F-wiuAn9E&gug+Zl;EbiAkE&{ogKMfVr%#jYYBs00S_^?|~rl z+W+{HePlBT)z#HtBQ9U!osuTq{I{J5_482Cf1)O++Gp0s@4<^aF|Z(msO_ zITBcK?cF*Ru^7pmaiRPGrcD(Pc2J3&96j6nyax&a+7wgu5Zly+K&m`r6UgS?xuX|@ zXM29i<|-C+0eC!Ud?5VzlNND(V_3MjpU>-X`DTd;mZb?g*4+7)Q5_bsu90m>xKMs% z2qR!sX6r~=Qcq9M;JNkE9(6ZJRMSWluQxeBf>F+e zB8y1L3=4pe&o*_$3c*7lMvFo{be$_3IJj@`#?=a&9t1+N=?g)Sru`?mAzU!wntISS zCITWGe4?UJjMKw48wfP!un$-hC|W!qPi}(33gW`vxEsnMu?R<(&9YJOS(8DP1c*>1 zujGl^_5Yxy06YmQl^SqjNT{Jia1{W(`crwNVG;UptUFl-*_laFP`rMjU>NL8Y0?Jp zilLq*z;nMtCXEE-SPu?VgBgk_S{AIqcb_4Q-Vzst4-0y)GR|$m^XCc*h6R#B$m&JN zlgfIlc31zj1n5!F4F8F%w9`#LX(<4lfMS!LZu4U&ln;vj^hYuxqy0mqaqszL{J}C0~gQ~6mJF@QXzB=oIOh~ zMQl1LE?w0ER}5gzqRImeV0r12UVy_MiuM>J1Kfj%L&1jhupk)TzUdeSL3 z2)dVCmKQ-9E%X(5pAr%h$xNmxE@J_ZloDCTxF?LBY4;M4`X=c2lhxan75{aC6|f&> zZRH@u>j`IM;tYBN74bi$HTSV1NgB^D7qD*0z#g~F>`#r1ev8Y#m&wlk2D71EPnG-@ z-?MziwgR8#I_Ve`FIbR=O>H5G2_|G1HdbOL%T^c2@??{&*|AB2_uwW9_0rFnYGwZO zcP5k@6w$Je?1CqN;TEtJ85|#Zx#$^E{CdQhNG(FPCwz8m5KN{}*?m(E6k&%t3C#aZ3 zTqfhL>l^L!O4xRR47$O3%AoD74bWHgxr3Zw(c)j9oP~2+OOT*H+0pZ4C=4xNK-Zvn zxSqEc!e8#-xZd=og#;E?0|TZ`>u@B_%y~!>vKlr!*owv{CW^F)`6{&jfX6_|Da1+L zy9Mx>529OXeyr;NNnJfq?;x23%DE8o7u~8?jQ7qBIvz<2GA{?bcrg4xDouI>+=rjE z{>PCUqz~44L7rC4U{noIUY)5Q{3_)jMym)CZz?{>K7NCwCP>NedH(E~^+5jd<2#^b zIQy|8R7rjo($-KBjX+33M9TDld;#aB zC!=oqeA89RXCOsj{zDPxao10KS?pm&p)!7?>}>iOuLmHRS2FIRN8?$RUwUrcuXU@J zA5nV*`pnaS|LCBWohJzaa|rd|k+|!kg{P8Titrw&{wNP&$x*rrx$W|(6;3J&haBPn1{!Pfg898_U{k_ODtS+M@ zci;YeQ*(iM7yk!`B6|Z`_SGO%qJ>yN=< zSk}6mxUoC~dmxerh1&0%{biy(MkDqN`=?vSZ;J4`{2qNzH>yQTE*+Z93j9Kvhe#qL z>KQ7l=LO)mfeHE>30*;#X5p&v8=~>p?5XVu~^9v5;;oq1Y{lQ zXoMaW;?Nh=@NS`+7p(HicFWH0$=EAsS#`G7v3i#KCYPsVWD~?q?lv=AF67LunPPZ$ z80o`jKpLUHXHgs?O*@ifl5HbEp)P+rM?xUxE+NRo=#WjSR9vVOq?lGDh}du(H%Y6k zj8&3HGT2Z$oeA^j3}gldKfmaR7E?PSB6EFI;umwD)c40Fr_zU=UMx9g3nvKLpANEP z11VTYR&I16Kx7PDp-K<;Agfjj;gqj-Foj7WEg{WNg2lyl!9o$)GhN0tzjz zTUiNn!j?e{_?V#DAUufDl&L3N*lWG0K5C)d3P;DG^os|qxPVf)4JGHR2jq}^EL65q zBQ9MTbU$`}d+nS{TQ9;Q$l$T&AjnyP3>NqS)V!Sr!4rU;UjtNETsT0|pu9J!4uvRi zp;GUIA<1Ee4ls35lb5fz$x3Q@f|P$i%IrBWl2i1^$o-PT_FlBW;q>(S0p9hVd=T~{ zdmvh>D&cUrt8R0m+j}(6AhYfpr~m~~5PX;@XD<48(&|^ucP$Gbf7Nj-17(B%3x_#8rnEkE;fxkqE^$OC5LfYCrY!!D%j_fbmo_2rpOQ4?A4 z9sq`T$(QsC7^M{-=Qc%C>5(jwinPGO{?i*x0)df}UEjP_ zYN5IZlvY{keZUGp6dq^is^TMmcKq+w2mo$VLG2U(jVIxJw3#G#=hZ+EfR zG@HHyO#a>fe{_t6!)Zd6jtWRg-yn~#3&|{e@U1Hh{~@@#A?m2@-NGO2cTUn+#?usq z8N!!x>aI-8f?A$SqIrW4We;~b1`k9tXOfQH_NtZPS&Epnn|UQXv?M~$@V2T5tMR5p zc|Rf8T^`0bTIHPm!Xii9zVD0ldwk^;x_gMU@o+-tJN!{?3pxgbJUaB{yapu-Nz{}3 zWqnH(!iLx%bI%!0F!$%hId`uR{o;){-VY;4cE(?SMd+;8 zzFqFo#})B&R_RMEMT3at8RB$%dJ%BAVpB{m=E^qm12@{~SO*~2k@n9caaieK?o3Np z=A&cZKpBXXxOK@eN6}ZFfAYowl)6S!@1jWyY3 zsL?DQ&USx#C7>K|M=`|Zb#R*EeOv6%EN)ITlJ=h~voi?Y5uep*hb+e#qwKmd5ZL{z zIa2R+6X2}7I{?Uf`}-=`9pEVYr<=Q{Wf$Zp1UAc1! z$dQsO!XA}OL5y2)d(w>GLE;v(Megb*{|;QRs|y}effL~FPzHG8D`(eiY}Rl!3;zmH zXlc@)PCs9hEMaw2anGzCPTE-kZRT%ir zQaxU75KjKpXp<0cT-d0gYJK+Pq~WF*C-XlKBgP;Ok(yI>I$)nTgww}#l=j!}pNN8!}lF^aX!lzr+Lrrt*TVh9y}Sv ztSs-|TYrreE4ZX95x6HVnx#mdyO742K{(o#N#pQ8jRyBw2c_&m(L0tD{M^Mq3!~%Z zSq|>H@{5`k&LAoclT7O03e{C(8d&uOpR|iLvmJE_Ri!$8h1GW=WxnuWfz5HWG5(fy z%$2jRnqVm^ro_KyFeg`5Np>&u#N=II%9N4*UN75cr`z@SdeB%j#DbtJ)#ERjxaW0^ z2DmWxa^3NtTE~5IdSiKywivSL-ntjn6Pne6jo%8ar+bWlu)8jK3~(GP zSriie=F%$u=7OZ4&D6l}9d{x4&a67qxyCSyE(s> zhKTBfI$GkHC#VD-{$R=)Bm3wvta}&^(|;B-moj$NEK;Ei>h(fuejNOIJBEQkX(Ren zQQyNP2POkU%|+QZALpi3d)#d!?v^KLG1n^WU!HjlwT#lA9;(h2W&c{Ct>N@D<=0L$ zhS9#hx_aeeRO3}0nPq|=`kJ@lCgB3}d*;V@@weI><$i1_%}=t|SH6Pd{jYvX;2MfO zKQk?6hVc5`yrt)-9@oO*)_ht!Nr1a3oZhc_)?@0n3(SFT;^w5`2=}A?c78?Sy;yJf_++zb zCfGiIHa|0dZuO#0-#7Y3^EEHQll{9HJlWcJ5Ks2AO!anjM5bzvDX}N|o2RQ3vYc~G zDc&n_%51dQ#}B;ud&Vf);W&czC~xe0{y0{3`JIGW>#w*)-EZpqSMj6t=I5ljk2V4I z0r^mCV4K9Ae3^kQ@bh!Rl>M(SMAtLCzs?Qb($$f&E4ttr{#i>WVarn9 zd@^C@vt+(wyIGLabuE5yS^oLpBF6RKJ3d&q;79y3CwBd>f7-Gewcou4vA6$UULuWK zOx%#*n&A1nsnWPdgmh9g$u`l5@FO&him^TC0}ebSW}@KVsXg!t;NM{=fk6Hnh**>G z?;>L4p->MvkkyPr$XeDNRzc|WPnM~@jd61EDVb-6`qz3SG4@*%71PbQ5i^Kwf9$`U= zK_E-QNC@yoUk-r-0;OX!GlD>0$#0{8NAH5W@qkaR|MQ8^@Kl!68Yg)TIS5A{fny)4 zEW^b4j`MV@FUS+5@DO;yN{^B~AyF5mQk0>lWM-6OS?z8;^RmDhTV5+CL0OssOU2zq z{s2LVmBYL72JxBaZrny$+Wi0Iv8u32;DiFXZyO}o=0joBc(Kq+__5>IwZC_(TFbb^ z;ioXGCM-1&)zKi4`n$4d>fq-y7QpF3|_&lT=k3|6xBb=P$aB==IPamW!Q>?Nt z@}$#vqx>zpvMluKAMlPN3(wzEGVSqs1qoD=68Pi%{P_tz*xGv+`K=q!9?gC2y%tzP zOxU?W&|0Jn6XGn^(vb0Y%$2m$Igk!*-o6v407uhMhMy{u8^RgKmCqv~_Z}$91S@|h zM;X{K5eFPW>T74owiXWzZ~eWA$R1`{58XWDsrsF=wTLTrJfYgH)p;|qE*UhXo&Gy1 z|8QZB=xYBMSEnX23u$d9PeTGmV2{yuIuEjg{iKc0clWoamtQN_X>JR5*cG)$(7Z9l zR)LN8R0^%LEJAaYFRu7{M22dk_nXw*7d+Z1;9&sKUrLJ=I=5a=oOaN}5y@W@Rp8?0 zT)H->f(U^!A+dz4!n?N*B*VuWM7hHKMn%KJU~RV^9?1RuR&NG6M|G3WE7ERzeLAOs zcvOv@OnQ?S$rbZsBx5hTWeqq3mntcWW@+VFc0(nhhiP{`7*hp$r8Hap{@JJIp0vKg z-!xx$i{b$Ge+Og#kr2VFrT#BGe(Q_aeiKG1>CD0xdb>|GtGsaeN%3+(R{;Qx(TLGs7`1&yOm$bkQ3t&%OOi-gyvRX}t~*)1i>D`FiLVb*H_l zcoGI>|NZ(mx6h3Vy5n73r>5Endwz{CHg(Q#F9q4clvv&SLpHOi+0GZ&>X|HoEaZVi6F0yzA576^f18#jK0I`r zjbwix|F_~4`?`t!dDVwzm$;k}9BsL}8E3IVG#QU$1WRvE=w2AaynGsZky0mZSu!$5 z*PTtO8ahpBm9Z9&Y$Yo2a~d+#ESe?Low%wb%YEBm5rRG1aV@AW7IJWR>>_J;bZHX);Nh0al6ZXwaJPm9w}W$G1SQ zMDF#lGzUxNeqJulm6*S=$Tt3BGUFPELaz|gGQFIAac&pxQ>yt6Z;6YBSR!n9!nSKP z>L!0d<6xA4tJ6Y-TvCwag+4QbJ-baZ#|PJ2`M0mw`y7cjjz~CgU$tBpMJ`G^O1%ju zX`8?IzDmR&ZQk&WCjNt~@K{Wh7B348U9Pb7`{S7FcEZuyDfYHQ0(7Ee3;ohlj9?Za zUi&9(vW5fgVn_t#SSW-B-g} z%HL(x=o(`#awtty<7-l8+hF8XJ|cUB!h{U8Qxu;N8_YUc)4he<_{@y4H=xA%&|XE@ z;Jr9wGbRU*eb^^cF`+b8zj`B8tyNl+bK9I$qew{Z7jz`}b)JinSQx_#3X2ulB8o#& zBqzk_>23^{c@*cGPzc=>wP?k2k-WH{h-sOw{>CI-i@?|Zc7x*C##5|ulN5&x?kuvD z>D&gw{Cy7gttL_@92L{2hujRPh5EaMdzt@_m4npOOEVS(1EmY`E(zS-P2_8liIy$6HsN@irFcTzx<(F@V(c4hTh%lfCsaQ@fB(7ii z8@*TE%N(eWR=L0O0J?^N31Y&bpeQzSTEiR z6~<+yh*dEQ$IO)!Xwed`fvP69k?WX)g179DC4;@jKv; z@=0#8^Nw1*aqV_=Br=E=%1U=+e(uF@OnPUvd|;6izIlSnV46A6E!S|=V%XLX z@|(jE!~{RZm~NvDS0ksW&RgHuiZl;bbWg1~wjYf@XcK9X{-R0g6xD6jG{`6rP}?CO zBL(Xc?UxUqGIM;8ku$O85nB{7z_O9ZvieUO9EUH9e{-x-gfl=KcY-CgAc?u3 zy9$JM)%*yZPYAC9d(jh?yzuj9@ncrRoKm@3_GanYHj z;H17f%VlpM+_1^@bXq;J>#dSB=}SmwzIzSz*QGyh7Cg+eUtWM7t}-m)2DD%q#b`Uh z@_`)puXsJifO_qtZRw`7yYKU1WO;pfEfMmR)D(Nx_G!n!8)thL@Oa^n!!?q>7a|c< zl)1_Q9wM@^t{yAG?)-n|v6{;H579yc=PQXPzJVl{2>@Y){W24zLLHGUko3*DTa+PR zcjH-O5`VbdRqjjbGR&Y-Az(d(pz|-G)b$S350}7pm203B zpgakN{H!qV8)@QfX=#Bdf^-}JcBsE5Fe&&1Q2bk6}#LBKij?pBPw6r#Ro^bg?Ub*#!qQ1C(dM+^qn6KDp(=H76hmKvEz> zKYxGUTw(L&&#u9Isds7Z{qF>I36~%mI^-yKl36 z=;-K(A_C z+qF9tm^ba}g)9h3qW9-uoR&rX4s4S}!y}8E;@D)O=6Hg7?iE{ucR40RSLb*d{8Zf+ ziIax#62WlP_H`X|Kw8E1%dI^t0X-#w=@5h}lcjy*PPf`cd#UXN)CNtANWtZO&-i*X zVkmDlqo;>{{dfRRhY$FSO*v_?GOh=&m0dfnj8jm6i6@A?QpiW`Opcg7@dh|!`p%zL zmf&4L&g#$XjH=7U8~jRHwaPy3&)OntPM)j(Ua^6vZ1^W#yaQn>bIw;Y0X=EcJ;Ic} zf%(+c8*D-qkN`DgxitFd`lVm^2|MOQjZS1w$ldo@osdMKg?pr%P?a|<`*(C;nU;+5Q7fg?En}<7oW;Xk;!G z8oVT+%Ym#vJa59G#sjhuq6^oUnFfRnj(TnCSOh!RWw})*_#t3rxSO>~J+cg(@z(%n z6;GD}IVFMJ08s-T*wS_O7Jg1`PaCL%b!jLQfhVxTh9!3d2q{EP^eBjiJsyi(KjrsP z(#1?q8XhPBh@#|5CtU&Re=V_P*6J89wm>9H0}R?B~>hvq@gNwxTG`8afpk6|%WFR(^ktDmDn;ZtNeHN*lzS1~!@6 zqT7NeokAx6-E&{od5(fAR$kg+OIv~YVdBdJYH<3cr0C`mDN$5FDfO$gL^yYqJ4IBZ z+Q@zRoFyAli(FyI?vl&`sceZAG$?IH^CMMKx8APyiSRPXc+Y|mNui32S_NYYdVvM@ z9kfCKD7W_GEP}E1@o=60xgdNnE8RSZN*2$uL9WI+AYQCZT?ytb##GD7T@KcQ*b(X5 zZgg)?NH+YLKpd)|9^mA20G2kE3ws0$7;g;ah}PwefyfeA-2F*L+H7>L=T@8u4dSU9 z9Oy^AkvCCwGIUM8=)y@9SeTYs0iw5DEysBtWU=DqssPfr#f!$~kSVAdd+tlT`-WZEGtxS;k4uQ42#1vtKaSEig^kCb zjqw5t>;&2eaHahIAX!yhZIWBR5mVlLF;RvxVrad7nr zT6MIyT6jZ@SVU6hQ1eH#X1+019nziMtPasa;H35NG!&<7Al0rmdAz1Wo2IT<`nySo zJ6SKk(_}E;l9mwsRQWiNQu=-tN1o7c3ax{ZTU~cYPPv(4go-sL)N9<1bI*vX{ z)bYKZhx_e5Tu`IImRa`0z$~7Xhnioqb}Q79de;pFa&4?7chXF`tx)%|7J*E#Y;7CO`5xzrl0%3*hqD<*b3JAkI59#J@4yVBu?#K=gfIw_CN0h4hL2zOkzELZ@Feo$5Epj;wL6b;TtGy+HdZVy_84x&^WY=RQ9cz6PvH zcBJ&ezUshUUv#EgVoG|}gh_COV{Lr?Ckp+F<9*)nB5T}xO5DSB!GrKPvC;c0%Fk%|`{2hni1OaF(KJT%WtONtjiqEB)lp?UVf+c_ia;C)2mJJ!0XNctO)+!)Pp1t9Eli$r6u)oj}pvVD~(E_mvDpI|P9@O%l* z(lVfFASh98O&9nlIH41l)HNwTX@d5+N=);`fQ^jn{*$U`_ATQDdW{dYtj)p1k3DeR zO8l;?ZbNKta6EL748h{yddVLAMr1J6@7eEwHvNFr^U#Cg4f9v0k*3)ksYziDu0Kh; z4&J*6C5v^EUeLE$77$qdxJR1!S@8Fqz5UN0uU0Ep`y6Ns%z=zw3&;pYxf~EBdOGau zLZ|Kv44v!OR~JVOSv}cgcWUW4yq={6YLvmJn}ZtnLAeER8E9znVu>W4em8z*<${K8 zCfu{^4avz!$^qfsY3-mgYIN7_F2&)bsZ4VBjUx9WWnI4KQ(!a9-P()aYsT1xd2;`uobV0*P{R`1PP#o z)WL(szpcBAd{}}pk-JFUQJ_#*AqS2!y`oI?FQA~v0pen5NVkfv>1|ZLJ`f&W-bQgg z2kHt>AV}xmVrEpr6gih$euyY6<-1xY5<*`4Us^=VfmyaH*e8pE!Q1p#b@H`J);jBI z#(u|M(sVib#FAA2Lf%@ zIv$}%0asGhWRYECO`T^=)Zdm>xX}_ddEmY40ZJtjlKgoT$@0Mf;U7V!r}4J>WC`wF zch7Js9&$3Y^%Nc$pcMeU5~oUgH<+1jQR3OsXiP*Fwh7`?Z3nw5H{o{+;8}Q5V$r5i zVS!%#-sHXVR^_U|TL>kSf?7h(OKPWoHph=tNF#JBr=C4ilBpc$FXFj_+Jg@~mgspH za}nFC80gOBc3V$WH)-VbN}*4bVAOnw&)Flttbq8ABDg4E^~^)FJ@I=+(&Z`>yopWP zFKu9+>%e0qnC&wI=eyOnv(gh(d@_kZq{s)*I+eO;7?FXAEoSEn`c|iRjm3)d4=6}MoXuC6eftSZRYc+g(|LqKJu^-hs>PRDwD1cP_Bi+ zLP4C<=Sy}&WHr7>H5`k1pxxvuMTWBCxt6uB91>QfOV{baQ>iPcCci`hp}E#|dw_Q- zH7eR))rntindwaV^;5Vn;(;3sLYIi`u*uyH|de+rE;+-`Ed4`YcZnk%rau8jY)=Sw(f)Ia3bvv zkRxnX{!bz}-sDm&@&us&{QlD*jq_E-r;hw{Fe*JgrZR5i zs@o@?mzySD)4Y_F5odAgoc{{wYVdyr^x@M*9#T|8P5JNY!!HOb#$7f%IZ=#*gq*)X zRwX6U`}c`%T~Eb+gDe)Vru4OX>NYVaj9pOlV5(l{T<<DGCv>2ZXd&D{(PqU#LxHjz*MBlQc9Oy2- z`{^zn|F)|rKSNvJuio8ZWi-(P6;#6=vqy*$c8e;qf?#9u9oC^BUM8(#Q(}8HFqLYM zl{iaY(8=I8a32RyZO03UfcyX-uCrQ>Sd>k0nIsZu?vU}%)Y8(8ggk-Etd=$sOzLLB z9wm4T(tSx^ru-)tkXcvc^j>CJdn8Pyb6>6P-kC7e+MLK}g)=Yc=Wo*u?~RmcR}t{@ zzw+f@Nsm#TEjw)~PsqD@@C#B)}In1)nqn6yUqeuEh$b<3~LqoOb zuVTqNeJDfU)eh_n2YJmukgx6wt--CL9_+1gb1)b2l+cO(Zvm_etdG@R-x5=Es{bmX z%VaX?>~3Px|KwCd0uevJE>hJ1&e=D3MG39^#Gz}3sh7Xo{8I$|=(p*TL2mjNgdM=V zqvf~#GdNanvV6;q_=0C#A|%8sz&_uk_pLSCr|765>bbIATy~!L-J+^j*QQxlSPyXn zd9{+J1T%=~S{{FRn<1L&(?#mYMrs#|mUG>CuSQQsg_GnNQ!^tcxmn~&m#kr%5yDp{j0 zkTo}{1ht**mqdRHy5aVSRl4CL3aLFzIviM-ur%Zg_s81)k``#cXg?1{Wqi|Ps|x-C zH@)fjJW9LyN4)ZRq#$bVfa2KD!vf_2)%_VKF;AX5}-LJKvKLM*hdc*!hAOkf5c(eJ5bYwpSX? z;oDEEo}p2iam@tpVPo*#+VQc`z`AJjfHkQ*}Zs6ED(P<`xRmo6#QPV2QuLi{y*I1juJ)40z$Z+RWN; zDhP7|txy0PGiGc}g!q&~M&hm7kR@b^7rt2n6d(xzAEFMmXdOI=<^}0IGGvh8ymNsE zW&rKqgc9Eqlio|BKkc=~fG^)F*}Ok2YG4O%0a+&Qa=CPmYD}v4S|N3@)EDprKUwe% zP%HqT-m`C#(F(5|Euyj@Z%=*`kSOp&NyYyX{pEw~7VHR};kkv+(JjT}vDSZPZ$)Qc zoPcnFB!94bYap?5Iu+s?fxF4}!q=HOM!3Xd)_w(iOc(!McGDhzL8cIrT>FYutb_eThT0^IE2i z$lW=rujzoui>pp+5asn8j51~n44mOSj`7;C2IhukqVqm%%&<#-GAxqgVqUqmS((o< zUgL#%`t$+8g)r6)3>5Sn%tp3NaO8Pr>j{8^`fjg1&LsdOZ)v_gk=D6qoQi{NdvES` zKBju;T>u1ivGoW>rS3JC4q4botW|30<(*p#R9(m{ZsI{u!8$f6%ADeY+2^R8K}2{> zqUj6Gz%M2%7ExQ}SO&oCjof6w;bZsHUx>a|H7{WfQO!p(!Xl{gz0kAZgW9$Xn@nk3 zt|RTan8@4~rSrrJjo;~rlSV|+0S5TQ0FQsVdX_8IU?)LsA1U*;f47pz2?MG1%2%VN zqyQIo>mbce5$e+E)?;M`RQwE^zuNXvOI+Ix74HVi2S9}296mdm8(_JZwDY9^tpGil z|C6x(f2}|8UnA6Q_v^F6^1)(JlGZN`10$oZa`#J|n*}>q9l#ppWB{z(CIZ=E(%{a< zNCGmT0R)6x-P~MSHBa`D>c46}V2%F#i3AA@0!aCP8VgD>l)Eea0j-oUMoeeuT;5RO z?}yKUHWDv7z1DFYNmWL}Jog=7+EQ=$uWm~1*sZ_F>BKPL8D)@ZOF@^w3JN%tWt7^o zXF?Z~)lA4A!0ZC(IUg97U3?wWxo`)#k>gW@y}FBJ0vhn3QFWE`6K*{)ST+#ta~`pa zItx6cW%|#%y#d*Q@>mz3Jp*)%(8P-h2n7a=TFSXuT6-Jk9{&3$K#~|}=9GqYkIO2` zAONQcs=bXmiRg}y0@%){WCmZ{MkN41J7`oMu?{eU#Ppe-L_*H4s1iXdgCZf_K$;4A z&V)P|AcwZj%)wzHr0WX|GSn1Awc7P@pYRWa4{i-)=3WbWr4DaI=RR}PbWA^&te()q zE6*5Z=($0%ZL;VlwBv1>C4F8&9n|sZYV}8?sKfW?rK~Lu&mv4e?rWIXOINC~H%k5V zD+!937#r@7bSX=i0cP8FV4F-XSyQ?K{gFRiJ;Pe6V9MwapT^K{NdeNHK79sYo01(P zEB|i6H?W!E33k}bzn6gBoP4D{k`u@lGc(ucm^g*w74xG$?t*RNjw}9LQ-9W)g*dFA z%OKXl-5dXuph_#x+f8-aS``@iA&KD`zr-nM(KmO;lMrv` z<(6i~S#UR6;sKKfTp;>{10Os+OgT?s&M?3IC%Cz9$ytW>qYFXLt(8SbctE)B?C5D6 zrge5HP6(9oKx1IS0~npyFy+pm`Q5e3iwtu_nY`_zadx|Ev`=hzd^-5;n zDN#eVZCK)W7W-uKiHV-d^vHhY@>%MR$15Qk(=4x6;kvD&5$Tbm%h-Ww5@Hl?SJrLx zC3GNa5>ntKdpE)z(S@xawl-aO;fZJx-EF=34?l{SIWK(8Ek5VxD zwBt){{nl^v6Esv2fKC0{yqQ@4sXLG1yST<}qDW0_zR9>lGFmfAs{^OhfTI6UVn&+u zp{6yC2?`EmwlyNQC_i}ZY`q-jjmxbtu-twV@h2K@&r@p3U9VYS!y*5@OI2-{)k4hf zi@X`VMPuh_76{Nmx;zYd1k9}^IR}~J@tz3%T|TS=4fBG=&R-ihi%oZyy(Y7!nt>UU z&l8hyzonYkPns^oF9e?}oMC#&m#2+PcrK*v?~{EFoZkQLF(if(`(6p}e~ma8z{+RI z>|Yv35EXuW@9iTw7yS|?t*>nSq#bC_YwUm6t@GxVQ5Il0|0ww>(Qh8HrsxdV*`?eS>}5a+vF9u*<1_NhY>6}{ zjrYMNTHfHw7+cqH+|_EXhB3?u-$I3;GIE%X$hD2{f2_aWqOUw5i5OTktdm@8?ff&F zFZA|uT5xoro)a^WjPXt2Nfy~ZdC3Q#P%9Efg~iX-ZZ3una6Uz-1n?kO;sFr7sMGJM zHH&|@v~3k$-JX#?>HK52de)eZS(2qnje{`ho0>3ZbS1B+id_-UXxD|h8oUuwaUVUB zWuOeZyO#HB#jHf=W=o)dh+yfL0bQ}b_1NmJa_$vGMsK9`xl1B+ZCEMrZJ#~|ElMQM zI$YL+p;`3z2P*c07lEPUk)$~<`dn4~jd+>P?`$77(M;SSD6cpD(3^kXsfz!*pQMWS z^)sxb>d#DANq!j~09RW%4V_ye=``Hx zj`wlMwu_S@c|wvv&(bw67V40Y{VMw@alM`5i#*MSa4bS+Gb*7Ol`L24>Bn~x+ zx)#L&Sp3u6v7bJ5m&`p6qV=PmdgI~JWsT8DpTTpi*&jKao+suf8#_t=;&CqVz<)VJJ0jGRS$iXyr6G^4llsHC^Rq9)2F${X`!yQr&Sp7;#2gG zO5bNAn}x-*^wa!*6qzZX7b-k^lkm2DW8u^LZYtKdJn+Z`C?m1J-j^&B76FO#Nbkh4 z*@c7tmqO9opT85ZEj-n3$Di`_eq1b8(L21S8)#KN@Si&kY6LK z#|!`9bE(w#GG2xgkYao|0o z3E{nGn2P5~#6PLL*M73`kwb%I^(EM=TWKxx30u8pOo@eq6fnZZP5>8d3H`F4fI5jo z!#??wSADkmQ3PkNzbDHU)!$#G(f&Znh+6-#X#*Mv5;rSdht58c=m(G(*4GnYxj=K< z5y`)2T?4-OUr<Eo0^c8GGG}e|c2mOnN7n@I=~v1oQVn=@yUq{6F*)Ydy0oabie_Si^;JJXIB0&| zi=ynki!3GnOc*+sMl{<@&-i1c58}T(%sZA9ekmp%*~h`H2W-L?+*Jw=7xm^ z3F>XNa^Q)9?~AktPB-9O(w`_KKi|1;72P#QK@_#zNE8j>XK*J<3#|UN)T3xL=^;5k zPgY5eLUT5cLMwNMq>y;$KYcHWFDk+~DWA873x8xu#qIG8RD_j&fHD$Aa==q);Je-4 zCAe|8cE)NgBYA7rjF@B$m1S-QcsT~jDs+na(h;Tjefg#fdjnAFgS+@ znW+ge6~SM`W~_w=69G5$F9fg1AP_H^E<3a>vP-5%?h8R=kH@x5HY>ekocK?t|3HU!qw5OlY z(0jCOt%^8$+-+2QMDpZiTwhCB#Gi-F`pvf4o$-i=tM7?x#n_gOM)Ze^Ml&2UQiPFY z^6v;2tGgvC;_)03^bbr1QZ+q&L+dEov{`1=sw5dqp4H#*zV=`{4zHw!oay72aRoH6153vs{aDP8}CkgH&(!2m1`=j>fv78sF~vNd2PU(V`g=WuwO`jc(%%R`RKa08H;~o5VKVMP}P(9 zWY~_=Yg`A9h=-S>d6%V@qm#GLy}l(4*%l*ihFy!rTkrkcwVy~;e_u>_iKoUxB`F;} zfE}swtHhrW9SOOr18$>=Z?Es88k3pD?`)gQRCqS!Jdeu#3IQ8)jgnzTR)-}SgfMrw z{`X40|HWMMJ2lFj;V@1w6DSaPI-*{7GhCR?<1st_lXZNC<{cOWW*B!*Iejd^F;Si{ z#gV=)mH0;bf&p3yT$l-d@DNjqg4wU`d2md!T746LtXFw%A{Ggm0mPO}a9OyT)ZuMV z%4gzaM2xV?zaXl5bigBBB0|Ls_69^``x;a8!W5Z>B5-BgoV)$@ooxc?nJAYfga=9$ z5bHoYIRDXQEj$nj1TLytylSwTlPkek61vw2yZSEjB~TRUuwUw@WY>P+=7vS<1oPY- zx??d$Xr<&oe9H1wf6_Y6ZBW^pGC`P1W%()rC3Pca?urbkY?zQEZ66nofJ7#$B*thT zu#bOzc@vrk47C4kIC@jd#4Ktb#vR|#6 zdSRWRe(|jwnAN&OnLvC)A^v`m+M*x!UGo_B_nrWAy_Z;%oHan$rfSVo!Z!sN?zzN> z@~fo%4`b=X6JMAfP1@xF_>@ICFcRzT+5zcqoeS0V4nDrcyOY#W<_ocq0IE2eNkopgbp*)17wWy)q0q z2^3p${*qoBV!wlNR+JCnWZKiUj#QN&IcZRgYlcEJ^bc(CbIiLKgQU(xE5k^CYlL87 zvXs_8-!oVkOrTBMuq>m8vd{Xnd;X-nXyOTv251Hc1|WMlA{Yb3;_jb-Ph2dB#DTG8 zlUZU~kys-v1O{-b;4`TX4V%;yoq?C%?K<9FoHYCO#f_yByBmX8nV%y21hl1g6=|N~ z(YbzoLoeyWr#hH!=PT>CZ*OruADDv!`BC9tGcMB1kw`ImbxqSxB&e~e$(JIS6i^cN z*9^Pd9zEy2RK8|HMF6X`^>_wsyO}Vb5eNnMf~mj|=yg*JNQXW2>1q3l)13!4k+Zof ze~nmM9}mR`!u2k8yBL=Z-__d>F^gyi?q$vGa0dg%gY}hqMR!BybVFg$8PN^(KirtF zWv7hqinv9uZig=oFS+M(_)r#4QV|fNi+~z<4aBa&mHb4>L0}eGG5W z#1$&XJ(wmUiqB(oS$T!kLF_!u zm*UD1L5^IXg@B`(E(`fN$$Oal1NX}E&MWHpM(Z=pRABKS1RkYP6E zIGS3w#ZJ#T$rTH#Xbd)0_^Zmg3lN_Q%b-;iB0PYG@fZHe8=^dWw$rU9fir=V*;j8d zTKvkcmEFB?j~HjBC-p^KUKeoO;%ZKG=rEEZ)$H=5Z9zY-FzcBs^k3yzT0Tu?Ka;px!t_A^+JX+I1s3kzJi92N+=@YmFE2P2ip1GyvpR2E5dE`1rtQCWRs{GP zzw|yFo$Kc!E!?FIGC%y!MS|rqoRqFStvDGS_bm)K_;EnR_XIwNql3BrK~1e0kn(ZX zbJiYg>k%*wAr8SD-*)|^6U_Ahh4tZJ@KbFHg7|*Fv|AzfHnvSp`d+4`ApA8Z+0VM zs2#OB38ua7BHL!i3j*GGB3c<)^Tf<^O*un&2hd_blT;?++cY5QeBwMO%?fD?1|!bQ zWRCvZ6)C~1mfTQo^dR>e!b)e#c#ELof3 zIF`cG$LL+D#j{2#Lr49|VEzCcqeStRK;OFs$&FISwTOr;6dErHbK5Aa`65{=PN!F& zL7W>kX^J)i31S|zK~9vB^X&-TP`x_jL?mKs;MfK09bm7))u#)Eh&-+2h}%HteWeXy z1V)0deaGC{v8`Wq_eJL$6gB(dT_}B}!z^kcnU{h`tX|x!fX5~qly<>yi47>0vrxJz zF+RYEL5h=tci_<<*_-JYx%8IWy^?a*;kGzGyK(U{{}*O|W~5aJM&jhiJRb%;5ulCg9`O3=4f$xJt51P?p4=%431+W0=Imw(4bp)H5AV z!`YzB3q73{3{#YVlIBGs#PFBTGne5QiUa(d1@^dRDc&nlXcsMa#WsSX;=fK!29Jt(Kh=B@< zkH+^9t9sGH?0=*+_nX`;5|e*lhkKMaR~L0iX5Us&Jde5P3lb>R!-%j2?WxMQ@2n$gCejn& zBQd40*;73fLB{!V5m=}>bw!Qe-D0kI@|1c!E)h#z_h&%Fh0Whf23#boXb`CHVCUvQ zv(u2tgB0)|Z`ue51XvG~)NY=#ywur}&6;3On&6(FtD@~7!c$-m zYtb|QY3V&VxnGYhEq>yUunnZMXf4pUA6cS_B2R4|ynd4^5dyDPdFy5pz?>ZzOm41q2pI>TI#I0Zr>}iP5OOXcefNz#zj3ih?pq_qNpIYICJCV3%r3Oj7_a~VA$dY}J7rm&fG{4d4`(?DRELPXw+_j9_;%sW%f}P`wr6{+ z&thuNx}EnJ!JKBJc6Pt_RA9aLYq|yY&O&>LMfa{4bDGlgtw_!X#wl?;@?UrvH`*z= z%R(8+w2w4Qrw)v)_Co)>Wa(!GF0QcVP7E-jxVy|Uy5}Z4f`m(zF}74yttiv45HnU?f}fQ{<$>PrD({Lwj|%Rrqk70c5TG62sVBk1^8|oncv&Kj^1Kc zSFS4X0G_4G_s=b@5kAkuGT*JX<>`8z%EX{YXS|>8l43L&rWDv^b_C;}&oHl0;Wn^IwpIuCJ zkse)#OJrV~*gR;DTE`}}gtjJ3z$K=r7bIB5;s)fbwRjPB*dNHk@Xo7y-N_XLfi^g1 z$D+uv>gfG19m`8KTTV*C?O`f^PM<+10y3B;p`Vk{iYMHpq>`{=-Y$|f!4bnZI;lyA zf?<1v8*ESGpJD-%tHQPKo2fJgYTJJJdr~{d!hAcwSKLy4-{2;f^|~AL-;e-`K&r^7 z4_$|X$1a5y3BjSbTAm4c6Bwg5#H5dG;oYJ-4-rO~LNA$`c~0*C(g!IIkp9o0UiHn@ z(Hi?Fq(R^-UQ;-cX=GrWT@3~e0lq5Vg*{|P$&h~PVrazD=?p6@dS58XwSNbfdp9fU zhe<#7_4_R5%Ct0*g`6~flWJ0GzO8V3 z29(qHcGY7U-?MzeD^7*t(8s(?a#`<|N9Q={PV}Rxj`xz`qziefAG+|JPe`5Nfpkn= z35RvxLDDsm3ey>P6{3JS?$jJ}V|9piSbYr^jY`tj*P|L5@WAl4S@uZ>8j z+!ll*n-oXczzzeGfB1W^0c=SI6coZvc&#ipFaXa_!sExUmQ8XH> z>rsq%y3SKt&9=U%iY=p(Xfu7Q^Cx{%dVj(;e6dG&Y{Pmo%d z{7zw>^Ck@9x-fZ>3N?XFYa?m~^pf6H`O5D#lGu@6NRg6iZ!E_{Qmh{{qN*zeOOazQ z&dVQSX4T}P&X%1qyFQWawM!f08h1iz4C~n=mKkaJzQu&)2BE!oWXG~6jKu#C(k;ol zU%LviqcHW8qr?&oH@262vH{;D_4TOg_UeD@GoswmbhsRCx-_x|6LQa49+*%o`*E3T zBFgUw^5LkQ2Po1dc}Wv}tXyiPHUKWKfRZh~GHa(Ts5f@N^e+5#Ip&x?9Rc?FVwNuMK3KUFsTabysik@|{7R+!<(lMUe3;T%yJ0JDSCw|5!4M~WxfUhB+G zOB-H>j2i!sYH2_(j5fQVYJo`6M_xNwy8Rg!ftmWo_#V~-tst2P@$M+0@Q0Fo&MASX z(7eV|e%{ z-Z(s&hrh%k_|y#sNaQ%*aLf6B*n87JDBJLTSoM@5q%1`Um9keNyR1c`gzQnsp6nU4 z2_f0HYzf&KOZF{bj3VHJf?s*1}dq-dQF6wuN?syT9A%+IVsz z-%7qa{8KorrzaI}$Sc`LA+p?Q{AG7*>ddt=v2!W?GHU%}r!zVfZU#?Q?}n5~O;$1f zeLvOOF$QgE4Y*y&+wBh&7TT3wus_O~oK+#%{R^yd<9kil%uZZ*CY*a$%MxcBbFxBr zW>8wKO73uEk=Xy>Z(KDcKH$t9ox9~}mcQpe#SsG&+wp7Nxt`X- z7A_VZ3KL_s9nad&BYUBuK{wXz&3znx+k`YPQ-JKqXFN!E@zv-?PV$u3ro$`yISY9P z3rwP>`N>g?2J+_yF;dyfV@PsQ?!3oUGWF1H1yUK}q&_M6r6}^}1!pkF8hL_ybARgO zjf;c?79!&T2Ou+KF05@&9(#Mg4c}4SxYZ18^WDf_z01dgr%x!mDL0HX_8V=X$;=Q> zhbHKM#tX?_mK!JxYxfcaAvM{9Hf^Jea*R!fj-`CL@cd8$;gJ+<@L#_QoN7#~Z+d(z zMLuQcqSRn@7;jOv|p%2_es(Jca z6e2BkoTzBS&avZ*WzRezbo(``72!mWPr?30CP^U1VMk-IN#6HCam4UCkOUG*CL zN}@DKwu4AP59`~ zmw{GivVh0XVP3>~r*f+q+dUCFZ^iSge6QN*BYhvmBwJCl5x0gL_AB;O8kdMc{6c&j z%3Lx*l%{WN%+AmJoY5Ij9D!TXRg216T!y?O~%`D%p!AoB)rdK1Dq z83oWBZ(y#HA?7YekSG8D_ClBXi|r>gcAr{U7zbB5&fYf+PX?uC(9wONDH%n55L^56 ziW>+)#CfoT3WoRcbvTbUevO~v1O6sbOdz8e$K6m{uK5Le*xbMJQken7KjIzZQ_y0f z==S?Y$kYp+7j@&u4$py-HkqvB6mp|g?~b3Td8~gOOf8U}d9t-k-+dixDRN^o+O*Rf z6Sx~**Y@?PWO&5ag2sU!8PVvGL}}y!p%5y2Ro@#KQPF`J8hDK75`e3VpXUCPB{hSn zL?4lHw=Yh>Kfp-p??2GM&sH6fiA2VJXaODonCRj|lr0J&nkpWWpt;H_Txml+*R|NayOw~6zX?Y9^oJhHc0%nIQjCxK}p zErW55Xw@aHgx_V-?y_Hz#|9>3o7U)RI)--PQoyNR@HmInd zW>w}%%@>!bg`C#bcpV-#hz_p~^RC~n5ZTF8d~+@9N%0%6@wTHZC4#UMH6dhNN=gw% z&3faZUtIxv=rS1f*&kdy5MSlDG9)H;>ub>N|gTo~ceAzS4ozBmO>i4NZL&$aRvkBj-7Qh6g%<9w)) z@KioD;^V>tFI1r17E|(7r4_I%{QlWOGi;BKqMiQT^KUAP|12CoYG+0L#f9RuFfvN$ z1AFvCF-F&hik^*V?p%UFVWF|MSLp@FMPSt!6?(S4B-; zrH^9^r{1vzp*Am)ir!;_2wca!g!x04WzOM~h!FkMbzey2s?FswbzkvF`tf26@9{;| zT1TWh!6B-h15|l&d)cH*(%>2J-{eTipK`x9dfM28SHsjV{%BF@!Ert}B33$=Q0V8C zdO=pUnRKOcjro^;vw7`_@jskupou5!hZ#o(LgSVgSJ~wevD$XqGPMl_IUC_d4+6Ao zkCtE9>5#~+Hf7gppSr_hH4`a~@7|m^CVr;OjfF+$*}EH2vi`3qpkBKlV^r*d1z0=s zmHNyc*1K^IxQw|oJyqAUwn>UQjr&(k%y)Zc_+El*T>KrpX`V!9cUv3B_kX%wCNGpq zEUfZUeb~63)bXOq{C#cPh*fUMZ;6`&vv2dD&EW6#y@CaNonEl4%c{9W1Pf1(tMfV# z%BfKlB2Vm%_P!LtjcehWzq>A@u*mzW={(2|O%>pjU?f?OWj=P#x0po=*FL5qPkF zf{ljvm(YAhTfmuX4`oK&rsK*pbkE^45fG3lD3uVTK!RS{ADR2TcK@Hl!T-UP>N!xV zP22A5>Z;-Jn(ay{=LXLh5jNU=Zxb@Q0Psqid|{8^FQtF>8;{8jBd}-IfbQ(%8)^ir z(*2yUk00@7M(oHaU&aGOCkqQ?la4ocef3O>Z4HTZiCHraJBWSTe;^U5&F^Ev0tEed zlEneUCR^L@Xlm3%K-hn1aJvA3llz@hGtk@zu>AdC8|Iie|Z3DUjlPBXr&T^h)Q-^25wnxy&Q&%7)Aj! zJkY+V7AqV`L$Pmo_VU#(;*d5l5`H}?9?acfg? z1n0K@{-hQanyXhS_VO_hd;&*E(5+5Ae%wT%ZX$-R}|ED zd6&Vn&5iFRE%HX_7KTb3Kzac(0vc)JN}g$Rx#2Qj(a`g5RgqnMWR=A1m(YjU1cx@! ze6!XVNq1)}tN9vl%18mDO89qLg6E6*-fY*O)H6s?w`D0%WW;r?Q;CuH;j=|h>O8AH zJWARqs@f<(=c=;>$We=8v7u^(1J>8fT8g2>pMb(Q7L9twdYzZK0HgWgXy`T>4q#p^ zuI*a8#Oj|nG*h$&3`%G1p#CX_LUU`ZNENQ089?h2$YVa;I$ue|H$yo+`(duh z=F)J27m1+6=JEY;!-qaY_r;Rwcy}U5N?iXOK=-N0?ZVIEbj;(N7IGW#_Yi}>F4bt_csiRP78m3sez%_3WtrIf$;@<_>&D zYI_B9@!syQ?RKdh-jjZgj*i8{ODX~t->a0CgHB%=kE|k2hN?+5jvh%wNb?7=d!)@O z&#FBo>FrJ!eDj3RH+as++3u=UZjJkcp5t>$ zifPZ+d#L9cybP3*OI~B%j9ZChZ;_o_(i=uwYN1E~`-uVB!RZo%bFU?q#R-TZh2>LOEPc%{?QYaMdLNz3o7B4mt zaLu)n_IalQV^6dS#LgV_#!eG!ceYT~u7vnA765IEPL8Hsl z0xzGA)p+}yeSS1n8eZw9fy;265!t@tD((fmmb=osjtpi!u;R9UX&d4CWX9uMMoREa zJOpbFY+f_JT)vb#(qAT73X|FOHLdI$FEE`X9-e8Cqmf(#AI;F#+{d80>r`Tc2=N6O zg3GtEB{2qr2P!I*Xzy1jSjf(0aqTx`ZOaygNe5pd5>YYyi2JeutZUv}YKRv~tZr|s z$TEgO1!R5Xers!MIRj$3xs3zWP=aIEZQjf_P+xhGRuhPaFLl%hG9x-MVJ?Bvna@Ge zq^K&geJaSnbt>BYgV@}qi2E98cP9(fv*7i7av5>6_Kf%3o_*Ruju0f=%kVgrM=jp@ zZ!(p_>K)LRoD9<#c`?stNQsb8FDWQ}YGG;g>0BLtF`r$|AoP6wmv+9}&hE08G=o1* zSMUO|pt>^Am6o{=A6^^=;SqdjJwIVLpmhx5&9&fLE4qSg$cOGouu8+lA2LDQ2O+09<=3k9O!enz~{+FBRQu@eBIXpO{U1(ClluUzjr_09aCK#7W42N2H|FNrBptATGAtwd(W)KdNsBd^)o) zs6^!WEiWu!HUCN=p2MaUWbt{SYL9e8;===aLhgrDtubILcNgut z7GGe>Yf-1q8M=KPR4h?z7hx9uP@+sRJF|4HYXfG)4GGx-*wZvm>YD{)FBY z^V_5PD;`;gxs!8LXe#;nD(mECU39!@dNR>eviyD5tcM%Lx6(OoF&rZ6+hhKF8&<%E z@ltD}4O=q_S(G8KJVeV-Xv?~tR(y+MGzwFwzcaMiOhS>av1B3qdCf%{)Vl$LkJ z#Kc^~gQGa`=I!yVsb14{5bb%N@Y=+BdyF+1H<0C~}<$6!|Zl{p!zstkZ>DyquzG34)7BRn1IiGJ;VIDZx+x z38mxy0(m*<;HIFdZrJ&B?Z+XSCshZX078yBcB=veyy|skU(D6ix;PcrU#ibra~Wu7$Zd7u(VO3C#V{sa{DEkKu$L;^l%6CT~xWw zr&Zg*^4f>>g)a@&FGQzF|Izcs;Fu2^X?q^M3j4_;L5$hXF|JQ5u!3@$hRa+4t`k2w zv>6Lr(aKzpcVC*FpMP8{0I(FP#U{h69MIF5q z7$mj*6riNF?I@D-YTZ76k}$frJ6cO4{fX)h6^oc-eZZOT@iYhCA$#}pEs(`8mCk<) z=2XI;X2-(9s796rY$ZMm?Fpo{G?aSB?{5ZWOiWDuD6N6@F(9DDy-29uL(&dE4v-V) zw5ybjsL42yLY@VpXg5S}g2l&pXE25PhNY3w=-!UkUeGTimtm)@lR^j}^cSd%Y8^vv zt6VD%(7ENW3{6HuS9W*t^|)n}Im8d5DdLJu7S;q?l4Apl3uu$@4~+I>U4;$NBCdf8gStnZipN#weKjOgWt^U}~jZ|}v59hEB! z-GEuL9p}_O-LOY6<>AwBM@B}v)uiKEe?iWT`+f7__otok_I@bd>fFuOHu4CsIhxzv0Q2gnddFj|Y=8XY! zHsL8K479|GEa$gNO$fX}QcM|UHq=TrBzkxR33?w$xw_E3y}fNM%v?^xM@tuzcG<{hkb%o{}7IsP!qWU&E!~)CisDV_wgOnh8Fts_kV?#v3umbjIEBiTF3wTYgFwK>!tRJ2 zEV7#xkTGnR+|=LR@OeWmHWTDUT_$ zSy%8h7X1n#C=yR^j(Y7OrIJFwG|2E0tOxSEt5SnuDr0PmR%y;X0egC6Pe8x=31D^} zk``-@FDWVBRqBQUkU>GDAjqGPbvfv~_`A;NkO#89j167Ky!}pb;TKE$9W#@Y-&AQG zluEFJM4VWAv;^>&J*AN@FT>72yJx*Q6+2g)W7qcjJi&^-tGBlh-go}2=*~jET+#`z z1N(ONmLA))X-JX;L-i`WnWnd4v^%E_ne;KQk|zkgl&+DMN0ai7idlGT>Y(E0WTX)? z-^f{A&m~+Quil0ERA=<sG=XX4F@o zN;ymiQ|u8L@<_0ap@jO_kAQV3Hu60C3|v?|);dx5k0Ri=#o^~(BM7Gp3U1eWLVpI0 z)YdsK9WSu;kF{OF6`(y9KZq~=5s>kQS92}_axjG#8(WN3TXoQ5vSme&Crl@Fr;A}f z%54ex?QV3$a92hy!|hf=7^J!Jhm7Q1$AU)q2cjtpl0khdnPQUa9Ix1MqzS1yVu~<4 zrjY&KdHk4j^uZ|vfEq~N_GZPm!Aik!24TCoVkF$~kAo!ZeFQ5ZK+L@&XDy!j?7!7_ z%44%^v;y*~trJHe|9$UR6aZ%Zh+~k~f&D@Qg>BXD#_q;okWI1=HB0HE>Vu@$Z3`n6 z?u{WAx*AqAxIR6rLHG>Vl5nq%&%i_YLf!xY{nDQDXNPvlzU9qT_5dUVhmOqL>#{wN zgp@lmd2Xvyj!K(7kZmye$!}SSZA?Q(4a@*-!c5pQYFb=qC$Nih0iq=Ashw8nie8i4 z6@hcVz4mr3h}uf4MT53wfnGeascH#rR!RDRYCh&Z8>>MB^0M|vVc30KD-21mZ(74n zLn0E=;$Lv@P*_0g%NFubp*H*?tC9RyAT;FQpeN!>na*rCw?$xJASCvgdDA@z@j#rX z0#j!7e;{feHS{D*9y@h$h3^Wx(5sm-DX-R-keAz45~dJt&*S{v8)I)9L%F)+VpU^> z5r|2oGF^jB%(g%073=Z`W|1Xd&JTY=FW9$^kO-LM!}?yHBW9ELyTf$b zw{PDxGqqs#R53|=maiQXx-aKL9e2Z?L-OFaEh^ec9Bij)dXG>O%%iTWts(?Ub!B)B z8p(Pi`$cF zT=ijb?x*qEsvWee8vZ`q^9jC#GCUzW&4>050b~`$82EK9TLVd;5NK~Ccpzp7_^0J> z6&t^=wcVHiF3s#(QqPAQ<3~$tuQ*J#02D`mQY-DwN5XoE+%32G4J$Lay=&V*^Q5W9 z){A_9+wq1klHpQt1BPxp2=rN`c$%F>#L#co$R zn+hAmCZJ{5oT&E%F&Ay>y{6+_@Vb}abi-Ec z9y;cZ{&SE4eqk zRoxB%ct7+3n|JZ819t9#srw#iiOwGLYaf>+aNMWAKjf{S;&uYehSTi;GU123uAUx2 zShJj}{d{2UbsvmC@E&nm+PeOJ^`nn{ubp8|nXy;6?sX@S!un*#pp|!ZW6&<&O=Wjd zz-z1^krwXc@+YTW9hA5-2JN}dV@RoR6C3&9{7K{d$o<~9!+a35GKOFgUR=!Ih2&*Q zKqeQiQ)pa!MA#?YJl^eHUri!}$&3Nm9Jk(Spq^g-5H0iyBk+ct4fU}V3w!+z1i0w} zLFlS|nK6lT?NF|Cpw!qvEY`uj1U$y)C-YPY({=}UaRT_@mKJ>_g;vk1Tb z!z$?#e9977+S7%8Upi85Uxq93?qbvCzml~yX^~kkMwlA2;1`f-yYfAy3O4fe+U`#1mylj6o-p*)oka>?NC>G#1TcE$`ihpm(9LDmx+d{L?!Q; zqVL$LVBIF5U;I(~COZu2oruS_VzF%WG?-%EypPq5|V-Jl7#WR!=QP6 zYOANJ1{bF^6F)3*>?Oc-IZ385KL64Z*^39>1;t2vNwHCe8J3#YMno_m_#W0Gu&=ob zT&}QKwzkQ*XRhD}XpTMNhCnG|MBYOfUiNGlJ@&{1Xh}c$*Jye{X#IQ1Pe(@>dlpp{ z_z}?DAkm1-?5fk;gAJOXhFb08r+qy+y&yL0k=^+N++hl3`8H+`ZnCwP_B|<9fdt#{ z4<+&x5xWry;XOcjH-Z0w93m(udT%BY4BXwZR>BAZA%pT?{W`-Tw@hOs<#sS!&W%Ea3F7Ij=ePkVJw9Be^+LYk7hI zrDyax3v9PE0)98l1y4UYB*O}mhXN#-vuVC0SXI-SzC>jitoBDvZuk-@LKLLJmjW{?D&l_a7jNiOFX9mryU_3~8Uu+hzu;5}HCY+gj%_tezXDd1^M!xNFoVthyuxGj=P;eGWIgBi~MKCL$F)Mt>@ zHTfLJtqx+iuQ|lvDq#GPj3k>*O^R?H9WH~7WaFrMO~;uawmIaWvkQp1IrAJiuqD>u ztpo;2O z@vKVFz9|e#!yEoIn>KYPM#xY5G6WCgci}f+l_FblM`Q=i6A`e{r=BLjP-ygMD>o~a zEA~h`s zlSvSN*|B^S=uzyc(UU7edRA}Qz95EMb*_?6Z!r;omZ)jr$C80;+jCdZvz)7AE{B@Y zFat5M%B^{-rix9T*d;;l^yLL(2hmuroeX=&f@HJ%?rIXzLo74}y58)(8$X;$+)ppC z5L6;~Y0a*(udlB|m4G-2Ep;jaQwcvk&y-|gmZ@&M6DhDu;<|bwKm6Sw3`+F>b`89mS?>wLfh37Jn~54O+FJ}ZCrI6(XK;T3{N9Ab8b5b$pcuE zSNfm|5P6$pF2Ryki_t{17UwlC{UumniQx|3Tr0}yo1#jfgGXD;;Wx=w@twFz4*AM- z1%}>Zt1pl?|Ene}Lb%C>U3Cgxd^ZF1S{i{5zL7A_=`N1zd7FNSgNf0Q!o_T35?5G2 z6DAGbDL;HaS3i<9=8LzUxAqBw>A656Kz6(?G*V1<*YW$hO<@h2%>~%;Drz2BjTu*? zBf)miav_PVlDBSr8>J9tmsj>b{szNN)&%W^ZQ2UIWPH47FFWtT(5MbLh1W02&= zX*AzI995_=Fugw8%)X!=OjQfT+SiDf>*VMPA^0})_sR7LPN`|>8#{{09@m&To(%G#1x z28zF%Nxsq;8S)G!EhgqCA1ga_;?tq8o`H7-H@g5u#IKpB#6>S+A=XM}%KyFUkyC_O zu7&7Ab5{5mPHID7O~Fk;7u=j&L0(kNBST5;BNvtZmbhRpJ)!A~nz*%|yIeZvg2#Xt z7^PRy+nk!P-{9#}3baT6Ago*TRj{P+Ti-ih@l&|U0A&_snju{pt6g(0W$-uz&RGa_ zSI`PrW`%gJ*%YI^{@sf4AL>`%1Q6rn;~~NCC<=n2yms_4j?qu}wT#TI5b|jr@>#&{ zQVTh?Gg(8R>=P3gcgsE+jCLo)N4E37YW(y5;q39N`1BHo>AWjn1inQhk?!n8wfGhW zILoW*1mWF8b43KrqvE?IsiUCHmz}|3DDC5Vo|^Y+0~0l8stD^!JCq zXks-Y4V=%gv_n;A6X-a6h9x9;m%^dUR1EbtzILPtb9OZsb2&}zTReZd!13tb=|-vz4~IJ918`k)G?A!HYB7M-)cEd>?v5wATD!%Q%Z zxB>CMOBiniWosyIzx`md(4SkLfpC?|;|SfTf=B9snc(FiauzlL3uZ#t;G!Al!pzJc zO{6_Eu|m(TBAl;YTi^aLdXDYidvU|=AtXz0QKf>?LnEX1OoTg>d5jc75z?acIP5o_ z2n|sH>P=>9Sa(7}s5yinpU}^Hatxh;G-w7whW~-D)=|7B6EFMsQ_8diI%q7O_+r%! z%K?*#5REb_{)_@FdeBFwhTq3sM5z>wwl8QTQrnK7>?Ys>^(yi8^-w?wY@ z*}cVkrx;AUI}QCf5=^G2q4F4oF#hEka8^hwXJ>xR*h6~yy}0SIk~5vZKE&K!lyoFF zc9H0@q<^=yojW}`5c&uGW{6%}pS=xJLrn+DD7w)AZ}mTP}!lZPXb&>56f`&P5xqXCBS;gCaRRM3IE(2NX$FdplL-8KrQ? zkKBLQvVXWvp}nH>O@&6at0udKhay-_1Z8zr-r~(uGo&oa>H3w2 z4Mpc3re+k!Q_gKMWNyf#m33n3F#9R4yI4&~L*{QqDm_g~`>uHOJ-Do06!EJl-W1Wj zB!7lu53o4%jxUnRYim-1z{?-ZT(9L_wRs#z0j}tyuzM+x>SX|`m+9IpZ9ERp#U?C6# zl~s3DO{-rYu0fr90=_Nf8PC9a$MK28%d z-T5OlQ(X#DF;&M62ROW-Ah7TT*)%c(0G1B}u;A7!v=#-+lUZIMYCLnS`uGj?=;SMH z|6Cd+Yhobu01j`ELNfk8nnxgkuqEnxdc{rBvp_AAf6Uas;d>t*4Q!n;HJ$t8c(4}| z0joO|J?Unr&ftWjIC`^d=$D_vui^J_L0$_${jTLBg&{-JXTHR_2+Q2|gj7y#N9?Ld zOhFw9aLfgihb+*w4VjQ4!yBn14S9x?&0wY@`#VgG4#BY>b3iH{-YIsP@5OP$aq`-t z@vXboHCsoPWR2Qb4@eKqqSK`Z*##H98&faZe>>SVFwnSvEzX5XqG#zh<^OAAXArYv z&*Q><$z?g1;$d~^K{V%HIQTb!R%Esz@XWP$+GoyiRQ=8!*`$+|91UGNIY@NIy?+TC zea6b28_td0Un-UlA3R{(d*V6dD!7blc+|({J=#-F47!dbTk-h)E^R)`skE{z1r`8-~G!_+uSXHMn#y%_?h~i}b zvhVM+?I0+i&9(7~5$!27T4(5@f$?`O62=SmK#+*2Z~Fm=DsRdG)p*BV=-PWusoU+W zf|HAGK=?gwHn+0kK{E--S5{)(LJZ#vZ0dTlf=Am*xkYxEG(H`2;Eg1ptJqw_uYfv| zz3{hf`lkGMQH<|){lf*8T}L&oUmy+1hk8$!cPMESVf72%yuJ34S^iIE4`ML-!`sU{}v6vWwnupc%@M4I5Pl{JWhlaPLsJ=F)G5Dj_9YSP z$1i4kU&%V%b2-yf5_dQ7A z*aN;zYAWjS%yhi-Z;SG#m+-Klp-$_qe`y9Aucn2qA1u1g&xhg54L- z$f<{CTO@n6PZ4sd78mF5lzQpk4}XuEgYu^tTZKs-g`uesfRa}V9KlGI!XKe+;6#t^ zq6UCKQ5r$>JL3^{xj%OT;B_jS2T=8^+{wTJ9?vQ%iRKT5qH_l*%6?{k2Olg1c4!LX z)7NUhYaylf<4`XYe2M5*tfL*vhRc<~*$crs>ZY7~h?WQeYtp+RD*R)b?;uA${R7&A zW)9vCb((GqaTM|W_kzOsbdX-<9XopDO?lG|z(xe5+HCy_;bC3Fd;bLz@0Cc1vJI4O z9mk@8Rz+WhV;^Ls7EX9UT@W#2($xJIzi_h>9&8T`ASUeSt4HBORb4(?kOYot0vEJ@ z4`4%5R$AIG6gJKf9|DZgnm@}pqL%}^_38ipL+J>Sq*@QfJxpf6k2Ua*^%j@d2##3b zG748XEM0n3-2JtGT;Z6iC#R$;7wtUWJvAWbyP4!s^^`UC%D(wY&HFeo=__)#f4V}W za^(S*#xjDsY00&H+j%-FCiO#;bb{F*Oka0TMr1$hM$vDjhj|H~(*1^WJM4y+dVl;$ zSW61s$CjbEzZLaY>73Gq)$8MM;hT1uD=rqg?r82M>uS6s~peont)=EB3l*EamZ zonOA9OnkORw|k4m?cH3w{cP=Ed9$Q5UBnmpl-qnu9jiIZXB)x5m)kgP;Utu&puQQL z(7&+Bi*1g#=a6QyUM;TArB|GX1X+x$M2=~A`7;1)D7?%(ISaKubQMVq-6aVw;b4Kz zrtLSB{YQzBeVn%NNWEB-nquU2iu~dglrr1+sPmvrIk$~@hefcYnumITlZkYxa(_#> zC_Z{ce=Ic6K1@wJO4#1}O3yvd=do(~#@_>pzD0)3J}&cIc!WjGzZ@;;E)2gV@7W@K zd#hL(Yl}R1YfCBy(4cUv`mv0AfkhVzN;0mR%8uTugM>#}7=afN)P}{HTTj5RHTK>S zf8wGiuJ6}X{HZwG(&gkat-I0y#-BG{a8yR1Bj}!dv}Zv>9gGg~s)w&DD# zbs4eNXBwE`4R;W!Y@GkfsN1KbC@iL1zGRlhJ2$R(Fy1V)pVW5M4cklqT0CWZCrdCZT_IukX3H6&~vPabG+^l3DID;z?($VqfZuYG=Q#&_LDAZZZ zJf(N2rpjBu)NWbp=?&(2Iw{{)?o8B!E=-WmwD?b&?A(G&61orKYSL^%p1eZ`akt^N zQhOeae>%>u)szO)(Lkx3uKu|+##4CC9NAa<0qfmY?zH+XXn>;9HvVZq}Acg(z+h{TZ5`@ zmf1jcJcEyyg9?AS(K2tO@6@gA@QpehzSi#GQG9a#_Rjq-jGCRAWqtAMiiP#p>7Rt} zMvbAGADyvq;HTzB3yRmKSI<~Z;?Y*x0mN$SrIXD1H9Hpx`JO2^JM^$nlijs`0ckS7 zVaj7oR)QaYP{GLTC(8!`5* zg-N%qX-em3kMeGOkj)oZ@0_oVe-}eb6$ocz$`{3ccv)FMy(5qTq_*hC@tq@EBD$8G2E^--l}>MLJi7YRhTwkuXo|p zn%!qH@e_U++1R?S8AG;H?^J%iUNZ;~O4zQ6f03Cl_sP6$BG`1CF?;i^Ms+Z zKy$3!XwATURm&7hBjj+&8#6kUud`5TYk&F6^4zm7LVSTNRXI~e4X2mpjShVcjnV%8 zA6*grf`O=;t%lz|^DN9dDSwK8_|j7-t|#kq?w7m$KV?DeYgX9eS|cvEbVaMQY}tl3 zG@L&yM5B8av$xJB=3RB_o>yBW+?Y^llX2WQzRg9P>E~?rCZ7Hs+kKj>^oa}W-HM&` zP0Cp)gymiz@>z*i(9*MZ*zKp9(egWj_?ppWk8Vb@}0Ry%; z-SjyOX5-G zRwvX@VdB8p?VK)+ZogZx;ZHo$q2cZHyE{~0j+kiZ^|`9LpO<{GQKyt5Y<&2>h)tHC z>&9|aK%S1dP4n5&qvDo$8>1c(#nX52Rus%0gI8yJ-+P%1qsCOe(BRBlW9kgG&e^74 zw|*T>h)dcCpBo_r(LoT9nyL8g9^ z`8B@RGQL%JR*Fh)`?`dwYd?`Y;nW?mK!2o}#GhsIw<{H>?U?nweWE-k}LzRca9 z;%EpgrZr_xjM%RS)X9xDQ{xaJhPI)bch#piB$?|Prp><+XD;B$V?^)RVkuSP>f+by zZKRAW)3u72GCt?HY7C-G?%Xc@KKGec?v{g0rAFbknX~Z)u^6A0-$wUzzto)w=QY43 zBwRQyZ~AT0pR-8F^%_sWjpP_DQxB&?jir&Fd%PcGEL7f3bVAV z940(^+$>+o5G%fY$kgsG`E8%zDyT>zxb?Q!YRfgZa6?UdIOl*d@5<$1wsNyc+b2n` zo>s1*MH&(l7s8L`#$!IYK@It-%D0ywYL8>vD^_B_B>-4?a{{V>AMG#;!jiw zJe76R#ftfNT~Ei!dy=9##a+WmmTO;iqX-d-+Dhl+#kTLQkkEX z>f5t{0n@w%4CCl0k1Kk;U-=kr$Zct;KV7ZxtbDD^of>GD_=$oiptz`dT68e9jpdG; zb<+czC6k+*bgg1z#*U{6Jmq z?Qy0&DcesMITciGS)RvMyvxKusxOL`Gu4%7kgINMbb;IC`sTq`A+I zT5TPZbH^VSUh2B5)#&HfAEZ>3KUt6dr-_Y_n}e zoJ+AS{bw9d5 z+wTT7$9P=9{dxb$l6^@Uf2I`ke$m=ZJknMDDbsnY_8J9L=8G+9%|K+ z-#U-vhK7O|a_7xM>XqIPe(2dw2=|8Lt4j}r5$j_sbJ8wIo(cCOKA5-lFVl#i9#=5& zkKKD@$)-X@)KwPknAcFXFG+Z&H;IEw{;c!T)9(Zwh7g)q!}1 zvU{%u=;yTLG`Aa7>~DWu%Gm&W_C+0?hroU%YsYaARd+A*;7^TdX_vB;ja%yBF=_?Y zIW56#G5%$Bv+I{yv`_0KKO*!C4GbL%H5lgG=!OQC7h=XB;8^a>)^iJQd*Fag3ki70 zH1sOl-u|}Q-{W}1HR+y5_F{)l3bW^{-Y)pcb-i_Q%+OvQI>N1n0uL7|D0_Xk_R=p_ zyXchjmr`k`#JgkqXvIf zxRYt{g8#eU)?GXJfmZ6k0dv(O^n|(8z_=ESXg&wAn4wXa+ete97w4k3xmDR#-Bt3F zOx8+1I=^b-uPoE#JO43y?fi>0I=PtvlNsL4JsZoc>d;IbvJF#7vWiF^;&(wPpyX4|cp zCX42y?*4s}!)7)cho}5`tSf4-4_~;eCLyG4-FmHwMOlrsKWCw$cxb%DWd6AH^5t7IG(WrgBZebY z1m>k~S=$fL9(OwO8=Whjnx~rot8)&Q>ybS2@{N;Ll>OU5uhu*qS&YLCI&FORb{Tfc6 ze7_`O*Svky)BJ#eacXh7LmZ%Dy8>+@U-k!vG^+x^5Z5$4ltWzEwftKReYt^evn8Fd zQ>(b0oD|-P@@PfY*5tRK=4wV_KGTfZwk&vttug1h^i-^4I^ zr;wfHI=Q&qt2%1;{5Z5=D{DK&Mj4H>L&N4mD=nc16j5?wHi&3)X*gOn%JYF6RwLva z$E1e?ej@z1^LM7dCm=ApNxPUxhs}PIzJJxDd&)Uj&v&aZ@rzN5Vv?7}?ALzW`!FGk zZH~qAP~F%}xrGIX%yu(cv2{AFTa`NkH%DWy)wIk~z38Yd>!jc5S!ijw?EIADra7Io zLh!u~ZiAWc$5>9doh>iAo^pTP-lMqPr(7XwETQW^Xw}*XH>{;PO1t1){P0=ty#2g) zv#Bd>DdKht%>%hzN}6#kNM}j&-`ldRZKaU8c1vI%4uXg$5OL+jN$;@^6azRwr~Pk(frS|B$tu+ zg~7Qa`6Z8y$~&Y^<2&52(m-5V<7cE@!{Z-#VGm((pl}Q4vx63r%V?&T1E5+Ai)ol6 ztBvv2nh#xj{?|R7D~NSzm0J=#YtrGASZUK{=HU-38Ly&+E}mn89VyElIuSMJN9cCC5Wvllo1EyM?hFuuFQ6V@Vm+8VP;l;dZMa@@G#WGmp#8*2ap5=uBZWV_A`+p{M=|%2< z6%k|Uc;M+PulXNA`|VG#z^=7|n$dp2^QsnhK6`%z%|K4JGvu7PX10R()Bha-J>kXt zb2w3b3dowsoWog}DrJUMJkO8wTn zmzA9UwSS7`q}A4n^*R{eBN=wL8ryYCI&>Ggyr(IBt{@{5Fb--4R#$@6(p0yiW~(W( z1Yel`#HhKHAa&Y!ZOx;=>Hq8#<%^MOHF&kM*3S{tQCOw6 zf~t4=SMQP^7{+*3mG|#a8^>SQo8BXFoxX-o2UV9UF^jrJH)^7VBa zS+7R;5({q+eZQ5X^723P7zx=^>#nYqwPK$t6{YK#?;2>2M;=$0ZB6)oA7yv7b9m%@ zxi+`t?eSEbV$P0c(t=V^k?rd$VgoP2|Mz3X;_e!zF#8fYz_~OoG;92z-I*(G1C>%S zQtOg4-lG3?~HYK}^6+oY!dOGhk*O&s)p@`gIoZp*n$C11A zH=G@FHggqCMZ5gB(^>0`zf6_djVzq$9{pl5f92Y4Am)SNfj*;kO8w#rRZr zKf~SS!3Ghxx8zsM*eV-Dis#=d&pId!)ZtqmzO@+~>SYvpD17M7&E$S1aN^_ou3QKKee~F6@eb5$)PBz%aJQ&l%f#@h*!SG_txP zCqz)v(X{4;G>p@*6Pr(88wO7dFEBBOWiYv673igj{tv?Tp{H zk2fUZ=EAK@;`xMk7Jo2m7diY{0z|Cd&BQD=mgxoJL zrRR4b`iO@bse@EY1ykQe>jY) zK%mZ#BA4B7wu8P5N;r|y;;hk3Xad#p#(VSfv#%DcH0$_Uvhr|6l0>_AUT?iC4MH(K zp?vaSS9@KE4Uw?3GIPO^WQp%V8FlfPc*T2`@y)-tazGV6FX4*t}v#czgrTd zAl648y#8^0$nGL86h74j$5=)t_3A_FwaO@|ziB}1Y?Eor7pwION&b2~m5 zo>rZUYSLbABwD5G$ru$1Ft*!9$t-F?+W<{(jO^@f=Im2c52f~^w5V1?x2F?O+2ib9 zi-p3TL7mgRbnbzTEvgnTu3AdX#EvfBIWOngN`B0rr$;0yl<#%!q5ymNPF}{HhLqnr zrysg`fAqhbTpL~5PI&A^!CPfCyY5jASXbyb_ZyRTZvPyDl*N5~fLgzQ5CCwK-6DPf zW*P%36|j1<5}sCfd~Fam7Vo`8)41*2bI?9Eu?%&lP$#)HlSy9lqIlmUGQf#YzQgx& z(newYhQVg)1jE+0Y16r)2y=Lbj@re&xYkLW3TKHKoi)7u@Q*k~snlsRQ>;aKyF7b; za#a1wNP~!O=(CCB*l_yf`$ZasJG z4VApS!p;q+3ZpKsHlwK*5``%%?gPZ(_Y-98qU&^uJ?mCQQVqRrCAg65NAPBe)Jg8! zlp6<=Tn5eOw<8yAM`77;!79q*zz&Yc99+`i)+?LpJynYB?N?G^=o@=Hj->+V!ujI&+dkY> zHV;tlbQT6o{d$Tml@cQ}>LSPhpP*ec?dt#-WzX{FU(Y9ZgEb0A2Lb!wOuc4UO%WI@$OD)A?`_LBZ*zQx5=B^D8e` z?SV7mYK)a@FHKlZNBe%Ke=8T`W@#&%>h{(}-Xc(8dKpu7oR&^Ma91RQ+DMtNx%Y>V z%=TN71^00>=%l=Tep?<3tjEE2E>dkTEcDi+%N%NRpBkln)(5XBergHG`P>~^PRse) zZjvI~q3r3o*2tvCVBtW6v<>1U6v_m}d3X!e!|8^^fT~<4MyqX0-psd8cE%~)syk0+ zrrO+Ina6rwSrOTx2U_gymp{)BuOgHN#ib-6zj46$ZsHmg@Se) z-VroMtDG(DcEOu6x#?7OtC2CzAd^pjtl)jO`a-7JMSm+GEB;_G^TDh9+k7yry!1vJ zC9g*ah$X^e$TLYZpf@x&(2Lur{cJ^R6UXl_s7t{u;~^4^p}@k|Rw; zK*W>flaGzv+^mMN-8ZvyOz}6k3Wz`4EE|3VC<8eU-YK)7zMfb41Ga_6n5_RP&N*jU zQ=DGn10XrL2b0e!x;VH1WnM%W9)ha6=$4>qt_$Fn{jQx7&dEbn_ zkjE6oH5GLLMK&i&UejNcSv566S65tN+mD}Lfcl#rST88G@X`sVM`le63JMM)LRr>8 zfyPe^BzP6s7Uh9qR~7?h1g1I`VUPk2j7W=-zyCot-5$r>zX5I$JF@W9CyPq1fpWh^ z8Q8(pqO!*_n#Vp`E1i$q!gw7EO`3uTeznZOrD$~_~WKbAi)D`1@ zz*w^@2G|q-^~1lBXb^^%yBu6`ALAZrt?X>#@pw#!fU+fTYNMX5=;IO>&*(z-&6LW- zuLIs}K0v_5?$t1JK2SN)a}C!|UJ$Sj%6u^0ahjj{1{Cag&Bf6r`CaWv{dq zDEy_q&v;t=xH{IQq*|1@LVT0z0sR_#D}MaPw>6EcJ~!f6@hOW=87_gMi1otS6;+7Y zGOk-)RzwVdiF=Pei$oN6ut4+2hPqjlckW}S_qSJp*JM6?%b>~YQQ6m;e_(Lg{eKYb zDa#35nb3bB?tN8xWgD}}iNCIbVez?V-4p9#G*W_-e_W$1IY0gZM!EfVZ(O?lxV+4Z zc#)XoxIR~#Sj%o3fwxW%NRP3K_)W#=#drnP7d6)5M-mDVet$8zE|#MPCYFdbD7;(4 z*jDM_Z}6LJ06*W2aLc}zwPp#X43Z?s(tg8Rj8022rM1j%@kS?)30)J)X1qTy2tn)G z>;Q5StUF9|7mcsw7psmjxbn{i6;~U5!V4=pxi30n>Gsv*n8`N$bb(;p(VqRMsP}5C zi=af6{+FyY=Cd$mZhi0(h_bVa%()wVQ{3qk35SRzTM;b2BTgv($P=^cK+S}FD?+?K zvrVOUSCvtcgjDw~_joH1+{AcC`4wtPZZ^JRl!W!$$Fh}GvNMl1$M8ui4D)2Y71y9u5>YSei3plD(Rco7}u3X5860H?Pi5{ zTZo=_1Ey$G)=7U@S*{{l#9c&AP{{X&QLdKexPT>*EPvVB$UexV_Y=JU_0VlNkuFMvHuuaO_Hd~pjrhD*x_rmI zXbvG(K+$hentF*@C~OXGPa>8Wg_ia7B{O=%f5MY4_-!ez$2^q(vx1JAyWb^!msIBKC?Hp1yu0ET@U`sQ!8Zs+4< z7bo_%d^353TsFGV#-c!zQhwybu@WXuLaKL{Fq*RHF%n1&I3G2GNltI7b7RcM!ffxQ zpw86&Lt&#WZqps(dBgbzK8Fc`Aw|3Ho1J+h$z$yUi$w1SdvEg@uI5OL1w9vh$T^eXIiP+=H8AT}nF~dj5sd z55q?8?;H(DhIjU-@lXo24*kB(Nvk7UqT;wukjo#IY)}i zgCNh4S<4b4k$y^70%ca7jVf;pa{hO&b!(~P>O|#IVoziYa{fzqyFtmUe7?x{L7NJ} zhKa7AMbe6gepK-ks27)`)J8_Y+3-S?#At^S=vbWp74E=*(48l-18m^KioEFN<7KhUCrlOMYm`Fll)j;-CVnC_>qd7_=RQClX6m@%Pb%2n@W9r;L0x7a zlT$0Oi9f>D!=j>~k>+y$Qn_5<_no|g{+T{buc9$w@9~`E6R%8z0%fNs_&*pH_=j6- zpJ{(AePJ*&ej<{uA&}R@{mrIA4eIF{b`fc!o#b3H#eHo!s;~i1O?JAZV+r3ndVn;G z()=#}<4@EIb7EY#0LGqXi1mX09X0^1xdtH4{0$LAiRX=jlawqWo4&gmrwgDK^@Hxv z2bYGY14MBJE+6?mGa-XrUfJX2*7j;!!FHMP?O)L3^=JvVbn|lO?su0tysmOl8%V@} z+{va?YZ`*joO*N8{9$(1Hqi}GeQrur5=k_LBd^|*u20hyl{PF~56JcI8jTpY)U^dc z_Ge@Sw|V#VmtZdg;<1%Aw+q#oqTl`fKw~bWcc3f%)esOG zR=>Ed9gFOBkX@b1t8DtSAs4l5T;I0058AjLPg%oe!af!Ryyw&S)5bj`_bYMx=KfeA zkFUHDA;JL#^82rBd#dnkjT*kFL;zFmL9D&Ll&$EtTVpEab+(KfWpXt2xvu^`WLd8s zP%5l%E#Yuh2EOd`<2|sx$;MJ<6;y1LV%c{T}VD+BU6B4E_{%tp-s?che6BHh0)Ee zur5EXa=#U3&w}jqtJ^G<|4TMTw=zhl-~%i7(xE{Cp0K6a7NAhA=okuEuh{B*-;!?N z|J$f>Vq|a{7UabEeI&MY=DX5k0G7WkNGQ}h)kpzp`WR^7uLoHXc+sVlYTBzRAs1P%FigLuT* zXJlsviQhXDkG(UwnlA-+)s3$qy47ieZT}`pXtQ8oalB+y<-7tRUn_`XFLkG^x(iZ5 zu{A4*kOZsi;_q(P)V2y+JF1I#@=d9?cjc+*5RA`N{5C%OExKq9K=AfOzOz@{+}uq{ z4q@m`o#6A7%Ez&s@+yDkL7;@zZWDpSe{&g->dWhjQ+mreVL{gp$hPCIb^*0 z{F4KRHcs4hzPO7ivt-flF7OMkqoh+WJLh;mSN*io=tP zzrV&NjpVe1enDb#T~u-Ao+P|^dFt@y)v zE))UTt3T#hIFn79j?-^yk$^Moo0aj2gv@4Ga*B#|ifWyGO4vF;iMN$fWzbQtG4+5$E7-ln;r4Ed<2t?X@2 zYB&bkerUF)!&KmNUV3Ta*Ix}mopk;KVapX*yq^m9>vSj%Ck%YpVrPPZu#?h;>27E_i|pzhspl8BE)U6dMm4(`B!aY>DKP z7ykVZ(l+zW3F)o-j_bl)bciokuf66rOFZ0|sC4n1ajMQR_GksrlyYn`tr zHQb)ltQ)2+Dpt@qJTMlja;tO!$f-4Gb9Wg%mkxy^lq8^$R#;I`wU zBFhvR6#>;34-TRX6NL;zP1j|#MgTniAp+uR)EfGd6a!fIWfpns-+jFC&;`#bg>j*b zVpY>e9DkoeuI6rU0QFILTVE)W!lqiwJSLy6t(NC&wN3QjXtYVA2Q4VtJ(vN9GH<{@ z9-~DS9tJo(Vc3LzPx|)V#Bi!8PEEqX{xxpR&(6n;Bpa`4g{`(77q?oFtOc(i=L60o zob!%J&04>NRFF0Hj;9;IQC{Y!^n;4^?@!uftMu*~&)w|$Fw#l*?sdNaDb=ypKd7}; zL0TyGV$_Gzxu%GJKqcnQofF{emtKHojy1Y>bx(d-+uOUlkwzBol>9MKOnbxP3TO%kj*y3i$k*FpSm}v zoE|Ur*$m%b(u&M&a-vSVXs9ow_w14`U83fV<{RQX(h(R0OYG&Xy_WHN{^N4KX6AHJ_zwVXtwe6aH=?RI@Miz^|BxWXN{>xRAjA7dCn7cB71g$CmGW49Z4f0F)BN)j2J<`87c2)q;(@9U zF@u#?d_LpdzqhN*EIm~yY;RSF7>Y~OxFr0w&@IY)`l;HyKXT`}j`GTW8tocaD#&D! z^6E2vmDrn%RzWKIZ)SLS@ES)lDf(g&d0+Zr`d1WS@wNulxqX8>s&XyvAtY>2K;=WH z2xIYC{Rz%d7UfyduNAGO5#gXy5-nqCR-r?pg;!;D5Pe#1wTV~eG+1gNv`fBJv2$d$ zzPCz2%qRPIHOr?tk=;%9M{R%xs{@$_0NkI8Ru$ppt* z!}mPiXh)q8t9|ZuL;l7cBUnD~vzWLk^HAB=(pF39)y$~E3#GKYTHC{#w6mHkS6q%#h4v>ZAUnL-EZ~Y}R$gq`rptH7%RvbxNm`R2}QB-^$y{ zVK3d1?)WIc)d{Q>gckNP-VHXEYNX8_J7q~$h%{O*Mjym!Sf(dbXL9%z2-(>#&MYPb zH$BR?=tT!chpjMkQ(z&Ch}c?QX{L#s6JW)D`>TGJl?{q4b@7XI&i+%op9L{t`>Sts zgL~WYX_VjD5g=qJy06}wE-GWz@ONE3A|;n#bD`0I1wN6*00S@63b#gpdqVkLz`(!N zOOI3g*8w-d zG3nXyjwEU&-u|duwO9fh)19jaBmgm2fJf9|n)Q;|A3*#So91b{e)#`BAsPTNixeRV zR+=J7?y6F#A4+Hd9@Ea37R0`jqjnaKPJH1f?bzQCPeAa;^V* znmo#-`H*r(lh9>V2K9qqHv2`A@aAZddf-1T?tdX+ee!y7>?ZX*<~3NK0CivW|MrV$ z%?n`vJ@pf#S|wH}lnK#LA09Fw>=+}R)$N-!ii&FMHElZ%HasGyE zP-lk(M%!INLNrACIn(cuP6{?!Y??jFb;a~#1f|6*eJ3;*^LsVb6agNf;BZNP>*_9q zdB|VJ^~m_F#IcVw-CcVLO*iPCD4{S~C^M5F8=AyAhLhkTskw;fWS*V~HJs56uIXH0 zp_%0{6~Gi(M55k10wkGr3QRctqV8Xb9Fmof`I?G`s@Gk}U~-WYzh-dc*`Zgua?T~X zn0L?^JHPgpBt5Z(l>;LG7;amglr*R&QJ4V~s6+kV3}!xqOv_23ZTZFPvpZn6Gau1j2q97%vK%wwwkFn3gy zZt(T{A>dQ4f&PSZTqWh@0K2_(-0~dBlg#pcBGQ_)ce>;2^BaxWb&_-E`uUzJGA}vP z6}ol~>VlnL9~iygn{AxWExYKY3K*8%tDI8NvrXGIU`IijD4n7*6o}Kfp6R_lIeDG? zHVFuZSMVBDSYWGvC(-by$Ps$$!&=vQdxu)Ae+H$`zi?pUX49Wj6M_|wQ%eBtb1yrr zCdyr9KBqME!qwYZa**3q*LcIT8nw%S_Vs-3x5OMwezcXaBBG9Ezgxl?AjI^(eYy=X zz#h6Nh-d*+8Ao6Hw@}uxYY8qljyytH8ja&+@Lv2#F{$|RBrD8*J#4jXYsBb=vw0c0 z_QFWaS+K1T!8{w>Iq8@uY5MNLfSq;njX>_6T`N5qH)VpBnoP6dd85=EP_av2n)$Pm z^$#5j)1O{;e9q|=ifj{e-QPVU);Zbp(Kj#69w(dm*Vj1%HR`fmru)|sdI};!U8F^~Fg5Q{~eu-QMuac0J3@ZAi4pPw! zF?{J_%#~)m)l3_1Pz{qpJ?%}>f+=wD8=!Zqq*jGYsoQDj0)5&mG2Td5H&t3bAXo#;~|7NDCLfgUfLV=ttIR5X>PZ=T#swikX%o zRLjvwYc@F4bFf#u?b`=rWzVh`y||fJiWbWDi!07~1DV>BK=m0SHZ$??P776E=L!jL zxjr-TjZCy7GI-you#(6c7Gjn=imivJr1Hek=cdBk8Ze(p6OUXT7qRR=Wf^Z@NEofAGJ*Y~UmFmtpz6XHHAqNx~Q`}TDev4%hPW_Vb-n#qOgcl5(~J5SX9`OC)|Fto@0#^jx-ydl2#K3-Z@A2y3yH(f?7&>MWD2y5T5O2(gay+O$nzjpvmD~;@q`ou zh7^)MKiwQog_cR~BMLSR*h+E^9dQ*A#W23e3O_*VxKNW{_1WZg9+emE`lAY|o{m3ZxrUWpk`p5}pNRjo zC#d&wURw72@;9#-&66rFepy9#{iq|eT-t0`&CJ)$;iZFc@!i;;LOWkxek7!3Dm2dA#cjt<(i?H9aF_3Wot!V)^R(G=_TUR*HMO`?V1~~B z8Jh2s`P;a`vxS@L&b!6iYDbv?&s}=Rx)yTln3fIp(mv#fGdvBApsFMU(-k9rUUQ0( zZt&LBkwX1>8NlI@Tw9tFLpLqbEYMI{%Miy%+oDB7_3Lq+pShr7Uo+?I_~hSa&S`Gi za~pLozL0PES#LcA1A2Z$&|m(PwG+7?u1=Hf9xweOexB6Pwmnh}KTIb`+e`YxuCIqe zfVxPHhQ50HUHivhMFpRAzH8*?4}aJ^n+1)V4Ob{VF82&lhar^~yGW6>tK$^BHup2p zaCQNyx4b&GHuL2z{FEl$@=FeOM>oekS0$75F0?kB5ApKvpVtX9MY~a$u|a6h#p5<& zRF0L1O=2AE8mrV6n!l@Jo}<}@h%RQHoViHub0ny#3+g55;nC3lCGln?6P@3J z0(0m$e6|G=K7_K1x#9a~A8>}C;xB((lW@|?gjBXBnl&_YltIRJdT~=1^${_2ytA|7 zAOnkX2O}5#BLcoq@0S0Wq?_1(7c0XGa|xJf_qsSPF_L)qO?%55HDHY8Uy&}r-ry1A zCy;a5XG*()NF!uh`1M2=g(Ae@h`-`FaN(XeQYIbEli_s30>aM18$@;uTt!bcU{f#K zUvfr;_h8nC3iv>b@;DF%xcdFIde0Cd=zV2^xx>=H&jj5-eYP`y3oOJuZ+L$&lS%iG zHZNde-4Y}_VDR0Y+rUu;8E-ojNjGA0SO)aj=az_s7p9nJ{`#Z3DD#3%jd*}@&-2RPTtq{IqZNKB?Y2T7VztOPKeU2@I zAUFkyY21E2f=rJ~p^~t`8qUgj2?Ru+&Ky~N|GG%y6Nn64&#X^DT>mSbVdans>AimS zz!vX)o0=_5oUUGUhmG#;wZ1yAF})Ts5o^o*LF+l&iu?iIS7*E<;lJVacO+1Jrmfiu zlzPXib1Q-P_QQZu5y#1LD0+oe`w|k;jghbxN>-nt{m<5F0Wr{<4OIzMj!St+31kF|dpGjH4$loN7|8ZbI*onSb0HK`4m97;TQjwleV1cp z{+<6`WBSpEHci;&k3$8v2FyO?saxLXWeyx+e+Ipnzyq_C^85S40YT6rpqBZ=1#Se6B9s1G=V-_6YN5oswLso8@? zWt3$cGs2OAp;+2Ptf`R@qP5Z5ZK5BbgH(pb4K4} zYFb+HpK)~_0nx&XN10&v^H)HEVm>HqAvi?+)L54psBs*&l|r%D0zpxW2in0=0!51N zFxm4NETADss;)_xK>YEU>#sS@$wo=?1eZ5HK0Y0elfrgmWmw>D+DMM0YnWd&Bb{b* z2Hqk2V&FA?RevY0vti`;9f`GvUXX()NpbfRxkD#mx#VzRNsgH#Cn27kr-4X;l-n~{ zI!)44aNLkqB6sgEyI}$eF8-s&Swz%=VZ4L~H{to1uf(0$7)XUf3?GB;-S)`reR;;> z2hK%!Yh(GJIh8G1T`yGzC6W_ke_fEYS3{%L9 zyxdWG)zjN-3mgeO6#zF`Cb*s+h-3$Bz?Y)tAUYCLqy-K8$d9%A^pzMH_}QQ%yNx9k{*k>w6ry8qB#?ax2I1KQ|;TS1sp6VqGpf$)*z{&xF)2LW!W z;W#5->@N=5hw9G(a#DZ(c_umMg8fw=47;a~ca3cqTi`)lQ{BLsj(jf!U* zd;XWU{ITnZnc#EZN|y_XvjN}l;ZsxGWZwf$JdSF+Y}z|khTP*`mz?>fSV{TWQ(&E@qA$27b4$)%6L zSxjT_K01&WJ`lj!QPA2}VajiiVHy+WFVN`cd^ERuk|7AvFv;h11XNagJyQ!07$r?@ zBEAvPFZ}f{H1rh{V{3Gh6FQVVlCrj}O1Z}d9U7wlBQ1TjHPaQZeGOq0KfvD>jJ9lF zv8@|>&jAhLs+cEv%G0>xEDwr;tm%*wo=u#reaUW_C zLNGfWh`fi<3p-l`3Sqw$Ilvt0<|MWf{=z1iTY7JQT46xb`zt_R%%`?$Z%t^v?OKOl zzLY;w&t3^_Kd7+cpW;)_useOi)_4eav=Ls|AHNB=WK0%vriN811YrbSy*gsvCMbMoXuf2M;eg8-(ChC&+>}NLm zwLU%@bPq_wmY?qeWxw>R2HaTRn(DrWx)(Y4c>KYa^A-xo<;KV5gq8r8qwpalZ8RTK zMmmqNzX7;37D(uT1I8+5AXEp^0`5>RS+R0_l^A+UCW-H}#Dxez`y36?BhNqb@d?F{ zJV+=f(vVojXb8>_WAxiF*Y)-FBkX<&KdDja_P@ZBq&fZl5Ru&s|FSn(zRBvcZ)xR| z36Z;?Wj>(G$hiTQewuC;2vUH|Tme1AeULjyn9oMHrDGEvifb)d@H?qIPzrV1u-^78 zQ1{q74j;rk$QI%50mYy|l5-o6OL_F5T@+(N%b8+iG$kVQq)+cVoxyMJAD~sBAGT*% zUu)x~PxxK??N|GflQa zK_X?3i9!lx^L{*3=r%}--O}ou{t#?)SGA%Nd5QEx5y|Q^sECQtfkgz!7Eb!>1A>{l z=Jf;@mJ3KYx8c`4G&lzrjtt=^7jd49({2?LY0E_5@*w6wHhikMwl zLHvdLIQ=i`nTvS3$K2)lNXc`)vl&ao8KHUBNxOnSn<`EMG?bFsUGi7DXtH|oHMlS? zG$7S)g&f5m%I8!f{^aaJ89h1nQ9SXoMKX3rFbMH=bpmadipM570L$fJe5+tlW%>gUXu4iXQBEUXe-i1u=(RX|K6V!J{5^(a0RP1DCXu zDljX^cFTm{(k62G5!h$v(Mi3pRCt4}{%cm|YrF#1%mh`O`=b0`s$#{p)335y49l#s zjfn8nPEKMZqyBv%ZU1J|z1b~h?(=YmTX-nNR1s4Z6<+47M;**ITZ=zU`v0aKw=sDW zhEp-WqB8-&_bcj5P|f@0^^G}VU_S5|stex3jSQHyf2uvzoF1Mvbi;T!J-%6xKT<)o z^nMgm;a7vh@Py>5bpv987$)=c2$QJ4QXX0t4lRj0Y<>(L;Tjrsyf0ucRTpGu-H#B+ zgN9(3x65O^+eMK`x)Y0|_H=Z2Di-K7#xLx+9pgZN>2vnf@wdUJx*n*&q4pXk=2v@wG0WR>58!=oK!;jn6@4Lv>4`~Fp(8?)Z0UH$noxOm(NYwUQq zv`?PQrm2Pk45BhWdA3Cu)ghW&jyN(UO_w&r-9;=fT_%iIFr^r=JmQ{ErH@hLzDG=x zdcJ9f*3gYCO)yKIkYQp8?@IWjzkT6X80Je5bLPkN=BfQXHh-^cd&X)@#|nRjHDVj} zttd$rZaqpSdgw$I_xmpC9rbvmH?J$9y~9e+htN+yxZ+~J?K_jgjGveV?Yjo=Y6ZLH zOWH$+8}$VzRI`#rs;rtCqv*`Yt|3;R0!8F?|AMvIP$XgS^m`E*4-b3tA%dysl5&pn z_d)#>+vltuzSpa6@f)UOz}^kNUF z7cbp$_lDclG{uIL32WjY_)#6Q^g{vJPwG=;r_QHmYgwD5r{EDyy9}h%&kgPEMKh9n zx7%Qbi?K@<4dY1rCdx(mN<+t7$Q+M(E(r}2&6E=Fvi?9KBt!+(={O55PL;#1cc|E% z(g;1dq+a~sJlyzzI>MV*y4Uu!P_(1RH<&DHIAa4w-T3}Q9*;fYq`6~3R?Je7&$Dy$ zaIaJU=A=a&x!F=nQ|dW_y?qxa^uV38&8!@McL{0_M{8R z?m%Dj}_=q?0Ek}3!1&cc+ z;5(>lOuKa0CEY`I!yh_CB;RZV`f0z%R(Qtgq5kkyL@l1xkoUaYN%7`fp}dQVJat!`p2OJhDo86_y;okllUR33vgLl;5@ z)<}xO)t{6jYqsCSHkvaPHxjfKO~+9;!Dl<=u9L2*i)kqoX`q?+=f#->b-QP-^D53U z`Ok8|1NMp0Z1)KhpsNMh(O=Ji{+;vh5fJSVGV@FMo7z*MO#A$fCI>w*tMt#DG8FfZ`ts18&}G~zj?HYe?YG38dw%b$ ze`=j>RtzjZ80O*OZJz6A%tu3Tv{D*_?wnojU!brQWZ#Hq)y3tguOU_s|2WTcKHf~z zLy0EuXFCdN_&+eT8$5qE@mD_x>+I0){nO*kH_!pZ>W9N5z0EyVP9%q>MC7)E*Wzav zV;%jTxh1=I+2kUEDTb)B($%i3X7pLE_{u?lTdEwQjA2YSETW9Fu#3AJDcGQ$e$os6 z*L#*ZIyrl0IZjNc?SELn7yJdT-j@I|DgGJT<*%$=2T_qRw&R4O~XXM!qMVJ2-ZZRSl)Z>kt>z@gYa6t;mQ8#Bg6oIcw*6f z{!YY7-&cLq_rvo1I4jy4$is}2Qn6sl_b6nQuSg3%iBfHD(`&0aayS)SP`{%0J_Tj0 ztby<-`M|*9)zCr2U9@{FvGKwzr|aZysaN!_4(rQ&O^$yFV}?F{Nb&e~$Ipmrw;sY9 zkLx3C-cI>UE^~Y*x<1_0)gQgM>}K0RO=_vPVv%iVQ#koOaU(BN6Gh_MK0>fELDwK# z@;-?WV}+0z6?YC&~rn%_}wTbZ#u;heWkCa!xBL^ zbH|9N>s?<5H?CVe{k(k#uCy-h)zJS*|0|wz|FNN=sz_%9(OA=gZtmHF+uUJig}kwh z&wWV&|BH69<#5{^eUptlX2A?ZE2~i1>HFaUlsjHyA|YO^;homodx1iVeOzwx&Np?1 zLSpW2R&!QK9mh~TaObHz85DZuJM-Bu;S$7?zH$?FG49_X3Igr=qy|!&7V~P{46*Dv zZ-dbM$uS>hpf}MLHu^H8dL(*_;$DgMktm}&PTr^ThQPc_zIut5FY~S3^s{d?&*DWhKghkPnU>!az4}52ya2Ia%HL}( zE7S{fU|24QL3ZB3L4J{)ev>^ptQpgyWxS4J$hvsMEX-vWV1Dg;L6u)dwVYDsn2)Jw z^257k zhA6K$Gd)cD*qNRCiUu6s`{7iK%?fK+3vu^5lWZ!JUNE_5=ZJ<0Zquhzk=A~0HC|$* zKfr$?spq#Hq0W>+;M-D_6?>C8j#8R^hdFQey1PAtZ>6s*lf{fIGL*$TY^#jd8g2nSu33pg(^Cd3qIfsB+TfXN*tuh4DG z|B2WZgh0|D6LKGL5BRTW1n0XTL5RHiY(f5jB@+TtLd)Rn_Y))YebkOo-NYQTF&Cu0 zLxOZm6v(sQ%hCwj3dQ)iU432ut(^y#W%`)K=;RBG*Y8Qd;WIPwK z-hmbSGL;Em2y5Tbfb~291}5ujbwK{_q)G(LPrh1YV_D&ZZ=Jy`W7Cal`eWz8_xeg= zj0zu{-50)q7si!^reG3G4a(i=p5WH>Uov+o+2(B;7Jd^7x$g6PMkKXGVPX?>HXJAi zC2s{uJ^T?rifLAOA{|s=`db|9_=6sYAia0Jq$80EyPLC-Xo^I2OLlv@;WIVA5$uEk zz80h&h~-HQjT21Wg1*J(lehRmhqn`XlMJK`SaTbJ3$l98PslA0s+SYuTf=AsNCyMJcw!>FiO8l(W8gbmYTAmqOz@W-s4dhzC+C) zC^{z0x|iD{_jj8+WWpx&5}b=bilPHptMqwDKX>WR1+&6j3DFfKFRpkX;q4O@+ms;~ z7Ftb-s7!jdJBM^%#)8{^Hb51JL#dCYw;JEe$>B|DCcTZELD`oVn|7-3lvv#fPpc$} z#0H0P%*ofI4Wr2M&luq)@{S928!)je4?I8STFRoJRtgyQp%16h{XAT|TmLPvLm=ry z-c?+_YsGnNm7U{olugntzPx)^@!tpQqt*?i|JZYmF&;{g)_M=_2>0t`4(0XfcGQyP!5qRXfB-?-+jY)5p zQnM{Q`IV1;GjHmimI`D$V=y{*ZCM4^eoGxK9C181@#*vAR{vY)onT-&qBhCm zxo|IqXqY<@z16kwa#;>9>DlCc*fE!dp3XNk(b6{5$ksJ%LHf}g!cig;H&k;PtFk3=w41qUtYZLA6X;J zB*HTA5v=)~x7!#AJl=oaej`K&@TX5LIlC_x{gr(f{cj#zMS^N zr}rNIqe$iJh>ZuiuTV>VJ`t&+YmNuTa=o&OTDoR%stl`o>H2kL>H#q0HF}(($N^Dj^fV$AKRUY!Z^4+JJh-yi z{ik10WW8rs?}&Z*1C<;}ycM6-T+y)FirQi~uN<@!^I~Pt^-iSap@324q8fYiCpN6(BttqUx&C7rI5v$Wb zmg;*gvcy{^j~x_-)ghgXH!2c;*8_MCRa#o=zK^j5!T?%V6CHBUERZ0#V8kH?bvE2Dss7L;Q^VVK8m)t@;Ad=t- zon|1rJx%As#+0*Hs-u>I^Ikvup{X0c#w?9HW+#!&!y0#UHU_>(B*zeoQLZC||vH=(TTO3f6m-5=hZseoK(MeF!7@dV+gr z0Fh&qdbvZ%U$c5UW8WUfnbxRI>a)8U zf-(B|BP=Yuk3kZn+7saU*k9CKhKt36gx3q#@pLXh)MKOT-{0ZWq*yk7$7n)jf?Wr! zBu^FcM;!No8{5u)&efG);I_=!sCkuvpS#J275vH|1aSk~(~7Zx>0_2Mj=RGPIbe(a z0DOsE9QOg2-6uo)>c!}w1WENiCAE>r5k^zAX$8c7-5$3Ap@z8+qOw*dWn=0%H>dga zO@eq3nGs>ErE5^0v#z-6MOR~30-^3P=wS8T?k@n*wyA;zi(^b5=3{9=+LzGBK4PjW zV4M4-uLQAFtVQ{$gXIDjBX*w#1c;CSZTKvlYUu%KgIs;1)*OQyE}p6NuFD$0h(nkY zc;(>g(sC1rTEekvQ1-JtHcT5XHd-zsoh|^F>Wu-Gf13Lv7`1T{2Fjz6BorWWBf=OY zeeYAx9}z1g1q(Z%IFe~dfT#?_P=9`dFfD(^9|%+Iuj2zI)PD=s-eHuhozAS2O@796 z-25M5-VKY2yE8`^cf_3keIdp<(5Y;xx*mjmAV8bBnE=o}ClEv8eFla&Ge;R24>GjD zm-XMmII`E{2DU?lI}0t67*4_exbO*ZkXsKid)HtCnjg8jDD^l1OuaGd`1wlBkB#n>HBfJdN>1O1C1Y_owkTmMH0K4QObKnq~xrXJ1 zzY*eIVleyFtw0sD<$c(zF^&vvK-X~nTxotu>AI6;Sea;j1*XE^kd8pvd)B@y0V!)3c_--Mx?Pp zWl(Z%7$M#QBqivb(x^5(HW+Zfpy7z&dVCm-xQWM|C|A z{Lt($udpWGy8$m_xDbk=8^NSp0CWhm{RrrBl@J)p0-Vi%rob0Pxqa2YvWL znvn#RV=&i-(f}~<%L(tiSLjQ*SH7PsX|H40GF={sgw}`9S!cWEUHSUej9~=(F?6o` z-GPoL4^iGa&u3-OCE6k%+{@U%REJd^urB68_}_ffcoV#RhjoXUUPMvh?65xAfP4%& zQhTTNGfOE)t_wqZr)(iofFP_fHi%$i2lmCtOgA+-r<|)M1i9dqKsSsTsY`xSm@P-t zVJkRgGK))eP7$T&UKx0zUO-=fR|0nwVRF1{gTqhMd7N-wSLpao2b(OH|QYcC* ztszD1=&=UEvLI~g+H@(hoTB523de`!x5hmD*E?8DO$~*yxnbouswy?PqORT1fZW0wF}Zl5D+QV{xfE}4Cu zbAqr*N=-eCcevgLO@4MpubxwwT$`+xi;kn<0)+P@Rtoo8I|#TQo`t|rnzTx$zLy3$ z&$eWTrs~zlhVnM`!yj;X2=fp0Suo@%qr-Iyj}1KOFZmm>(@S23Z=To>zb=X)|ei^~Cj_AhcOxyN93Nj(xLH;SiQZ7d;j9V6o%VJB)}}XnWi> zN85Fh{$Pr@ljCsUxrh*n_ucMR>{TiLt7VTRHz0ENe`33q@eyZ%!)HYyeyy?2$P0U!}-6;bB4}EG{)O52XWNAnT#5I zfc2i2LQvCgl-x*%?bZut9hqf?Gkib|GRI48FaU4ftSovD_?iCc0ya<4$EN?%*^cyZ zCEIS5R^WT|pWB)5iC-~}#h{|aRtBjP)Q%^8JO4BvrwC14hyV6}{R0E+a}0ZX5B#`mVoxonfi4$47vvUH>PT}Bdn`9x`W7R?>%&vntPNa|;jj};#!msM zs@3b4Jm=d#M{n$(=ly!IOBHJk7vfz;v}Ohb($jM8GH}L~2){)w)Ur&s4xYKAiuEIC z|JmQbB|`}mwIF{q-01chhl|WmSOjL`@KYbp;=8PT%bVE&8LK?9qFH5Xx<5rzq)PnTE}o8`u+PxWXG=_ z=c1!ev)&F83uM%Y=!Q&wPm>!r_xg+3I6lAeePlyu&C2epsc+F&j9b+05f0T-;^-l8#eb(R$qd3<^<#uc>yN?s}sFo8(}2sYvI3H?9X zgs|Z-7{^q|2AIajm{qMeZX5+(kWN0+8atpsU6cCd)N_{z`B{@2suV*Avnn+viWAkJ zy!vQNg#+;>v6|v~s4sPc;X`ATm&#sG1EpnRC?dYWe{aay=%ISE4s8`?Rf|9Kw#Kny z2_`_7*iFM{Pz6)3Xsx=%Hi8H3YfRN2~)?!-WpY zsq0NPGxY1e-L@J@uJtNSFQs1bbrf$KPMIkcT_fCOw?0piS+H899*&_vEtC|`;>V|g zJ&om0j2y11;Umkwdu^yP{fvMw$__)1*l)@+rF@aGW`0!DEPV5p9M7#SiuF&*ec@S~ zBGRG0l}FBHOR3uBR@M?hF7$2%8SzEmcy9RYs7(-)wzO(RmuDJ%ONKqlIM2Y>^|D} zIQ0Bvfu*^wCN{^zwD@u^?@6BqG@G%oEdmD8m9D0~NQ66dUetj29&HXEv2GFTh=^)R zqx)SkDf4pxyE@47ar{tNCh2*j8gkI_5(~I-UOnD58vlzqw|WB-Xu48sd-ec9%R)Q_?1VTzoEJab3Y1vkh5lESC% zR%0ddWWQp?p@DDGtMh^yR-G)BvG3NGwuNqo=Q4-%GLc+A)Q}AazYB+3o(nUiYBBVcp2fnh}yZ$4|Jy0k=XMa?^qDAm!La2_La z$uwq2sI+Boj_;WzTBFO&qR4&2(Uer4W|j%v07{zLJ^K4;ONMklUI|L7;MSVVa#okE zQ=Eyke68OeRZVNLzR6gISf`rLvBEL@c$+&_Q$!WPi=A6t38q2kvNwLyJq)r_F2OV( zBXiSMk<^<#t2p-$R~C3qKxHfog9A`w3X_!tI;DKza0;(tqf%%-DVl1PCt*s5A!+D; z?(sEhZmziaWc`YG8Rt7M~g5 z{EZRU;Whd9>=%&;7e<-KIjwAXbTaqoFMR!8Zfk&uc(J?QA$_xE5%-`E!rp4pE4X2n z&iF)4I-nqkOra7^b*?2KWXW5tP~f7PkB`++D~s3L)Au|t!h&nw-K1hpIvQle@hpax z12rZP<2|K{olv+!y`8aJe@5m8-FLP>HurPH-h`K=5_uXdymZ!5j#7AdDXdFSw94jA z)(a1&To_Wqx8rzM7b~sUch~2YzV{8Aie@&X48C3mh*n&O6QK=4UcuXsO|x)W@>VR% zQ7A;6x_iB8y<-N37rxRa{gUE_c|>SP@w+R=J`RTJwHu+~q|(15^LyH>pEM2TM+~=^#ChXk zd^J88v%tKYq93D=CiRA2rB2t5%z%O5lUT;To#G7f>5AUF^~pyIQo}MSA(n8yHxUuT z3rDW*1W*c9A3}fu8sBl;%~xtXdOoanU-R<2+0UrsJGF;`J1g(iN$g5+pwxM#A0Tp_ zR~i9+@;9jCc<+)oIC`tcrTk{b<#PFHM9q-4j&Yi-go8rwqs0LEwkN4X^6yD7ieGoDT~bK#{GCs7adH0%z;6HV4n0D1Q=hL?lR4gu z@`#bfcL*OIh`b&*$^F-W5|9n6m7_xnQXu9bslI5mq=RXQ&?s*OC#q=3R36Ht>)){RNBY3Sne$jfCj#|qiR_mpv7!O@D#W*vRYQ^0g4{Mp2jPx z;RF1mqaX1EX535XLB?8cZf*$kC%!AUtS*^^&jRgx{&UscvJaTn2u8)KY?B5be0nOh zhlYo>??HbA9511Tj==D-&KLK>r`DAaSw*D8;Z-=Z=VH2xfzL(&>jnovOy5({9N+)2 zz6choH-=TMK)t4>W}ol%z>pGlZ~PVOXaX%n`|k$__mv<}+!l!Z-CT0f|5Y0Ar|oT2 z7K;&Vz&Y2j;hB~J@pxyf3jkTk2?XCjF~~D zdYAiFEVD`n#OMJ1{0w`zcItizUdtC)jkc;ljLIS~zB`2`WfXdqb_~!SRxmJSfUtKr z)&nQ_b3S|gL-pXH(b3T*?Jk-VP{ny3|1n|;{2WQx*nHbU3WLzveL?l+kA-9xgtEnU zRekv4e-DjVY_|?#oq!oF-4&y5!OqeuNF3)ti~M3{`Z^!~REbGL>e0Tq zOeF9Hkm#!yJZ`#HYw;RWB$r>7?IdpivG+5T-bsxMgq&$G^7Q?Qhag#iW|{8TUJN?}9_kk5{#pa~q+bh@9kK>MG?TRA; zSl7wQ$}p4d2<}n&-8Wl;QYOE~&1Zgve%9M1^}#(31s6|nwJn>p>zm#8NW*z9hi!XUWOez!YsIfUm#Tj}7Q8eY;m|08ZPkFfEa?4&;yQx1f zPdz}F=NwVUFYQwBG+Y_wgWp~iqs<;=PlI(}Tf`9ed`tcQ^z@!F7E%sjrrxTZXBAnz zqL*P5z--9V@V*YSV7~CvEvbL@WB$TpePN{ptDUcRylzVAsv;^s8Pz5nJtbysMG>Px zp-;@oyCV{)Gr>gcaq5kq~YB=$P~Iru#>328uWn zN&r+KM&rc}_}eSrGTn_i?gS^nduTGje;TFm0JM(7>o|ODDfThJe_jm+`4%{mqj#VZ z!^rW`p+^ir>MaA+WtN5$mcVWWewC|I9&25V#fD{fuVY4*6iC>dm0)BLzI@7vfCL z@l5K=Vk;a$z@VJWG zJPmYK=+tAra*6^$Vju-A5MLesq3OL_;a+seK_EC66d*FOx{RcT_jj|fAGeN9tn?3wXkMYL?EX9X=L!xSCgN6qG~>XB_d-q z7;uHW1O=12;C*-n2OR$9%3w;LzQJhNj*bk_@yOKPzP6$_3X!3qyg&}1_TM;=^jq~( zad0wdm(%?;+;4nx{17zh?n^GHFa(TWNEYjrVN4Ia!8GCHfgqwtfE-NNR+#=0j0u{x z=0c<1-wZg)dcgiY_7@otx!*;&`ps`v)4SQk@o>Wh>r0~4NHKu84>qqn*JKQ0J=Jk1 zn07%>v7>u?@9_uWAnp%9NLczbupf#$D-|lqx8{(fp%@W~MubyDe43coUq4PY?OT)aef1S*n08i@=BJmQ$p$!d zaeC22ZARNIcUhDT=4jGcpI*UE13Djiwp(NYt}E&Y?t#ip~%xzEChkF8%!YkUzJ zq1lRZ@_J4ZP%gjROKTq&Us>=_X6+SC)k#^LW3z07Q<2WTzXJ3@4<%Q%0*>J~{^ov!C~v3vWgrmAtInv8qh&n?^nM~sSh(V@2t`g3F<9j(O)-v%by%{-;U zEH%H&pO@kdCJ*iT_>l}6Etn&q>XTguEK^X)hx+IyC_z4}G=C0kf_ZhHmj+M05wedJ zbIw~+Yvr=X8Wpu?b8?wH-%qI?)vOC%ePM7#X`V$^rDTuPI`(>yNxnaeseRDo!?S`u z_g|(}jlcG_PH!KkyN7)O0^bJTRx!;SDyqZ9X;#Y6tUbQUhuHjw5*_RY*f@d?qxx!| z*hRP+MyE76;Hx7-IQv@)>swH3z2qIMr$^SE_H$k9<^T^s)*P5aSiU9&vwn& z-;qLv!ETfiO2jPC4Mn$A*sFAYl?xr_D(rA1tR=d;P@bELI}4%iPzgQ@SwX~vcBEIN z;yF-Ku2V5_!9py(i{A!e9!)((htp>UVecm1Rq|S=DUg?=-HP!tHGDSn*v{hy{c31e zq0R}N=sPoV6I=PjFKzKf+aie8?3UciM)&VyQD3EGiAQSOS26_L~Ak9j0i)Z#X&-K0s zs!Rz5rgB1+kLG9X<~N;fDePv$^RF?~yzw!2FAaw-@FIe590URD^H(NLt5CblgqkeK zJ!T0U2!26vtRVBdxI|c90=@NosPP;z)$tEZ$2nD>^B)Ug37iLO>G>ReQk&P7f3m-k zx#W?)7ZYny=l}fV_d~_{#@}fB=S2LIyT(g$SXwY9yt%V&d4XBH@7MJeB+$ct{c$E^ z@1nS?2nhHOl^)#F{B`5L1loc8O$z3+^3CHPEg@I|5}t-LPjq~yQ=B>D0{LP_`j)6z zEwgX0C0v~7L=};PCu@EDk2BOCqxDZyK_eXE#!|66TAOzr~Am^1S z)6H{j7w^5%N7WIx%E_4yJ6WW#Z;@H6&EutQU zZ4QyZZgDL|%;?jWaMZ|6GviEXIrew#5SfozJP<(`2+*YsE1%mJ9qvpA((JC}A8e`+ z;|CTL2Ti1U)R3|8PB`@23hb4|Y8-qSIhS-@(6rh3?2VHgLt)S+$*V~-vdiK#Z8iQ` z10I~Oy21l%AWVFpnk-rN&JU-ioR!S2CdMorYP@5J-+6*Q!nFpPcUq30)z_JiO~;W5 z1z_MXjM6S6&PN9NnG`R!%Gl}~-^VxYpAoitGSh3D%pp-+_ss&&%-C#}tUn14f9}>S z!CiG`NIB4Yz+RaK9gfmF&uS-7mwXL(LFfB*-%s3?=4Dc4-@bmA)BkQp^l5tzWtCmk zLOTMvbrvZY(1}FlCq)Nar);v^^j(9#D%q>LJpb_-x~lCKzqEre>}I=SqsF0KQEKgs z(m-Hou7q%|Jm&IiaX0y@cY27a-GX!sP%OabnT zbdOAc#EgHO%+;p}l1ty| zqo{{W>~R#nqg^ZKUm$f}xN8)g(J!*B;5i?(cd_qUF<0H zu)Az+w^RI>G-ztzHQ;n8EzR0!Um^7fuCZYz(7g>Vr`tLRSsRU99VRLLD-5sPbeTkB zuDdD5vscQkl6mENNYQWNv~j$IZSQ-*!7D9##lkg?;c-o(3$AnQTR(YbBdN~z$80RF z*s5y#$!4nV)3=WLRh^N)oo(=`L`If*O0R zR`?!P`DRW^Tla9Jn)CkWrg&&REjQt~-70hZjvFL#{_8%D!@$7R7}TQTJNkXotYw3v z@V__4R6I)Fdk3Vy{*3)}?%!8uABKxecN{?ppq_R67cb_O=+1!dMQZqTmlJU2?X&sN z5TO&l2O|Z2_T?FvNq%)xf?ng{C6kBkCdV<{dY3?F(h^QD+TcHdG@Sa%Xw`c1B#!H- zXZlA-?{8^n#kK$3G_$`h7uOCY9?#G>OeukM+J9RQ??Cxdcl<4tF*zwo`fx=g)kft% z(?N?JM)|ZL%0RU-ZaupHOvkec-xF`QuoR~jhQr;}+5Ck6m?sVICDy(Mc#j*s8gGvM z1g^jTGD6-0de7YSd+xD=fa5N;t#4Yus0J8R#P#ulzOh3F`THZXhpIa3$(z4b#1X$( z)nMUksqPD755?eJG2029Q{v5oDV*0|N|@0i77e9QfiqOX%F_Cq}oEp4q>2{ZZl#v=SW&m=jMv6L*uZUT&K1{dQaosr(3v zZQJ`#F72h?L{8FdLCKnE+Z$SsR++JP$5S$m8VftT6#122l~B4XY>6d_-m3Cg4^uKu zH>iVANtvSV_&z@PtoB5FqWiaAn{i$V-W>a);{$E#9U8#;M~C(#6TM)|WpZL~zuCeg z(_7y&+tv_jqMT2Y?k>cdD5HJDHleegQv?<-i&a<8@B8LQyhIOdEo*!c7-+>Eu)tkV z?1v)1Y4wVA6_+PgjuFNiy_=2FeQNmIeTsU2K*G=a-hksuS=l+oa>GukEdjFcETjI_ z*a;Pi5u03B3bZ6eSKGQoH*6atVCOv>S2OFoS)@7HXiSEBYIG097ViU{q>s(tI3vdu zxOG>sx_fZO<~`12Ht=J5CY~BD3_^e7>u3T5i5t-9S{gGD{-Hx0 zCunhu!>)wmYi-1Sq6F){t>FV>E1J4YxyqaH){{@BqO!=|{vBqog}CF!x80a9+(#|t$>%ee59tAc&J>Z)^`nwNN>SB!g{@k1q;H~#g9 z2Wiz``F{9hMQPzQmPeKXYctcPly5jti3SY*CQA@}#D3@PS9%C8dfSuoory;uuwl+c z#G@w0RRQTyVrnyFlGgea73$zwNuo5>=cUE%gm0C6XhZS6gkKx+MCgUh!=n`gIsyHr zSUW;6Um#adKY&5N5QmJ177Bj3Sgg6mp+ojE%yuF4B7HSalY#tR6E4$;qCBIAI*uOQ z>+$S6#zJZAt|7Lb%Zr`mK!dKOB9F=m*yOq6FE*<6qpv7Wn9Z2`0K(PQ9NP7IbmbfhrLqo9Ewkw;Ao^ znVI-YK<9CIm4GKi-TMZdd+TeIkLnVgwXn;(Qsf&qT5q89^m%0v(J!7+-K9v>L=06m zSmeErXBbA3^2$!%P!%pJWjh4s#B%Df0uJ4@hv62$?F7Zs(rl= zt+z+$^H*44ZDR)FKXd?`?(1d?P?3au0Y0 zU389Mv969C=GykBtLZ+!M=z4!x~TYO?Ma$vF8$*XZx7m zB;!^$RCzr_&5PJb`;=|t2JD0r?F0m6+5eM?dDu@ zrc(Ly=b=06NledBHQBX`ii^UlXQ4qn-quTQK|S1ZqOnhMYpPnTWmJR5g#WTWxfeY-voBA*0#q@#UHbh6 z@{|+uBs6W@B<*D{Hb~r^(B>JYpvE%Fn05o}X@jA3sFc&D!pFw(Gl;h8WIS6;;&&l5 zdCK^kM+bqVT5EI-(e#`OmLD{dVn1k*drt7ecx6;B_Zb{Kh<@k8wnFQbtKCmGx9IG1>xta=Kn+NS5NtH>N4}+Tf#eOE9>QH%EEFg{PIZdR;0D`D` zVE9P(1MbJ2Z|-QzXN~pJ*%Jz%^zU;ihVk(a_YF8YdO~dd`AcBNSXHZf%ZM||JL#V} z)?BJeiZ=Mt2@oWg5sC3re7{ZkfCuhrG-g<8y~kd%ShmcpS0Xcgc=Z$cBNL>h)^Gn< zf3<@LJbqw?S2>BY|EwM%5`NR{oBXU)b(kv+nebQ7QWI~idwIzqNVgk4Sb8|kNlu9bZpXH zkDD3O0!yF-dvUz7oZM4gWgc8Y3Mj3r7g8k_(*hQ4puoD)EXWzRdX8TS|240=*_gt- z?}kj%o#;spu5!9||J(>I-iBWh#1m#F<6GM!YR)dk#U$H9CFBaOlV+5yu{SYZbB{j+ z-M>R`uJ4LO?F$X zWAo7ltz-ww*W6~K>BhFMD>KNw@mVomF9(0JDC$t7gXVcZbM?B)r|2ie$xg2f7dJk7 zG3hO>F1S1OeWE9`K;V1Mp3<*>&U>lJ?oj@O}Usw%;V3BmmwO*U*6n#7wJIznejv+ zh6TTO zfkS<>#wsZ|z+BXQ*3juq^`#rMDNA`mZ&cI(alYcB&*TcltI)O&nl6!}x5Y25dt09I zjimtgLMO!mSGB={g4{Do?ZG>62hV5$UIbCd%T7JBLPIOp!`hdhsR@O)t<=|o^;Dqrr8IL~t2 z&wt3f^F`2=o%Dx1`K>Pk7Zk;sGi-@h?;F%DqahE+J}_o0PM`Drb94rq#mkub>ngEx zT)BY3GsO)Oug1fdMGRlNhr|rZc=B59LK`lfpF7uUk0-=y;S|iGUkW{FO8>q%Hbs|y zM#TYBZJmB6U!>b0HOGuRC}i(y%f9F2h#vOJA#x?m{i@P*&>MDM(~YebZPb}pPLE35 z#qQ_LOu;;zaV|M~;saF2xQ-`*H`&>Dr0W zWt=+a8W}g2%~r@ebY9GF1<#8fnAOA=ONfm63*7Lmsi>+6dodJ$sz!qV z>mEMlJ46;j$%=58luotL0GlSw;0Gc>c@%y~M}!|ku{7CbU%M5jDI(Zhgv2C$D6wu7H5 zDgcuihXLM_$OY_LG&_;nELUWzyGMofT)CxPIR@P&m&tUhf!EGxBO^_+?e?kwvpzx; zIR6vBPax^X5ML|gH;sN}Ljw^?t;Df&WP+!9i_b9)gSxxGR9Sh-1bF&=5?*-BlQSi| zt#NA_IaOz+UYa!vr1wWZ38XuBy_!n%&eDU0?bkCp+=GWONON+9KK^`NL(*URJm8%D z39p^*JDXiA<>y*9)%atY0A}73Ok#t1;lc49$a-732zJuPY?zmWJme}!p+ zKq}c2F<>=slBdpb{HyiSdJNq2H}CE+o=KYf?!O~H0Ad4bnw6mdf+0Y%4Wu~zuRfVN zpaFiP!IG6p!i^v5%mR+6{|Vm!$VuP!N`NnJmenI>>)*50f;6dm!F<>UNM$@?uKz3g z)!2f>myJFmm13)}XKHAYr&8qzOG#K!mfR$so`Rt6*F<>4Z z2GB65E8o9+rG6TyHSht3mt@=bzlLLoWsMa%IsAoBWzE$lilcn&e-sCQ%tdnHiY&eK z>X+W0rCU%}A@HT25MOtbUOk>G`j}PNZU=~X~*?>WPfy9{ki7LV_itRDu_S0dd zQ+{^xNx)Ss*)|NCxmE%Tw9gE+Fz^HR0w@L-$m&H4uu!DrLH2PC7VUJEb08?^5d%;g z%Y<6TKUW?Grp#`=%k;f@005MH?e%f^++ab(K0AugqO1yomO#=uDKExdA6M+TamX%k@aI#Z<%yY5q3c;I=Qa>RkCQo@ah7$K9z zfpTn{q%ZZS-uE3<4Bq$FpcAl!2tPw0w!l*2O`gY*Rsr;F12PsLvWMANC4<{IGD zlResBHaQUoP>O^fyu>6I>p|zQmD$ z;sv-u{v&d<0f6brB6)m4fZJ{W6dw2$E6WDf+G+9 zjILz>9=c``=L-mz6O+zf0nY}PdF`!P_vWA-Hdx(;Oi#9CPPRa4$j#y3e?6TYHwpMs zd!$$>p+w*wPjKa}%NR>oJ32$*uHzH|d>q_Xj{vpm12sH3{Vtx&37_Qr6VF4K>klIR z1Q&u`Co^5WwV&TiN=ixvK5fJUy)+y^$uPdjz<`PjKd6s@i<<~QCkU=QMzkL1!hLG2 zx-)!^4$3rMI}ckV)&nttRZ{~Y2WcMB&(N=2UCW#VuQDUhQMRC90qV!$MNX z$8h+=Buf zx;?S{g!}RPZg*X6h*vw*|2a^b(U7RI23P#ZjERye!PCy;fa*?xft#wF}Z)$Nm(zEar z+yZB!VWW)fLqJUFyIT%F24BmR^*V>i z4#_t^Ca^iZrGJuP-m78D2^~3bOscIIbbG@=e+lMUR5U1+xd~Y0=O9E^oj|(B){y}l zfb4*jO;)R5%i4qD+O?V7Gpc?|ag+qAZGhUrrB_AFd+CwMGmA-j#KzgE)^e}C7t)M< zzd*+CFrp|9jAKKKfcZTh#D~K3%__+P<^h6%dBBzVLCBN{LddJg?5e4#_(x#`fa;A) z)3p`U2RqBHGL$Y)0rbpHrZ?=kGeyFKPZeym*Ub2h<_R|8=Zt7)07sMonQ$^A-K}(tg0=)I( z(d2TVoMNAdhteiq_N3YvY&ecb6$S!EC;pTtTj#m&RVA(0BJhZ{lM7v!K~%wOAMApb zUY-g7A$#g8{NYy-AoOdZQY>}e0;92UaG0#HrH2<0Y44*h#+e)&ZTVW9(sZeC`As71E&eJ8*ZuN8O<4#{q-om$d8mv-BI9 z-07e3Gn2H{^GFa)GJ+>N|E_Vb0w{$$;R|rzW1zWBEqKtjKuiDbKP7m#gxqwPkVJeT zt$#5k!W58%r6wsy3!Pgww0y=7Di);z} z14)u5H%RJ9r^=b{!|f08fXQ&8h&!NU?=Yf%CJf?Eoksya5CY@{KIS%R4`l8#M2nY( z>#_T(MrG2-eS$70*Q>ChHECp+^NNo5H)hg#= zP$GS>6a=5~=X24cGPwldr{7+y^0-WW^TSPXhi$Jj9O~`D#7bGina`1DmkLT9h@K%24O8oFO01pwo zHkJlR@~1*S^-H&+w&GjWD3>9Xt0s7kWYN5cR#xAS>l$)CTgj6fnrZlLDGSbjuTJr0 ze&g+esn6{fW66oczbRl||MWO^XKaozh|A{8LP1kP4P_50uJFn8um zy_=8Uc3RJMMa4sOF!kprmjTk0AX*~_uFt^K6J(FCI}#w@RbV11U1WvO8ZF zepyvzu_Rl_*+8xr08gjCW6Loq0|%FOTu?5NH&E~dXXAW@N}zTpL#f*_c>tU$Nvj#O zePBIWX)j5;cn&nOn2owQfAHX-^w?{7Z{jDywBbt6tw8&gz@)fF;jU^<)`>+@Q zpeh-oz*5nhz}}$E=v!QorKcZ9vc=Jzn_leg7Y;V( zlO5(;d!@iUn)qNXq=^?VfEK47da3o3dg)G9BG;+JKAWi32xO3jr(U$t(_F{f)xj#vT#vqZ8h)(*N`@}) zq`*sJuZ3Q{=MHeLwmSa=w}H0{jLY~|7cGD~{8B6fczc(k54C!IEWC`dIPw*u@khJM zdAdQEr6)g4cz3Z$oYt20?Eps$F1gbpvi0~$3n+T@ zv!{MOc<+j+-4&UxKtp^&u>#d-l>->V7S=iN7hDF8BG>^Sa{^5Q+jH;5U2V>G5ER3QsE^5(W@5+&~qgS)=>C!$oi_d2LK8wObn)GKflfwFXWd4?ZCXUwi{lqQ#rk zfR-xo$22&1165K1N*t6-ymTTYgjc0$^gSWl?$kmzZQ@WYZ4eYV9KY5_r@vpLlFnxaBarIc~>DANz0MuSa z9j5;zm_0kH%rzC1Z40;)&M-07RUaU>l@P(1jt_x8BuZ zo=Tq6_~*A~q|VkuV-MtoI?CU8f$Aa$;J2S~CQH{nvQ?SfS+4+l~>cgkTY5 zP7A!p917-Rj-_02*@?!=%y@$OFLTKN91bQCfCvFDqm9NxMMVWjuF+@n^RAL6XNAP@ z+C-4Q1Xo0Oql##xv34KkXEZIEnrbF@r+Go|?N&|abqm}5Ooa%5XqyKY`ziNu{C?bS zyjX#A5f-j$ZjKilJNf#|%T*lFCGg8(H@NoGNZNRB?^g4%$z`$F-VzfTH<46vmg~QV zh)=uXW<7*zTDVBhc}G%L4z{$z(!=y-9UYqxAQkg!KkY{wkMnk2X7_Yw@#%DwqUr%4 zPPUxq==?MXG$w{rwDyq^g)R}FyN&nKea~+Rf1c9VtN&H`ddLl`pV*!SS919}9!>*8 zA_3nLKqnI{2BpM7m2NoDW}B#{qwFZYwg$!(91j560IAdQ+vULcjyBB8xJ_POdq+hR zcDL<`uZp;a1%`z9I#w$PC4V0S2l}W)0-2NV&oGm)Mq!50o zQ^zxw(OaN6xb5Wj6H7GreF&{iYUrykUj%e8QuVhOeuuB3HSPwY0l+WXy)9-KltrQ0 zV``o!WcaY(m@v?MUVgbMYFEK)Uo+w#>1jbu(!S78_p&L9{shN}7!d!6*G@GYU9-8! zwI3IVm;LrywPFmY=0(qzx>LXOLB)sOsiH)y#wxlDIP>OhOme%L;D)^lNyN&D#I*lR z#Z`!}t*qLrI7EiZPFo*aOiKH;Z}P5i=tVIGZ5Q6?P_=rRnQ+>#cd3{Z-n27nb>D#- zuk)5=l~V6L3$l$;_WN7tmLW0>AGpR|3F>=riH4AkF$#ztcdA0zeuSU-II&1OAaYk- z7m8|nPTpp38}GT;Hbk@8bzAT){on%4tBbkIr0X-WTK{wkALCTr$oTj%_z|DU%IKu9 z=F0fs;x7M%*6`*UZmOj5mV*6{Og+>$VYCjs$+b+8;7L7OdDg!-_k~Yr%fB&~+digz z6}(-_JF7P5?cz0Knp->u=Pr2-2#m|i$R%1wJuGZjHqoy!Vq>{a%4pN=}~oW&D? zMUUK#?4?B)HkEtzM)yo3vUKa6oCLDtOE##^DvOvhiAO!@&0bw1Yx=a2eZGv(q|PTDiV&L3P*n?P5JfJK z)9+FIDr(rj=@<5owg`wrt$9E+Jv;>T@@b{&MXn}|OY6FLoCfN~yuHuEj$+;sJ6nN@ z+-8^Q=oI-?^?iZA;V+$b6A4vp?%cGWo*}gPDopM#w zkwsPsTbG>fk2$PKEKzmR@jp49K@b^YnKHIACk9mpQUZ7J^u>7i2 zF2;W`L{V+9)I?qYK6xzY<1X0ECd!}py9|DEl!p!gQAVJpOjUnKHw;c_OKwXU5bJ!t z^ZbeFp|AWNCcL$Ab`L}A!1RfpWf`I_(QsMyYL6n}u25>v!EpVBAh;&gKl7=%C$;M+ z$#8vLp*3!}CWf>8a2b2G;k7t=!E=S)M16zh@a*zjY4@Jtabx#hQ^xey_RN>ww$1cC>!cMm4}ZM> zxPCTJY^FEJ@ct<0?e(zMA=JAy<7HdgR)QB-W>?2JN=hs%&l(sOe~%q8Fp?j~G;>oK zzMP9-y6p5geE=YU^#HHhpDmHTSr5}tTT0&M3pkCX%p`A{g*Dfb~O99()AQ$p_Ue?fvk1PSst{v6@JjV#MYr_bwccnwrCZ%{+vFYOky27cC4@QJ3d5AqU)G{P|AyS2js)!8O7Rh#CRGH!A#%gy1#s)e(aK74aqp0YM~4 zLk0ga;dded!5t!SD z^!0tu@0|7h_pNo-cm5FJdG6W!-h1Esj%#0g!rv&%;5;CI00Mz<O`0ijxnD~W?Zl`+^iCTPIlSWdFKE+7z9*zbQxF>8kw zAdnf8+-q?S52L+iyaY;Z-wp3enh&e0KNMhn+_9FKakLfgGNyewjK88^R%U(`sk0cG z9}&DQ{oxgzF}3wlV&M^X1-*Ub;+)l3?0d5BUS~Sv+qu2cqp7q)=nvUKFujoHPsGfn zQN{6eu$Ayf(dWPmRhs*Y#riMdQ}tfOO(o!&%>9j<;n@7*E92YC$+D$e86rSaaTEkP zC}0^0f&vmd2!JAiXg%8CAP~OLG8hEHq;}#128Y)`0RjbKz;HmI_Re{D8ZR?FJ?8te z&B60l$3^PEZxI=Smo_K@<74pmwD=0W6pL*mWUPfyB@S`gQe;FO1}u7$e;NEf5(%=< z846YTqOrb?k11~O4oy&U7wSDkek@3fuP1bbDJr)N_MUa&14%Ri{Yor@!`k3k7|Y;g zVyAk^W$-A96Q2`y!@?6MzD=YCiUu^;9)1JG2?&M*Lq@n`!f?Dm2y`$KL=Y8$E)0U8 z)CWMRkRid;0nkAGHaKn|R2;7j-p2NSbhBxNkHPfBzxsVBr1*_V*Ra!On;S$}+TeyR zCXCHCRsHuMr+TwxFj92L(XWrOq}p5-uBPR~GcQ#|V+{wUd!*foSV+iyY;psj&^)gf z;y>nZDDMf|;L3qewjydJoZ@xHr<&i}Na~#2P2WGv|0vyXfC_=gd`;KyR4$zzR(raC zN%v=dG76O$YR4h=3sH^BzLn|vBnrw>*F_`%GwO}#>;R>$2NFUAIf8>?ScuBM?xT@Gy24tVdREtC4X2(6M)@apAKk!oqi`fX1Oyx8&IMR5rd(4Fe(b&|_<&R)mo zk0>$L#M?^#y59O@M-FNn`yYE}j&TJmholmk$(-s5b(j}+t&{Gw!hJ-kep_fg)k!Er za%t*A>BkKVA|whaq;gc;=Y(&Elxx&kR&HJYOtQkn__tpjihA{Yw$FcpuM0Dl1qt{3 z1b!Pkg5fO=DzJB7<>8nGQ8K_BGUfenO)meK7#-1jdtn`_Ek_fAe4Tro=dpoRFAXWJHc6!{S0TD+s02osuqpq_cTGHBd3nq zA38qtSa>bv)AzDi<^n%BN{0@TGvClPBZb)jasvS5=ph|C!4lHP+$TBBv1m6@2nJn3-}XkIm|>=9l0unrcW7=*8Tnjlrmpqm$grZl!ox)^Ybq9K%;ZKv)g>YE#JWLo;+f?G zt4m|wb$ib1H$|45<^H!_ukU8RPwk2}uV6rCmNtLA5A^5xzn-!K0YT>ogBl>e)d*E3@wvs1(MNGPK`wP6lmLcgF;Jy;WqXKw#G zTG5CKn96*HlFXyz3hyu+%8}q=#$1vzNN$ClQfbh5dy+4)6tBiWpzo&cwesFhG|!zb zQ?J)u!`_YVHcNH_J7iKRC8wLDJhPvnhmpmoA$th{9$c56-jLshN!Kz}&)ds5dhJwH z`*#si{V3oRO_ckHJxr45z@5Ypddu#cG+vP)D`New#h{oN zA#eWC^FK%KGs~p%(;ju$T-f0+%(Vs< z7LCv%Fse zDQD_s?g>Lqa{?7U46UIjUKi;ZdG_GzhAG1h)9?&i;fw_bz+J^|gIP%2|sD@3`RhX?jA4kR;Z{~5AGUbdz6PHe7 z;)mkai#e65f(zsfX&WYeaEsn<2bGe0N-J|KT586yb^6<-?jxpeJN^#fX<_g42{;uh z+ZcRTRq++a3)cF&e2lcWYg&hM<@~Akw%+9eeEc0+pEJP8;9<2>rqaG}N0NDpCXeJ7WF=U<<$ zP+)m$K4)b&w3}(QXovAXf}BfmHY?xe@XL3_FZF}OLSPq8g? z(_{uF`0rY@g7Wtl;s~Y+A5VETO>&y_x8$^J>52EZ2wwdJ`xVzzZ$dvJ$`i4n>hvX3 z%>${0>kq0|;=$|fCx!Fh!yFpCx+Z78dhz7Fw6h8H$9U-}^+BpTJoP&ym>k6cdC;gi;dmwR4?eMfOy+=zJ`cQyiuzB?cY65qZ}sXSFzkh$`qBp zIEaK%Bx1}RT_Ds99VE0fhe{T@<7D)goRf@a^5Yx6>kBtZx#usHJ5lI2x;tL;4{C>x zINs^Pi9l|WSi&Vs*l2m&w=9p1RbjjE{CD%1dW}#oAu6WB6OE9}r!Hu<%ajyzQ}f@i zgzRF(M>0D2_ev*dW~SE^%UJuNbEe0EuPTH|bz$^YwNPh0r{XQ9DA}v4lrn|dqe z)?}R^$6#!TgtUav>OAy2;jn4=PpET~wqG$uANWxeX5ze;?;dj#U7be_|9 zO!aGXLz+T+BQ#N6&q%{SYo$-gB)bAT=offe@x$Io-PJ=Q;n_*soI7)`PsBYmwg}9= zrHNn6*-*i!&t{AAUyAHIJ^tXY!GSA zn+(R5w!}^&yDMKs7m0t;7knZkRXudi@Dd4)x5mh~SNofYp`4(MyKD!x2c%abm4$I` z@&T|R?#Fb8TWh@^aaZY84&i`26u2)cb;M80#bSlrmwoz)N1-190Xny^P~}=dSt)gn`(!?~Ll#9GX0m zzz`0Dpz@J%YQIQzV1r)F!o6%{H)#)b{fdnElN8FS4PGX(4r*=S**Y#4K@*SITdHn< z8J#$igurmM6DqAxqX(D45pNT}1QQ6zme%!L5>U`MSkT=r58i~txeI-F<8!=a7IMGB zhO+!}_fG17-*xvQrd>u?m~_E~-Jwd+)$v7%?`vhpmLx=(^;HD3q0o^OQWa3pE)K_z zE>EQ%@@=K$=r1w5VQYAlxNk-Sy4(_z;Y)pcCb2az&?)1bt0Wt5Ye}+ubeB#f(N|qs ze-#P@Ld42c2nloe5_cmQr%71hRG7XTHVxmCw=x7Q%5NC5qf*szEX|$puEVAbP47B@ z-wD!t6NQkkgkI9J1??*NT2knbIRpzWN;fobVYlLzfw68dB8JNM<9|$PZXXdWjyGhx zvK@sgmN5nULLkE<2h|-NaGy`a;%A|m@8P^Zc8;D;!5vh!`&3>|YM!BpLKmz-w4&?%aZq4cHTPCRyo8wr7clLOAK|~ z9DeTS&qkF4wpmW=u$I*pEv0A z5_$h_SfGk;-O__SgzsWPW!fbq$jj{8^ZggkBikr>uV}AN)*sDJLKPNt?6A30PB{tm z6MaU9@l*AXzPSm}zGPsblz9`p|KMHaN0KVj4t))hlQHnc(q*hB(z57hiT9Ab!P-;;4hPijzs`k-2zuu$7Dz4bY#1XKe$ zj&6c)?8a59`0(G)3))!7?<~)}5{);MY4r@N6l=^VyXq>jc^3KFljLLwetH>J6UK7N zJKaRT!&FtQ{$oCayX;uQdewAd=BYE&WOn6Ntf?BcB2~Jh7hV(Il=}NcvtF)C;gnxj zzq+S~~PY#aJmxwtvoZB@FCu&7HNNbi_c(eCR=W|B(Vzn9l4VNm014-19 zZ|ss8DQ|CKB_xSG$B>9v#YPAxMC8y{;v>|P|3{RT-q_A`sh@L?JM&o0&IiOVi|y*Kg&ygJD%LqGw*g>_r7mh`!;bdZ|0v?phr zyb#srIg9XPOKCO=@;y3YP7D5RtB6a7c@tCh@hEWGawi*)ODvVVTRxpv98rz1kzf+J zIqyiazW1dTwcQl$FwcNE=2Ol&$nwTbo)ONnzMi#2zCDC(1 zBt?(mA~Ma9I`G#NYm{tWP&Yz{1TJ{AW+UqMgrwcaJgn2L_HJjO1_uhl6loi)!6s@I zw$ISx_p&sYiorDvK?1+(*Iu*jUSl)aZc#TtQ8pNDw&v+P^BCsfaz;}z1u_OUTP8ui z{sa8ElNi`zlxl{5AO)~g*~p`F&+KomKj(Lg7W>OiFZ-ZzT7(K`sLivb8$xDPm^0~C zcVSZk<$Zd)E-#CG(=0Nol>kGG7c6B=+C|g)Sebqtm5B0ULpL)Jy3p{Zznq&->V=4!Ig4zTq~>k z=0V7Qf9>?QMpT$X@OGu*IFHB$-D!Vt-()A7+m>&b-_J>pxg+d-r4*KXIMQzb%R$QQ zZno=JsU}vVk>4|`64f$P;t2kCf~eBI?p*#gE)xj=mgV4_EyM{LtO&NJU$ z3p<{URw?d=6ZU*2WdAdP|G&k*{zL03HxTLt07~fk6Fl)79t0{GAlRekxXeQMkSopG z;Aa6)UlJy)6tMp04rVfRYq%h+tp@)I2LK_ux}Fd5(VI*Ac82Pka!UM&sQ(^vw^TDd zGe))zo;dT14%sQ!KMUm>l{~ofKjc7m$CMy+suwPWKmbIBR3H2^XMQdJNwD51OD@6aOSuHL3&!+szM+J zsR};|3W`TRk~k9Lj6=W9ek{ETy)(e0r=ugNE}nT}V#0?R)Pe!Lt9fXxp@&$gkKsTo z4?>62W&xUovgmVTv#>QUk7i5GBl6B=c{b78kr8Kt((O(INo+okoKS!#f#|>!7Z;aQ z<7d__>vYmIg@7h4m;uWm2nOiu!jFi@tQT~n6xKZ`LAWp~P%10iq3e>MtN%DOt4?$* zivpFXL6DTR^C_Et-NRa2VN;$>qZS{2+W>K~KqyMEs*UFWTm6Al8~@NLhV>{;5H^g8 zbTqQ-ag_dYPc+3P9S9i{M)fo*?DjFM)~x4jj53rc&GYHkW}&9m;4Bmf239de8sW~H zWJB-0+A*7CiF5SZ%8Y$Y*Gy8JrZJVrs8kJtZ=cUKL|@&uTq9QXUt8Xh&=WduWI*&D z7e=GQH}7yIKrkvH0mV{oVPS_FzljgBmg)C;kM$bB_fxJ|#()ehQ5(EHm^d$beX*xY z!1X6vJuHXKp7Ft6ds?Q{!_DX+kM%5ucdJRnAymjYKx48|@(K=7#n%Yv$d6yCA0+(R z3Iln>&#~a#uEFbvbhZuZh`NQhy{+I|c&x>Gk=3J^k}&Cl%h-(MgLUZom>Y zCr~9&U{o(Q)$dbZ+BHP)K0*YxMqd)zI|GyaIv4&Flg|nAws2-=-~ax)F)$q;vBe&( zE4(L3qq`Z5?Jo>!M@8VGhvCuGWj?hDYgHh9cSU3>X8a*^k+uOa)v&9z-$vfDmGrHy zB5Q>1hQuA)H$l8a9RKw)*jTdG)U|UfcM%wavWcm7pTZgL!*SIY)ca^T+QA z8U!EC?tUlCM?%|sC@$q_BkL6W4H*)KAt*KG^vh?fV!5@K+>2W>h~hnIgP_LqIZq}^ zln3L+?fx-nfaoX+BrJM31Jd!N7FY7klXe_BWEhpi_B8N>Gb?KfUjvZY1VO@xO*S%| z9zJ`PE`bSX$h?erK@W6#f`n$%mwF$^*fHr!1rlt7XD=siBr1$xt^P_D;d|l4z?a^i zEra*D4ohqIx&-5bLp8hg<$T%{;A6vX~HpKF%i60o?kT{|>ZrS?XNESh0f zuh6FU{;Ebxm2-zVcj4HuAF7=)gU+gT!+8Hh|L*fG`^0?r^7TgT_MPgz|H(|kRy{M* zT>{o^BaPS!?I_^GHs(Cn2El86Px7C{f1&JY{tENTUqgxrxxHc9U9dM=d%iie?{Y8g zdDZ%?^#*ro1cdsaVWD@jCD-T9u$dLgk9*5c=7G(BnY_Lx9==5d_P_64%1wv0RjK@u zOcuIkTN|JlN_xk<-jjl)gdI$9vM3q3m>Xeqe(+rMkB=oAVS8o_O=GZ9 zau%2UpmHC|buPZfz(at61VWtQaNKEJ_TsT--YHeT!H@e&s=j{yjWHB-0UTb?0F!>y z)ykuQI#%xyebn`GmkblsiVV39m`~POcv&rie~at&hJEnJ3^iaR0D3*dov9Y_BmUxL zgf_G2m5`_!=mM(&THymx;}>{)hSs5dSX9rl8Y@hQ-e!HLpBqxZd&S#0@ryk;Xx_gl~n8>4!%iyh> z3kCZZ%<&}nfgj8`*BbYaS24wm1EFh2yA%EKc7xJlsL#GMNPBKs-B=@!iY|l0zaL1h zb>Uc&&?1wZ(B{lAZ1OqPzizps(Fqfd$HZ@soOxDoPcNDeI>BntCtmVTkG^%R1JQjx zyhuyP^vj^p1zjP**g~Y0Jh$QNJ2a0rFK(c-4{iU5KPaUb-WenjrJFq@&+~duV+_hg zfq0LMY>Ym8k**rW%whJ;ag=5mywB_?{@B8LmF6x!$6y_U92DLLr*^sHF3fr8;&Y^( zXq>**`_IGUDbC}mkC?hyn>awXdki(N&rUX;d&-kxvc-1?A*f=keIGSGVEzY44lZJV}ab9hRS6 zts+|eE1WM0L{C5-c(6U3WMkDYlz-ZEfuzc%R7NG>rt!VVc_*r?z;}_=WWj-0REewk zZoh_r#1bS{$=6Ae|9a8KW<_B=j8hC1J;z;RIH^QE3V=2T8uh4@Cc!e)Tp;uV@lU9t z3@Ej**Uw89-1ZQ+Pn*?zn`@zVVRBjup*j>PiLj|4&u{QsV1q={9Dl_NIB$b59<0VW z`&byml~7It2wNkfjAZ09bzoUnB3!2Cg*d5Chv9~_ ziGmmXHy11O>sd>@wZXRqaQK>jtrUyW015YXLY<(DKkRmvB9X+eD@Fvg2pDN6cczxe zZQA<4!s-jykbD(ODxs2Uldj_j(=|k=`3jXizvI5RuwNLWqMiTt0J%%M>^Kv72(cqGm-c4( zX{5#Qs!{v4H3Ygt=4NB(9h0ut&+A|_09Wd^x@%NwFIL=NzbAp}d$W!?FFqU-McjFY z(D;JB11V!@fy+Af!nKZT{j{@Ww}zY!8GU7%oaFUHF-&;)2!I{*BKq`FO>=86Ba{FP z{9*r!j2&mWR?4R&ZlpwOa%oF{f#q>jo03-8V!ab+?z zns605%$GrRAL+y0D@O|SNUb#UuXq8@z*!))8OV-!wF!!cPW7Uu=inH!hgwzJEOFlx z6VBuN7W?5m*39~n$1h8`nrj(CkRd2}cTVS~3_qhwr3}}D%k0=-hDzLhsSmoFp=qIG z8{GBbzIjPr0}Sxnia`eecfz+=20I>vL$S@*1*Xgs&dLRd5z{>+ivC#^*{ldKHK?Sy zx`U7NT>1=;-XZY@Un_=n9@MEbeZKXQuFmS za7j6Ww|VdLcx8xTet{O z#Ww12==9L^j4D-23+oNLe8CtUQ%L0=D*N-R4+R^Ge@|VM;n2-+QjLTA~E;7U$$fg_r&Qk!Cp2Ugo(*t1boglxE z5$I}wtsx;k8vGqc8$9-J5C6N3iU&bO#gF*nJSQZ&Ow^R)e@!X5eDO^%ub2NEhV)Or z?Pib>bz%7M##Ql>%8}Wi7Wxn&hwFzkIg0-gZX%D><}+@y9L7W0^Oc@0DiU1zS*KJu zi^??z6P@$1y#qq^pW%LYgnN*xr@Gncu1X(V7Ulc+u8 z|0tzOwwk6szaiv;k0C4XO*moDc`bLu(vI4T_&Zy{R*3TaIs0e{3ijcP+NikK@p*r!`jo@M;FpKAP89}U0ReO9_2?&$NWD=9WeJJ0d&lEeTGhLLF4 z=bJ+#`6T>nr%yh-%X`R9iIe{|f#i06v(mQClK79Z^$?!V&?PxFy)i#)EGq4N%Y~K)Iye)v8e1B*rCS@=(q)j!eZfb3JVYdvqJFUL8{&R;t?XoIIhelQ+8)AK-%hXlwY zh(&eVZ~WY#<%V^cx!T1Q_3n_p7k##U_;&QwaZo+tq-C&I=8sV~Do1-b`afnBjDfHm zxu$aIO@(gxAEm05v0zPgDIAK)Kxjc<@6*#uht6lot-#=D zYK_4aHN{N1|Lcgu1et(eP}gZ)zt#p=UBB(K-3_v%v?)-N@K3~%kW2(mQ{?D0UcX53 zz|bFF@O~4slr8T=#$3ls_Y1uD29bY1L*e(yWU;W4e^x!nFXmA)H0D!~t^UMO4^rGz z+!ij}5O6kZS6t2iuK5C@yjsRry0YlN?m;J94?Shfh@(DP6H6JPdGGiWe5YHsC+uv? zyik+J2H|_0dAY%h2Ke5zd*fa3UfJsbPwv-Yi1mGG1C_MD8%e{x&Tv5(jf0I~AVQuz z5;3g@F8s`b#cKA{y?T)s_Fw#w&;MV>A6&O>aEZ$PE_khTd;6ox3&X#PcJhhB+#Sw| zq_s)X{zn`dc2?6aYFv0K4@H2l?*0oy$-lt2+E${ly5io8pM?L$F`sL*b!=$;@^;NW zBL7Q7XsNErcm5H(UbxH;pQK-*eOYxj^$1HTd$)ydkxW~Ybj=hXm2$7o5KOAR*joB8 z;TSAPTyadP*rU$DhNQ7#O|5&eN$=QKqU3t8>bgCaTa@p=?lX5>1C#3sl0~fBX<5wA zSBoAXS}qLzPLK|$0T1QpBXODbc-;;f=9JE@8hoW1Oxzr8IT=Zzqi=r))*4lYwY<8a zZuLA_27mp@KM-Gd{#bn~&m5xfrn1Kclw3z0c$Z@sZQg+YZ#YTkISu5~Q$h(se*QUaT*sWsci;h-}K8En&1x%GIvqC7?Bo)ug zR=nJbBq1}41?7z9>yM6$zpRAYC6M^{-pPH(Bc#i=>YQ>;7<{i{ zUO{^DrOhgnPtdOl`bLV$JO{_T<`_+Gf;-d+zpzctmhWzEEWOcl<_eRmt2K9c zHv4*@gqk6QV#+;YKnI$9DRYUf-JC{*For!5s+@T3O%U#e@8lEU2+tEKLLN-v|MK~9o1EZcVS12!g$mo@=T#3#cW-RV~973zNPl4#z6%#k48 zeaE72M46mzulc?Ev&*MTQcj3lwe7-esX!buBSD#0*yH`!wCSOs+h|){u#UHo|LqN^0dMJO#+!RFA4Ga8m$dv|b;jU=?oa?wM3lc9 zJ(!mabu8=;?$KXub@2|VEg66sa}G!P`R-^=()O*0D%4s|?^!Cox5Z$B{D1_GhNQIw zSKz}LlQV#4lrECuAc>0wKyUj;l&kLqsR!M{t)pk2Wqz07cPlxH^k4C&4xj}=uG6eq z^Mo?Ys3d9M&8z|ll*qaA^pA~3w3d1#Ok5|3*u@$o!T-Cq=ZZ1cs$#Q|CVw1@vgvlM zeZiRF`9)ev)yDa^7FFwe3}gVIe$LLyF_oA(!aPE$KJ4>8+UE39R3n0OExxrseR03x ze5u**WU1w2PcOnJ*hX_{W!90^*rQ%;vD9Y=iT4;*tx_WVG}|8EV;RxreEE3oh=I7G zrK?A_>BL}Zb)Ou8S3CZHna5sOg4f$r1D(8Q?tkG@maHz;7#K4y8aK zM<6A!1K8TKK(sFh(pXv`;b{O;i~t^hGffVZL*xJ73b+@|?D#Ll3hsYbcR3e6hjo(Z ziY-O7GJ~>Mm%(q7LqgD$hMG|zD5LZMk-P2S;6=S7p8Z@^NGla6OCL}0FMTTh7RjN zf25kSBWth-7(opHkNp4f`_Er}@dkZ$x;i?`9O$5vhrny|%cC_|KQAdc`3eULr~&Es zqwBLhgF)WedPf?6GEf^1z*c7fPECMEVjZX;2*5o*pH)ozt@94RY4<+cTj)K7#RB6! zBmsd8fNBD6=E93USOfa=o+&gmv>}`A_ehtBK+6CNe>0q6f)Q}6%V__v-xG1GIp5}^ zfrNlIn#WE4X+s9+L%Wo*vTHH{Z8iDi~*%;dJy=H z3?T6~TXcrvqz;SeG_HOWyl4k+L_2C=iqj_HZgk!ORTY4QQ#VljlmPa0rxS#Z{QKpQ zW6M!L>m`kej9`@l^jh0rQ|>7~=DRldU6&K2pw#-qY>Th=>->D5m2jpSdb34vN=I? z0)YVoI58Rmmld!%q=p5+s7%lh=q<~hdsPyX8;V&Q=kpRp-L!@@tBEWBlgjI7s0LG0^2d@q2b%YT-H!qpw3 zX`J?1sFy?vI1Bn`_Ui(FB9eH6`n$c{+_|ca?4G`g^DQ1umI2uJ{|SJxr>(4p`7Y|l z02Jh(m2~)~g8&bw@N7@@SB58OA@b4osX{z!_Cuk!=zrFq!gUxjGXPzkw*~GBgj-%x z{P)SxIe3D+UvwQ3Sw@68tN4HYw8tT|SiEO!^*>qj62AXW>F0Zh?MJ>1EPt5%e^mlR z%Sq~ODLzeB%GMG6580oHZ6X@EPxhaqHs5OeYko)w_ zq>gxNm>*~t%eL{_wASZE=YSN;S3J$_-}vAL z&PM*&HNkFrp9E*E+Vz!|rra^%OniHpl)luKG=D+3pA|Ez6x&~;N6y^mWz3Jrvl={p zz}pX9tsFv2e{<;DNMTF&PbXtw^ z>T9Sy`vGvjW8!Rs(GvotJ%1niS1UyXKy~k)iQc42iSIu4@wEDLy~7MO2XsF1Z!oRfwI-e)+APSvvSA)FGyUiNWdGlB;+~Ru$fXDsq zuM|jOL*$rswpXUQ6g}q5$5muy1bC6R;D_2oqre4_e>@fH5`3y-UB?G9moWF1nP@Le zcZ6Ud)fuuG{1wkpLMM}|<+<3uBX@#8NGo$y$u`cwe9(w_ti_zsG83Za;8_)nf1F(s zEfi`*O%oogF(!nX_;GANI^m+6Mg`(kChf~Ov_$J+`0IKXapm(4Vr%|V5L7uuUnI;u z6&XEm|AeoOWj!;ThGR(cXhJ*X>1%xrHZ6?N}-gP$1fWl9~%HMeYLTptVPKh#}#Re;C&;{!vFg62@`?i%5Cu?OOH*_g<7y|OR3gEU9fUA^2@ zM7Aj2ugN?6vtGvYbTep)pJiOk#PyE$ZRk~n3gezbI-7Zqvqhwfl^;tyB}rN72#GtR zR%N`^iS3|rLAznpcac$^~SE*_uUUgYAEAnMK;?{M1F6fEa(9L z9tEFES`mGo(>P_%<|N;O_RFCE|ra1FQ~a6cL)kf{9i%$!#lU0ud|xPWaicZ+&KNLMuzdB-Kb zPsiTe-~`Rm<=4uLxj$*ERnOWkH84d%#@bk2Cj;>Lw?<#pgW}9Q>#stdHbUc; z><-0kBh^Zbw6{mwbF%G+N#?5c?D*K-E>LB@eyPn~+*Xw1cLQii4O%qH_ye)gx{p6f z?I{B;>Q9LCb%uPciTAZ$`c2Kh%aZ>>>|EAlP^sGY9);yV8D$-8Fn@q$=WKXgRh2wd zn6{98RGc~%l;!xsZg&3Xb%h%bZCBiXP4SIrozB7M?d5auOxLss2rc@aZEx;zj=i7s z`H(hN_vmLy{7m&@Vs!*th*8ED;ck-61O2RTj)A--#Zo55w^+h3g~7j#jky7I`7)gK z#XE6WqV7+=Wt1no1mym&ePsft2cVNL*U3rrdA{@)au6p}l*mllI!97qUGoy7^-FWZURbcHFQ&et^f6zxriA#8-iYg1UXUM6Cfbz;#k-VWpg>EKmA=<|5k8-t08H&X4w8G?-M#AkC zh0~)&)MJf>&#}ZbU9AiA&2(jz*#UF%vG}LbGIU!zR1NPNw}v;%vfqE1+7HP&!EH_c zAY)hfIDT6m6594RlmxAS9ZrH8`U-+Y2%!23j5@M)l2;NOaR3h1{AwQt7ot!e z&)HR}%^DJ;Q^w){8-=k#p;j+5-MR8#b+?aA;Ff7UZqgs2#by{-aQo1TH6j6o`{w{^ zQW>38XVd}jD0tufIvT)fT2%C?-!uW4>o}0T}!VyXd_`|B!S*PAlPtUmXdNpk_~{*4WyiZQVZmt;Fj0+WZ1Qa zg+vf0?J|QznBg;yK^5Btj~%cu_LI2MJHoC#T*1x-k_eF+vp#YJx)a0$OS&=5Dka2I zA1)s0i%D^ep&q>b&Ge;6(C3zD90(+noAV1EDvrc9_rm0iTcW`Oi-nV<0F)`Mg*ldAi_PX7g^ScU`2}Ua4C9l^SZLK8lz{mb zz3sXSo1%QvlBBM0;tljsw{(v}>HS8Y*P02-@!27CgQ;c#t9xCsVB65d$u7H}cRCb} zBxb4XE4=5qm$K~^)k!#>Fkp4Dd*H}J#TwL*F?Q3~X zaz(6^j9AV1ls9z4GH_Bl=GOOq-&{$moEqHsAqy$6?lOyUTGhv?C{|@%s|^jW)C`}_ z)^MFIv;Opuu`Z9`jKWti^`ng4C76>qv95%ZSQAKtJLNN-8U185^`Co7Y+Aka23LRk z&NJ1w1GPSB=GmSIo}z^VKO-kfI{p?Z^=R9eGeO0wkKNJ{>=e21Zw4fDwYL;Bx9pAVFq@?>ZA{LRmAZTb0I_Dgpw*6G%p;2G=J(1drm*w zi!vFmE(AD401|`(+%)J*lru?|YI45&`D=N-90R3@cwcTKv()R&iq^hBiMrI`j3p#P zQ$Lb@tX2!|CWBM@Z8q0XKIX0a6x|Ibu=yNA3H;S~eLE3a9#|&uHw2Z76uw<=$sO#e4`9^OJ)Tzwrr$Km|n=6}xNqw{Jnv82}Tcm9>X8s2&6CKUfC|QZm4! zYpioT)FH$KPNbh&d%iiw4k7NmR2$`%&Ib~ zvrnnh#-kGGN=EiafdmJLOL4z=kri&a)Z**Iycq~EJFTx>JUl*gAOl=e^!DH+oR>4B z<8;&q-sWJi@lC5|x>T}`Hnf2yV(1Q;42&<)nm9@;`Rl7>}{PI1}P zGHZv#@h7zeeQ;x*Zz1Tf3|_}skcJ6l(X*iLd&;LNTk_WCl;5<`S-hc_O?!G6fdHh00jCdtoQh4nRkjcy=LoK5SUvU9za}q`1Hwu%xPEv>pq#cHT45nU_6( z_!9dPz66|xh+317k^}E~!3{srhtxB$bcbiQ4?DWb4WSp8{=Mk zB<$GOB6$7jD@^v^1E5k{(0ZTWU*1SPjtfkVP8Gwe1ODF=2=I!fjm2ejYJ+hpr5D^~ z3;^zJ_8iJ1;A|75OG8@kY1aaY^LPtjW~c8gm7g4$zHvQm_YK!w^46mp%GXep5gMg$*L$ z_ZE`elcef$`1J{2@oK7(&j+LCiSY$AAPcaK7h+VChJuawCs7-%T~A13SvG-vBjlPa zyv%{iIArEH$7F8yOqvidiU@Y{Ow^|Q;jMitQc^MFekZ^RzHtf*?-d@(WnpK3oNxTf z_V#lCR2(xqw5noX>^!R8ST&rH;R&GhPjeHKcFWaxF>UR6N`ZH)J<<4(AAqQJc3qYF za~y=&B5&9puDGD=C{RC2e9Y)!LSl=b-@=JuWcUqN{j};cT19g38f(mP1~&H!F0e^I zpk^$Lqt!H3NkX3TFk=Y=PaaSK8oJW%zqFe^pc(BS9mUUA$Nc>QUt_7Ghe;#ICsM4v z)tA*8oAeQ26>Nnw@2}QNf`h;15Gcv#02?ev)_XO+O;fv0yA8c5{*e=aq|x;Q7T0*J zfz0=*e(U;b8vr*g3nCgTVRQ3kH<$6T<}<# zz;CF%gXgr)R(=_CovRi;P%81ECLg#bp;)Pl_}ZH9v$n)4dXeC~<=G-C?gZ!pvYAZq zziK$jvr8IeoEuz(8I$A+o zxZ&H<_tE;Fdk*W>{#iH2$(JG+z@Y@Xc$S~^WrskuDgKup?Yriqj+l!(4B(Ihy=B82 z!?4eep0c;V&#>`8(~syki8C;!@uv7lv@R_q=o6WGh1I$8GQh@I)@C=4BAw< z>GU@}S6yZTo+^Ctm%iLx7(y36;oS61SX2u94gB9pklLnB*A!#k}u}zsH z!ldaLY}H~H?u{Ye1f16a2FJYw|DCs`46P4E-p<%jZY$6V#`mkLqi&OGDtwWc3dhwj zyc3iLii>~>#vvmyRJxYtOd4@oD&YHA(n4B!DUco|eD=yX+O5BMhw2Qt8Drpd%6Yeb z=r>uk<)Zai66+VR<}GTM*wxthfsy;7Ju!@*055UX+r955tWDM-Bc7T5!Yq?xyRcI{i}|=#%D+o@3@?~Z9Wsvhj*GE32@22%7DX3nYNy^FknSM zMeoCI@KJ!)16}VntcGTMvT|IxCu|M_JR1QaJIyBx%n_hKx-C>3bC28nC;7gt#{>lxmiH%tGV0R=~ZyYFx$ZR%Z<#vw5Gu`yG}eDQ8c=C5ZJ7@y zoPuJJ(P>UCKfJoHb4dE=1WoHH;W#$K`3|^l=gW50<2>s09~>KhzFcLf&Ym_(FDZAa zJ1w~@Ozse68X+Pf?=LN_U0hN@SES!JsC9bZ?tI?8fB3fne&~o#5ysPPG6HUAn75on zA`2Pg>GY7IDy3!0p@Eizy2pEs@u%g@-{A0Txos8#ZwzLe0lZHgI6eabjn|p?A8G8C z8Gva~c#GkiK}l!VOMGvxrj-JNHjEOj>Cd4zUh5NUA>J5UI-h@iDigWDDt9@SA5xQB zk=P#me^qzp;Z(hAAE%=5Yd~enPrqoBZ3v0TkPI0^Btw+U!#2e#Q^q5i=V50kk(r%g zV`oT)%w!6Co3|;pA~I~8XG!n%p7*-Wxz0cD^`7^?eeLzE;knnd?&p5)^|`-`i28#D zNmar2>0)9pT9jlN_6{vd_!OPE(4VLv!@hLU&u`QFMyGsv1iol;45IOHHrj zY4aLN9I1@C#G>8yZ3*o_qnNS0ar}+Dyoq-}JN{Dwbd98I^7~AuAr8clsea}^@= z2?R`@^1_UdnZ`H@fCaJwlul5%q)XIH;g##5W)L!CzR{j_TiH7?U%R=`)IaHZdW{VW z49VC#+}oj0zMtW?sdZE&=~)0#ExXLcTKi@tTPb{3j}#O#Oo$&G(-E3G{L3`pRgd(4 z*q@Uw&@;bj2d>quKvL5~?HV*!A%pr%_t~(Bm=?B{2ff$7H>&TmU}R=TEO2V)nK*DB z((2Jztoql{Kc~#5XEaLE^_w0FKV@ezyEAuL0*4doMBJLXQfYWij`pi^mc8Or4ys?H zsIiD7Fs7l1Dl?z{+@bciG3)82lt5`QOo}sPu0|*wBG{ZBy~zMQcHN6Sb#OA1DqJ!G z6Q|m7v?x$o$^0K3Re7c1(^#|kD6`&%L}}UzL-1B8GUOyds?)-y3m(lpPi`c>3yrNp z?+u=Tm~MF;H5AFYEwY^9gUbZ(?kYFB2r)O+8UUK-C$9Fcv~+vlgH@?Y-~f)+_FQib zysh0I6w-k2RyEZLN4K{HYE`2n|I+S%qx@>mVZ>tIu?>K6RZ#`)79DNdU;(yop|nbz zc>9lp&cp;d&f1p=Wnt~X><$C(xON0a-*ZW;z2GQz`$@08B_dpx&KrHy z{f~NN5d|UHYg7l}`f6_T4l8GZonY@uc^#UYU$--6{l$h{S zMC0w|Z@2d}V^e$C#LCqx_98)DbqK@O5#A{{E69|^XHoblPZ67xK)1ll@s~lI*~4eO z7B-Iy^WH?=863&CjXOO?zAQ6PrwWh7Em`}jzHO#J$a==Gm0;tBlj817PRTyDHGd9a zAdT3b0H!0MJT-R=e~2ztjkxsx7^a<{+9Z~H>WUQ-sdi>k0#5?^}=}XgUf51sQ>$ce+BHp{lz4Lt& zUO;cLiU79{!Y6I;iUm)%LE?S&`QD?h`|p ze$NHhw@df;{cML?>0wxiyq#h=&hk%P=)>^}i-ugBS;k!_uY6+$QxI_l+1i8b zVDhY$>j=_x=CC%*q(&wRn%H>W*!Qmee-$$SckkN7C`+c+mU=BHBWLj=Kl}+E$^EfK z<%Qs`kW6*v2lnQ0y&SWtbgwhO zKr#U#s#R&II=ErWmpkiS7auk6l1n;)pae0L*l zV;Bf!)sATqZbYL!uSWRt@IsMl!rWI$LQLY?abHOWhWmsJn8Fa z==)vmcS6GziyMn0XNbZF?KmudCY|v)W<{Pxk`oUB1}*)Wk+;INn$LEW^5c_8?CN1! zuKhsQqm}QFeDNek?2;h)IfwdrdN@)R9J3psyddi|?&L~CR4uZPqhV>e4b62vvG`KiC(_nl*u2r;W|Ff{SD-!Wrz9kN%_8W4r$IjhyG1K`eE-UnW_)IMD|w9|YR zqkii1C|yuR_tLPTuZGyi4#aq!%nlAZazql8JDyw-(C1bRMo4#MKKX}j ztJt3QQ|GWDm7Va-2q0i&yqYWB-1jw3@C0&|Ll+Q=%L=EifYS)1Zr7~z7L}B4KpeV& zf{T0h9vgvqv?}r1d5)7!JUkPhfBm~##*U+X`Uv&=fP{>@iUTci{m6|P!@iu z?jlgzd=ChoQDD&VGhhV6>Kox z|sA$$aq&}dvar=E#d*$fF=yR+mOMFpTNKy`r%e9F00Uth|O&T5dNY#2L1 zF#<04aaTOQVTF&!1CSpiFHsa9#~+vd3G$LIfb0T=^MLHKgBdVJz-|34C}@+7a{NdI zn@=jhek1|bl}l}%2!Hw*&r8_* z1<=fagdO-TU@r{NFO!o#F2E2*53H=7BZsbnn)nGp>N=3Nq7Wzs$)ZBD5pX4|OozwM z0RuDL14&br@}mrhbS1D8{)5)=5gJCAPNN`=^Vkc6{Gkc8EHExb#m#EtSyST|Fjh85 z(ajH96k?pF6bom=_pvexACxbS(lHDM*rXu%Ad+yChurW5I)dn(LQG%S-Y@AajI2>q zv_$ViTjRjfPzdw_XiH6<256X4*O^PGnFfPBK>CX!VLlRuwt&p(ai47fZ6#V$!f+b+ zZTYB9jjj&hJrsB)akj%*Sw{pwz&lQprOX#~nUeJDH%znD-3M*1zq+=6W&obaGCXSi z($U%x^#U1G*oWUK=XVe(p!EtxPLb;l!;dFdHZEB1^_eEKKYH!+>i=}a4#GQWff|S) zqU^}ZG4Og{6a;ty6Qr(7X_A>QU);(+XYh1neMF7!=DCwRKws4vK1wp7gGs^1YkOV6 zCX1@V5TJ;W@>dA#WxL{zSn$N054Nh3v;oK}+;*JALjkFR;T`RR%b>vX1BUU|AyPO% zx!{&G6_T|7j~dE-B@LaUcQ$<0gzW}Ou2Z&Wg&Y{4jH9e!Gd>i6OtMeAg*H?~PV%1nA<=> ziU&X>7d;hphu5RBn(-uhk-)PYUPRD6QBpa93Wl$e-U<6!hg-~QJsgM@UQhuHk4XZZ zQ@K1?KJD+z0wG)FZrG~H4G?(^gtkqu;Sj0l<4_Ws^4DDig^Sih9=tiCETPK+Vtojn zg=h6A53XAq&^`uEB3Q^xPr~AK-NzFL|C@sU=|rqY?{lY0vLAQ(NMHbAAJ>Ihl)|Ou z0elK_m1+Xe)bQykAtlOR1L_E#9gutb_Hu+ygXIM963TT1t>1l=g2YrLMIr8xzt}QI zSPuqaAD$qtH@@R?jKwQkyS5=Ai0xoQ0Nx%#q?|f*(1!;#l79c0!>-;C8k`M*HT;Im zj{Xlyr+h?AB>etwZBYxGOmb)a^}P)4UzpBUvLNHy=bM|8EFp1HhkteSTQ`BOv2Vd_ z!pMIY>un)irkyf3K3KGMe9dI?BqY{(R`^lnyDv#$n>lLEMy2nhO}-@EYeXr-;zrw@ z1;6A#Cp*3pzss-O2#Kpc7y{o~2|#9+-TIP|NL(|ugU7c=hC7p_COedwSD|Or-jK?^ z)+Vlix2>xd7Qw8K(84L zmBdFrL6s9$O><9jf0D@aP;_-Z92`PosXlN%-|m5UfW-(Ja?yDOPUD#6}!>WXwTp-QeR?&bB1k8N|0&Px`Q@#N;2)O|>KV0|!_ zU%{wMb?2l%YbGC3$}(^)-AV3#rK%N~evO^VKyX=m@JpKWE}qMyXaFDLyM80szmQlx z@#{GdzUsswGw*xIZSOS2aE$m;Y#p(Y>Lyt_`cm7U;3LYu;^A!m7t6 z=j%%)vrc?X?v~MNYmBiMc`fH$oqabm>mNXwd($_DbEdzD86h<)((dXN(qF{mK=ucI z94FG8vBTb+@yvobOJcE2o7yCVMe&D%nAMJc_psI@hJSXYXs)^E7>}YK0NKaJ)dgya z--wJ*s$gg?QbC1(bvydjR@tRBSF_UYE%y8g$gj*p-X!C$S}w96S3Bk4k^B~d>&HKC zf72L3e3|_7xxE0ZgN{t9Mab1N$qcmSWr;Ss;@^+`JCqtZ4Ba!6R%GE42}vSCUx}30`0{%^wk%Hj#cIq0()D^jB31@P{)P2w;6k{5eBPIMyI^Fhj~z zcwmn3GH{HInqlbTCP~w|ae2`=I%43aa>wIQfag4VKAbdXqKzkM0xcaCcv3Z~*v^g* zJ0+GkJ|Z|UFFJEFi#mAt0x`N=UMyuFgVBm@9WArVOrKBPbsXoCD!_S|t(-IY9IzYG zAe!`FJ-#s3;qEIE(7{`}6o&A!dqclv+4h{{1m4~={u7}pr9cHoE?X+lyvG5(*zt}# zO(4`)sq8SlDipSYCz={Z>;T=%3p}4~l3oNDixcD3WQOoZn}Ts{7YoMu$td?6Jg^IunS#H{}d?t00O0%g@|`wU36M8QTH=vv{B|oRcAom{O2L5p&#nO?HC(n z97p7bjjDhN%?$gEPx~Tvv7;mV1wg;fiO&so8wcHM38Gy;@Y)T0iQ{fSR$c z+g4ac6+YwwPZ1#iCR|!r0O4{Tj>_6d<;-33s+!FF$g@IaL)mc%C~I=~4@rzJr=d$@ z<+`Us^?UA|F8!Fn)-PebK+DeZW|m{QeoR=azg^oGn;EDkw7&T@Lxfa)n1iTno?=1l zASgP%A;SHOAzBe<`U>1(d*H)prb{8wU~;r zvF$0lj{O~kw2~}f0XKb@N6aZeuI*h}oN7^b1P-ZJzFPs5+BzZt`nw-w$P);m?whah zp$a$dRcfX(r#WSwH%d!s=Sab7x>S=LYo0XGIf(p=*yP_1V*U>s_AHXFnV+C&KvU`hNtWOqOy5WcjOxH1TYwhIDb zJimnwJlQL?Aq9S*IVelM1{L;FpnzX6O-1EJL7?Jr+zUf2;P>0MQd$lm5M{v4U$pS$ zU1Jc)az|QRRMka)vyKp|qXxV701**o85`t@7eu>49ocl81oOk?T*la7IFCA1em$3^ z-jMoXjp@bHRWvew=BrU*)K>eGn!bS9Ptcd(fbgjZ)AtQo40r{wIT&ad}9Jq-EC)Nka229f32kB${e_`m`g@bo( zODbT|$B99NUh!TZ!~x>ZZ4p@AFz_w(0af5PDu7J)@&Qso^06ES7{>`6uJe{1&Y{^T z3+mr(rfxzts{dic0toY0raMrA4}WCJM~7o1{gEjTAZGHv89H+>d+?HPmGy#95eI6G z222MFybX%E=(+cLVLao)MiKX)r#30H3228)EfujYx4)tNo-;tdQr^yiLr8!wQv9m* z*1toeWQ0ir@d{0N9PiIE@7kxL!F51j9CaV$WCp|_M3+}zxw|m=@tMult+__jBX6XS z+T1|}Q+C`AqB_OW~p ziH>YB{-MX$_QA+QX${1IEA-C}a4S`1hzcHC#r3Tq{r8yz1&6V^^C1Tk!5j9ldFGW2 zut$4-FIbZ3`Y)rA@@LS0cZh@X?0_*#&-$7fv)@0XR$rmoHH)?fon{QDzsuRS12+5T zb#f(vuR(9HxwtMA@ac&%JO^ofg_?v+selvV_a?Lk`;zK~y)QSEdb1dS4IC$UzgOXN zf?RFVfi<5a}f{`j239Qwg?s*U(!3>v%88TFt>kN=P;2?Up#+vK)JCvx9pB3xCwkqE(_w)BX z8o3u3%KLKAh=FU2QdS|5PeVKT?0z+P^~vwaH?kB8)>Ua~RjANe*DFF*z#Q3`j`$6@yZF(ys<|Z2im1E5F?5}BPaGWN9BjB>wpQkf!Bz!|Co|Ra zdC^fRn#|c}Qlp)!CiVz9gRB>@8rs5XscvWE)DmYVqZx1239@w;s{4NqtA%W&>|z|# zIb<3&*vn5^=zKQ6yW~`dAev>yGg*&fY3&&hUHm!mB2_#AM>RkY21RW=Y=xwx2D80v z|Mgw?LB>%2cwYR|Mf%sBh31@mZ=Tg^2tt{e!3~ZK;=IV8);Iq+W@Z(X3znhFDlJe};Wcp7u5%zE4`4mLS6n`BL%9gcMvLiqpMjw$NO0g} z`-SJvNlm{5PU&D66Tc@2$5nzA_BPBu{HU8$kF7`65E4u9rIb3RHeDz3LzA7~L*Qm)aKRUs`;r$o)h7wxNL z@jD6dsBpI;JgMXEWLVT>iFW!K9>%QX!Zor7pY2o8z7EP5nmMC;W2jjd`lUdxH(!-% zgNNPUI;>1cG{q=Z>RASVI@Qm_{2k8tU2?n)v)g|T=%6BKZz5)f;BxV2ezu~77YPYr zx>_|a3`Gk(*ADsMNLH=`Qy;;RG!9|j@mj&*M{_X=XN*j2U!&Wu1oTjMX}ZNN7k%C>2Dk>k;2EKQqPK3D z*IfH(s-e7BH5ArMUGT)s;gS&j!H|2_5=rxgF%)$If5}q|T!K`R%@oHX@S+r(^Rb)};MtxfeUrR37I`|sUz?22`3=9V z>?Lz8yi`46VT+5-h7}$Js>Wi@c~yMkU16_IAs$K}UXUlV*TMX2=+waeHP4cJZ;*_9 zoG0<}EKC*|l`xdZ6lcExf+0A!4!b3F_tfuJ4UX29MAGkeCj}4&d@jK-uLT95aI;*l%hQL*^eSevwmIxtj~k+v+^J;9IfUE&cnJeRWS znW?t(mg<%Dm{`e-l;Ye4FN5%eCMR~~N?`fxX zj<>&p`C*jt^?bMicV0SqR46?cUmvG|2yP>U;Yibx-LA>xNIxOHsW-U-lvbIAfL{p+ z6dD~~{O~B3ooSli1n14(ft!&+!KKhVBr7G*rRk|dn#P@^w(<+1l!Z4J_8(l}3A>~# zuQ{v~Ovgjs$t%wM5S;r`(BpN#oHIa<_OFuI0sWK(txZp%5c;G;;<+CVa5Zwcz6}+M zesjc_<;;7Edd4dl_%(D09C}FDhhbl;(WY9=;jC6${^jN~_DAl0sztiU9BKD^91ogv2eYXjz)v=5|O z`7SBM1B87t1>3ZB@4K+?Cvjo@$ZH==78gS%_RJd(UbOGa`p?6{LwRbg6c@?^iO<0A zraLd6dM11yDI&<>Noo#uJZhuRFI>n+_Kaj#pU!XQJ1wHYOKZZUg_N^62WL6fgu8QJ z4?_kmDkQT31SMWJt@X?ye^zW0YHx0do- z_XGYi7RxFAz5Pd0cyMy)`zeqPB{}hNQ!uNX)N<4n#0Z;7y-x526Bo*77?sH3 zi4~wph-2JKsKI)myW<`&Z8@lGHoc8kYVi$;BGoGzv`r&DWGk}^AHaauv<`eM={sOd zErCKRb6i(5%V;A=dSXISwA?tfMr_V%qC5*8wUxr+7k`H|PaNEQ%gxEyA2hr3GwF>V zL|2sTdZ14e^X7)f<8hgX@IF^mq+V5-O_+L1_LRGsVE8HdBoHkIOSKlO?d~G5C&^4o z#PTBZ8DE=z4MMFuZZ`Au$JGd{{hI4C2RvQdCX|T>kwxo7J1Q(+zPaX8V5v{7Z=zY_ zEd$I~OxE(p3klie8Th8eUv!Xl->=1*{Y=I}@+rv((cv}n#%`-)CupSxKFafj(b7q- z!QJmwoK@k{{j`)M3fr9PGL!7UWmBuK;A~+m+yj~$q9Xb!RFh0f4ylAq3IvLUqn8}SJqQXLxM+huM33g=3jGV zb;qpvQlf0J|HKP(hH#-OUis!68CQ`9yAfurK|ByQI>)bZ-1xG=ee;m{qM-55;g92T zVNnQbZXjO$?d0ziB4N5o*@k}F`>7{VWp~QkQQ9%pp&t?bsE?vjy9pl_>&kBZ35>j* zVZE{~N6(a!hs(_;c2Z+|?sjOn<7Rg~=Jr1JeOPDKoX8I)0ALPlG{R9NrEbzsx2>!* z_t*WRetmCE*<1^Fy6?=?pR%%;fr*bGL%o4jSmhFci@KJu9kT91Rl+} zQ^dW~n5H+{9;4K*=EsT+&!>;(;_W`s(XM@vN4Sy>J8A6JhOpE=;?URoY=B-WV5Qx% z4j#2$!l2 ziUE=mXOerJE~AUecC0bAc2@_eXHH3J@dYfYVs$!@0z=af`7ZWk5k24O8M!w>CEQAy zE7)^jg)PIpvK(%QUS>+o8F5&wxloqab}qz9EE)3193Br zN;Lgx$Z!iII{5gWt=mhPs|iQ~ zNmkoZ)OtOTp4dUHDtSzm|_mX=&K-* zm0PZ}21fN}5d{3xl0du(oLLz|)(CZ-PQO zC&P)*r)vfYQVI_>yFB5)y6=^2$g_$}%t5Hp;K62qVVke^r&=F5jRJlH-pjcYDXxa^ zI&@~>?e;x}pIhvD7)kQP(EwM za}Q#@4-)G9)w4jIYKsWJL9@_TAT_akfJg9=a>|JuEM)t6_%`aG&c56=%`(8W?>~rJ zx;_Ysd0631K=DL`FN)Kh(Eh8!yM`n+}v zFX&s0T9TLQ_bqnl7T_zsjY$=lE?kyZ9qXNhxSdOW+cYQ(z}^pULJ0$qj{5CHB>7X` zCexY^rZcXs)$&XFEFVYm-O8q%CndivBmIKYcO0#i6ps0T6Io^sYCXEga7TtnXyU5u z(2Cm7_pr1IDS`NHb_DtNU%)L(bGU12J9cGCqmIE@8L)(x1+BE1IN!*(RGBm7Hu~PM z-G;~iGyE02^tM-`_Kdi2WGHKf<638zyFWI2ApB@(6k=KuXna`!PBfE2O17^g@`BgW zdk+lNc6#yN>0@(a zy0Mta{*Gow2knr$Ik9S;tk8nq&`}(G{Nw!?+S8%s2y9=y`5QyS$io~$-ErWsNmu_6 zCn0tGYWn%-wp`qJroK%BE^cdNv{etqe8`?GP00r)#GXI=I>HWq7=SO&lQj5bsJ0ja$Mkv@x4ANPAU7Trg* zOzY0p@@j{!0o3aTa-M?}?F`Ct)UE4n8>P%~e4NMq&(8>VsjqjbHy&J!RMsDMFs3YJ z3FG_&ce@7@di)aG5W5%%_lzb!{mTpl-D30Hi%Nk)2BkL~EOXI+kJ0% zJh4T<)2NET=v*&ebjymu$1829g*u`W5=iy$coPDw2rh>^bR7U**e3T$iUyF#W^l=@ zJ1H6ne(sISQTvs2buF)GMIzPV!+41wY{qqeMLM!Y7Hx;o7C{)FoLmS-FZH_i)W0K$ z7ISiOx<6h$U13Gefz?>qh`O4KO1Yrh?@gBoM(4a8;#{?bSRV`AM4{#-Kqo-LfcnMp zl!GbM4!g8Z&w094R@7ZiUcOjG=0-+<0RHJ7au&4t#n|3 z`THRfiG)M#?{Gb2@G-}h+eB{yZ28?tfleFVp@!F?Mr}LPNc%x+=1nH$?>O?PC1`i9m9@J9Z|^3xFAhA8-j;a@2%w$$4_8k zlJid=fBYpzZB&6cny_?@=)63hHK4rG_Y-;aG*{po%l>K~+$KdNBuDR6JaBIIDF9n= zEr>K+_>74yAu-XY;}NZooRDhUU6f|Y`;K~#bBm-ml^@2NQZQZtX%pX9FQ`m=wQF5= z+k}sv9oanb+1UOun{;~>>OgRgqh1Fx1c3?B`(EsKPtF|i<`!7^O^3|!M5&WYk?&;l z)dgM=tJHXNg5Wr47I|2>k2t+}7+iWX5!i#+P>-YWe!*KWY!M9!2??hGUMD#01uJuX zcz1qc5B}7J3`oSUU}2K*PZxdn3>qG3Sh=mqm7|4W{_aEJxYjiL^2R zpD=rA6lAP#-+ILf%<+w*;y2?@+|*BAl3(a)`6bKIjf;NuMqY|LYX5M$rOd^AS=ThH zM&6yoM@Z^)mjpO(pYiahQ7&31obIpgPoJh7+1RW8@}UU8dW{R+;-ZtbFj)7=iPg4z zT=2S&PFqH_5nv+iIH75YX)>sXJMEdbe;!k;b0_gp66UrFOh$`=|fNld~1)ERm63Vi$ zmGd=lnlPYLOo=HjBilWhJ$zq{tZ47D%pIhP8js55<;*5YCz(IaT1C^-eKqEd#AqKH z8fx;xtsd*%b@X)YJ;GptlzsVh7aflCU~OGMsQCn;VX&Cvk8&sMzQ6&MVM3`izm1Sr zCZ1I$MA!3NQHl4nya1%=2VQ_;yPB>E*GJ5zHgKM}b9%G0KVE7?{bK->u>8acy zH{oL``+^RF2cVZSV}4A`;N<;k)Gq?U7v4C7h-eXNv4+8~=Gti0!!iG}Om=);q zNd{3TNju_ak{l4eVzWhvxSkU3m0mN8oZ?rP>0J%O_7twwJP)u>DhuNxeGI>z;= z+iU2aP)%pdgnKx5vlq+H1|PJ?An@^um9J*JFPf_^qLP+TWfw_Sdl3Q486sZ|K0;IE zKk95=>7?UpvF2!cTb+AuuBulB{_1)J#R znWNht)vYdbs;3czo#FZ%*X*nfw(%*pppSs9StXwbUSwQn4>3`HEY*Kw)i(4c{m|n& zm*d#w0GdTnUHBbix8K};OZOCKw0+s8i&1}dl^WE526vBH z)6@918(*zwFgPF2uO%EEYaraofeESy9DTaMzUVs{_>KRb2bJC1psXn zj_U^B1v=UskT~z#8GK-GSgy=BWR5lO_HiHmlu%?L?luU(X5D>7gc-anbM)zNZMSF; zcPebG5aWUffe*al#~p+X3LN)7k1nJkm#z_#u?jnR*xrmj$mtblq2+*?>`|Piq#Id- zuRAy>K*^Y3=NQI@C)@mb90Z@^#0Jg^F80voUd};|U5-*OVnjw+G3GB_=I}d*%t0>q z`KRgipLV#%{xbI2k$nM4ycnWuhz7X;U_QQn#8QXsVdAS3RwuTvDIzT#OrT^murqYj zK*ZygcgKdjgjIQV+OwO&uXoB0l#c#+5xnS^5Yh9_vHtZ49 z5#PNE4diR@+gK$g1`&ebLR~|q(ahoy{Z}V~RHvyIg4s5h`T+T_rq?PnyH0iY?$(D# zEg4)3HMD@L@S$4|iyBxIuIfZ_zPxJDxS|NnYNTyMft4oMMRpkKaPRsAO4`g3lYumR zkg$byudGTCjg+X(7zgG2C>iM38%a5zu>CH*aIWpZ#IqXBmXLq?9sxuDH2q1$`g|Pu z&4|s_mi4{f<6jNex01vVp&@#5h+anHW-+H$44ak@XpN3@kaO3wn%=Uo74-_aB}h=Uk9Cx%&iEvY-l>y`K3%(GmDuwXUwhKfyy-(R&{ z{UHrUo{Tei4xrfrI`;G=8TH!V3>j=o=zdt@yiU#IUzThmN2)tHWtDnLLjSDZ=DvZA z%Mgik6#pz#oSiOf#TuiDsZjZN;X9JrGiA`6=tgv_l@`lROizSOeL=CbV}i^9S#yTJ zl=FTlbHT`$^#$ulFv%S*y4GWPDL!U0s$q#;vU)c9Q?1JkHVJ{WtyV+0LanIh7URVr z#vHJWJukf4QB)kc$VFMTj2^~QcYN|8ae8T*ed!Hd$HgNkc)b;a0lbPM*4{6FCb_=g z8JUTLY~}-7J^#=zQ?+=|Exnh<-EaV^LW5d!g?njfiHKhJRC1}pS99{!;H}o29~2)X z?8o5$@*<6e>5q%2w+x|jk{)WY9(;{7^tE7mwDrVw#i&6L#+1XGY2fYNh~nMfzjykh z@>j!Ah5s7FV%1K4-m0Hp;I-Ru(8nZjoUnpXZizgsJ}}x^(I7P~2Za=;&?Kzn#FQ2lzmqIetvF!wA6%Hc^bv$Yx;S1$k_+wdZ$(ANHqQI)Eb1)K9k1CLkqQcuvbZTZim$o%V$(k`8m8r zGmLUD>m!!qs3!wQg#sRJlgo=7yMOdjVc+DJQ)2<`b&F*BMaDR6J-3*DWVHoBamnas zdCBWNk@fv$eP3QcuP%Z!m!pTK22#ORX$-dvSR#Up*ydhzI$WG#%<R|0|=9&T~{ZRGT%R5oX;-lGS(c;+BpkNqXzlSp@Xty36zFK zNg^+upK%yK(4>oKd{@5~_gV6JpVCdpKd;O;UX_?W#sx!UN0`=3TUW_T=xJeaf(RDo+nCz&kNGaA?cQ$ec zqFA*#h>qP5O4q7o@qN}~KtKKv}-UD~8t5QaOvY%km+>sqJ$n$6&wfh{fE zdVILRCFROoxA2Sdxsdk^_FAIFI-^Za){$1sWc$~N7Ba0@%Il|;Ogt%g`Vx1Bd$}x* zqyu&Xg7pBv0=NRKT0^IBo2G$KAHW+b(BT0bbC5X}z(gBS^SJz!RX{BitF7A!fb;;x zG?M@jQUCwJP%yt9B8IMLpa>N3)tdGlaa3Vq`FbXubW!EC_>VtYE1pLA>DX1P?BTzD%>Fs$4V>0kceZXAR}uv}AB{F}TkEvY1_g^!yrhKeLvJYrke zA4@pe5!U>jZ%`e$kDcDv-xzk94r3X16e1UQG*=9=-b4_zFa_><;Eq&0`dsmrTjINu z9;=dW)>iBHpAh4yk~Mv>j$_gOKPr9$N1lCHGE!nlGg+?q#p5rD|E%V-R}yaRwEUny zIbP^E(IEn>OJ_aZ1%EcyNcSBgYmRlF6bcrr-gJy=syo~31t1z-EN~@J4U`?9*^)23 z^f1~nP4H#CD$m5nNE@?4nMm=KsIls`pPK@(kzgD-XfL7v*R-UA@R8O_I%&c;de+gA z6Ys}TW6b6tWGFwnl3JvQjOJg9D{ey_N2Z--R~Ui&RJ8KI!P9TYQW>7FvXb{-KYuEu zCvA3D1rE1XeqYcpCgYEvo2FS6uv8vXWaa3F@+YYM%>UP9#ugwe89wD__EfpZ99Wx{ zU!~~c(k+x4>zA zi10|Nfw=vCXzw1^!_uD}Q>~fKF?-?I;J7z2R6a@t=bh;eQn~d)Z~qT>kLpyCqQHco zOq1TPgYnFdo4wm~7`1Yu&V=#JC`iiMZhZ2au76eq8*a0Z-J9Upj)a1*do~pE);|4| zE|&+BlkS2z&U~@R=O$|%u+ zA~Y1kAnSxD7;eg;N@eO?4#xhoS~0ysuw%Rgo0)*jpC~mEGIGX!jj!!Qx;?<*SowM4 zigB27t>VK^)UZW)v0Wt3@XD}IKn}7;f7$an-3__@u$@&W!Aqcs?PjAcf4(Sh81h=3 z#8t2PkeaIvFHNSqgJ>z9z^|RK`q_=DmW@ej8!O=PTkB=+ZWuP}M6HE>PFpXI(i5DF zoh(6?cgy90ISSPH%yxp?wErx9lIxPq!R+ag5%f^#!yt-B>}zDlxK&Wz+F1o~5#3Eh zaygI0@coagsPh?I!+;4nzAHI|Dw4Wb`+cSL$a!jB!)!R>8xoz*Hrl%_7BjV7i{vf% z77sdMbna|{Y;!u=5QH|pscKu*V5hs4@}w)A+~8Zm*sjxQ`JIL{5I+In1mOs0U-woL zvk6W2^XEWV^S#{O!Q-On2E-6m!tKe?Rxoc-{|3^+~DlO`*>ydgf8_XmEGA# zP&E(08gA9!gaY-{Mpe6v!0=-)U4xlyq|nrxGhfluLyePRoJPAjNRsOw`JR{!(ra__ zi$xNgQz89xJ8JLCAQEaw@re$CXmPuBMW0Z}yzvET-VLTOA^9}ai1K3{#wpC_7$w8g zc68gdaKtcN_2{~ZTaaEc3bjA~5mQ#7T?h;~#gn($s9~4s^G_b7s8R8drSO4+9*(r5 zK8H+a6InB!=D8B-pjhpNI?M$~N1C6f%hx6w2wARWn)$b`VK)5Sl~jY1r?hUH?wd|( z#OPkLG$p6iXzX5o>@PbPIB!*Xxzda2cxJP;8b<}<1APCL&$a7yXEvVhKm5_Ngo`*d z0Iyc*P>}mQ$!Mu`H7RiGMZPhS92@OP(WmSYNx@@hL@j4tqbt{{k@J;#i^uh@l?cMR z8ttD$t~QEeMp%x$XO=VblN|inS`4Aq%hu1cr{7}k!)H4JB9+ruBR$>lG?@4ZW8Kr* zNOfdNo<^0zLM+d7)(s3EILH;hWn-9mOnr{uu=m@C773quYziLqI-R`5pjcXFPk9*A zze_k0N|ZbD<jDh`6O#(Uq~^Cs!uIbQ69(qcDnX~i0)XI<41y`X|Ep~f*^J@_@shs(ZyWS zp~00?Q+{NynV{yLY1ru$TT7pC*WVF+fulmno;g#QOWcA`zFR6G$MO01hrx1Cn(N)w zi*$@R0M?{^aTeH{ZD7XK5xTFWR(ZdZav-dH-KinPz#24#0e1e(NR6^RlL);aqH~L> zXM6Zeate%1{57A_Qto==(cfT=xt?^GHT8i~ZETAOK>;6SWmKARW~!s##`}%<02GN2 zu{RL*9lI~#%w!UaGN#*;lh&jmbLJoWmjz~C*H|&G!ObYKl0iUtxsCrz!u!^n7gO}q zUJ(YyAm9Q5VotB8B9KIVoLjSP(t9>ri*Y2NN+6IlUgO$p>A+@wd_ghGpJJd3dI4aY zPdS9yy$+w&J-i8MRid96gIuurr|mq?Q{_XmZV@lEr7Xk|fowqVnjak2YHQAdLXSTZ z&(_>m$-*c?gV&%D{rG`-fBh<~j<_=heu9P7$km8q-`%5L+Ai$3M4yivyCg0-;PPZf z+X0m4Uc18zfFWx@bnf|axo>%|@>Bw|@m(XTaIgD#0RN}yZe!BYef8()Gr%;G7#o;H zE`;m&@6Ln-fTIXQ_i%A;fP$Be)Sh%+`{;GTclCqywj)fRUqqtK0bZHSPE*nH_9d-b zlMA2yYNvwyUNDw7;P<|4?&@p5cFnkGBtPU8t)gJ;IK9@1_3Y$$0Gh!DAKx#k%fyo$ z6vCRfcvu)=AlSrl7nBSD)Fh1hQC>n%ysYl+aay-Yce0c;oP7Ftf#qoq+QG0zFbW(Q z>?!cEgxbUOTOSUnemuoH4Y;Peo`2jx0Xo6qpKiXGm)!I5YU<_;`{@hsjPl|T+2!%P z4cZ~MMO3;L(1`4QzbFV^Dq@eOcJbAJ^#$D8(y^y8^TcaWY!5iy031f-6m)Wzf12@y zyJ9l9Yjz{5bX)v&cGQiMUdk@zTt5*h4^OcGUEo2t_>!**FG_WyFF*yVo>mdHT#yg#%cpx=)~6BsOEw%)UKi%4#|CX~*N>$dQJyQ+=^vN!_R_BDuU7?K z_nP-+aF^du8U06kVNv0|$F8+}6hg*<$ijLXum9iGg^c3LQ!yNC+kH)&cS? zKwbt^h4ufp2M+{+j8&uI9Hig1YG57$q5u+QY^9c?g?&7q$zfj=yv~rdQZ2D;Iww?U zAWFc2Wh=Fs`NyDW0MWSfCmoq$o}*k!|6o|BYnRfLUO0m&&|p z*JwNu1i`$$lDq$ou(o!1U620rbTy?<0TZN-T3=sGVFVIT{iI}YgO#47wn|*q-+erk z{E~|YxUp0Pf|q02=GgV`6dp3KP_oZWAj zq51v(w1XnNR2Cf_52zyoP+JOHHxhsnuu{nxYX`>-*O$kB-aSctNIvrLw$oXkm7nA` z?}>?lUJU>a4%3G3r>6pGs|S%pH}dh>B35suA)C3J9s(^x0OlHYSMcQvV9^0l`_SMx zBtYF106+@N9RL|C3}6Xf;y|h1&K)=j2*OgN7}bwBg1&P-)GGV%xY+K6vs7-oy}1#m zqqhx-_CN1W3$}Hg_F(7@Ep#osbIw^7|ct&4nhg@Q-Y7FY~8$l zkUDT>iojgPcaI%D6DI(j7HgF3J0eYY{t>>idw!Qhh}c1OkTh%ed=3KO*^MYPY_Ltt zf*FSpQQ&l1{L6HYf26LHkXelT&NYzqb7HXv(AyvGY|2t44+g}aQv#~{qn$dQu=m(w zf!@d@;BEj`;zka!PWu1xBr|@&0dx>f4di8?0SX85D|pZu8f9xXp$G5HY#!*5qCvMr zf2+7MsEqN4#acvC5-NpY4zm4w3F;i9?_IO>WDc?V3H=OiJEfin5|V$Iu_+3~8%1G= ztDDNy0yQz`ifgaNEcgH6VU2HY6KQ$5SJu%IqHG6wEIcrFng;s!T@`_7QswN)EC|rQ zVaymzFP--B+t(y1e0FlV?Y1SfJO7ZGp^Dk_k#pO8<(2YW4e{E!%*)v5gK06v7KIy- zdHFM)n--S8GWh^oGZ_O+$O#xUkc9*e8$iv!jq8sV|4(e|k5T^C(fCC{UxLN%_~Ow~*Htmt_15`d;ReNEpY?j;106-W89OZhaTv!Tm9@^a0{ z)<)pG&NP=}^{gYBAdIjeBsZ8u;*nl5Nz&7OoRFcZ1Sx08+fB`xk2<|Ky)!Skny&}b z@W9SzwI?m{U8UGWw7!;m4My zo6H4s76;j;?2;To2T&TadMh5;^&sDxl3b))mP}>h-snhTi0SZJX0q73%S_h#<~Y`t zn)QPPlV$68!N%+SEYAG;WU-sJn?nv&0jYw-ZYm?^-5_orDpN<>MMlYgnr?FPeb@g4IXWiA8)t)n%6Pyg6$* zzTX8{BdsRDB<{Pj7!BWQQogFjx@APUz-6W4rTv|C>-t7Dc-&78r6UvqnFX_J#*O*O zXCi~x*q`5xcT@BgT9ebBqM)2AyasiT?1rY0YIPQbe(m}xFU;Kvp$`0Db}_$HEs6_< z7bRyk_>k3}MG6?y0p-no_vqjn3{ho~&)XfA84j({vbrs-58XUimj~uQ_1$OZsaiXD zEi(=80&B3e{v=L6w7M!HJCQdtv)RPr2qh#hkXIU(4$KUy{hTor-Q$=)6peBaU(NJ) zf);w>l2J5z9Dv+Za0HLi{E;suy`lO*BTmZ}YjGbpeMLFy?6CNCq}qN!@LW1pyvN8 zidNdr>SD)5#AB$8X7=~;h9TjDb+y)o#&5fYM`)rS(H%->DxAi$h!hIJTp?5%&_W(_ zl5h7cY2k6W3G*)3Mu+tjT0;#CF#?a3CwJ`)X^0FezDzfSdML>VEtIp(L9y*M>^bh~ zEo7lxSgq}mbO!V@xrIeyrQaIl5u9*9==K(>CsWNbe}oyy2(Z;H#CneaSCFJ4w2?G=uh6`w|n+eJ_4q_i;MW%4bSod9S5~X-xe)! zz)MqPlwf-Lu>FEUk9N}&A^dkoMjXMw)k?%s&% zpVY%#%{37b?TRz2fy6)qQX%A48Wq>oi$PDv;e+Q;vi0zPT%wO#PmZa`$!)To2xL9Q@Wl zDmRToXA_n8+|KVE+8XzSGq4wKsA;n^1Nd)^2L18=C+#xrT4^zB0PRVp1%Wlr0&Z(( zX%b?}L@kDwg|WBwAb8IRsRi`p6;#B*MZ9jf+a+yTs17*`eDVdG4phka`b^cO;wkit zdzA$P5u$1WvXP0L6ppM0%p_nN)_Wenlb*0E=}4&;wsJpl*kf9GjTXLtzZh9ASXGGa zQ5H{0r8`t-X3MVe{frx2xuyj?@bv1N?9CErcj+ld)V)SbJZfpJXTE9S7$jx z@>M*IN>@jYs&z1E&`>!02(7zH~fK*(*%B~lq^Zng&G=Q%Uv>Z7V zk2Ef(B3q^DHf4Qy2fY_8;dxw-cs5d%3v!b}MJzZoWV7OpGVTTroLn_3SY-}VB&N#O zq(Tps1xOCD2kR{{uU;lJVBRG|ISfmA^uT;qMRJhs!DEjL|=>=M$%p(rjtRiKFi zaN8TlOhiZzb;-Nhf4g8;UKJL1C&xGfoAO;Q67Ph&cqgZ*;F@AN*9!y`Hw4{zf14{PbjP zcR|Tm`=@qiz~18ec?M6)D_)|kDmINInm1GKbp$;q|LSIbuIeXe)uB(xYQ-*aBfVBy zDP3GVb5w960RwA>gNrC&y)*esZ)ozkkdZsS1j0(yz{!KQW5h2n^H+U3&?L0Nx$$B~?er?!f2xm{~f z|Jjf$CZe8Ek08f$cS2eb%l{uxqGplG2S5%9W&}@cdC=#mndICi!8i z5)Kq&_RY(}E#+U8e#xa9e(&}YBOw4^f$yr9h}}N(F}T9*fd&k7Os5CQQc@x#5rOU%|fpdXp)<|GEXB`<64@9h(f48V0jok{vSO?vb-gPyem6* zIE8I!u|jQmfZiY$9ObfH>k)>cnL%xe$q!(2%ZA6Rg;(mF>-?#}zj*zYK-t6}kI~sz z##5}N7E8IHy^^z{8m;x<`C@nDT0Y6zR630uxepr%CXfihgL2-SgtQ!v= zVHc4~CR0tGLHC?rI1B7&U&dcAcwfuJgGJkMTJ?}#qRS%hyJO`%25L-{&bPJw@4i`eZiA5nm z`rCDQD*fyP&t1J`s>=I zKHmRONdr9B?z%b@v%WG0lKn|@Iy5xd-pO%v?z4#(zX$sqgzE7S0(d0lA% zHb_V)g_UtX8!PAKt6}(W3w4-^>O%}4k3%Ote?!?@c&$de>9~6->>dn zUo}=`#OCbo((El%V$cjHAYR!K{127$ISXb*Yf(O#-t9JQ9gd-`eouX`J4L^jAeRvi zY>fX<$eAupY&N66QE<%Gnx4n*P_MaI5_1;tg%W_{;!_j!4Nm?q>QK!(dILY5``iT3 z1n|j!k%?y14R-P8G_Xbd8-F5jp!yLe9XV0=0quZ)&unzjU+$$nNyXoo{__N?@g8u7 zDmk9jyt-@nFHDk)6a_;0-=GQ*$zlMJEKsrJp!zBxHO2gB9?TF`+b+Aa&IVv|Kxw$> z+tb}}uWsQfI8gD8R1oQry#&7H1%QQoAMZzmfBfuGU(|5rroV*q2mZkX>RJ={>UTpA z_C0{w*L(j&ak700nE@$W;|tJ75lkxR%C^Ysrx&qmog0CKyb zKNkgtkHar&_j?y50Cf5{g&o(!VU5}S6(848|qb+O;KudDz7ct^l& zcd9S;y2jf)F3%2Y+oN>n_-4*$jt5KKfU0EAISlGKiAIc_z9<1(!5zMav*9AJ!16wADV;^t+leWQ;1YnHAQ*^AP}wKQ*oj9}BFcn%oadea5Pu z|CTZ8%9j}J8U!rC&O$Ip{XT5_J%iiHe1I#n8Kw3!Q+ULuquV0DguK5yuB)4*-@s-? zbw^|zSZpOXRqJGSA}i@adWDA@MLRe)K$tSIuSZq-*<;|=Ubj0V(U`o6OJfkn*8YBg0ALF}_KR$`1R z?k8QAzs;!`B6vOlaNxia{$Jglc{r5O`|yPjS<-^CFG*pF zkUjg_h%hCLN|Zf&lMEr*m$5VUCD{@p+aOEUFt)OdWn>6pWXbZLk-op{{p)wV??1om zef{C$%z4gp&i&laIrq8G=e`HTEGRv4a*F&U#2K;9dzGbf%$ZL1sl2>Axa>5zEKD%f zvnhfuu9}95?f3~$5Q*H5M?QER-14xNCc$OgZMYO0kVr-jl#%?)`d7sRYvgb1t~)3c zs`kvA2JS2$&@T%y8JP~b3eVMt=9H?2)NFmB{0((r! zS7bUWAEh)ow^TXV!)1(&H=Oj06odMJmT@jKVNzm9vu7=53upKLtd#r?&KF?a+kO}1 z%s~_>JDzpz%O&7k0e(%!jAuv7iR!)*s@Us9Id zHUPAT`;6U0ZKAE6=+^07T{l+!NBbc1B}(xaB?tzilUe-&%m4DvV_9+#{<0Z+< zj}$b{5M)9v?|`gRl!0b7?$1Ojs3*~iPd{LF9?ggOt}txTgHM7NrZ<;xi^UKT2^zXa zBBUNXc%>|~zu0Gt8Ju5hdZm#W=(%6-9s{Wlv0{U#S&SC9Di=d4ppcV@)~kir0mjaq za9jS~P&Klq{w`ErY5q__0<_=dQ(yMbSA{=z`;-U0U5rs60_H(nu?6^` zx6g9et-b=8P$tBR?)?njiY85QlTEXiAD`vLr>1h!$e#oq%W2;U!Ny_0HHcGhh+n|za_ev!K+<`CciKyxRLJt9-N$Tb4JaU6hoRSeR4#U zL7cD8tM{p$T4~dE5Bv|L5XN}Qn(9#3qB^i^ zC74SG^Qbis(fCbb2mE|cm<;n#*x>Fsn-DnxHzL6cqP2P=xEHoJcyW(rKx$e60k;xnTsd$JPQ_+ zu)H`70+WcaKg8`y2}mIPK=}OeIm;+oE;1natmS zPT?I_J1J`comv}4ytAEwC0wK}@Le2>ev^aAO;>u3dsWO?&LQu>lXm}O2LR*Z`N6UD5mfZ zVt4T}A0UHnxJcKf@cYqieQbt6HFjD6b%Q$wIQfa=3c>o!e1IIws3iapTw>f*UcSa~ zeB?j1Bo*{(&lDR7$fnJ%3@Q#-#++W@OmdMyN@H%ITd|XPmh(4&`i)s#%&PQ>n08=> zap(k%TX>d#qN&0nSC4yV`0@dbu=iHTtn@)E4STRLSdd<)2aGLoK1szaRI3ZM!H zgRgu}JdL2+5?BU&0Dq!kcd?i&>Ot~FUL17NgALP5GAePBD6XOV4T3yH!wYkRCv_Qx z1|9j{Eb${n*yfEZ6;aJ4yq_c%Fac0Zpk)c~o{dNR7xL(;ILi`4sozo9*r(@xfxX^f zU;k0l@mQK_^$W?sRE)Bb2oE9Z8$IPS*`PEn-XBrB=!G;7R9DV!ka`|WGHQRDA0Z?( zQMCVDPz&8wxq-rRA+G%E4r9HV<1s~5fV2Bs8jW7Qri?K9rXGjV?zVD!2GQzUcrGP0 zuo;=LNp&2Jbrkel*P3nG2|gI-zXl&4(SYWyc9iX!kc@BNf?6*_aLlM>*&`TrWq~MP z9L6bj0oiOYwYssh)^D=$bOW0ACy&*3@AxF%{D~ZFHvYm0PzY1}HhckdP)}1~TY38|6J zb9wfekPbo^aF-QatHtC)?xnTJw)?y@o!QEIhfQv;b9zR7aT9DO!Gvwv;sJ%#kTUWa z4n7mR2Q%-kV;$A}gtfK7%Wd-7@jd>7F@P@UX5(KiNM6DC=diQ(X0xKk6}E4CSW2eweL7lCYmDpBvvKQdyL)99 zU>?jhxw+SVbBg8ue#ZYpvbMxi1m{Xk682UbA$GKVBp?`?1&ea_d+N3_RIZb#XV6|9 z>|QKmL)2x4L;GRTBjnKoDV+CiPQ#vjsE^OC(9h~9B!fj0oj=U8N1L|N)%5TLNxZI? zngF{zr9)vnLrBPrg}jaZkfb}TKRR%4=VVY?R;bcrRGsc**3T3Z*;sCT5<(_qn%>Kb zRq~t8^O`kgGkMFO;;Qe4h;>{})qX%fjl71{gWoe9{evT31>sWMhy1ntN#Hvr$rN}J zR2m4R2?9`$pZ^D{wyZv9cW80X@ZS|v_;Ep$UJ3!3n z;ecv?5^J;;$Q#8DTECd-wrXqybgMr0=SRc{K+eoLBc!#cSbo{t!e%2CbRM!qPRo}G zq|zOUG}UcqiN8KwD5EorUtM$MvdVKTV2Ew1KOqqA`PI+%F6+eJ0+_jTH+#{QBjL0p zLoxB$)mjg~+!m#+Hy;P^?|aRspK)|E5^bGVVOs$8}-KquO^->j_OOXT-H+g_x9AmX|s&;Q~q0ajC^dM>iLg8xNM zb!jhwEjp=0E&(9HHS9-Hmz#4nr>n2Z=bR)ge`81N({2R0>=50V6;Y}^eA}{viYpfr zb+%83K}+m-_tv$D1_rhF_tv4&G7*lpbHtwk9ezu$O{D@y?~G>`;i-f;=O+`mL#Vca zR*f~7?;0;}5kJwJ2n{Y@Z=Db#PNb*1>?~P2DFm6`g|7mlbRelJ5xXrMyEe2#NACjhg89$NNl>1E zdVpPBRR>+*#w3!(ui~2IjO%}f5eV7tgsh{{+MP6#lrU~FhhSU!T;#gk&gw> zIYT;#!d5TO0j_$1GqjGU;G_M;&8C)d_2riDDGxt3H(&U-6tJ7(~!jvCRB< zr+z*EIQF*PsHcT~#k=BB-I@oiY365F?TW@~t@%9=F9j2A&XgE)RmkZ0imVo4X8x_( z3&?p9ocXV(j*v4KP;QJ;#^r7qy2o-e3;z2V_4=Z3PpXmGwa=$l_l$OxDKvn&G`j1M zH$g5+Rlxm~Wx(Z1CF+`6j1}>+fy33DrST+pp>?gx_h5ZJMjBzMTGO3cJKD_mZ$>9! z`RFuvX>Sc}Mwd(47$%;@X-Atw&!JVKJ3Su^OY@hiPj0>rOI#ZoBg539;ZJZhU@lA^ zhDULEg_u(=AHN`zZTqM&SHtSmU_e2;D@VmGbk5hvE$I$cL&^CwDy0@lb@b>F7&)+4MegcuWyxrXjbEdn~$Tgjl zDZuFC#8oJ^GuCRAP_U?=?QAWIGR8h<2;-+3D;CYYQjVITn!Hy$>|66pO3VF)b2s;c zzp(1?P(sh11g5Hiw5rg1m)&xt9G>A6b3ew0>At;OQH4G!!WXbmh?xj#lUxmZqmNe<80GzXFyj^AqJ#yF9JdvFixG1?&5RBC3lb4!52fn?CG z?{uz7Wg%X2*Pg36Q1;$@;0h{`^I#(c20Qs1e-&g~1jX^Sjih%Vt?Mr`YuE=-ws zO4rAopUNr@HZX#XF;fgifNLvZw|A7Nt$a6pr6f8+u1gQ{i;=XEEWrqa5N6~G7l;Th zlsc1{nBeIrxBcJ~@u0meV)YK~F3A(}WwE3Kxf`%sgI@cpF-tKl^!YYM4!#1BnoeDO zUh9R~=V5uZU-*1h?y*8kn`Ao(WLDffjnAd=0{#ePm7K!sMEQ!ZE!^jkJ4;fojMC@z zkc^LZ{J=nbT%5A*IUgUYEFi%2Z*@l|FVgMUZF#eGH4#B z9*Fg|2RCpHT`aB%m#vXEF1*{gNplmNh=Vec=jKXf8%UJ_c+V#wi3e46P%_Hz!O9zc7FNl3J$G5Ua4fXpsqkk|>r^S8y1JpnHh*?p3Sw9?I`@P1{u z;%OgAaFh4PwW6zvk|{ugU}r@3^C56!nU1&sE_5EuPh{?WfcXUX0L*7O5VVgOU_OD9 zeCq45bP&8vW|y64BnG;)HAGepJl{THfcvBZqXp&~*+^&Y^YyCQTQ{eVyMk4h?8Va* z??zhIbTC(tjpzWp)3|{QP2z0b&lg7f=4dEc1ChPwC^HfR^5Foxg$DPob3$t1YLJeG zOzCSuz-x)0^*V+EJdzNZ720r1)C;b>sNUn9^|vb*J5+!jEDp4eLwUjVqh_m& z%5xfWs0Z*bRH+OSR^lf2sJ9t;f+jUF8QKRu10JsUW8oOo0Rz+nxH76#im=c1Z`=ir zz600|V2%OBTxK%K3#7j>4sc=|gA+pYs1#&X0HOQ;KNT`rkBS%{9qj9yEWO;%o!9mtJCKB6y*E(E3MtJ&QfLP zpm(;)uYVhRW<=$?LV`|0W46+KL0ksFUaM4C*9fr``Ik^T9ic&GR|3N7>MqdDb!^Hz zRevTux28UI6QOXYg0_#eg~ggktq)AiF*+dX^ibAQPBJ+O%QMDYid`W$NQZnNk(lgzI^bbyv|NE#5G|Vsi1vIQ%9AP_y(Z>fM@=Tc1sJuRb4(Ke92=g^@XLNIFGbJsi|)E2Q8VIkbk;Vs5Y2;M z(YEFiAH}-5&)l8(7&JEZ_i2p~0I>x!ir;{6#N%i1W3Aq9ZW)!rrk=TLNM#UT-9c>b z#c4LGmK!{(a!or;y7_yC-=#{P!cF z`n`lgR#-zLeY;NW=VfknmjXp}9{ybgOX*pb37|c=qajpd)t7#A9zf2*%9-u*kMTjV z&dYO`2;tV&BUvs6LMJt{q~>(H%sL8}gTv0=sAm;fk|A8U>#)r-ij)EX2l5Y5!K@}V7mWoN3t^O)=zxJ&}RsX;? zn-2fNE}GQb>SWWzzTwPmdEFI+%K*jQXy!@E%iG!Q$H4b*{8c_5v?zyeTZ{~VF=Dk^ z5-qdt5_HEXM#H{E^lu4tFtop44GPHKd9x+u`=;L*FUnM-m+&ZmR;rhpuf5%BanA)f zi%4Sv=|dsb#_t@&Z$J8_4ux=i$sQsWaA<2d6Jlq=HHI1KewHNhP#Q&QnSf}zZzrKMC0F_-obM$ z=oGa}j%HYi9=#x*`Kjg|4-^Fu$79KT5Sa|qB!0ATre z{0ha!dSCQcb$fNEI``73`WQCWtWiO#{JBn9MQba`5Ri^Wqlza8tSB|fKpJ_ygHcdv| zQ*roDcV1)fm1U0*Y+w!njU4LT25RvkvTe+E3Ep?0j;0wyx|elBY(YLnfUbGaaWIYE zhz8W@;i`e7&$HcUmNDxTUNpD_as`%Flol`Foem!(_SL?<@*oebz^A}`b}2~2uwaVF zeJD*$d9>RalXcwjq5uQ)#8bja32)uuvw!TG2tzKA!O7FF*ehcmE#+O@;|q8K#aMpK zPp9f8bbEx`(WEQ`g5HVR@rGGPkx|H($eR{pgu$MG!obj;fo@qdK>I03E@|r8w6%0E zxWFz{cm1N&*D+(1$V)Dd`rzR&X^;jyZXUlhF9>msXysI4ag}@8CIZf7Rxix?!oxWp zBE?Od5RX@wMn78he7QIB3yR_^>DkO&sq0A{Z1RH8?hXfyeE+_cYkYY>>}YMUKw&G2 z$1)L>bCa{)`<2?#H6{`%-tM0PspYRqRc3Txb1(evEqLpiWw0Ja`pJ0+{`dPp&$eb7 zJuGZ()aWc%_dC>{x&igzLyp>3k>xt*1GBlsxyN$Gn&*Fj-|Y9lscFLO{MyT_cf%42 z>6;i;yk<(tjs$T(Mf6rgX(4&+>hI{Dg?`IuyaGaMwxjKNMianyoz=RcI`3@rkwCm% zP2qkHE=9u{MxjdM4XGeu6N#7!x$9MSPCs$izdlkur}k%>&Hl2B0{$G~{759H2^*G5 zboJA*W{#(x{bS8rv^Hyk{b@;tw!Io5Geif=1jV`!f#ZlX&8X_Dw#i-e4UWMH7f31d z8aQN@Ttzer19QYEBZdUk!B?;lY}%TO>wTL!p1Vy2A;NE2OE$+4OnE~dUJxrpm)D9< zA!2F5<%+j5sIDv<9K z`G|KW%TN53df+b^bcMU9R+8|Iyo=p1T8$;Cun!-8<9Z!t6k=Y(u@#FkLJsfU(JP1M z#5ck~5%xZHjd60&{f^Gi=6ugK3$I02YI+^GsDi=Om!JB!dQ}DC#UPI7@NKwZHudCzU2`^zbdfYy zNBFq4)l^(0?S1UGwyxiX9SjlpTQV(`jQ=Xj^B>JXK3Em<|F;SQP4W&j_fPS@Jz76A z7F<;|k9G*aHu?y7GC*Q1-$d_ku1)eaoj%-6{C8iJmCoIHpqPKX*0-Op@+Bjz$UH>mG4qgl$e4r>GG(44VVjaEvry&;8IoBUBQqgFW;Rla zNYZcZ_*~!f{$1a{zt?rnbvn-3`}Mrmv!1o?b>H_R=9;b=5k5UW1_lO^hPsLY1_ovU z1_t&I#CiBhrI~*u{NtRLftnIVrYPuXSQ4rQ&5%F{{z%%)jpVwH5YpI*v2U zF?{|#P_Xie?Vh%T8>h%;L2)caQuJRjsw3WH|Nfk~fu%!p_CwwOPk)HVLBaWPrvX!| zJOzo>@D)McYTo4CXDj^WWA*L1CLaPC{%kSl$-~{{r-7TLp_sToHz!^=PgQ>TRPQ!j zW0%aSLoe=P`sV&0X~|xBdNHRDuN=osD$M4*hGpT?TXW3^&lJN|n2wiX1^6vn{D1%aa%U(zv{R|jx4S9A%`aQ z!XNNGTl+`9eyq#?-mbN1@^0LpyxD#wxIj1OicXfO<9IQV(1*gp!d%&VKT+c)x8B@O z@JtLzOlLoQi%Bx;_c^%YW_{aMWwX;%CDDc7J$LKpRK?RZi`?G)`AKstX!qU4khghk z^QX&~)IONH#S*5%0S~tI{I<}3HI_P#L8|8Nr`OMtEtySb@|1C7RD^xEKcd5MtL?ZT zK~9*KkV`N3_)Gogliy$EKi;Nl4BU5ebVNaDer`|KdMx)8X{8TkiN!X!{5|;Wt@ z=x7gxjg8&x_m!WACp7Lw&c%@9-B^J((?-u#ErHg(kGB)LySv+2nmpEX{D|pvRofMD2_$qUJUxyp2(v7c6<9n#Z3uW+2AiTb+GFf z%4sl${)CO&=W&L|wcuY4sMPA(~;O9}}-*={d%Kvu>iJC+fDOsab$4BNo)PSTE?iHXYH};%#re16L@oc zJ~rtA{AnPC$8h1ulaTSsUT+dxm88w-@p^{pLBPrCJ%s9ECrsB%(`oKE=QCcSmrxpH zJ<$jCgi|*gF_m?D^F!(1livb-@e2QqCOJJ6rHN;XnC|?Mbke@tB=W}?{fUVl$A`Nm zxqe_SZLm=Np6nbkp2IVkcqzH?P9lB|Zm@7irOy8TxE~xX~IY#;4Hn z;Q06$id6i34(wCLh5#SJ?SEDzctPBy0SaHK`u>+s2auRm7cRF zi#$$E034*sY67lS=6TcS}!;xGNamEGxD2XC10 z(VZ=|@vA|R&ns^GPD7Jari^i)5BVD$sCSih6e>RpJ2==E7M3ItHwfvlM-Ww_|EL7koJY8$owXfK* z{K;N(qH&qQj$$j(-BLKb^?XOLLqW>?>0bO@y|%X7YhGPaFBInsXoNF>g`b4zac z$Ge|U(=|)mqjI#yf=S}xB{vs%qrnF3wV~U0yqkQ7NF@}oLGA?>rHnsQ_S206VQv}Q6SwON;W<*kb z9=qFI+Jru191v%!?{Ko_e5Nc`zAnW}@4y?es$h3yr1v_~jw%$reDm_A7GcEuJ!0=! zo==Q#nzW~OCcpKhX1Zny$zRr$QM}DpZTgUC6K^&_8%w{&(`{civF-24y`$GqC0bJ*J(Yhe ze~i+R67$8J_b^KAYG80)hif_2lS-jC_0byo(s86#Q9T`L6Q78`WMF&^`TP4K66^fw zbF_+`=U!;PPEca|YcFr9g37Yl@%aF$Vx$gTr1xT1j^BL1swS=NTS$04d5rB(Tv2&lSh&)(Rmt`)IVA;)CG;$e@hEEN zF$>ka8*z>1hMJEP;#A3?xh>`7=d=EF3$?bq|H}?>l{?#fV<`R=)QpgWFqO=nd$J(;XE0v9{$}taC_q29>09!sqS^e#c>Wal3PgxaEkRZ;DFJx-?FvJKjnLh}^rEg*KMBuu&6vDv99k6b zJo2g2Q!WVznmyzdw&>@{c`tWM+1#K;RI6exZ)#1wll-ZxAv_l3Efx+|O1_PPUXI3a24^ z#dKY3d3u_h{caUM+P<%%1plOGBYim8J8ODSg!H9 zEc}oA{Ri>_srUjE0uzCnIvaI5^#e*m6;<`9N4ek*avMKFL#c_=?>jGAyG^vi*IvpR~;KfA%B>fWu=m-TU|DkR2HP zoOH66NQ!UTU#rnc%ncru9F}X$dV$KiUPXs3>taxx9((BOG*`rYPa}Ih*WpwC;In&o z$miC`6U@hR#GI*R1@o>r75oRAc2QiAa4k&xF@Nu?A~%CO`#EnBiz|b)Yz7;$Yj0n# zm#=2>J2=p5Cv7}pQCrz8wio-9pIt78{W3@B<++HFqy56Q>YQ3JU;)1}KQcHbU}d+O9bd-j0oW69I|_<5WEGg0KaYhS3o z|0fU;foTaPgh08Y<$eBCP?O0xzAx<)3Xf7Lv)ep;z6$sD0-eVz&~JQv9295;gGS`F z{nPS)iulO%wr3lY&4wOxXrWgnZt+5npu1Vzx!=iqO0GZDRdhL*>P^T_+~yMw)OqL7PwSd{88O4#;Nn|-G1G= zF_hbbJ-KRy#tJK0l}49}aD8`|t|E*D{^yi-)j3bOHZ0&-@# z0bj9_)WC&y)**R(0tUHFo_9C>qaorfkO6)Vp`Lf5`G{l+EbdNLBU^r*s*QEs>j*zpE?b2-Lg% zvZ9#P_=H%Xpa!$cruS*ds-5KdOQw{35Hx|7Zyf~NpSe0S26v|Iu{&Z>GrFZ5Rd0*> z6{Re7(v#c{QXWk0)b;T)th*ZdEGZ$`GfQGZDvCKuS?fjbkU2HM99E*1nv+A6=Y)IO zfegkkPiih3{z#~Kib!Mh&L5ho*m3MVxg|JmS5ljPWyUR=N2K(fdjCz2_%*KBj0VDr zaSt=?tD25(8OieeJjSoD-Rc<38@3+RiMg%PSB9A@ca7)zK$Z}Bx38Y&&G!6f07Cq& z0Zy35*_)J`o?~g>wQt&fCaU%{ToM&{a8uF;^(r*)9EM&1ED)4 zuU%?}kF-cRv}iW9i>f}Dxm?p@A;F2cCor$6GbOa8Vw$-#+vtVT$4k{tKgDi%d8jmp zG4n^0EgHFe50hK_?79dmP+1e$Ot|HyM2&r=t8o&#DRe2VhB@-B+eCu#w#xQ{+)4hC zmB{-msq|tLram!4<^2Mk+HgUqL9oyF|AoXTayt}O8?&Jd6g+2 zZ*uZQ;w04AQ?j|Is8@BIDgQvj-rfxxer=jv`7!S7?(d$X8kH z;pKUD@m}`yy+7Gr`5ESP)nm8E-LPZN!|=aL>|y?TM|c1LFd@si>WuG=@*Kp~W33WN z5kUf%-F$qQ>7NTT)gM?|Wi(LUI5A0%_&w1*x^JjPf*A}`p_62})4bi;T<*85l8 zMs1F2c`DoEmFYL1u2@TDef7V^seyW}?A&J(G5&#yoNE1ZnAhPq>p+!Zx!-(9|E~^x4>F8sOuVQ`p6FlrVe)Z9--cDTYzCd5RkOu@}`+OO* zxkKkK4f*^^3}^rG(`!*qCs8%NFuFAGya6x0oFpPw&|^=90=PBG%|8>@ClFb3g4;>?5NH zSsSUM{$cH+PQu>J#{)rgJoUKI1$P859!mAHOf&6HPUHpp?{?FmtX)36CQ>@xqi4t8 z`vsne-wj-hpxL0k>DLy^G%n^WvLw^`(pWw`s4aj#MCZ=ffnZ#lK947Ti9y-*J?d^9 zj~hgRk{|0=T}+9c%!Ov%igSNU!K;wMB#rMl+aOfhf7$HSgDLe_KY;A{P-YxEyLazv zD~-~5^sE#upDiD9(n%tpEf;(==l6MVCWDKeJf=zS>F4md;;lxa&b-`toqAifHa?w! zm@-4QINnK{oJ*bamnZ~u2L~i1XKaR1vz!Vw5}7GZ)3Y1OqD)OwRMlGVy$Dk~eOnB- z!)TaaJlzgB5Ja(c6kl}#2qj}DWHXy7yc z861_H=jB)4;)|cebUBafYfAe&yr6aK=HsyXNB#7 z_(r5MlO4JeUs{670V8x`|LR<`A6LRw4NtEM@WEJuxlT?hI6X z-!F<<6bgHKc}-`%69haXj7;RVgM8n!1e%aRspH=s;7B{ zN`~r&QZ{W_t3BO}C3ta<{QWmBnvb&nMURcofLtTJqon>}7gh3fta-Neo*x1a zA3dM1ZtNe?iP(E@O~%QUWTQS-F9X*9_U7-OuOE0W;{26HV2(%3W-@gRJ+q<_L~Y6L zv2ahn(6`fU&}~i4;@^I!9P+@Ts_P3kZ@Z-4HcyE?)dONGrr*5xZmq~WG)0q8Qc+d= zBv=RTZ(_5sZ#KY)g|e7$^x{Hdp%IBqj{M1Cc+jz}z5Q#~xh%O{hG><1ATLTMoRa&n zoUf+}3R>J|n(}_~#r4UE?beZW+dnuE+JERZ`}}8-&QvZ=@0|Bsanv`ma8K3Lo1pqj6KxkCQzrx%cP0x?aZ&wk1DwF_NqtxYwpk~Ny=>7rJPXXxwiPgZA_nDeFx zW|!KSZTta@&3>S4IXz{^TW?e$Az^|0&D4kGYy^rizLXv|>pVTX6I8>&GPY(UIo> z7?D$K&(t{sElDDG$Q(>A_2XM7$#9PJ0njO7#^G9$kJJnLTpxI>rEZL+}_Qp}$lafulk+!s5R-z9XGv9Q_} z6I`zQIqlu=Az|@kBBOs0@99$3{l%25!R1wl=!l)yt0qoTWl26Z%CpDB_9A!h6#X6^ z@m&7I7umYapC|v|pzoj6bwVw8YPZMJF;W0DOVQA-HN zA)+rVDS7i_NE)(~t(@LT$+pR)Hu8(30y6HY$K)%OB_871!-rfBkIdJxwxn`cHBz#j z-O23=F2&5H#H9Du7=I_T(Rghl8%h+7^L{n+k}_RA_0~HVZQ-jNddVHlBlfosRP4JV z8l}eG-}zHPc<&a0Uh92}DhH8`hs0leBe$+>2e7_6D``He2~2|%R~@^hgnpb z-q11-#StAzf!=3hm-&`oK(1FZrCTr;b4+V3C`aHyMsrDIOM5LdElV!DI5)-v)| ztp3X4%)7Tng(*LaDgN}th@@n-y8bbW_4!U|621Jmu*{-d*37rJ`w_RbN?Y|TM@8{@ zp6S@<%7G^=QCv6oMSA1ym<&FQxwN%$y_79Ut+`yMN$UJHg;HZVLsCPv>|=1(h4f3N z1`8T5c}*#S-HpB_v;VAg?8KI#IX z0VZXa=6gnd0^gFhWPu9S?yvQy5-uMT=bNi0%+V5(is1E1^MJ3d2V-MnmCZX>jvDgN z8bUxoK&qaV2~k%(%AjDMFUiFOx5P0jJjSP4CUr_IUt8cn{EgWuV8VM)ef!Gnf@>q% zi+qbAwSaBd9|O&`RA&4FRFVa{+jl^Uf@6)>p+cg_^;$c<+nhc9rC`9d8D9G@l#NIJ zRkl+$=6Oa`ndbMjOdk>0z5l}0jCH+=n-=1IZdqMcRb5Yy#>)Jh0iFDxz3~UE)rz_shwV=@{(WBKjr;n zMyLJxk&WVs!kiLu|7{}k_IxYoZCK#|_R0f(tnK}pEHmcK(ZwZ56kLF`_46M1GLqMM z^5?g;qCXTd=EWhX8v$_*ezP9$Hbx8TUFYzfks{TOO+tt)esXqx^9DixZ&v;vg$5v; z2m;bYZeCumy&}#>^kSPp4u#oSlR)&50#M9JWcNsA%p{L~erfhbje%y=Nxw*PD*j-l zLYu|R4I~_|v}gvMfRvhfVhBp(6ESDiuiZovvsJcz@$3642z^o|gw4RP zzl*y-!y`}C_~2BIYmb!KZSwm{OrjVWuvuE7`&5poKhbIN+hXU?m#qZ3qvw6LVRlKF z=rxh84?hk0Qa3g8$`Nyz%GIQ{)UYHnXlQ;U^z^IpLS@Jo9ea~LB zv>mdVd-F^G&c7OKb)ygs6s+5UN29`z`0=0?E!1&wtpj?gQbnFAcr45CQ!Op6a4xn^ zXrR~x)Nh5^KORFn(k!Y0Rn)zaZVg=<+VEDC(H#jfu-F;JrIWRCY>I-tIR5qHa#|t> zvJ5opf9hRV5a>uCtcI4bPf>q^J%^>y9>W^ZWE^0*1#%H>(4ru93{4rTA!N~3&`s~` zZH!Lc3XzpDqUGe_afBGcsS6d$$2b$AsDZmsWuVE;ARI5$NS>XYjinLju;~)e)YPo0 ztp)kjx1pR66y}4L-#g7!k{xJK;@n42t01(CF&(==0jj9r?_3H$4c+d zZLpa#1fNMlb+rzo`S7rgT|{dSp$B!>0OaDuF?W~?v^bN4L7k~GNI`Qmd3K$VL>d|z-^{HuKsxe=|E-FT zTH#co6I@v8j{WuP7l0F{24Bd3uJrzbzY?Mnawmt&codwVm$gS!Da*xx_(R8*88Zr{ zz$j>c;yQe1_&R>4sN1{@wb#;F%p`XQB|d{%G#)_Bq}dv{59KvY;MV=k4?Lz~n}?7$ zG?&fxK<1qSSycDa5GZldpwl99ANC5MYs{f}gS#~-dvG$@WGray#|kyb4U^JKCu_?|zt^eEIm5HfsM->0e&S$~O+$oK=S69m5$ift!KDk&pF5-Q^JuSE+) z(97|JA-gz9uKB+e8~|Yz6jHo?8`9Gq!gGTTbj5AH1tzDr1=Fe0Ut2^tI5?!G?=APl zfnr{zL+b_0+TWVedn(ro8Xc0 zAslZD{v8%}-jurb>;;hSkb62$#C0--i@er{zE$~x-nFwaUAvBk*bttO?|oeMwRGat zVUPtmG64o9~5%nBEYo9I`f1ryKsbt`ugQ_Zyj3ez@ z+D+GhU#sZ_wpTu5oJp{V70{(WwFLYa$Q1d#hgz25XqE=lX$Zq9lttcD&07JhsUVLh z5|J=Sx~~n;#M&6Y>`%o>%Zn+uJVRGnL{Y<|acw9Ga91k4M!mazv?Y z|4fiSc`(PKi(bmJfn3?&uRpnlmk@V8q7`DbZ&nU?Yuz*~8^w`D0JZxU18A}^ng*ka z(xJw$XVcUexB^xMVYwcpJ$Pp*#O`qXot!q|jH|`b?*s2C_fGLd(em*E2=%@AkW`yv zMFiFBFhEW`*T?i-!Gwt}U7Z~?;#(l+{nY-34ZFR2(dN}kzo^W8+O<$zrjJ$Dn0(6I z#N%;qJ!dbIH`!Qcu(>9=nX{2ygVCBh|J5sR5CTA6VKvK4OH=LI833dQe};X z@K1ZImGVsC$@8uEc1x7xp>&YS->9+e1M`FKQ#Fui(OT$B%U1k;Qd~0Uri;SbvXOK{ zf)|@FFwm?^mV>9 z1|K@!5ulQ?>o4ao*Ux%_ur*48xKjzGwkA5rSJZI~`TE=eYJ5G0E0-1m3L8tqXL`-l z^blaqQ%R2{AaRiOD(BI2J~d85MgDjin$r*4(_NeKv$z_)BoR_uKl z8y|OEff1z%Rs2M~5NKo3FlKf3uwq_GfVxW*z(em$yr2oTBC{2yizke>*=aqn^h}3o z-v9fG1|lLOiYYMKqyj|3991Yu4z^G*(cjP&dio6pQkPQSCt(*ui~rj8o1af1KMBCf zqj~32BmBG(V2`vud-px&5H&D0RcC|f%AjAx;~5Gc?FKRG>PzNg)Nw{ycSh!1@xf1H z(SHv~tL1h8Y0W|z_(=6OfW*EhKQ~6hRZo7cU%V1!2y%ExcYp*byRhOo6$3h7-$TQO zQ|fh;2w|RtMh4MrHRYq8Q55IlDb2#wlx{O*lncsGMJrMC0l&Z0J7Os&_MX?d?haL_ z3lqfqVQ_DumSaXr^qvO){@E_fUe-4}Tq0pnYNYhMiJP1InVW3zkvF}R?>22t-Yczi zeslvcJEwC{e6(8LC~3{*t&nX7$qv~ zxuQ<5NXj0jaQf;zQKna+nV<|yVP9eGk=W`k3mY~0qYfZ=Bn~SDV#IefA$R7Q$E`d( zgpgpT13nBJO;tYt+tk=P*&;Z!-UR&;=2Nf<4>|c&V%ZwFvg5b?vAVpRFn=PRt$GJe zn4?OO!DuglBy`x+&z*)130Z5IN5ZBA24@Sdj$@QglZ!#1VVO!2Et;xfp<0N%;3`NqX2kO>e*B72fd>z_Z5R=xlC&jEnA$B18%_H|~Cq>71x zxDeWoZf_c9bsr(S_moGoETXQnwE60UUg&Y_NYHgw5@Fm6m2|O>N|=GsD~DpM4Mq#m z2Jpn8Lp2X?t=o`Mxr`SNF<3JSN0Bqc_`reHVnZtxNnp1bMT9Y16Q3p>BFrdmQ)mug zW=`82FvvtKPrg7tl(d6UO#uPx!9lNQ$oqgz-NF5sL7P^}6AU34C2-UWHANjp7Jx8C zCel#zo2zf>?Pu(O_zQ^&BS9KdXIBgb3xtyv9Zu^zO@IhUl@b?r0hv#g8b(c4!vv^A z857WZhopiET@{cmhJ(dgu#-No7A&D+f)$P>3D)!{{nq_FMJ&Y*YEg@J0I@bR6l;2b&9<@JqP`_TnJ{pek7I4yVTd*d&z z3z#*-*atrqzYs7p@^Eoh_SyNe&)$EJwgl8{=HL)S;plLVL-SAy12VG&%%}hN?EGv% z`r~E$K)fm;6eX$oOcl0zwqCv=&ciRgKh7cfcnNKfBCVNLeMGAm{xs-29-KHlt8ZcdA~XSu60yA#ONk>{7spDpttt{)CU-b z|J4WmSf#vYsb%m^Xp5Z+8M4zl5oU+5Yjy~mCM;Mtv~cBKdnou9VrGMW3_`ho zJLpN=UDh~Am{sWW5jN{UA&VN{1?2zIyz%_#d(!vbv9YnJ(JKdH^7npjqkDST!JOa4 z|2|qAJ*jMwFG5`p=MIys_FUC81Dz-fo8=;W91n#;?Nxwv2wfN9_!qT<@9AIiojv|} z$F4RDSPgMWuaaJOSZV}jnL);%R7sBc8Hzl;$nA@aSeE6>l6D9!dewqwAKOrTF)?PQ za_z+g`g759W18Q)DD)J1C9p*2UR-}UHm#Q>=4@hOf}{VW!td++ULtzjVS&BIJI%9v zBrJvu?^&ICk%1zm;w(_xuX0;}y(0ZibWSH-5Rl~24!Wu&8s5qQ$``$4_p=h`?BB&TVJr&k`yAjI~V+b^TS+%m+uR6{}_ zwbqhz!nFxT84Y#y1siQhMk3Z|3Cn<0Dzlqud^gVLl`%J<$c2T4H-scZoGL9_vvB@w zJPu`6S3m9l>RxU?=1mjc4^?G4>+1 zl(wF@6Z0H(=}j>XE?E8^feMxmmK8K7@J zR`b_>O8b_>ieH`rYvR(9&Ca|C%oBn@V`OCPKyTcz-rJfYQDVTvA$+zA5J?fCNDAEl zzd-RxmFSdI_6vGAUi#DM`RN#Pq}gXnm8SLkdtR641JCKX$%jW$-JNgva(CNia$6V< zwAn8HD$Hp;DGqt;~vcA}kJwHRj zjma+rGfQ4TTRUCQeL*3R9q8`|iQ~9s=KN?NrF7P}8w|=*IYu=VNjVbwro4=Auk-1af#qS#4XwSUs;X(pQE#8lAv(eT}Q<+o#%;rpui!t#p;Up+*vCCIOhwwJLY%wE4=qOJa} z5YSo05;YgIWrBl!!LX{8hJn)S=#xeSV<=k|_#`3QC8L8RprGr6P8!K~NzUE2&ak*9Q@Os-gqR{p+>YzC+t7)s)w03PUs1G+^= z8SIMau|eI|-T&SDEA{gVym%B6y>i+E`Te$-Z5gh8mUS{5EY?Va+Ip4YeM{g;uz_;C z2F6}b8YqGH5x=y+W)ROPz0nS0fO7Qhk!Omcu7Xx|S4kKD?=%aQkA4d@1cyCxGBIm5 zT?gu`I7xz-X^riK!t;3x0 zREwpv&0jaMwTEaEQFN>)un)vWxXu%0KHDFKVEGN1J5Su&G!pKd9RGr2egYKZp}%;J z<#@4A9NTA`i=Ubn&hOTB_InO$)eE_p_38M&_8)b=y{xY&`-ss7J-e$zvHj_=Ne8KZz zG`z_}iGMRwaWuyI3J=k$nX}QTtu1SeC+^^L=Q@V5-p35dc!m)=M%y)RhWXcvtMWdG z+1J!GW+dCcupH_c@h*4>u&qX-Dkw7P0YYM&1GtAUrm>+MX%>|e_D8DpD>|2<*6yDj zFG;0O=5+d=2iArSAwo9Qhv{q=d2k;_pL=0ZB2i1_%rLI~k*r@_>dFyVF=lP_l>*Z= zlX%GPJg@ulcv}6jQ7+Gvr`+A?Jgej#a+Z1hd6JGAjiI|dee%R)>|wxd$!|`*)Z4iQ zzIgPo7mcGpo3#UE2a1KPB#QJM{CC}DfEl6a^Zwv86D4g-liX^iAOdrN>^9WwZY4y3Tc+O znwYP!q?5vln#m$>2SxxrtjsOgG2#uh#7wcy1y6N={yw>vZD1tWy9U=$&_Ds47yr)r zw{9UuGZPbPBx%4nF0jEWW~*7E4UBDN1gDQ zm<+3w_^d4&f+5}^bz`P(YqdX}SC3hD0a@Ga-7E!&4i|mhuABHGZiY_=RZJ=wV42f#@GgktHbdqo1puUxfxn3)EO_ zYZx7cI&o`jJ{~z8dwJsL*WTfR>ILzcAb;nP;4|w9!W^| z0Kw}Rbg(lAK%p@3?4WMD6BeTtZaCM(`dGV}0DTio^58m$A~!9v0+kDf3Lxm1aVO#a z3&MkvntBX#&CXtdv~cF^Ahw`67nG@I2V5n=(8nspFD*4^vW@%&S_v~HVaKF5UV&Od z@w?Y{{nUUBf&R`5(E&xdM<5Qt3R2j=7Dxl!-}wqWpc=zT!;B(83mv~ixpz{!=+bnv z-nHl*XR0Pk*Dkc=@xvOse)K3Hc2>e@7uCZ*!>j_rhBBpMSqCsAn$BRmj%B6<;ny24 zCT4qu3w;yB!mDaYqnlOO;@O=$&SP zFPZgbeE0@R6v9LD$TJA2MRfF<&R*U;L(+q_0u$aa5*D5W2{vjb3K}~CJ3&xl4(6UT zCZ?C4deCSu2*{xSew5E0p$RY%#ySFs8mpfCAJGc+f73>oTE-Lp_9PR~h=WO&JS)Mz z7mo#|dO+{^P?^er-9yh%9bsxr0Z5YvaMMN^p;*=n_6RgVW)RC1^i2_go%&5;4(_v! zuw?_Ygp>q1!7&IVYV;D!PFJtu|8ICccrCp(bEs1!Q% z%`neZe3iSe!H7P>rl+e5ugW}e{&b=_{{s?dOlTR4K`0#SpB>mqA8c*l3A^`i2^Jlo zYiuZjnfy2D67wxKN0|nxNH{Z~>!IHvOtX|l{-$#nja;5V=|;scSfHoD^^($OG)v?$*h}HY*C8D7sLsb7~%t#dlD{^z*Z`eoGwMNv&JGR*Lk7e zJVs7l1W?kvxqg(bW15*?cDaqo@JOPxj;kNo{Co3(KxLG_&mfM6!y*0M&T&a{=Pm`> zdwj*PbkJ9`f}653jLNwXlPb`YRvqC!AZ1`=gq3GSK~}8gzmXT!t2hs%8Jea^9E3Tb z4vEAfcLRRak!Ml~s}x+E|Bf^9+Wr`lua(O3++?dJfaR6VLA$-W?pMaC?{5es_hjun zq7!ML87RL|+aMMeHj&ubR^b^n{eIu*Ra%z!4+E(!I|JBfrE1RBdhP3kFB7eH`vu;7 zfT4T#GpJXJ+*m3PQERKKqs!TVsgRPrx9H*oZq#6*aq9oY7Q^`!Yi9hEu2ESs0P80w zGn)Gk*(bgIn50dbeeZgX!_oQ!ZLNLmLA8s`XOp&^zpvB2;8!mcYqgP=YuG(bQ9T(2 zC!9CR;0PmA_MSLWNue-f2ht2&&>Ha~JUnnF0qr*uvCZ?LxOOxcT+T@hN6 zIayUYp9iztO66zoaUHOrMqEQgQ$j~$kb2BiX8AITRL#tcJ`N`Qb;xD(^(8b1NzT1? z9N`yD8GEYu8|qT6bX>aY98KkXP76Sy^rFSn=& z@W&r^!lg(D3wofOkV?u}nUy9lilQ#TM9_t{1JhrWaAiQxJ{+f|3JdW`sZS6ev&Gxy zr9s!6;VxE1qf9MVk;-hS4&ti&P(&9o+l7p`cd5A0`26>e3vXC(=S@s(BNM~PE5^c6 z{{{4ibApdH`Dl&QOw@g)1-GQaIk{vN3vC6IGBx^7o>+(k=dJ7FPk!;N;W{vJ?~-TG zBZ*VR%o`z9EY^CqJPf|vm(JIKZ`5FTSHz;ef7VH@uH~#orZ?*6GhQu9wq7zcmeypL zsrOh`rDu}y7eu<)&tX+Nq|EdNADZboj!o-yzi%~8pYt+vy_wb^ntEa$agST14Ci5& zi1*$CpLB{%4IuqxTG5^_aSXJBr1nyU_I~`DQ%bQeF>-pPy;c%7v_*MFFV)GkYK_YH zbbVXJF4H(Stho+7N_yK6^ve@Udf%$ccrlvFAQ~MZy;1xdUB(M6(A`gxVR_8GO%$sX z=j_lRZZk=sl;dgcG}}z-KBMks-Ml>bxqbfD%!|iXU-=x+crM9Y&rq}qQf8LE;MIvy zv`qzH%%|+DDKeb5z6J*)}WQII^IrH`3_$GIYy#F~4eE(#x)^RBE+i zDvg!r!jGV19PH)+<#dw5vT_(kThm;Be2xt36`q4$Fk0~C^BZ|s-jAYv)Hp8sOc9>e ztMj46e{CLLZ`h}B&$o?|t)*3)^Zm0S-xJSB;GE(VL-o+yXS&9iT8-T&_stFb%9M>S zncb2qeTJgPnHPvlKN8ng{-JwBkbG4;`xPQ=);c;B_kBV~V&~3~53f^YrG=zJ=UuMU z5;=Svj^?%({A~AyU`E<869(EZ>RN9!AFxyWL)UKwW3$AwC`XPQ0ws*5=N!(u5$OQ^ z4#jvB5{KF_&wOd6Sz^ur$mrYNH?VKepYOIAv(qhBF8`-a1yt=TQ(e@D zk;|cOP1%i2KUHCrNU)&XsA?R^_T*c1c#YP)StvDGhrAhy<^darlSZi)g-iOvGg+!jw7h~H^H*@YEPU{)+7~UU46NDY?v9Wevz&zv#%v(IOTdD}Bi*p_e!7iJ{e?@UB;wVWV3O#%Zew z#i-e!eY4tO$)U#z{<)qxTl>jx;@f|>=`O$}jsy#{icfV-->seGtb3CJv=_H%@2-8* za{8zpIT`HpBd{XokGzD@kk5vxpf*iBd$8=J?_ZC?NN7niNF{CdtR28J8$u+ES16Sj z0G)`%l>jq?F0W_tR#4c9aM8O`e;vh_1U5SKRqwv0ZNj4FhKo6j$(q>)S|6wK_mrk4 zKMIEn_lHNC5DMu#CauT$J5rQOQ)X{AtqSu__C5>Eyj0$Qt=1}o-{PW|09F^=_v^E~ z8aa)2BBGbUA6k5+ai?iZn{#LDNe9s|P~;qs<@XoL5_Hul6+~!#u)9174$Ne|bTR#PWCQVg!iTzq;~I|K zpNsUbCMU(PYpNfdjX4w5PS^8{@)~};cIlp_iI))9poy$B74mDb7LB#s9L6DB zO#1kuUl90hLyVGWseGA`I)o*^xyKC2>wQU*K_W!V=H9 z+x{~cI8u=Sp=X~vo6KXPpKstN_fy;9*ahI2KU9CbuJ=%AW=(V6r0JU5>$k*w?|j~e zZhtBZ_)$qU&T}anJr%XgRW5*GnlF{rkag&q^EyXu5__qPVdZMjOpS7ToeS3`-;{ce zo8o61a~3QHMrnWs%Y=wCe}S?!AffmDJRX*JaJV4!U0j*qoBHGG9usminE0a_;(j zohDvQTJ8cb?ZZ&(t8IPuwP_jN*q)AVSrm_Av@}yMf5?>eby^Tk^4Rh)B>kk8d(N)s zm~3{XjgkGk($G-CiuN0)B4auwh?W1Mv@2 zNr>nZKL60ZpDJOOJzJCNMH+)rN~H;w7cn%j@Z`=qIqnb68{fG}mdDT%4rq^0W~aeQegIeJN#kxnhRz$z zG}G*ousVsx;Oa{tjA_X3Hoa{9NiTZ{W+H{?+u*tot-QvKo<_c%l_LjJ;^&3aftJ@l zwzS9u-Z8#Jg|wlL6_PhB6^OVV9U|%U;h)a1$(pZo00jnZeU&-{Wfq~tKpndZf;a0f z5JF3S`dD~Z&Q?hktI>RI-C?>Frfm@}96SBnu`9N+59l8e|KMgsNLQ2AwarI?*IVLoeAwQeC62bE@@+k}My|9G1 z54$bUWt&TEj&AMmStig5omZMrFMP5Cir2>O5BpSf*LFMxxg9NMD4ua6q^#7jeXK*xZPla}3P8>+jPzGGtMsKvRX85mHx2+xw6 z3P|XJ1Mh#+Q6dINtZ160R3F$ZxJs6;E@2U-)UMO@B@JP>yVNZzckidm|9l#fvfyA3 z!U|If)?1hk9pa#SP{T*L1uY6j)%937IDo{a0RjtXe~lJs?`YTvroVqRjQq9Bw36G# zxSl)_tp5>OL)7$H=_P|ZRQgs;$ahTl9_4n!@0wul+6C|ENCUI1GAX;++3d_Nv)e_n z7rFDX8dC{Dj-kTu7)Or=S!FpJf^%Sd38V%+CyF$xs^HK6^P$5axGAycxLNCI+89Of`kY=qqPM0snG*p=$T4>@&9nDsKPU+M;GOD=ikF^ zTwEolsOe^X@T44m3H9V12 zB@RX2OQY|}3aj}8h8SEH;!-2cVGn>A=ncX+O%IQr@_ys@<4X-U7ZXsqG66O4+YM}& zA!1*_s71Q!p;0k)@c_cglmo-_~Nk?s)(B40^ZqnG`XVd@# zuSN^#W3VsqI$gKUUS~P`Y4J@nvrFXFI(6{U5^k0Hewz4=5-~XAb#$H|4e?&X%`6_m zQ7~9~253wifgZ>F?*Uk9Y7uawA8HRc5JXSl#Xu_~4LQxmE}{?}KQLF#%I-NT z>5I;3R&ChYmI!?=Zno~k`MD(gNSiY}(Oo4IkQ-V1yf8pmJ|LZ9oC zQ&J`R!bz4Z{2G?VdmH1FCJs%h*4Jw8=}h(fyx0X4%5*;|lVKu!4>@>Mlyt!Xcm-_* zM@r`2KXME#KGREg2~NAfth@M}4dWA*P-_)*qe(7d!lO1!GMJIhlk@wHuwdFG(K4^z zexF_xGu`f*kEM~OT0b)FT{@9k;2n*{Kp1t=%-m`0a*D>1q1bvaL2T$BiD~_8cC9&5bZ5-o!78-@kP&~C;NfT=*G4pj>iSgA+ZE}g0gltdywmeGe!^j(h zgTXiN^VlRLbLqkTlDHLCwBwnSJY_>5_GQ5V&`%U3rDjtLW+P0n?C^Kop`xB&7b?i;N6CThQ<*yIhw?u_;#WGB4bo~ry|(uvQgtayZhq%>=p(vo{hfENH!H(}qWoIX}h06#$&Am*txR1UH>ZytbPl-8QT`J@+ z!DI#O6UdT^WoMQG*9WK`4Nvdv=VW}vRH*Ec7~3ZdOip69T*9o4ZI6N#qlm=ZX1HIl zHJcqBegEw+$4}GfqP`G$lPvp*Bg=ZYQNQzVz6s5_{i*8?&K@@3GlB|klSCP9-)T_4 zQQ2gNbCpXqrO5BK*%m)`y;)P*T@suDUUrqo#!(9#TA8`oOznN%6+PxcZf-pR+d~>w3e;Ei(x9Vp6MsE-)8lu z%7&K!{ShssWT{{5{FQC;!j49002=S*tFsSwzULXK8gr4by%=zAknlhyT8XVxUE#g~ z{!thH#Ay2l)7!>^aEHpGxJkU2o^tzM`X`Y9g70Q+IR~87cX666mt=ld*A9JYHz6{o zF83T?HG%tK*WdE*$i5i)nr8m3VNZ!9htCEpNB5nD(1}0KxfspyU_cru4V?_Zb4mrK(flo*eYyqo%JLwK5|)Oi5Z-aM%oVAHxQa69`Q{IK0bRQ&r|ttN)8x=jtN z-1^xAbMk#fT}+Ce^1J9aqQ-(OSBma&&9u5-$}YK=!xDC>JvpbF+a~O?&J2oI+nEhl{{%?$PoW3*uJY*Uh?|i7UzFGF)TCbAnNQI_=(6 zdnt_;LVcAvya*!>`ejRuun+&N}|%F-Yt+Xft-A zKRBk*m-f&MJk?ocFR&S}I~Xs@JeR?e(W3sg--_fPvd zg9^%&DYqb0tCf*w$>XM;3PV0I3rRbNai4nMT$Gu3_1&%(GRLmaf z{`W+%u%)aOb+JWl>b)@~QMIPGs$9t4x>mc(kbII{UF0Tpzw+Tk?TZPKc%#WHmoCrS&TZP4O3np)EHfhGWD^$B4)+&e1csD`SX|_q~dCqR>sf?a@lV_*M11HVBO?_2oJ{@0pfbMY zGJHdX{ZyOWh;5r9^-^=S_FYa?NViRx=K4QR_8M4+Pt=eLPw;g{?fkU6@lB;nsjb$i zwNHXME`a!UyhOvjMhm`jC9*D++VYO^-Un`DICq`y`^K2O)y}1&Q``8yANz2gv-VUvINQ2b_m3%VfX#1yU$n&+tdK>*r7POz2KYfTI zvMoAh+PQP&%vSxgjM-70-$U;Mjss_^53i9}ajGkCNhy`0mzfMwb5BsmWcf@RgQS{m zeqkmFcW;#OHF5`2-F=UlZ^KjTBcZE&jAKpc=xDz0%BK6@!zU+Fe;ng0p;2|xLT&?> zijPg1Gw1eg|Kptxj_;nK;jbViM38*kZUNCX=sT~<5ukqoDLc|S_{k!aAe_t(1?T{c zN{DryN6SY%vfC4=Cl|@dZ1ZU>a0nIya$Mm1;7ws;8D;F<`<+yxOR*=O{qbyqCJ0Rj zUE2FxW<+q4FGck@5tp1FU8cQ5laVVcg;QbqyNHB{SkHf#1D@-YfXhIZH9gb|#RVc# z7EzE5VLBs?-frk|V*lR{a#oFa>%MD%c?p-^bLD650lf_(Yq2O*)8!7?T3__}W{ep` zbl0iL+SzwrlM-dFqBvsQQn9dBUxF|!0Phb=DoN%q>D)y(tY{ueMB0Sk z1MM(7BhbJ?uz3C77IFQ?*&318+ujZOzzgU$46919TSl?|U4n_Z2HJJ}eu~F`j74u; zeZpqk5u(St07{WjSJ1}O28#MZ-5qxPm?st%^pI41LHqYK%wc;t_T8O%gCT>aX)|~y z!ZKMjb?Cn^=}?1FjI^smI)rk{Q{;h6M^}M*wV(NVR=W@t(B*7y2*U_AKbi8cX zH+6>?S7H{iC6okF1yQn*l$^CPI3;*R7O z^hPycQ-aF)I{XC9j6cOrz#Mvvbj0PI@mu9ajj@?K(0PL9hU0=bMHV7ug@^br?9*O? z8Wq$KI2zT8i;F?&`A&nKE`k4x5@rwwpLG-X+8q$l$_$iQB<`$9|E9iv{mu$_4-n}w z@@D?q__fV|mw7A+qakd>^_Oq3h{0C86^zNWUn8&6IE(qp~(6I$W2tdRiR zS^(sX)JNp~47G%uMR1t1BA7N5P(jdGrX-3dE^yJ+z5ndl&E%m^fTxPtv4&0^eu?it zY{$d3Ny0z-@@rYm5+5Xwu6PL1uL{sk!i}zXGWF_*m)jpd#Ij`;1f!6BhXP^^dV3)^8wvZ6fu3R!{wq3|8O(# zecCm_1*GC~PluHlf8~*zl^1BQk;ixbv1DHo;^T>30zgc9GB&1(d3HODnM->$X>0bx zB{xaD*yxZ(TVk@czcc*-+3SM_aXi=RRZt`v3V0xCRqEV|&; zH3une$I5&_m>wrH3$Zl$w#CSKQNh+2**T-hOrAXNMqHn;_s##?QdkZT4>xgFd}g&t zwoR<7LwCAOA*W5CXZ8#tH^F+QpGMGQ68c%hc=>}=GPJfcS|hIY=%=N=5nBcL} zhns6-c6L&>BWrn659XiGX9R!Uj=Xb!po}?&EY(bEb`o>?X^c3=vQJqzQ*k<@_+4M6 zbm?xD+OC6H98zQE+(&~$Ze13$sXQuSDqL0`mNQ+y({_Zo<;#yMuSXOsv8k=#NypRn z0UyH4d?Xre{Sp11FZPDp;bLi_L~d!$(US(Uw#=aIMn%z>Y5Hqv&?R1rQQQnW>duln zcz2Oe4+?f*xJSfGy5|dMmQCQnC${|Y4^-}t)hqh7mEsyWGA%0|bd_TB4n5h$Sehf1 z3UaZP>V^^uJy<=OgXOAblj5*Tr#THP&v; z*2d17T)K8-CHMpCr3xu5NssnZ4ea>KUWQ}z6CV+45v4Gi`02c{(Mo63HFXL4IcHF< z-TwO#^z%fu#vV+LbQ>t7`w?5zT`WOY;N;+MZn!yNadXy9Y9WF>BWQJ=;V(=Tj6sYW z9#Bh^VI>UJM;W(yi%ER}c{kL7;iS&}0!{PX+?!&#Mf8xC3fvUT#_OEbw!6bjJSlVIOLYLvA{(Qw&O0SxU9Dqr;@>9vQyM`8>5e zYviK3@#;!$u*1ni>$IIoy^;ie$G1OFDBCe(gWC~8MyMKApVpZcnU#4vWA$eR+i+=a zq<7_9T=<_lUU6ON#KM%nz-QkS2yf?@D`Mpe6h$=XtKe`#$}?avgPNtI!_ugm6gj2f zvG}E{KCuG|*KF3NtMFHNE0(Ofs_r~jO`_qg_;B$d{bTg9x%D0U4+oTLjpO)?oZ*!D zUnEaC&U-&_#o9Q6gre)L-hy-{>3H7LV>jUHjaSYos+ZyYOaHeg)BcCj zR`-dG;vJN@?u$u>T@!W~{3_R`JlVsbxdufeUgKnxOKs}1;1^;>)S!)c|m%Nyy3-ET&QLZ%<{?x^Xu zEy&(MzmqJKOSm$9*Mt6Qh8tF9xgp7c8xIh({NO^Nt`>6?{^CW5#v1-=0T`P#y}r-P z%;2v)fgj#Cg606Mz<*}|<&$cT7v0?g3l(kO!+mbsO(gkQZhne3H!*i;N>t-Nj1{5B z70sKZ3h{H%1E^x@5{;&_ZJy5#@6b;!3Qfyg+$rRDGor}6egb%EAX`-TRn*hygV@lYm-||Uq_D;Qb zi`KQY@?X=P{l+b$#!RzV-1l1m8`*bq*KX~jL&ySSes<*3-T%}?{-b%i!*^0T&;LdN z9lkX1%W4mb;{c&az%FTcl|~ZTe?|YDO7LC0>aS%`y9>=Cp4xG^eQwX=QlTrkfb((K|HW60 zwi>4r@3Hs-!X>LZ+_Q5Pq(f6%}|N-gw0-Wx?S2d^;Vxjqcv)#-~K z(WF!n(`OfYz#O{pN#Mh-9s) zM)&*m7$>K$UrBp@-zjc<%@(VAw-1%K?t14mIOOT_5B-K)a_Mu*d7OXf@8+l#R?QzK z^Ce|A*xwMMO9uRb2jd`UjW1@0UVz;F?JvS;n#Y5c7R{h4gF?MhCG0@M)RX{9{~H1o z_1v<4lTH=zYs#BN>1Caec#QBwy##Xuj4cfQxwqG@s6HB62w?$6J z%qnSLJ6t@E0|%FBcqlC2i%9}a59*{FYYEJcQ+eK?Ni4g+>x|7-YoaQ2Bs@>U_J}tf zy4P)&+@z^WcI!A4d#=1LzLiRQ$m{B2{iKOZTMCCoIXus~^ZkaH$Yf$=9D78pw)#Ru z8HQFLz0P*fFBH^&HK9e&*N4y4!o z*#bocYEXc^MsTn#V+FWrrlqoTHVzjVK=yMD`Z?>rJ5xYiDa#5#Aovwk7|>_UjpSEw2l~xWx-Yb2UM}N z6whC6iR-mI;cF#salVfz0;8beS_pyoG~(d`m*Z8vGM!oB4vZhX4XcTg?KtJG4>}3K zC*C<3MOq>E=zIrX`1jG=sp$v4-Gk6N1oktpxq&k90sPeA@_#Bj>qjS&wQk0${ zt`Q_zC{}?@4?kRQ#4W=)QiW37jWtjdh#j|A3B+$gI~Zt?NV+aLQcNVF-w9?)_phI| z!SrR?0&VW&I}(V;l>n^KS6>|EQ;R-&b{XTZMJCMr{DCdsGcr=$mseKCMlP4;$r74o z66nv)Kn-_`6_DBM%gHyj#8t?UR+Q;C;zI=HA4?-o! za7IXLxs?FHL})k=*gR~~Y4!3^`a@sJ^mO7dUepV9FTl153dgxWRDrgku@+FU9RsGy zFgbb`{n@&QKOmu(9=8Fe2x+Skhi-WwzJDjpc^VS$dUj~-MF#KtRXRBhW$b$ac3Cu4 z27P{T}(h+beQtcvR_OFv_vaE>^A`*yM(af-0}2s`kzzGB|JQQ(>Pi1 zK4Bm}8rwvlTzkz2Y~HoXZ&G^uk733@^IlWL4zknYhhvYpwN^d-=ICV|e1-f19o|ZV zpzpzwG$gB$=j|RPv6PgQV{Y*LUy=#cK+uRYrhyEjZmnvp_*=lky_I#tXS2(BSiQlb z3!B3GJdFc9KjQHVzBw7&O0gb_emDBLkFya7`JnSPc67DsrsSLHp1aPhGhd&y3#y%2 za|vzXBy3U3&A;M^DAwTKdbh`89$3=1P`3Hml)Kjr9XH?Y4ZEdOdCXO-s|jIbg*NP3 z_ZS^JS~v^n^994U8QGOzT75eD_;mWI{wPBNlk)i@oSrZ!7yJ?LNUw5qtD{y+iLL7G#_2-sEN$P> z(v9Q(=kz^8`P)C94H(|@w`)Rl7~i-D0V_mH%mr)c9sGxx24_nbXCUFoqLbwa&w*5L zOa2FJ&&YO-{&zJg{Ab2{ySSe|#PvOXj@!obvysqk3zfQ7qlj$RwDO_Sy@Br7p0#c2 zQO$==!rlB9;ZoWfDJH}HhSlGR(YbY#_Y;s61v7-mFG5v?3=pH~s(S^7h) zIf*XKxLw|SIlc?vwr^V7d(X{M>ic^VppC4G!1?a=>nf1#CPa))e0d>9`@CW?K}SJ1 zqF|)EaO}1(`L~hsE$F$!rc1^z#?XT8Q$zPmK@Ra-sIp|_oY;PYltO15pfp`54Ily7 z9GT98Rl@168Qr>w_SC6Liuhy^8mG3i@w0$f@ostz>M5SV4A|r(A z#S_uLs5A+p);$xP&8tJh;z*@OKxf&52q>RoKB?woFThT^$Ni~XmL&m||AqPvg+cx{ zm|@W?*r3PfZa;hWoB=?ZIx zhbSS@j8hrVmlU0=j)>5tEiU}s3!~7B$`W4_;s^?gV_Xzi5OrVFAuTke`GtrA$iMyY zt!^2ybKcJEsXNuNu1~xtF65@Nl*aAhRbIIBMI4}LxMgEsJ?zbeDwb?svcJGEYwApE zEGI_%c490GxXETI&8?FhMVX5Lbrvf&Ua8Z*lYX%NgmvGX*=6V2;)_OXvPOWA{nWVrQ`E0V#< zgI6(O<+MZqj4SXwLHl0N4lZ<%pI87S4vzOIbQz~6uV0siJ9!rrTQKPoZ>c7A0Ms|n zKM5Kd2PkVG_)!))j3n!eNDngh3A(2oO+S4u(Rv_k+;oN?)f(qyRxoh4n!&Vxo!0WL z!kV~;%|YC&l8$4#bw1CQab*MglDu zZSucs5fi&NgWnyOvGg&Y)fFdvlG!3SR`@dCrcw%Y6<72+` zMx6qLaG@rvJTCIGS<{B#=Ur~y!KMtA7t3A_zW5DWF!`qnB3vsLTnYpaz%G}I zT!ZM71KL()z?z7{;SWNS7oTYDKen~H;_;)0d7WR(1zO<_lT>9j96|!6{y6`H&j}Za zrju3ZFI@fYHTSS~2H*6-McxIfnE8pm1}bxY>QkD?QVTi4bvz|Jd$GJ2Cb6F{KAYuO zB>Au~c`81K6!wZI+h{7 zvhGI<^?}@Ws9IhJDb>aEO2?_?sIBW8Y&XqKY6`cmoPm{bR;G_yH^v4^vf3}n_aUTO zkYJ9adN_Ylkg;3~7n-*+?s}l!n>>C&qX1HF7c|n_nN{<(%3ra$Wpymv&m?|e+NxXV zXOF($q7vB;VD-$7dkM**9GQBkbs80^?Wpyb>useEIU6~X!>h5^7;mqqxXi9kRR^dy zyeyi&>-#KG%}9WS0gP@j?2ja~hEe+sqS_6=&6;)`$htN6YE+_9u`CMl`kysEYrAd4 zE;11$Dt})TZ>~GPkN9ZGhl=atm_v}D+I`v!L#;so&NLm|{b-PrCVjHc|2*_&e9HK| zF2&t~MBm+m0G2H8*oC0&S3g)Us!b{CRG6soAQPv+W4;T0r^{KMpU7bsxN^P=EJqn< z_V@9Xhg_JC)Ke?7v6T z>Cz9Wsuc+QB7CkATjJiZppWyk#%Qln)Hzs%0xgl>^2T7ZeU`?z1_OGF70segGEd(S zfyb^FV2SvKe2#^sf?8YOQ3in%B= zikVve;wLN#DETvyv4b2OVo8vLL;FF5uoP5>Usip+aphlz{tNf#itZY*qn~&E@NfQy zdQ2HZdJ%X`X*$`~1BueC6w;`#)Q#!eM;*KXI`Rk8IDFUG5}@87)f@fS0}&Q4o`6>y zw}RkLs5YwBpbUKebcNv2-GyXARuK7&45zlUV3XI<0hgtFZGJ#I`93euu3*{zGvit0 z>iUis-94cA@QMcu}>&B&|c0^AGuJVRY$RcY)0q1~{SN#@gP1q^0D% zfS}hCvzx`r6Nv9unTlg)sa^6t{_PnK{Y!|+Cdv$x(%pgdL`U2T`)LAFNE?^{)g;oK z>7@aq1`NJ>b{jN~7tU*cdMs>VY7TQS+6>rXGQE_JVi+w;ru1_ox1JJfsZa2$95}z_ z=R<`ogH|Hf_9w^srLsH-En!K>|8x+n zZHQY<)6cQxi^7JL<|uZmqc(!)X|*5>LzfXw8@0=8hkOK1nw;}QGV19WuH#XS@Y7uLxQzhElfC_>S z_$Uf9d7uoa^b44Q7XxzuX#WS0d6ydPoSmIP4oDQG$C-$=c_e0 z9o8iImCH_NKexdxTYd%t%G3B9_`8!*Kb@Q=d zwG|d@`e5ZBTLL0S?*Lf*j(6vhxN{H#1jI57*KrRhr0I#yk3ttaKT#RroMk>)34?+K zkcTrM73I#4NcYp0!^fbH3DtT5jHWvo;$I>9#IHp)dpH^Rg~r@sorRnE0zkDR+mj%k zhINJ*G25X~fVz%>WZ_MA;K_hcKSzZ{HThNmCSmsZ(~Uoxcl|JmGk{ESeMK zOR(8R^eS1Rb@5G9;Rldd{{hEEb*Nw5m%0@Cpl=phM_G?mH z&AqlCdr*T|fNcFBbv200yMl27%u=QJ}+|) zrW-eI2nl@!veD>+S`aAu@5WsFRFLNA-@hyeGAHm2W&z{O)N!^3Ycd*b#O+r%pAf~_ zUYQoVAwVWfb;7mSgN%WC1MT1IMb*B~$-l|Kx{Q~V3pAN7ahZEq|9F4#7hDZ8i^bej ziPL$#pQH$T@jmn`RcmEmZC#du!)$BfH6dT@JG?MAnBa!3J;~FnD?Lp>cmcfH>9E4a zaJH|*D-}#l$A`?mdqt?#3y>+!1Z4!vmnJU+_v04(n4wea`YBp{$(8FjsN4f-%~fFz zm)1pPxj zCHAJb1 z93U3mk>tlNm);}Qq-NQdN=u+!4{F-Jh}41{f&*c^UGgpZn2`ZPHqm=HGWyWC7{~gE z^hL}EHZQ5wmjDUJtLrB-9Ua<95G{cW$f90-*HHqGxYVkQ1YNbVvU+EO98lL0c$bObpHOJMP}3;|(k2aLf7j`?uV}qYNlSNGO@vtwe0{Ag zBu4rUjcurk^xX=y+>#zPC-y2qY%C& zb0yVL^#DdHjSU#0!AlfSOVDYP1|a?MNNjLs|AGJkdrl774V|H;16uG=62%wGR1k?SR%oEjAK`i(P zz_Vl9_guDXdVT<6d0i)% zK`G!BlVEPJtIZ$JY+RKrdbxAy? z!3i%Pps9)IjBL3bV;@b|+LM(&r&UKtK_t#Wo#s(ceG}1xH?^TwFoQ4j8Usm~aoSAZj147&dSZJ->vx(hXJ`9FneKaNC&a zmjW`_2^&ElF_w z`qjrl(7*bRw3FAaNqPg&0{|{0Z7qpMfRhM6lZ*<90ZOB2W4b}Yh4VEzK+8$*nHB?| zWs_C?S_fO*09E0Y;{BGN{7(0YNol6azv#YOX%)Ob^519djVnH z^Pm_62>2?$;l<)49ROedABbA**XW+fQR)7!G3K8c_M-{w`rc6oP?WIAHRh=OHLj>U zm#v1~5|TZoJOY)10bV|n#loaybFW6O@b@_r@5?-H-B^3FZT7{)F{J;!7ESng0007fVL3)RseAv^U^zD^36hf0yb74 z%YEJd=m+8)IE;qC0V?6)=#I23p&*c?z=Y|?3W=ffax8_gaU)MjuAPPa60Qa zE|>KGSO0_&=)?z05R8BlYij!|i0t+tR^z)rmIs=!cr>H7uox%`0gro&Rq_Rn+hx6E`zBjFYoCwX?ww^s8IkR3RF1p)n^My>_+jF0E zuAJpzq0aPgQM7M-oMBQZpD*%ku1?F-;e;q?|55Qy6`;!c8vAtfR41NLNui3|YlJB}Iv=sFV#8w9TuoF@otkHEhA z#!R?FT4p9-k6+yUdC(CS7S{P@zhN~R&a=!mUG4j$Ib94dBxO!i7o8!}lv@uQ>0r85 zBIlgF8;5aRaQQNrP@gg0r{PU8qH8s_}yb@M_reEasbODUfEh& zx*hmxG(N|pfFZ~fa}Bm3ee*X(69aOjPX#{g!+z;tZ2)s3WNZL+8h}4Al!GXa2FCM( zvf&AYq;OCUJ(^FD&P(r=a^41d1~U5c;Bu&l?}F1@LN{4~9p{H{=EMPK?VFry!Lf4) zkiVN0JLtl1k}_C50f0{hH0l)-IV!s*x>Z=A3b`?QTY%c8@!HIWe?{S};8x36Sge4d z_V!QkgIjl6L#rMM7CZJNzs%ZKw7(TqWsYk%dq(yip*rBZLMZ*(pv)H<34nP8=$8AB z-RQ-gOuBGY2LwZ63Y}IcP6DOKX`1@*gHxT8&Gh6#@3%ZwuB>4*y!FF?yN10FnM9K2 zS1Mt*t_%I`DKzR9AZ>Mr4YoEd*wEz1-J$mi2T<#mZptl^MxTNN?x{22k>HnNwgrH)28u)X&*Nc9f>DYfx`jc$_hFuh z6xKk>*eWR88HTs!44~yL z$73oKQJR%g7g0X(0mr3W|!DO=K1&stgTNBO$e`Hc5gy!7-1=~TeE&PV; zVF#2SQaHB%vL=`$RaH|%JgHEU1f>fYgWZD=(m|xZTrT`ro0R@Pe44d<3;&9 zJ~8n(0Fj}GJq^9#%ilhz8yFbey-QF^R%iCnh_mHz!B9g4q4h8Vb4T7{(d`AbM1cnD ze8?G5`-GjKsGoEo0`bFz1Zgin16sPsxN`&=uc`NCciVfQ-9IQ^U|$-y81DbyhZ<&% zg7FS_?%lh$B`uCjoq;p&3^u&?%1VruReBI;&q7+z(CiLua7q)kmYa?%KX6Goqg}EM z3Im{#jZ}ddnbkbt6oayJVTd9@m>evE=)<3-e1kw5LY%TkV;uk*a|U2?W$?3~o|AQK znLHwTm7@KhgdLdv-<2?-e!{T8Ftmp|J_YxH4kVngUCD|;})_! zpepqhVw!gr=*}>pxdM_04a;i<5In$O9Xd{&OMXeS!nMn+7{Ge+h{ZF{dSF0kDaUM=+$oo z?eX_k<_x%rI4$m^8<_XLq=UI7fe>a=LQpjf3|H|^UO+}7mI_WB`mDkOrpq{q$^(U0 zWq)CjpMN*yh0>=9eK|0_PoM?QHK*4Nnxm>8V@80~j|}?2&-)DHAT~gi;{Uyq7fMEm ztJ6$>>8EqJ6Yt%-%Aa`qF~~zkX|u0A`YClyz*xY+6GUtMU5UCvcK`f%wE+`kD7PVN=uh74f}G zoyJd|u{j&Yad}3%Tw^HiDFae(b7=NGYW9r=uJC>rE8+1RAbLbR>28uOlfw8D%?NFB z^NH6}KhAoB9yY~45Lf`;?3f5W%tHurCng|#3I>Q~0M=`C{2fzJNsE7DuYW4(5!WZx z;DSw}uV#a<&s03SBvTEAU^XAkz_oe*#aXur%d;3e)+RiS2q+hDuf!0l8I(Zc7|zP3 zs>T%=DLiv4>=7!VuYWEBhWq${KtS&qt+X}tNHsoOW9H)8g>bF%-Bs?C(>=mrYbM!0 z-BvF3@{X5MpUc`dy(YKDxmvJYrMJNwHnI33CsmnMf)}<)no0yMD!9O!qt&wr*4THL zwXjZfHtrka;NX~}`3DAr)&_CW)U-)~$qZtji=0L5wv%7Fm|7xYC9F-qUk;pgi(;5r z_+i&na{6Fj@$SR*d4(bsI+)wH-0@3G4KBd3M#DzqChZYO1O`7hx!5pe7Q3 zPYEJ_iAmSXYU*D8N_{tc`;l(`6+nNQ+SQB+rL&E0*eXwxTw{irnVkGHQ*5Jo&Qnv; zc+EGj-S5n{y4j4#osX|pq$`n!$YJ@RKyX0X&-mpyBCzY-?=+)z(c6i`vLe$QOkUM8 z7<U_sBV&+dMeuAF{LMbzIXcnlb<-Kw@E0j@DO+$Xcrk1u=T*&I$6RfA((0 zk6qOfV99-K6mF!eI)xvN`ImEt9AA4aur!d`8~oi~+{R6eF-mq{o574N;#lgnJ*&%e z05?|!1}dj?_0lY%*U&JKzuV0Gr1SVlWq>lWn(WB~8Czz*!~xBXNZd)0M!&cZy9KJg zj%VUc#GBkk(W}OZRr?(GJMBoFdC2$<53%RmEz<21FQN8r4yOs`n1vC zha`gBETEEF?7e5j&G(WI*GtXUw05o@2(?P%tfUWlT%noHh@x8e&sX@$>dxdG-A>n; zNL^C}#LmYjTwJqP6a+P_AEo35KH#g{x=j7LN~1WHe%Vo`EwZ0ZVncS&AKFlN5J?nN=7o|l4fw^mk`a6Uop?f1{4 zJMdae^n$j~j;R#(=LAD{56OOHB`vB>iN+F4SQ6hO=kRPy9mQRlWN1h&*YECpY;znj z<7Y_5hrNfDqAu@BkW111F3ywHGG+K>PqQ~V!-iGT^czS0Ghd}_#g+L2`*^#-g@U$)%SQA3q}Ao%-Mm~gzI=2O z1LrOr;RW_iTIxGoLpPU1adX3ek!a;f{#Ox1m!tX(;v#ir<(%dO<$T~A)AbGy7id>O z-O0vP+ojCGu?M+f;QJKM_^PtVEA-W8zC_*kzE1+dl;!cTK7kAQ3ED8PL35W(R>%_t zv3zf#*G8$h?1ZXv(rSWEsZ3=1Ua5sE@ z-^Cv;_G?#b@rezL=DIT4a3f`q!tqfA>r`Xd(1YIPan{@SIVT3B^a)ZOL$kkKYIB&* z_R#-i=ddxKR3E({Y8*Ii!@V}^CBHRm@kU&}bbh;7SH-LVCxhNdEr|x2vPsIeOE*|O z-wJDq5Lh<_U|cQu;k{GU1D%og%fJkQJ?JlnTBrYMJ@URqD597i^wyX~suO4qrXYW9 zANKak!=xGwKCb#{{dxPTiOq6C#U}LD6z(Wp$0m0u!Y|Ec?{<}#$$jIqDO748^D#Yt zqkd+qJu|v#j4?8@jhQ;(Gc%Rx?D*Z9ucU#QF@|*}mn3^D_}Yu4q{a2aSDy0xln1I! z+tjt2{9;oiUHN2Qj=@+#TVC@l4>!O5i1L?ROWuR*=#Iy9z6#kOb8sLQV3PADT%ewg zUqYN&8|VYm@N{r*ZxVE`)KIYfB)I>YT#W1>)Gpaa=#I{3=SPOXti!wAuI7|I%yIQI z2Crt}r!`F0>~x}QY9c?sbh$M?_P#EzQ@WgrCRP8(^KKO;oq=l7J%ubMw7J}*QXU!5 zC*c332tBn2#uR`?huXSV%O>TvQX5;}oRD@_rRqHEedwjJWXsS*{Cm&0>jWNNSLi|xp zWF~+0@uim)+|&(xYV^^-!KRxqbX-oJ$u z_c^PGu{dGxV>t@5QIbs~1II9uugw1TfBMos)<1`lUz*K&fUIP5rvB_EofXEPs!8I>$(xlTd)31ua2X^fh#AFz~wP`T4gIZ29sXA zx_@OCGc%}iFmHHCQqqJzt#EXgwx?Ko?kGhRG(_y+Ud&E{W|%L2$$$S=fgP$f>0^BY z+$yy)&~}MI!?9IpBqY7)IrDB|<{A$tL=OctN-^RO=?j;o z-8uaIp8oEmR1?WNwQ8zsRo;tksvR|m>gOy&Q3G;J?7Lx{-y2G?!o*y-3qhcPf$IB~ zFCh^T9kFlnluDy17GufbJcYY*6m1f^xmuFXE9b|)GM*18Bqn3a#>e%cm1H(Tt5jVJ z1rAKl09!6|(|NU|CyZsBE#8;*yp-QehlezM-OSjmd03mqJ!kpcF!PUFUX*<9B{dOD zz!n*NHTv^SS#`=?S&hgq!i%HkI7D7}+0WO@X@m5JEr$gi&7jc5@C<>N8STN$%$8f_R^2X;+vgH%wG$8XR7aRx~;9kJt+VkLDw+K7mk1@=Op#y;}36g~>(!-8@< zLhU)3<#2IKf7ER`!&6APDc)v#N0-*Ez7Z*d5cFddN-*) zUtjGkJtLyJ|4lyDttaPfB=z@VPs^pgh8UTSUcZ^tC4UPQr1Ky1%srOt7=mGw zoav#%>Z2O;BMLX*%lpNu7EGN|CCY*S!A+h@Aol-o~IAKNhPu?Pb2eh}L z&+6+_akV=(vsX}(`a}n|=4xk@k@-rohLkZrJ$w8TJtI2gy?sr&!uvV55WIW_)8y5j zjJCpHGN#$P$JaQ7jUBNm(bT^EH`P(dkIL`31}{#>-&SCN(HfXqWqOh2HFcq^L^6G2 zz_`!uGA~y8e)!X>H?LyUR`x6p3;9W>`nr{gBTEG~n&yQ*0&2LN(nEPz8{vqNqeCuXB62Y=ME6G6oI*0%AFXj{;QXF>ly*!Rr^gw9N|I1&<1zD(@zn$V+zQ8 z7{FHZecV!)Twk(I-uE(5HU4Efo$fwsab-e-^0eYZm(a`X89(f`(NYB(>VNS{Fsi&6 z|L$snccdJat)ZW3mK;X(EukOh7^I>G)w;PNS5 zqFKQ^YET!$$Gc}>wzAD}g^+1KMsbBY~u>W)2mNOhuzA0CiW~< zTZing>3{$Mqlq81fdYX~4oWYnJSOSTQxNhWZSA|JD3J4{d$UfalNh!aIJy`h9bA6U z1URr)X)h>6@MM-FrL@b4a@P3HZq%lY&71sWyT)aZS{b&(DquGwCb^a#c1N4^Y#j3; zx7F3SEWai#oHt$vv8DP^Mi*QZ>u$W`BfTfs!RPzo!F>6|{>ZvS8&}?aRj34jsQQAG zwpGZtIFZqW_crDc%K14M%}yV0V18NB$4UImHk|wMvnO={DLJzDLxmYw*f*9EFTBSo z$8uu`ek8PjjEWfgEB^=*vt}M5?wQBITVd2s4VNP<$g5&mbAO-N61kL0E6K~*tX2H8 zPE%o=Zb3{pOk#v)zE6sbXXwe$a{7!7QRLi4VFdSR0qW9TbG2=VnRpnJ$4&?5Oix6Pm|(Gfu>WGlV03c+|`CD-S| zli@MM(U6zrBegCoS{V0c3rD_>1dsLzBSMkQc4f>cmP;JX$6LyVm1Bn8mW7CvNy3)T zqh_H=EpkO^YO`P;_?JFMYWVCuRUQ4+c~~D{6y#R6*HFm-EIGXrga7jmy#C8O9{RR3 zb1_?KIlN1+_r!Lcj-rVFFe*XKWPrQ>tN26x&B{#F>#0b*&Ps&I|hWf@RG$g$AkK({{pntyig4`+weK-M_}a1#fU2 z;D>W^+AOqawJ9Bnug(hiH41s_Tn7upj+rRg>BYA@zgRD%eO&vS7YpOZ3d>b(Wr{lc z(-gxbG5o9WTDNOX{Ga#6NDV39c@-m1e~cp#2H@89o^6OITj; z6z07(ok4Mzj<{G)?b0y$Rjc(B;++#_rTt4c^1XuYtvXtmoYBpw)w8#fu`+%C$tRb+ zs_9yz(3^t)tCb9fT;6_h46Tdj|NoZ_E2SXq=3gMU?iI1E+5XJVPeNwSfrsS1R41QC zcl`&%)V%U<7~5z(`W397j@K9Co2UHn!zVQ(3Fnsa$BPY{A`1Xc^U7n6`_U_o_y2Qe zl92hjF+=-}8#GU-%kc@~oi^^Iz3VCQP|ZthTl_uGot3SdbK?@<)$}pei+9T8CC4m- zF?0tn8F6F!D{URD9D3Wbk?gj&C~PUY+t0ACrdD-1&Y*HWx-k(p1VfjYX2p5!;#~zi`AdoJIW%gh{`uA1%Ko6NQzaGe; zY$otJ%eGMQ&OR;sALBd!aRxJC?>awOmO)tP+v_1kxA-b_*Paf~I8NI95blZni20!D zr?^?w2QQvAmXvSe8lyS4fSIOS_ASP) zL>5U|>UF&csVucRROBs4aQyNL*>v#@P_+~;ab5c2?ro1>`?Q@l&x$M$EAgVZCO8yW zaWjqnV%xX%EJoGBEam!2E~;md!Kxh8O0F@6Q!RU3{EnxeU>0>u%-y?+HwL)#=hANx zN6_i6n9H1*(Iszrm0TS1Vys^&&(_GYTHJoNo%3=W%@9UOYh-^DL=}JgOmOA>v>O&M z5$K*GW9?h;R9;cOw!MU2+DvSuY@qNy?ex`jxd-|pIuZCH9Wwd%hl;c>NXBMK50VH7 zN=B@CO}RSn*>Zm7niAD@N;S=it2WA^9`4_7HRB`cpKhGgBR<}rH*g5{kq_EJ@pbZtlS$s-s>nP7%cZW{`W<#&0^wVZU*6%QI^tvIZy9`J?<0CU zZ#NKmcHcQ?(e)Fa`3Kg#0{eyRNw2}!d!-5POE+iO{zB+CJ~CxXzvtr*$Y8t&rWv`c ziOE=MWk>%L(fb6YW?8d6WoR8F)mdxL*>giID=A!JnLN8+_Lko&*v`jWb1KLx$o|QA zl#AEEg{@6ZI~KQSSa38hxY_E3t_ndgP+vFI!I3P58FW1w%}}*P(61uIwtBrtJmZJ3 zV%u(mJ@edyfrq#+os@mz;(mnRZ@OGHWrC}nH-AIEZ^oe6Sll^ZP9%Vc=594heVrde zEyL$qc>H2wSa_(%L{IlG$#|DM2$`{9<9j%ibP}yHtfOQuon!Gd%=0z9(A4iou7j5) z@33)`4_$zV7( zKC^A|Nd*lW;bH;U6+ez;)I6B@3x z{vo7T{M}8}lVz-k-O$+eogI6K)A4CT6=w1$4RQu@$`;Y9EbPMuq~l|GIw~XC{^*dq zOM*h+_!_7)F8s}FDlHQEkWh=Xe)=uNruQk@0&l!A>cb)q2l;H>cEks1fk&IF9zwYe z%i7|eUE(3PudFE*O>Xelta$mL>ZeJq*Izq+xdk#ye9Y`EO%?oiZenMTfSIR{k)eue za9Qe4&igJ-J*s9ofs4h}o)ohAFFq3nFe|3cKcC|^Nk@RoB;!wg)Z;R{tPg7q21AZ# zF9dONk$<1VU+L;2cloMwRjpNWdia>Uk9PldadnWm)0|%PgSF`B)0DHzKUN34SZ^k@ zIJL;t;g4s1({4xm_9iUXl;`RlQ~gU!H*aRq%)oR(x;0$bI5bK5w!U85BCZ!@bKP7% z&GSJ^DG$#6)|=FZmMv_&s^tFdMrwKZc*naJor^*_{vtxrv?Pli^#p&A9U@d5W$pLV zI;+8P|EV&U^H9v*vw^y}%|!Yho$4DI=4*Jid?b0@VX;Rq-mBn!w{v$GNY&u4aLxSx z*m}#bsP?vR7_b0Qkd#)G?rsC=?rv!Wi6JFaxzAJ>6v*81lezpw@FWluqK#)T<2wwqbMxT~#@0Tb9yqofS3Jn8X?pyO7R znY=^Pxri}Ztb{(@m#QRtJ(U*3# z*Lax}*En^^MP*$LVwC1>avEa2zm&by@RqlbK1u6H#iO`hOf2bfhdnIY-sjO9nH2Ok zr^JI@UB#$*YK66Sr(&VTAtW5BNfX#xh3v|-zdx`fdOSanL!9%z_yQ-`gP9C23y?rX z=z)$fc+UirVCcBzd0M^MF#I7BqoD@ zGK)!g{*?n=#XrIC`SbgVaw2g0W`>OrA}+-mL$5Nwcs$uZ`f%C#V(`0z+SYmE*hkMf z#SB8UuBGY8e`*##?|(9wa6H@Qmh!R{KYMYyU8Py8Ks1t#nRpg)sq_LF2fjV+o1kr( z?9fV|O1d)=JP@+MsN^kva}xy&yhJL-9wrkb2dv%=BeSHba8l%jn%uX&q^Bb2O8NGE ztjSd*no5o@&o|nVP57>};M=c$>D~LOLwrmX{BO+$&*qTf8ZKE5oA#zp7W;#s(^WB)Fe-O z#fI1ifoCV6lpgdTHI++Pu;5d@*PXlTuHZE--z}SY|E!qEDp`$t>)0P1nl0N!z1zqb za2jn|m=c9^lj|C5Sf))+!v%u?V^ucu&Lu<$j00y$D%Se*p>|qiST;YIHsLwch9!Z& z(dx_d=@_o|V3wD-F<^r*84q+~p>SP-5yMkXq}+_~e#tviZ0|bv%0))n@`yr z3mvnD|1dsHP5*PxS%fdIT+P}m@&WTBm^`BMBLdx$=!|;e)=HKYxX3x# zw?bKcN9Wp$>mFYAe|COsEVa!&*J?aF(5~Bf_?g7t2mIS{=vwFcc+XX zw9<%)B_r|bu0NnRlXJep+@kl-*D?a;g&7eR3~fU+)V3PBLya4BrjBYR_J@A9yX=& z&i?a2b*jgu9mH&IE#JV4zl~+GypC%*{yNS0i4Fy1HnE4;RWyWulWx2EPG;O>zcsI- z*Z$8(;Y!gyZ@Z*6Ir41(cg81PlTNrd(f9B$To0H(418s?WmnaZ+U(gc!SFvXb-O4i z+VCeIR^AVP#%HptFSN2)b93*8eHm5!>x`N0XU3z~%S?%`)mDD*Q0)+!!e{m2Uo37sF8S>t#IjwhN~y-=PPh5LT@sfTgX@0Cl)zvkgmL$8fnFw&ob{)T zjF&x<@8)7yS?V=e&v3MDL|@PL1ob~x)1E{q;geyjscro&UoMlx%{vcnLgO1;VKdam z?kKwOyWvamZgh@dEdv8su40Uu7M#ir>j5QqaqMP$jb3^ByPbDW@V{KS+;tT!L!JVX zygU*VMqU=hK2chUilXo4(%0cL0hQK^3y9VsqK;`TU-C_4AN!;|@?h)2|F$p=g;-^i zlpAU${Wwp0ADG}{1IGF5klPnj>pViKZa%h`tj;-^S(fK%d<13nAa5L#NiN?2o@Yjd zOp2Z$`_9GpXfev{*BJH$K?xiyT+(Eo9y6YzjWLniO=tMqEMsSh8KU=WBdYpYXto zSE+OX*kJ+4JpA72W{o^nf+LbkHZ(-63*(`kR;hG>m4_KFTdlXuX^ZUQyk2Ttc{6hP ztA(z_*5t=P22ei(Rq!|B%mA3Yng<EgdjZC;vYH7&hx)9ssui=@7J z^r|UQ)phMUmb%eUBejpi|7>4n6eEXWUHE3smv=6=3U+U$%cvVSspA{25G+7Zzu|euN{@p>+cls4; zJeS@w+Jsv`H>40_#CQ49F@oir-m&pUfv5-3I*CnK$Zmz1AKRB&SBi#x2j8u?`W~ru zrE`H#Fc3+CJ}9}l?EnJW9Q)kq_uIN*JS{fLk-|gfr_G zdIS~!_3|-w&=>m1>mS_~m-XVLo8AU{yksZ;8i~+o(O$nx4|#&K7Ko)bu^BNb z#FtE$&NSuUtQJsWGE=jj1kP)D(p?=ZC9Pjgs`d%FM#StmAph#uGu{7u=`(x7>kDir zRwkxe_}-a!V%EiImfiDRy~{7;6B7k8-et|Az}16eduhE!#mF;rU#~lVZ;ql5c`uIKCo&&!kbRs($RLR- zsC?;h{ZUSP-huKNsqFovEH{oURYKj^IKsyWvQy2i;#B^JItI@?-~Vh#MGTju$wq7T ztdS`PJe760&xTf)b;`jV;vqOtSG^;vRGz}_>->!X^>R!+-zUaDIqc+SLPCvJE4hb_ z_Jy0pkLH)G8j}f`Jf0ky3Iyxj6@w9tZ07No9#tlr{Z&NBSM7 z1(kn*Mb235gq0O8s$Yc0JNFM)iTwj?N>VczWFTcSDtdbjp{gHq6-DxiK9c2Od;xFu zH4mo!2RBH#<0wSy`;vbqv3?`rRgEuTkImLhKxU`@=wq8vuE>_|;@ABuvGUke^EKO+ zeoVe(o>JgSVX&V$!0atDB0vF$1lTTA$(K6s zv6`8kviUks@n&hgWo2zKaAM_>Bq8T#6Wiw!epG4CcrxeP85C`!YJKOUanxJx;H$$y z6}o)s*vdFRnpaFWIP9zv8c$MBi=(?eYO0AZ-B}+zuK4@fJg&97E;DzPmY+ssRUd75`-^HlY1jUQ-d>c`5eAv;#lyqP4L9uRsPCI!OX|+=SyNg%JJNHa+g+sQ(-^$p^a)Qh_AN0NAdoSw8Z1GDyK&L)v*14XI`L_s@(R4x!hh@C%3`i z&)aUT>!$%wh#9n&UR^Y%LX4e|eCfiPez=HyhDn8055Z2C0NMFKBlE**s?{s5?B z&<}1Qqalv40_si`0Z3XGE1}T>eYceWr`UEkkhR15G5?2-BOOI1zPww8q~vma*2D`; z3A^uH{NDymJ!&Z8oyfuP$O=f;ty)o%*CqI##xq_qcMEfTLlXITI03P2kFU`Ib;yT2yKOJI?|$k%bKad4k-pV{gST9%vq ztwXfbmVHag>T|{Pa>ChLZ?wd$RX=2VEk)Mva0pH}C&o6(b8+*8NV`MJci6iUV&R<& zu^#P~gCp9$0av#s@8AN$k{F03BIa1M6*$TnLcf6P^1K^&B?RhSt$?^!w{J9U0-{}z z^+RTT+MA7**`1%{-fmnrBzR0Yi7R`^^q6L-KC!u6ebUk|T?1K|k6U+|AM}TW52%dP zQTF~Hh95_zcjud|352p1B|aG$S^73+t+c1KG$sKt$&#kf%Dr` z@};yuI@c=jXE%V;C3#s2QaRu_5DpM^CB1z468cjQO)OHOlMk&us^pMYl4{$MOs{KF zHDa&zRQ}>M%9LxH+!a$EnyyJ7du2{g;S+U%Gruz!RJ;C#g-#<_0Y42+N1~F%7-Hv- z!L(ff+_qcZeSZuHKMD^0(f3Y9Mf3z1Y@hI=_XYB>^rh<f}uYcdlV0zggeIe5&@?oWg1zh`w723e#ZcWL-gbs|L+iC8|4%?z`dTHo<7@b z3ZtVg0RHFQBk2EkD9Q`$0ILgz6d4%kyl5m2NpuEqc03xs?udMNUMZ+w^&7wV?j!)z z|H10>+90n3#QSyT;Ly!q7^^V@@YHm?HP#8&PYygBh^}PL{Bpq5#8|C4;tu~{^{a(% z)A-;`^HbsfAdSmlqG9Y2*|(lE5JxbnWw(H`v_4*hGG^0)2&w43>V0nES zz9u~XU^8**Zy)*hOv7O>S`J<#sr$6)a{tR@cnu7pSM>Kk0PTFwI8#AU(HC8ia_`UD zofmSOy*Dv}cfe)>msr49TQriW3OLW}vn3IC)Ehn$ZMKpZ<;!>~p>J9B+Tj+_|9MB< z%dQwLXq0;s;5yOwy+N(q8u(opMS}PHp)WxN%n4WvfZYY4d>H#?KTuYMfk*<3l8(!9 z!)js}(m}DCgXqM?+y8;mFRDz7!w3Q^heGlyW+Q{Yg7*c5MKvvWN2&62a+U*p7okE29!x;G zUI3%VW{~`Iqh&Z^-Y43HeTZRH0jGrOet;hsHNZ?=(i=^J9pVox6hOU4?a|}=-qCMB+1kk~j=;dKFBb?}S-0z5%F9`gr~ zI+}Z#HUd}~)}pxcya${wQaWq1) zz-?_bcT>x=$$i0NS`ZlhtgNhpf*z+w+f7YPrC?SA{ZG+PzXx}J1KGySnmjkh*@4)q4MTWcM1L5-F*hQQy!t z(BJ--8RCOl_2_E_SwdKXN-wKKE)p3nG(24b?xxbbGYIm*!u8nX&|n44g}X>)iAIp} zkY?!{1L($3mbfrfZ#Udw%FRGZ7pUt)HoJhA2eFECSPeHw{=(|Sh#rXGsJ$+*%Brd1 zbMpoi0>);mkM0$nBoZ)xd1$H_8z7B<%N%SP)TdRddw$-Eiv7R~nSsV6@P*prjvqkA z5C0_{9SFod(D;KdXOP2{d#>F-LO-|Xe^e68D0Q_d>04+hJ@oMyfHlRvC#z?UY}v1t zHp>+jJ5Zi(RdNPPTp-e1jw+?t;b9I%I!vJx*R57)mR}7>hYj9LIIMh_@4x3&=<~hR(rQeAVq3*mirq{zt=I)CQ!EPoCve zlL>y$AW05q$yjkRvz{Gpx#Got8ZL?ICcaaEgSz(Sw2z0o`BHVTeFQIN;A5hqPQc;C z+(5*2xibVrJJ1MZ61(0o(3F4!M=#=H4kYzY*d#`bQh3bjt;Svp9&k4`!|DVBMSz#X z;?K&;k_D_=6ScVK6It0F;8=xSqn-i> zdfyeOkFPJaor4M-NDuD7QoRSAbSv5#CWD+OGVd{(qQD|Xf`kYxu_pjUjpF_w0!%T$ ze$UsIi{3iP71t})CXy<75y(Bu5%yW4jCseYhhuKj;=LOEtI?gdV!U5Q6L(j?&cAAr ziPQEZY@khd7H>-YmE!_jUWdB{6w#EZ7=<-P#&*ZMSa9`DgW{^M0Fgcn;z39hrpKtcK1k?h!Do_T+*MRmk6J(>pgn->({T3W~V75 zbk7dk&d1pHwtJUn!UV~IENyd_7_YSiG;mVjJQllbe1GTD9hkqDmn?vnPlU&g^4N+$ z0Tx=9Wq@vm6Gj*oA0q)E^-zW_Y5ebxZ{I*K-S1!3)3=lkNSNuMY5B<}2rqx_Gmn|3IpN9v>qIZzrgdkn;qYSdH?u7wId*Et3K7c}iUFk8{YV1RB2%PQz>^|O7 z3M^o;va`F_MJ<61SR4z;;y=7gqnt9$cnCzOq2b{(KqFhYD(`?q@o1^WL$5ikG(Ky5 z|0tmDV_xfD$w|y(9QX$sT(fGYIMnQR9&C>f0 z#PzsJee&b|ZN;PCwoZpze@NEyduYF3eV>2e-Je#x5Z?8UeT@Ime2EF|ntY9o<;mvl$K7q2ZW#jaYQF^}4d&NISR|0CQ|FTn0`_ph-)!6T{O< zi}_IdTk&&n-9Y=)?JVF)zh z6B4>*W`J=KA5Ux;_vN3LO05V?na3y*U;sRm4`Lu3&7{!Q)FkHYjfc03@I`|TZ2=Z8 z{{I3hJah9f(iy~=0z`?JH%0s)}kh?2L zAnDqR_T(29M&h$^aTStcz4+hv!X(5i$7zian5x<`sQfUw&9lHlERNDa^hus<)P`NB zrD`6t@aTaL8@XBz*I*sZWBjXEWd!%v#)34lE?zFA_A*EM!vA1oXndus4C@_iYSVPr zU^Yl2LePJ_`Y?`^4Nh~ir4MC(Uu_1rmzH@1q^2$Kx_3P+n7m~LT0p`Yp&06=cGBouIl}D^ z82h3EEcFZjewb`3S6K23Wep7tU}(zpovxdmi-Cen`REn4-nbizO-W- zUhA}ioIdaF$sYPqG21eTsgP~WRhWn?V!yfkKQlO;1AZ3|_kMo;ZS;2YFJU3I44Xaj z%)|{ux;(i))KN)a%Gw8Ophzqlt9{BOuPR&}HMRE*98HR|m6^4flA=M7 zmqkYhBex7TXkVWW-M3D}2@O7f_@*u20cmKAb!qBUXG>tTFNM2B z-3-_*^hc*-y0rU!Vxc=F@(|a zD8HQa749CQOX3a_EQ&g}jVrsGN74nOG}~W$GmS(=L#iH)!XG?N5Gj$|?vl;B!I z{fi~{2-vTnU?yW&6b>3cO5Cm5nOiz~CIxMk!!R8rwiJ|b;^fi3coXc1jO1#94}$e? z_A+2aU;yw^cO$*#8(kAiqymnXukfOk;V>F~`00}btm_rA=;+f^zu6n%*gOltVdwr7 zG8EWoG`cQD`YX^KS9;H3?Txyb1Ao^LOjle#7>ff578X$#sM9d1^5C59R|bxmK9C7^ z{%7i1_ora`l68aDI_`s=^E>4N7g_Y{u zS?jT`tAV(|0|(!%k`yn9VY;4z*^WLqHCRvno(~oQ+Iiw6Ewo3G{S9>;E6qd@n?@oE;;~Ody5dqOqT3}NCb_zXS8n6&G!4Vd39fp zC_3Z1dcN% z7+s$~g~gNo;{qJbwO}M@m}>^KX6|8O4j$5?`=X+v0Kdw;WcD(36Y51pM>QSy&7K`> z3gpW$Q6Ahx6y~&o{_7&N(AN~k}Jh#tZMAS(kq@@ zi=2ww9GK(1QjG@6)sN)11|G*s${IBqgTY(&(km!*{DeIAh3;B9hKHHyK77NLR&4uE zhhDAs2LaH%&d@Y1O6O{zIr}lt{QD`a1qf2c9f%%-g^hGIfHI>sw{P9QZ$=|)Oc#BD zY8k`}$WtD!NdP%?&ytAW94VFyY~?(xNBN+w5HaJbE)iZB=}i~1bRZFnU2l()O&Q6# zl;^dzxYG5=7F>z;EWfyM8aKY~m)SI4zfY9fiGW0JDC8(!iR*^KFqxjQMYaAYV*^_n zkHt%t(?&wtnLL@RuGiL{&B_o3!{nM6$K>t;$Z9O=;ox)h&tS|wgF_GpSL!%BNf`cl z_&l77d)&uSJCO3OCH0tZuXH1Ja>wc&^?Vs$hgp-=VtIoUi$J;iS0^OoCk2PN>F~uE zFRv_BW+(iTNXWemwL5^^FH7zsRIyTl2C5Y3wV=-WkDuH}F>$3f5t-}3XIo%AFYWO?{y5OVADE3J zLm{S?qD}}s>_ccf+6jhV0Z#o5k81xDPsQ$7NL(UY4$YzRW%A5_<>7C6s})n>LmpJ1M-_{SG8AvK=$0m#nm@s~g0RJ5 zVbEak>N?SQISC2w5J12-WrstCI4Mu8!gG?}XUED_0)&6^NqKI98_SL92QHz6^s3xg zyq610*+JfbJ(3p}GvrQLx1@gyu-n4s-SZ6sPir;x8y7FfKExfg;ZKrJy_hbFqb&)k z3rSL5TtzkXYJpU_(i`OS{@MU=9J&N4GojuXZ;E9}psc6oUHF`^6RuGeBW9e__ZpBi zitX_%2Rhqv%XZq-jvZh59w$eaTYfFt*0~&C6g(u40-x=zZG#GCB>Z!6C|5qz?gt7U zE7?NfyPw-+Mrn1yy+K0D@Q9kPD=ViHW0y5f8#)0ct}23ungotLWg=3RYZt z|GgVd4%!$Lv-J(Cm9e^8?)fjg#~}IAV67{|hLndnbO$7D#4+>!HufGH*T4gb-9i33 z)^JH4lh*B=^HaIRd^S$Qx;MXwUtlKUmAED(#D%~vNbt112|4$pACNZsyM&s&g|S1Zj05|+*yQ>b?~Vh(kTelTb_LWW$SEG&ivEmG!VRI1BbW%7?q8MfNA?L=Mb; zfimZfgdZMjqc1eu&5dCV@0eXBJDTR7#nr~+F`;#v)X|IR-1`CPX?~1AtZip*e|y$< zhz1+xe${DxB(i(1uR8xOK9U=wmPRIhL_jB@e|n*?l(y@u8K#2D=@<%C=cv-|R6>I1 znVB@M*hjl7QQYIQZxrF^c@0v3nxus=#q>pPIb)$woQo#{!PIp*TD3EACI zXKY7!e1g0~#)8dze^Sr>>X6nl%exnA@PAoY-2Tg<5{z!f!kLw>m^`_>x19vY(cIEE1noC@oyWkTI}cEbCzG$xumR@Q*i^t#Te zxEQr3%L&mHH$qxtH>za_*Yae+{6LL0h8MA>?iviWtwK^Bf(+rO!aP#(4uEMhZ9 z{;`5mcAkXfV3v5wlzj%#+#(+zAI3@2_nJD!@=jad8DtNo*Q2|Hm;6XpI6mhFQDO^p zLKRyk1|onYO{&N6NwzuuLiaoG`TxO;dABsi*1eGI4L+n_$)F>pk)QcLL50-E-zU1i zU#Nm4-g}I#*HXlz;H9(TFIHefsK8F}Fl>~;y^V~L-Cn+k4VGkClIPyfaqi1!6lGcVJgSc!By64__n$l!LTX;?r4dVLU1hS z-Z@x>*kY)!4a0$qSxafLL8hk^is(1Vejk8iv>x8|w8z5Dje(^q0lwlaUzR5eOJ^T7%g`%%Mxu2oTj z^ggGWX11@WzhF|JW&CQZID7(g(tT5UumYnxGBR=>3g+cecLJYy{7U_0n6b+CVQyc; zaXO6Cm_5DeUgP5a*<{6Gf^mo~`-tA-v1UOs^j0Oio;|O399Q$O4x(236Z`qWINE1w2+!}xbb?0hm)@?@)5i9;&xP; zfo9zGC#8eE{K=-PhUbkF&R%~mCS?*F8U_?$)Nw%jlk6O?qYue6 z+<~37#>oz(A(I1-Ts>!_x<3{;zHz*2ZCbj_WvXVstFMl#|K_IWdSyB}qGGAdh4#Mk zBQ_!r>y5lNbb~3{gGy|YwMGw z^@%0z#{nQJ4E-_>=rF*>K7rGekiKFj`VmBEg2~)wTn^tbO1X);+k3++!s3dQgce8@ z6BXMg8%nnHg}E{nU1}N4!?E^Btitt2dFjrdna3+<7QL)vy8StmD4n{SDVhTO{*K(I z(azrPZd$7UOEx{uz7WPdr?b-g1%#|stof?HX3c5_-t1*M|Iotw_aPSiFy=tgxKr7( zgI^#p2s(M~g!Q0Fdzt2v%oko0Lm>jlZ4UaA*zG%b+7U^2v(a>aKkpmptzOfAw%yj7 zNXaQXHdJxJOrxVG1jU?xos7QVu5?(i%i~EFYgy*%qjO^;W`)>F9WvNC%akVHpaT6S zF%-m2lSF1Q*MR2bk5z>yQ27~ci@vC=r@)Z5$@*@B-s_>ts=r=>>nW$TY_rAsmRrx; zgV@u%9(wvNuXenN@NS$sj%Tw`Wx@3)XfqW5%#QbI%|EV2_QXQ>VE?-l5 zOGw~|?1AyUYoVM{I;$T2lS0y`c;j{hL-_?52%&Nz1 z;&zSJ%&Jkf$5Y!A9t2@Eg>;M`93Y9~62Imeeq4bB`6~&tUr!D%&VSFtDm*!{t&s6Z z64sOE$g4uGzZ2Eos9uR)KXKieu?X0DB|oq4AG^M_R22Fc3zl*;t+?*QC%6z~S%)wJ zt&xwj`I)BzR|T#pPpiZ1*exKuxTWq7m%NJMeP;JH5w*YIVsg(f)Wv_I-$O+ABRAZ* zJ>fU*NXL{xOHniZ2k4{yleYOmFsacG6MPM>+lWGy!)e0b>z;m=A6ZIXvbg;TUY|S9 zliH`Df1hacJ_;MAavNCZQBeE189Uw~m)_{XW9TPr&M^B&U(+Khf%q6gak8#U@O z(qa4+PTX$=_0G5LeL0Z4n74q^oOBE(o=8-@xZ!zniNd&@@C0nX^AmZWqp9FhCXPwp ztd`^|Rq!E^(YEFgk^6a@mNUR4bc?em+@YdA$e1}+y;TkmMjC{vV~gTipB-wlJvSW>J4IR3upN(gK;@ArKTS~GPP3G8dFXK|{T)tZZn}jj%`5yBwhyk;Se!(5TDQyo^=o$YPFE6tG3wsy0W8vutIBdv}kQ z8ut=vO-lFB8|7!Qct(wSaftNpevVSB#KMlyuDb==}{jGe^>4b^AbUoyBcf>0qM#YGZ3*; zL_?V~_c@wvEE#>~-d?|kv&uEU{cSgR@&}%vT#KH-0JqDio;5cO?-w%#u?fGHgx!gcR52g~=$C&b4OI}IoZDPfXf9HNEb(2zW zY^Z$PGJ0LP2DSc~?x+wGx zW+y6&4TLu}|A4p=X0GcjkxG^7i??p1M>#-*py_`}&bOwjVrk%pYz%EJx-*;pWKoT`^ob|sKz(YyymG4Z2kJ_9bQRx`C1Z2L52)wbny`5Eo z>P63nG;4*|BtNNWHO~*VyEZ z8$2c|+t4G5{JAz!Oj1CM2_OR)QjFdDU>amuC2>%Kq&y~!={zRCyiB}<44zd9peYG@ zsA{4Q2KiLS<5svV<{UJ*lkdbxv&MNB8hPt>)80Hg)0tJ??raS0HuqfjG~CAxiWY$=l2ZVc30UAx$x2)&aKSmUz$q96AaG?G5 z_|iu|emn~Lde5w*po-VOU!WZ#A!LpCiSeR?f%zCli0KBO8e=j@{-h(mLf5^e!u2Yn z-u$#KkG}LD6>jQ_SHhfH=ELLSoj$)ozEFiN8PSFgI4|2NEUNybWK%|2%f92T|Kf>i z+|18CX`ux*RPl6_0ew7QGk=5rT5GxYR@&>2?u z&LRpb)o!P6Dkqz8kEt}IlaifL5#5BTI)dmEeiw88MGOK^q5Fd8ueshO0?S~HKWMt@ zw1R+X&5?`Z1ud+7N($HRcKAZj2orT;akcFvFIj3$gO6R~+iF!EUpCs!+V#>|%+@Eh zSz3>W{DqEA?DTRhh2mjps{wp?ON9g68RgvLGQ{^_g;v`PQEAX(#2N~SGa;FZVTGgw|=^H}Rna>5Wzwo@HV;8p>0_s17QJqy9r1FDEOg+_la z4mdD0ED2i~w5Ang)D^UuDP}M#mFi0-FJ^z1o_4zN;f|0U~h0R`i{X@*#!75oR_)Y+Bkw zIh7~`?{{50$YQe$hqH4R?1i{0r##Nd;?gTr+b!R9ba!}%Qy*$di-Xc3eu|R3b6!*2 z$m+n}I83~6MVVDO-%fIHyN2x;ndo1YsmZ^P=gLujTd6ss02#tmTy7R-eyt~WtLWhL zct1#5U?J~S{1yD-PbrvwtCw`$ zvY1of`Dwh#>QkH*`Ib4v4<4nb3iXUF+++=HzVs16)Kx9F*VhgwS@aY$djBxh6NOBg zI9_QQ&MaKNt89);w?vN@4J)MX5n8J9 ztyFv*NmM_1;xK!s7Ot}gK5`RQiF=G8Np_zG)h^ok?K_w42ai$-qExmX&UOVg$9YF> z&>Y&N$F#B+{W(3hf92VRyvSp6a*H*MSwOFgCi$zUu3CKHGOx!rC0TBML)|q;qX6FU zdxS$7Ri}8X8LK9&j6Vx!C>*pDp6bmMDUMLS@ zCd^U*V zb|6BNWP3AY{LAMX*r-WTpkWYAZbNZJU6DcpN?0o`HY3?8tgYq`Q2SJf!3TfIVImoM z7I5~vQYoKUyre)8I3)Hg(LsgCZlj^zItlNVvM-LZ-$lb&+gRdPQ?sb#cag z8GBUap~>|Iz+w49pRo` z1rrz4J{c`%=;kaUZR@R#1gS6EcRekMI9c8}gTm3a(iuc|K5^#D;_+I${Fd}3aMWHbGX=&{((AWQv_Yzp$?IcgjOZq5;C;jM=-L-pB}h zWS&W~mc$7A7M;2>n@?q2e8CM*pH<(Oya(lNr={|NHaT~gZTTF9b6(Bni9akVT&Mfp z@g03eJ%h+NER@MOw6m7znz=#4c|VE?K8pfW!ZkE#p+NOE@zw9ltZd)g)N=@T3*?AyC{<_*i}27V5^&DBo4AKaviVk@T|pk+A! zaC%u~yZ_LD!)7GC4}Z2FRfQKa?tM28J)H2}pU5iG&?U@0Q{u|Slm#Y^b39!M&GdRM zt<5&}A+qr(vWP9t%?O{??KJe`d+bD&4y%kqyZKYPuZp9l3;2>EME%Cat21>44ll-z zW4k2>YV8ab_YiWYTsHi!r>)Cwv)e2ctl;?P-{Jl}`iL;%=MmI%Sqtt!nVWUkx&1kI zqM~249a8o1^hTM_aEni7ESo@eFV4m2)){+{lzw3H=!%i)fK~4Rfqj=lR z3yo~htQG3?eE2ad&6r(nB0UrL<}bp7>J`n`r_kxG!fcqA%X{m=G^!W9Qp%(Wk+$x9il zw)`go_J86xXjU&%VfLQgm=u+IAv7}U%H$ySVdiRr>be$N+a50uo^|3CjSlf*#wk-ZCHP6vG*6c$@u zp#Hy&Yy<-ndxLjYwx}PA4qwc3u*DKf$hhbe2|pk#t+;prDK z!~7%D{d#2-5EDmh0(=0|(jC%U`QZQePcNyy)0*gpQrjcezdKO6*RC|p%6S4!9hRij ze+vnC$`m&|+YvWzKw*ah&KUWLZYTmPv6{F;!!K&L9%{^r5B?u>5snM{|0+h5T!R7< z`ohGDV@yA9*b6Y6|NnmrXxrOA8Q9sQG?Cw-F0W_odg=bsw{PDt-8!MM|0)3g&J_a# zIdja9oBH$VEbp`089#ZIkKBTqSLGjmTD0`y&6;?u8EWEj-oX_&61nM#Z-- zV12Kq%G0)+Ke2BOGXhW(*PW1&I!~u32VQQP$ZFTGsvG_&k3_>>+X5z~?)b#`qz2?* zbva9*ysxI3l7#Y07w6jVt-2d;7@uYqMD6bAm6jJ1(>U(zk|&jXX!X}{KKsd%LJ1X{ z>ndK?+3k8c0))?MjFT548T**$8u11w2NiAkrnQr)wJF~ckNGo(m|R+{ly*5Skya4CEdN3JbkHqZdYbBo7bur$2qi`1DmGxkSA6c4b4&N(*-Jxc}z~ z5%-q*J=}sj-5Q3e35ye5hba@R;(;+aQ=2tb9z&e>9w;q}Or@geS&R0t6EC22@`O=Y z9|u+RJG8AAr-G_|2QvpUcMD{GDCU5SsWc+n;q&z*2vQWi8SwyB+KJ61sIJY<#TY6a-D)=dV#jo4ILC`?cN@* z;Ym%VfR^R~>G77CvdXl9ah1Hsq3T>$UG(RYbVvS%0hXDEN_3fK87Sq#!Ow2urL|>t z2sR~+>Cw?Dqbql}B}x0_e$1%9X5E@+tz@@NVXQUrjGOghM0+E@CrK;7D<~ptkjX+E z^_6_+PZp=obcb2mPFWOEEgg{6-i``cj{-QSEw!k1zo%4cYbT-IwA9LbBPWc7wiJtr z(oZJq6_sr%$KT&LFl9lrw;d_!t~B)Dm)>>QP;TH5k9O=CFCmuW3tKhqo@0uX3^0->Hpin;q(1EJvT_(pr(}F6F-*4mHNNA;milSdbhH>Y}JsH`Xz>{HIr|1sBpdnoW!L zxjtCxe3z$DlW&tL3K$CKfz{p4|8oiunyl6Y&=+D<^@E0Eq9RS6J+jlTbP+*FYMqf9 ziSD0Guh%aidGeXTK1S!W<~j%6(79Mmrp~<&27FDTj|(@f*-na|(_Wf)FbKn;SGs4u z>SbEEvc%}qz(aQYU7kB)4GVMmkHtcHqG|u8g%Sl!F_(S4M1TV%5BvFG$9a-nHDGV&E2Hr{s{_7*TMd{vGFYa?3u5 zP#8LCcp{7=4Ba>{_)s#8*cXb#z- zEqceM?DoLV<&)GmTjAG)j+Zj`xCS>Ddqm@`KPfz2Gx^SIBsB0uJ0K*H;Em~*)jG?J zE(7o1VzgTW%C6ro2`ea^(GJ${5Wq|&->gSvwGjG>m51z$^=p$$sUvnjR*~2JEq+H z!8qTjq9KswsfYIXqR&y@_xqV_OmuP@oZ3p8Nx5u89YnIF<1h+EdiPtXRn1h>*a8n*Yd31o&+>=0 zjvB%X_8j7AjP=%xHMn9-$$E9M+{1Atg5pMJ#TNJ2Deva!@zwU#7{w3mh-vH_6jr@_ z&(SJ5u)cYAurhbmfF|OQbxyAOlHjq#&7 z{sgR;d!d`cjMSsrvKwEKG8FtEBHYy)uM=!LqC7|k2%nMtilb&DOwtlvtY zz2~knWOBPrTY|nPh4~M{mXkeETKje)L$MprO10=&*jSr_j`K?$UbnIH)U?5M)*HkV zQe```d)@7A>*eYW2{{h#Q=iHilbk8SU@JwSKXi0W7TP7U5q?&Z$gjUTgY$F2@@IrO zl?lkuly}<{%9feOi;D)WMmEUnK67z^^NYN7xq+<)*Aki0E7OfYi&oL~w=hq@1v^mO ztNcMyUphgE8ChsFs%$Vjtl@0CkHjr^(*N^jM~3wdq3rjCbNvtZ`9J)cwMVQ>>GJ0zJ(Y>Rxlm_XfR;OJg$3x_q1TB zE?p%L|97bPOwiaI$(!PPFHA)2nm)yI_@?E!bQt-GHI<}Qyp=zrE*SSoIVc`icz#jf z4Z1|ckq4)Z?F)6&lbVq~5*@PAGZkf+`sr}t`b>Tctt>f-JYQtco@~}gRKA?BPe;SoKe^2kvSKThTg5E5mDjIQwh>A%k=YB!LTTA4M!Uq8}xxMQ;sMVex zjqF0By|;GAywd2vFe2-%O9uL6P6hd~rsV9}I@M`2Nrx#3&kIXET}#rrOz`>Gy2K znu40o+x;Zg<{wde<*j&cSx(a%t^HHE@Vv=u8C(p8o>Z?|qH5Z%HK8r?HzsELoTG&NM>mu%H=r7d&p_{&OzKFOD_vLl? zxW=L=K+$Du%C3N^2=TidVB4-(A}8)lou~0QOjxCz1%1uts3rSu?Ui@WfGmycmt41s zj4;;ua3d7ve;VYDlm1Hn^7~Z>;XkFDiuGv64$=M$%cF;4#D6$_E)OK=JYYym@~*$w zJBpCe{!@M-t>ROLFYB&ES6J^kUN87E{%ju*aoseYNQn(sJ*6YmvwJ{J;dw%MV zr`ucJm{}I4X~XvZk>%BeX4btD+L+3S#RzS`D6M{(ElPF1>x6XX)H*PaK%$Xe zYv)mIQfe|WMVBN8^GmrpUWBH*+0k^Hd%;REN8i{})07Dvt&maiIPElL zqJgC9qK7wbLu{GAy^1|PYrSB;($k~7^O>uw?_z}(*zQ1qT%!c@MJ|V^{Zobw+W(8K z>kh=SZTog2Nk}0(*;{s0_RQWpSs58+k5qP%J+g0`viFD(GIA#o8g??XiSM}7Gv4og zpFf^Ip2v0F=XK8GJdWQm62_8ULAbKo&R-62eJ^mi>PxNh=#;pIcejldsM$a zOB=!{mi&`U`Y<+yU9qJD?KeKLViRn${}{h>S} zs7~|ry5k4Vg(V%6swmMbZwzI#(;WpzKD@d7^bvgJ^2E#ustaO1J3OdUoq%MKImvVr z88^pKhBXw+h?8cf*K*nIU)CIi!)^uXrHb76Lpn1Vb!RQi_$0G0VQ*cH7>w%b+7vP{ zZ!P3}Ex?o1V=fjxlcJP3B?%aM?X&)9V|4pU?s-3tPLowC*3c8Wty0FUY+D443nRUbxOPT+9d*ya~GF~o=-}Csjy*~AJ695!fK%@tn*{b2xsy! zs~`IccgUK2!aVefIj!X?K!U^0%4fhK+@W)9~mvFJd36!iD-S zXsVAMX2>l^o_)c*BPJC!fOKm0nQGBUMx zWt-+fB9{ce(4#mH;_B8L5f+^%H#m~11DiD;(GKUNV{Mh>{_d$v2g6T?sH%) zlB{U>BzRFXvF=eUN1{eEifs97{WF8I#z_u-eHZ+%L=YSU*T;#ZyJo z+s*9BM#b89NGbpNk(8Xnlgek8k&>KikJX~-?iFL2HY@GmNZ0B5xgdERonV>et=8`O zS79zmRg%4TlyXzX;*(AUi_|PBC-(-1G^u#J?N&NCo|(|ff*==6|zU#J}(Pt=`pZKFg~ z9Y&_l?fE?=$=#JGR?T-t27Ac|a!N@xh9~5=i{59(pY>GcolDCaC(|60e!=P_G#%P9 zr|`Vo_43@U+?+WsuBeugl%e!D-D%2(qPj{q;@2dsBD?#(+g4wUuE-kw=ugML+`CR~ zYg;iaI9$di6i2)0(@^;n`x5R!j)qIP?R)+Uc2V*99z*4fsCAmR>oloW?9JGpJzFnr)u|7~@{Z|K22qn~vTQcMuV*C_ZuQlyy)D4v}ZzU&TTh` zn=4q{J)Z*VvU~fwCo2fNgWsPJ+L5Yrc}k~Q!_z{ALtrSm5JLkc|0QR%%8=6;U>(tO zQk^XtX5)Gzz0i7w9F2N~;}g9%yD?t?9?{X z2MI~ySQe;$=nK#cJ!RTV6qHi?{Bmjz*UeKJ(pzFomz9Km0!Jnyo+uQ(&RN#6+oA)@s#kS2qu5p*6!AO)i^ap%c@^XD=C3 zuk$#p!bAG8!M^=NVcnV48>mu>=8x8S`}fG6KxP?fPm<

P}dLcr9E~!x2#TcNl9fYFpW3@(%oTM>6OHcl-u$)coWu0OK6x+3oZc)w zJ+6v{ZJpKib^Ir;-aaFpitXW&IOpW`g%y;z0591DY*9Gi6H=V?A3k z!R?=pTKeQJ-<=B5gTBo~Kel|ZaZ2Y5=Z%MFpOg(*KEY*7@@42P&f|+KyZl4#WvVmn z1R;$TE&373@dO4e^m>@Qf@s&yN1<2_ ztpgim85n_Us;DJxU_gmF^)`xrz4l7zVBjnKTy}LGBc735wmhpMLW1TeY1NNIT`L5g z?gd~Q#-+D#w6YLILlGUIKr=t~(${GUx^G^kqz$l%xL%0b*=o^Jp?8_BwAZ$uQT;}< ztk1;8G?+>lN~Iwgg}FNi0N5 z_sw|17O*bB_Yx&>Si226ew=Yz%+RHYkxaD z43rK|z-JU_<~@6s^U9*?b0Fp))R9&FiDn(ZQ5l3Rq4OWK%~X{37}H0Pe!wfumPE(h z6fvu(GzS%cEQY=96*X4NnCB`%2KJ5E*nePHy>24*UTiA~Av>Mr({J|mY4jE$f_4eU z|0O3l(q>Q^S?6{^i!!lP_G7g}N`45C_hL0VZ||LUoxz zwJaldaTTvH`OVIn`-oUoT!YN~)vGJBqMO9V*l|EfAyHIRl$JThiFAVI2fqayQRIEI zKgsvLB(6ET=@NroWoSkVRRo_%!XVR9tB;wn)xhXZ%PvRAc(tRk}F8>C>`vb*+ zfz6`r;mKG&aj2b~vw~X63qY6u!8OXNBvkO~dY%A0VR3UX8v|SFo?$YJ(%L!vUme~Q zN-f>Mz$}$rjVOvsNRVS~cU}GR3it~@`M@hx3|09Lw$a$y2m#i}k!X|lmbyOQiTO@% zuzJygOvJYq>5jUYBto;85=<|Ba7i2mytZ{7njjHSJXLx@fSOE{ttG|9Zi0Rx}>p!!Gfe?$@sG6(c-59 zf1}T*;%7s(8^svbuG45(j#hDXZEMvb{5}!{@}W)wm)onTfQa+<#R!MFnD=*uhlv-^ zE&jb#U#LGfa4;0EE%{q+perb?jZ!)lCq&7+0_0!j&+g?sU;Ow&tB4oMcO-9bSLBCi z7wM=Iyg5ZJd^7g#Q_0TUMO|1z2tM@LLtf+lLqR~#{#vdIQIz|tU;2D)-o3{4$>J~j z*b_nWu^xHVi;Kai{*5w#1dzD6y2{9a=J{g{sZMl|e#*f2SI1~qKpb<;t$a2z-Ruy+ zc^9Ew({T9ldWYfWLEVJaVsPm7vz&!>@j~vu>N)ORbR~yb_F&F$4ZNtKRR;i%%DP44 z@dZ=S7J#S-C{uw4O-^mJXz=ds%5pe9f4tH1o+l=`tWh47<+&33nu zq!_LH+E_3K2Xm4~B;@J_28yg-K-T(7<1J8tk?%FY`aMO?4!PkdwM@@1~QHq{96ZrMdYE*97fp#)4E5{3|MQ2|t3wOP)`E4YtNxhAcM zri~60y-8oDGSa?wW9Yeu6?=MvU>j-gGvm*abE13v$4Z9s_%yK7mZEEx*f(5 z*&FNa-{~kkFS|BT%I#hEL2`Laynl1bzI0-x{&Y!Rn}Tqy*~L#&q6g0b#=QzH(KXsJ+XG!r)`~a}-Do@cfDmsuYciK%FRy3_}*cxcFWh^RIfh zfDN2!WyXX@M9Fm32$X)Z$q@9cZ0(l^gdUvZUW+A_r{3r5S1@y-+zVV zHP+hPPlW(UQu1)H$JzKK0TN31ZjCmcTMcn=27V|9mu^G{-l@~j5pal6ci8JaK7EnR z&eT50o|&4*nA7=S=dL;-Lc2o+P@?nHC~y&gBApW91K7!V`^vfS=wDKUJtE?@MMN(xw1K}~}ORL^<>VycT2Mf~! zc-u1iZ~{OZ#o?pAKtZ3ZLkt+?@KLKkx(gqh3zPcuoa2P3I-%OtT_*S?AjVRu>mXH5 zGK@|*)$kFbkb2{RjtZ;Ud0@9O+4;9WCi*?bSm1S1m;Z>@vhyYZ&TMf-|2L~)aO{SW z@<2ukKo&A<$$hus7ZH&}^?e}e{2t{4p=Cfo@;;7yvc~SaBHugn-L&En1SJTELN)-&{$92M^yZL zVN_wH5LaUf{~y9?5}WU%=x=}u7aD5bAhG}&1P9-NkF@hG9Lx}Rf!Q#)cDoj@m z4gt;zbf0EW2LO_D&C8doK)z0fbqt{s{3A0=Of9@x2d4F)61OARDtQ;E#51~WK$)QN zny2@ZpqB88y>|i&;O|%WWRXA*(f9{R2AfTZ;$w6Yy0kO#oxovlmMvmdj( zeJKpc4p0CCsftj91e?#m$iT=LbPop$_{&&nlfrPi5Mv0PG=K)LNrw2S8 zmi9LwugAZH&qS9t=@A_`TXqwfrrJ5j%{{C7g{}nq;=(6 zB}Ek2SWn#7iY7r&=b$?Q+mk^nKWPv28Y}JtKFiKi>J}c%5e?2~Vp6^x&OZUrG%R154pHc3$)K`5H z?_oYbrq5IH^^p8WwEhRt@YK-IkVmNbA-);lTfN%(S$*`1jq0A65T8woh&WaFE($Jl zgiYtQ;2ul}Q$X%LR_=BD=i}o>WRyG{2(=gCakQN}CK#KXJevx`p1>w}i-3S&z9WMH z>6D-D%OHpoi$$)krNu-*^ivK48yZ0m3J38lZ_gkRXwzAys}!a_yu8h#kCi$>EKG0yo|2*W_ zuyn-)7p5i`XVBnz9YdDxArp}t;f@|G4e0zharjX<-`k`D^HAh_8DBr@eHs#s91i8R5u4pR{q z1UvXlaN$32at1{qjCGH95*vdzDo+If$>@6$C})|(+R-7-ttnW((P?R_?$n6@I~w{x zCjeverHzn#TH6^ES)jp5AVsjA)c~-2_+Cr+9DX>S&Y~zo5J`^IMEo$Su-G^cYI7qQUvS_pgf`MTv zSp31ri>A9{U<=8$)3Jozq>>dMWRS)|UMwR20btS82_C*BBqn9W2JqIf2rQSE zEPC1qW@_y8{1Z7E&xeKjM9Kt|pte%$3g(~cxFEI?i$z}ntXOp~1X4zZ0c@rs*Q-Yd zuIeujkEtspgNz7?Ie80v$vhg-zd3Q5`iwTB(xVs{wH9$z1jO0kXuyDZl?Net)*bR= z?mg}xYaopRNTg6YV76g`j1O=oDAaL5{xO>jKNty65@^x}F%Yn-KX{H@>d|=xlqws- z3{4nXk-|HsAY@@ZhV>8|1be`hhB%7=kC;Z8^~`A?MDueu!TB)=y9zYcAGR*_Jk^Wp zjWkHiXF#;k3p1;rxv3U7`?iAkq|iyDFU^DUA3z|~5{vb9ut=9%JplN}@U>ni;9LR$ z1+K7>8?aT`UK|v(wYAlnaYXJb1}EBZpy79}q~t$90Dc4y+kvMijp>MN7Z~$Ewxa*y z|6Y2{>)1sDyS|fuv?&-7qY^#C3_YxAWW-U3F`y;WnI#i0jcLGf4)ljWrb#vv&i$;1 zxS(PpR)SGK3pmGb7sy`0EKy~b0v5njKgit4kOPnEkp>+#GN}p9M1E@(Zgq=n15Fi~ zldc(JkFP67FeQq6@>yL&o_La{zz0%=@aN*1PRIOkqVQtHKL4E;V+;)xv_F@&vf1G&w>`*Pv5Aj`0E%6ck`t zOa?CC1p&T`ElLV!<-tMWqV}xC<8bzwEQPgRDrOjbGq~Po5rDxT!~)D9a+-Atk#}Lm z01}wcd7cSr*h9`{PiJ|);&w`eB^CtfjW9O7dN)>-RO0mPY#5(W!#z$E#-Je2gqer( zeUN~0T0n$;C+>D#tjrLTwt8e!?^HL*r-tpKKhZ54xXMDp5ai|gUw}(p|C{b^675wO zHw}V^-GJz1?0E?XZ7i15NUK{=rV@IXooC_;wYW#rp%oLz%UHgNqR%3zcAQXTJ#!UQ zSp>On#B@W*eEb_`^(S$Z!E_Kkd;wXu`B)7T5Q|+ku6HBp@BmFK_!v|87!YRykmna- zWnJry@54%3fw{|MC%X{dChxj8HS``g^j@nQDSjf?DqL;(kCC)rpt77nhMx{nOX?9v z`=5^>((!}g#Li8TYIO^gEx z;F)A$u6yhX!JQ@kS`H$%7M}FqFS8pd?dv@PSwNK48CFnTJwx#mB)E~e3IcC1SO5Ns zihj}Xgrnmt6P*UKO>oV^2CPAV@f@@+K}swzU;)b?j(ucj&N@Kc4HX0BK~^w~MY(AB zzuy-MlSDK2=osmGh86Twvv_v3RsQQ|Ly>T8-GBAf2?#i65Xr7bA`r{K6wv&SDWC?* z8S00KtvXpqEbKdBZ9mX?#4u>A|1k+J!<1v|krWe4(V)|Sg+i;I2+SLb|NgrVFHE_6 zjt~+HSw0RDPjJ75v(*4kDi7u}CR&BgtiBbhO=+ucoRF&GKuKMyg%jPw95(Da*dc(B z_^&A?QRXtEsD6kzFv1@~m{`Fw0Kuv{0S0gh|MXqq7Y;7HoFrl`@zanhyo_)gd;CEq zH5Mz*3^Ht4;zAB%(lBT6{~DD=HGOkg*-V|QH}8tTMKkTC9!J)}+6fUhgl!Rf`DjYv z-+x^+Eu~%^YTMy9<;DX_SL!!!()Ao6hcSGf>+D;SnOzXxz4`A3?K!``?P&cCm3a~! z`S}&Z%C8}0li6(nonB~DD|%%YV()A?HsD-9CLL&{O58Nqu)F8w^4N;l)cIqM|4&90 zMOyV+6jUCr<;96kgIXFl;!~3aCk?zFe)u@wB`YMsd4Sf?NxT;B80Y5rW*};7CDLP& zhn2mb_E|LBO3!-%iTz=QhM*6tmb|_Dace)u#ca>bP*nwm!i6>sw)}FIqpU0j#Xv=~ zicx7#i)4Wer$4j8K+xD)AyZ@})NRy|pW~AvAzn@{ad#gTh zV->_^r`eqHBNroavk*b5*EQ2i&tm_5@?oxm?1m ze&hW&tXW*l0n0`1Psfa1F>XMKNmiaRrLCG#>`F17o@x$r824LXe9h&ZIsU=E zr*gTR!c9@oe+iw;p8wEDG+gfIL>~E%3@f2aokfs5Do~McRDzCKE3e0^!^V4)=bsh;klTvER+llrL3$KH&HLRFT2#a zotu+inMZwbyknU|UYtG$%m#S5fK3$35wICF;pMPaLf0E1_=U~-ukM>~b(=ZSPxc8% z_WQfu8X=j_ZZb6l@r$NX7U@6Y#qVSDD^qA)L{y<>S`d}0K*d$aCWWs<76P&?h*sk9 z(a;2Hm+65RKgw@iQBnN%f~UZ3XhxbVVX<%HAT`}JX5n5 z=6*IPN;ZnGfM_u62T%X(2X1Edmo!n;S{<%tSY7207`9gl&ko^NJ^K=`97n7A(^D}b z#E7f6IB7^#f0;}>-OtYWPls#=V@mgX&OPtytw1uur~dt^#-_cI z1&0om{bzsT15rc20cD!J&nj@}v>;P3Z{ynnag#+Jeyjp5bp zu}zvfx3SK=-HpBpOLA=9TiicAA1J>;d3%&=81B-rb=H}jicMMm6n9-zTyuBIhMG6` zA**$q_ne#}=M9aSe&VadmD;mrY>+}%L&E3A)d8^;aRa=#SrAd2lSxLBVi@xz5LJLf zTOGZAr}dR9y81}8ng83garZB0vmz1147vf0-U>248 zx`E?wyZL70sR~_P{pw@+XNp(ZR0|$VtMsmwa?2G6BuGDg5g;1&c#FnQjHGuOlEj{9d(=cHd~1Nl&9Qq_L4Cso?qBYKJV9 zgiD)m`=1+eoX&rCf|M6_&%0<9p|!RbNPaFz(G{e-eliFqDgPNUMst-B6M=WBXl)(B zD%xmrh3hWG2&6wrSSOrWLUT5YuTIIvuAi^KBvp*ViHsxFVI z>>&(x;5)kB!5!b3G!_2CCrBbb;+F0AFGJXskCd&5nfZ`|IoZDS>!Z)d8KJmbPuGRf z0Js*z%>miLBVI2zz08)Pc`o?+im`rTqfc)$$K}4**d~i~2k|3Ky=Kz6#|NjHgLRwc zcn5LDmYXoXA;W3}BKwW2FB+CbVOVoPKdMoFKxB^k&$@{Im|L`Ec2nwg=S*K9&7Gke z8k3w)3~+ykMWJv(i=yy5PNr%l!!;#bX%B=pAY{A`OUj%M{mZ~lxaYH5hpJgU9WTUD zp|I@M+SHJe>rGoEnB!W27$OD&)6+18%<6fJqXyxaddPABH^6s%`2vVk9ICKDj2q(} z1V#}!{1e#K2lN-#W0Z3{ja@l<*MF&u{HBpSOQ_qE;J(B^IlH+ULqVxy@;hXtI*t%_ zAe?|>QsDRssoh}gt8o{@4a8y31K+ft$a8^Nb1X57Z$Y2mdNj<)a~0>wC%31uddw*m zW8K6<`-JRcZktc`bSeF{_Utah**oF9vudus1EE$1yc7hI5?O|UsRd{61SXnDwk>fv z`WrPVKhYXZ;b*IKPCU`grKXq6x;?Ml-sRe#0<5t4%4Ad4>i*`U9P3TnG@o?^b%fTG zM}|=k3b7H=yc5s5`!3k0)dZdkHMaZZ_UK)kn|XaF+zzP#t3*$5^iGZz1-}eN1th3F z$h7_S%NB{H6?$oYJ=W{mCxp448HL#8xViAq z_fsU0M5R0~h7A)z-GTg(q}PTGY`NHITv-fcZt)n?7jWe1@vZ<12ND@j0{Z?K2xsAK zw32?}C@ECn0v00%eEjvQg%*t>a-+ zgkZYS^^;H){p*F0Ez%|P^YgpmKWTl#B7qYEl1ECE54 z0agxo;odusX5Z#AyHdN)2XOjFRK=e2IayxF2C50u6h(eL7po7tKMduY*uY%j0S#95 z9f*!$1INoHj6kMcsAWC1oUQqj)smdp;>KYsjjo{SjUT$wr^PO1={VCYK!U)-Vz;X9=0irIKJf2l92m4-A51IFZnpB}+5f`pnr*&*mi z{d<6V+#*%Ql}Bu;K94tCu5`WOCtJ?CG)&Affno%8iLtqJ`Dp>Z>N=K3CMqYt`W@36cPNaC|DvbC`hZn=`z?H%PdjBL$uD~*ZGj-Fp+65_8`NiP}yi1!2M z#4NNa#L84@{)e9GJ8-LoSY3d}*bL`SsI$gnysIx2*R|QhApt*}xghE=Yv_|M82dFo z$*zY>!ww}KppEDwgq_^_;fh8bvYEQ8M&BZ1{E`c-!>>u!X(W%1fT;=#gEEne2kUoG5h9%E*c!-c;RpdiUD{q!X~pZ* zZ$dc1bc2N#aW%ReI!J?06zhv8JYY@s_Q3M)Ikh6M zx{YV@r=Dl}i02@Hdv+(>-dJdI(^; z99Lvli7$#u!?ykJuXCvXre0EsHM3HhWs2Erpqx9W7yW!%>_@<$+-Ds%e+Z(tVl-@< znK?;4;_#Fc_5$-aH8PqC6r3RhcE16cRGF`cGVK3rGdh>u>v^G@7Ajt8*bnQ;MHyU1 zKC0dc7j{15HADX>M-00n$leM_RQZpXJjDt2Jaybgm_kZ+>3sVE*O3$=ekljyho z`j{Su6Aa?@LIO@)#>VgqkZB{V~x%U%V^+B2SmCMM@ zwxCrJQVFhHK1+R@TqVcvLVR48->~uK1+}~*z}BHn7CLXybMhFeohIBzM5KJ1Z=W{R z_g-=(+uJo-GU(&>I9-!&rK_y0q4C!QAj^RqRY<~PaEsLT^dbhUC8cq3;-za_BAUx~ z(dP&j(cSa{X!mXE^Vu#M*L2=d+dlW1lP8@+$v=MRvD)t{`p}^6z0|&PmQ78;sxQCUUDLIe{jZ>|xIxidz*jP5@aKN{Iy2{^=Go?)lfs zGkx*b%~R$|?<$$=m(|6@_9PB`er@~t(!-l~WJ$S%o2wO0(BbuQuHb%^iPGlEa}by0 zO3yAInB7XTI~-F|zOP@AoUG>}dx%GFxM`jsLrLuMFkz6EwLe-GU&r^p_EU-wzSkc0 z0r|!aSGac}vp>Wag7>fe6{Hd_U+j3s=%UeQPuB8*;%-fhPDG98iwu1q&2mnv=}Lr_ zSy=FV^q#Z{Cvxj+8G`df+GIIu<2@$shwk;KJFNO5H(bVDmF4|79EJriJ4wsNGrgjI zH$x{kYQsTti(Kg4&;vi_cnhxh#!>xh2BA1O91^^o4Egla{1Qpqa;(;IF66%5y=O~& zLUH=}T#wkc+y=c)3S*yDFVWJ{14tnffj;Ir?p?9v+Mxd&y?jNqJ2ByxAe+Pu!sj*k@K;m2uB=v|mR*JD42u(nN2hUe_0@A5z0?_@cf9lxY;%I{Mz10MMXZ0NyFWuT6pQumHH?+p8|{UC z!nupC=(?5Mwa52QjQ3b@&)@3d(6XSirReNf`_mW4x1+E1$ z-PYFD|J4@aCmM@IGT)c7pfR{s9WQ1m95fx?a#pEwy)1Y-bOwxdeQ#=LFen}#r||w` z+WyWN(;j2#rn^9Dm-Rokqu+_Wt2-dwCxc|NfBye&JWlEc!4*C9S-Pk@8|pts>UR=! z&kP)XvZjF(xf?Z1nu!F9#o%kZ1k?jAOwCg}mQU?>|}b6Pmfo zB$rz5jPm{OmjhZvrrRHE5dQbe)w5+opB=#!ob`Xc93lI|<4}$l2gs6|NVcFTFA7Zd(a@LBmNzy{TUEM1~B#jD#_RW=YZIN&@uNG7?53*^h?!1 zl#2sN0{)$4^>zN_P_t$s2?b1>3(_b5J#~)@0K~~x9D2f=fL&R6Dhmn>$9DwVw^(@=BTvz7Vk#`=C-M_s;`C{yY3qW~D8MUh4Kp-3}!n z^(Z&xYmd}CGo|DwJ)?)IVlU>l%}Z)2hWesI&*ljF=doBB zJy1#Y2d}r+s5fJJtZRtEV`EyIanV+?n@&PPbD zpKKq36aZlS)g&wAW6pmi?5mez^blH3bL|%ar4BLAb*7Rc94HySbozW%{Rd1Z6j=!9 zMIp0=jvzyy!iGp@3UTpa`&>(ijwHO-U`)0TW+&JGc{7<*1=k=mOoJzRcWY!LDWT$h z0D|kb#z=Xv4H1beBc{3%q4C#PK9~=AN;K65vv4$hn<-{kV1JhdHhxcnbUEIU~~WO+%qaj=MEhH z+6Q|HRxT}7fPX-?LuL`M7?FhM!ez+z5duE$pZU!d!ff*aH?$2@u2yWvC7??{PB;!A zG=udR$QL3~U17hD%K(4(aS!w{69H7R)*Z*2z;!Z62pQ-Hc^^M}uSh5c{xvzWP9?iv zv-mnu>kUZ~#NR^)ka>}W$6(v7289w2JvAE^M%6Kzq&D@Xd8l$E0YtC0>I%p5Cv=@zXq~EPI;f1ttvtnkH7YQ(nu;nJ+;j(CZS0K}W;gg!W$nIsqR+P?^eP7EE1) z-M^Fl(lC&2%kHL?;*Jz?TF&F!Appxv38xDD>aC`0!3s%Zp*bpDT~- zdmFo>W!XW#5Sl+*i=7_FIrUI;b+DvC)(02HAWtE`9EX1SYjmwQh!n4@VU}Rfqg(us zz`SD{fNIn{X04V_CE{aZK(Z|nc3-d+>xDno#|1jqoSaN`AzR6CB4aX)ksSobkkf6cfng zfO)K>0Qko5y_Xpv1>ZpxgWTPK9&(47378q?Yy!rwU{yoGEqdYO4p)v2)wCSx5OA&`c7}PIpm9IwW+Koqla6as zj%Q3jH<&FqN@TIP;_%?7nwP}iqvU$UaonSWTsLHo2yt?;wa2Jb#K~p%0DuTA_{%1} zN4$UmHNb5A*yvfok~QcF;<^hRbO2d9ZoBqNxb;F1A*)r!26Fz`4g>)@V1y+C6T|*< zbCZWi9_%!r$c=Ju?L**sxMY;z^RuCd;7^frFCz>T2ISy$l;Pa3fKz1bG)9TW1vCOp zfn`*0aruBYaYgDCfM;aBK7_lv%7PW0#-HCFUIVYc&)|ogl|}wGRV3yLmQ<0}q?-u* z+@|CB%#rL=X)tWs4|*=)ATA5s*gyh0KK%=00k~NLVQyAw(*;30G6zg(bl3s`$PxO! z!lejl8V0+$jEn%Che2Ha4K3jJB~zw_e`I<61?JzcNJa|s06h4?KlTG&w@afa!Iv0XFG=*c5GLZX)JAcq`tJ#9={EOak^)o$?h>k1qV{O|*xH`%Wv9 zVIcOHLyln+r69CICMA_x*-08U16z*toZS|TU4mHR8e=CGy8BPi9}l54=K zK(7BK2CC`KzkU?~-Djj(jpXt+7`ra(WKD|5Ezz=Fkz&#~DC?LRR~xBQ4UJxK=}#2_ z7mW=V2h$%%aCYIZN6dmP*6Z6e0O^rgbh`%Np!wHWJg{uX$h1I_49O&(Z%5rjpC2jw zP--<&KUp$Y&hwCAxk>xI_$P6NI)dn^waH7RGeY%On|n7k0~dGb_vq#yfHmxm%Vuie=WuixuRs7|Xq~|YBHVsxS^NyHG4s|i{7GR;tp*7js zMe9!_px>d@4IYy%pC1U_uF!8ZIx!BGziv8ytJgCk|Edl|(wu0yTe48l_=@rRlcR*v zw{#olR)+68@HVK@pq?scsaxgtihx;(iHQm9;jhJz6BgvnwpMB~5WT65JsiaU$4#;M z%e&$E-*cRb=n1Wh1ND-DbM<=j6Dhib?0M#q{)ef!HEb%ntKe$mCA6&b2I77YLu)~Y zlV0%q2G5J{PRCT@{&l5{o1Mc2QpU~6UJ`uiRlN!Iy$ej4`CjN5aQpNEvuMBvJ@Hs1 zr~}xK*>k-@E8~nI(o@y|hp+L{TQG6U0 zI$b1ywV$oa%!ll|KkvE%O!l4VUx76HtN73!F(Ir#e4%Qlx2{@id=pYroH$S-LaP9rg4ZWkXl)@=kn#XTCw~3C4rmOAO}VpPltH z+8Pa-+c+f{e_{|JhCyD6zz|0tKU&{f0p;P8((mFV^kCymF{aeE4vQ4s&2#0JyoJw2 zMG)sfw!U`juXQ~s0$73hKE(^yXn9LCHgl=2wct8cUa?;W93cSu6Ch+L1;auJc~SsH z06Tdru?q;XADnA#x(CRYgIEB2gM-pL*lv&U1 zD;wCp0&Yp{{_->f0_S(k(%09&DC%Yd9<{(jFaR$r(Of+Zm{@RfnR+i(L1O?^4zJ*)O`^!_@b1PPrYhLEZ6XwQL@Ip@tFIHfPdZTyd;-J^wylp5O zzj3J>yH4?W5tmzfq&c{O_S8$F)WFfHf{Qz=X6`unKPDL8yRgKLG0s7L5wBKIr@6fq zvM(JS$>-W0_O8MGczQvR*JhB8>lN6|{Q~5Y(uopTK*ZN2{H)L`zZLqy8?F7PhI8iQ zQQW#JjGF}|dvb%S?XvYup+7^UE0k?ZjbOsH0MP4ogWTg`JCp?847|=40zwxT7r+B< zAs{)EA6?*h_YNj05G~gMUvC8diMp;{zNdDkW7OtN-%UaMyt>g)3#RfxsS9urc4_Ye z{QNUG#Wm^4{f?F%x6Vz()tKX8kt}``PrDkBOwX0Oh4bu+L;VTo(RbAFP@ClonCFXm~}kr6`t^C>v{A_ z_8sfx`lCKGN(`qznca3&YIs@!vQ@OC8<;h;89a>1AYhaX9C)GaH7uxLC+0vjTCU2 zOw}*}8@~(yq5J59VO6d%b{(rPY!rcGjVjEVa4{ow+q3%9)H{^fjhl8%xA|PljZ8T8 z7sM4snafj|UCQYzRDX`uKI45l^TsvNH9;A3RN5)6 zm=Z4kV5QA5{Mfm)5nxikAV*_^mi|+^;R#l%6u+2!%OF*!_^p^EC=#;H#ps9LZ+4+K zh@P^k_cU5e;cZ}$3?{FPK3)7>WhA;?IbtSaz4xg_0nfyIXW@qp&-WN7&Fd!-ehQ>f zd|(Uezq)_t?lcW0bPtfEc7M*Te5!d`jc2|NPVZ08^FE;9jw?uk5dNx=(0w?zDxW|8 zQ{8_%yVjFZ8lS~s$i@9=!9Y*JRAfMJQaatoilIMO?Anmxhj#7(OE#sa!3zdf?K+%M z-?(@tY~=JeXq%o!yylyL-%o8Q7B(XW`3sg>6k4Wm;on|Km+3~yUpHA4Isw=~S7S@GkZq2d$)QGT5LO%Gh z`bKnwP^bmS-4C?DJ>zwz*Zf!&`+(SU)n?2Q%2Htq%QG@NHQ^v--3TX;);u3`j4e=? z5wFk4%0D?p4C<8F8pB8#IyB0JD1uyhvs?46b;mFL`u)1nVwz&NC!epl_N4z(RJ`q3 z$DuoE1Q3NIYPBiU{h`k3iE9J@EVHNtf8AZp2?s#o?mL#dT;m%&3xyrP{RN|8Ev_*Q=ll3-FMe;&?WK7yp zcGrl3{8O~PXf&yJbESF&VCm%-F|_@f)HF>AjYZE6hvGp-FSJ4O zzWhi)uEk8tyygWb0mpi{t58Y$9Y8R|by_1TIaw4!EjRUi3pkSD<|$c-Ed=*C(~8?f z4w-+7d)O~GWoDLiEHfSJ(c>B)(?lA7GR(>klv!E#7R^>XyH@VY%%4&-uWUKm%~8N9 zu|Jc@B0<2+;)M1MN(=tU`!eahjGlSD%*CVvqNP5OxUJa=6@0l4_xSuSm!0^)6A3Sg z=}}vb^C0f1xm?cX`(VaBtM9EPic#5VVBM=C;;~ENg^B^|_$$w1_nDgRy}=#RFjlx=R8CNa38cKJVU(o zV-(T?=VKIn#zW!i;)=u=GL3&$LTlLV9`BOQ>*n?S7l!g{Y|H&n?ix<0ifZ$PARqp9 ze2YHs1Z#q1*8P*ha?$ag+a_lslf$ZI*HW5mIusupJSF$OsC<{gmg+;U(cldcHkwPm z3p`Aed_PI;*aq4sLKe}eYa9NnzjSUo$HXexcZFHk#_;RMH<}yR>r5;3n~S&L_nU2@}aLk8vj5~e(9cUMU z^G}?R!P27vwMj=lFCN2;t&Y%Qks9sK@49~6m`)gGta2Bp=d{}W#jrBreN9?EmpweY zTX%w*5oNlH_ENg>RWa$zlOED13Nwu13IefSIv!4RGH-GhS7yvFp%Ou(k1_mY&dT|) zL-KWg*F%~G(*XKpM+N`eg<5yc3XQHU{kmh)RR?=1%;E^8+Ejs7Qzx=txxwci8tLI z7NrWw&si9gQ>;%1Wg(T!=Ekc|B&}|YTuVD$Z3alQ40tAhIWq1s75R-v7_le%&rFO*>H{{3f)M~0d}X#F+$XW%YOGgNh{Rstu~kUCvKhjdQbYu zfkD7{*Sdbzw^wNOY1q@SF1Onk789m%{cMd)kr%rd)@q6S&8 zqH9>?x4NH!n@^$o;bpKFK~9tp+8~~VBw6fy#Ed)>hD~--xQM;{TGGn7nTg-E6?#xA zAZl@atWN07qk<;BCJG;c_ZR!s*To*YXdQe;>)*pLp4}&j^>!FebR^c97Me`LU&~0z zyY-sX!`AQe`LcyD1@>Z-_OM5e_4IMox2L2I8ag}Qxk1|Fi-gZduxR}bsgH~D&JfBX zHoJueow5<xa znfD7Mw6v6|A78KSkOv_5yMfqA^)}rEcH4Tk<~Np4#PeU)jMVpidY|i~hu6gve~B2q z=sImVv($*U>>iP|d$GqmpmuxK)4sXrj^}yHch}munj`T6ouXOyZRo?DJHCNe)7G-T zt84?Wq;~InQAdi4>=SAQ_5d3FTIa72#9xc+`w0I}p4|fP&uGwoMm6@@an?4~op9|- z&oNp|vVQgcx|c=wLXx#P<;i#<+om3exDzJ@25Kxn80p0J@qMZ@j-;~Vt_iODWFKRo zfM&nNti-?)5hZwQbV9Fe$8$o0<)P4mukU2pw``Zii~ih;(%TDfEvsJ%1V3n-3v1{$ z9I@1&(0BB<#lmlhlHbj734Y^!PpvGP_v2y}zwJ7q-@}=Gx63K_xz#jS*J|bme$eEGBxi=;PNK3E!&AhnJm3`dcBr3y>TsUiODa=Cpj7}^is}l${Huil+KH* zr!_`3RHT03XcA+eXSw9DG+mi+IUtc+MT`EG@py}pBmPK$?9~3wfQey8t{8-zYxh!N zB{?=#V$COH%RRjcw{fk-2!uFl-JhE2&igwC)g^LBDVcbYoX-*GZoLS8h)! zb(n;B*M-QlF+H4^@t%{fe5}Ro z)%ETocw#5oOUoLXVN9c?bf;60tE5CAuXG@uVYW8-*EdmNy$|TSLv@K2`bpzq3uT^Q z(_qeok~n37I_FvaN&&0uiJILW;|)NbQZ3fg9X_?Lnyz$SR@1_@{*(O}8SqJD ze%K^3;)L7O{5Y%DcDBJ6aL(q!xbKSg#DSbF7hrD6@Hqb(F3~D1f0%c^#sK?*R&IkD z!$6<~W3>iI@ANZB_gGzE-(7g6eIC%WYQTxDt?dgqgJ%+a{gciEFj(e??Epc{2Qb1w zvch(}6zUdIUS19oIz#9N21rVuy`&vd#Af`wGM5h~HSYW?L;$G=*Yum4*fRj5Z*&G$ z?mtPNfF&R$fJ208jwJPmodbU?fl&c#7*}1p<3?zx9~xN==QKDHdoFg@9Ie#?EF`(s z)mJ?r)5*e&-3JOK665C};D7E__L2BuokD1$p8~dSqK%eT!WjEmAP^Q{yrLJ^3!h)} z0@&loSh`1&FfEkindgaC< z$N=>wypyaSM=G_CKmW5^14!f$p%+X$FKSZOssURSC^!(5xU6c~kJ}+d9J@adzTe=`Q98F_K5TNH*kX ze=q?9Edg)?yVPL~GW!6phFg12(2E-?wDXC`%%RYf&TJC5`}xbx`88N%SwYSFX>e%l zRL8Lbm2d>v&7? z7HRw%=;byapfd!j)B*J$qjWx#qN_N$i0C=G18q981_NMz-f7;VRryjf_wo4}oo=@DplJd#|gGY{Tjhqudo=U_9w##a$5ve^>DPe1a+`6+J4+VBO z@YrInpPf#iTrSrT1#k&V;FU`U<>J1P>N)lqFUwm{uTSx9zqGeh)M@U6V^3^7e@?gy#ZI1~xI%s*h*OTFf?nl_t3r+BA5bK;%qLl|%ZW5MZ6pW6^#m zSb!w6X>xAQHqL4G6I~1jOO_>KhG0cs3D-EdtD?sYNvH>xsC>n|O^sRg!PH{xlHvd( zRu-%{SO{hAewz&MW8Kr$^s3vq zrDrMhg|mIrdl?wngpg!>FB=aZrP%SkN7&kY<5~`{)adP2XuAO&@!Rlfg6t4mz=9?h z_lt^a=JACCN#nyRGuf2|+`x#$7%?K5d%cH#-oLu`$5w>;$Lq`tPOjw3MNvzG^p0Xp z|CB)UaG8dZI~=NwQ54#RvZUSH8HZx>(@B81cArFvDJzgbPgR%^HBFJwbm5m#KV`$s z8K0RF$Ie<#==|&`S~j?AR0Asa-#vp0P?+1ft?`le7sti0R_wDr-OA&u!iK7CUjckh z9FH9uQF+If!IAs^UNJ_g@Hv*sQ*tf4aQ zCfX$AJ+Ru~<$^^~$ix%sk2^DN^ZOfAD$2ccjHw;x+TeWWzoZVzR)h-lCg;mrP$Ysk z<$WKmaS$8J!}kcK;sB2Vw-kq%J6pywVQ}&8OcOJohg(<`pn+>62EVcz%n{Yh?tK(K zZxO5UvW3+($K%BTjQLz;olIE59$qPdj;Xb-N^N1n-0<%baJoVMv)Lh<=-4ZC4WUa` z0j{NWs&0Y|y{3!bdLE^|HpYWlP$U=U;=)LzC%M~8Up78!z(v$LBJ9&(04s`9ct_Q`3rA7R_UN(*Ls>*W#;2;H?IdY|4nFq=zm&dtpc9L%94 zg*GeMbcON3?-Ve@g>@?nI}1B|o@91v=d5lEsCyfPoJ?%TV~wpPYHZ{yCiMdj9~}YE z$oMDFi*`Br8=zRSsc`5@6b<9GE3Qs32((A{Fh(titua_HFnNm7*5+ktPwDodzPMUZ z{(&r=J_=s&GLt_?SYSU6Z;kdB?K?$3x9Xz@JaBYAeSn>x_g>r9@YB&jAF5dTH0-ff z-MlbqWkS>s(^d{@)1%1gxb$F;j+F~WE}@8XY2oiuq5=r%ykrQN}^ z?Ea4-hd_fPoNwIm@ZZz|d<%xoL>Md#MTJdPpPU{`=XLZLl^ygj_kEaGeU&0{yPnTt z>?RBxtdf}p&@C)QjQw~%cW&Y5b`-2_Gp)fKCcBI0Y7@HI5(h+rQ})r}_izj?=<1m* zWfyB6PlKvaumC?2o;irV!x-D`sKtjqc#`=(%j~X!>K;GsI^D8Bq`D*-aiNlsK)qR6 zEZzQs<_JHPfR#;d#4GYoX78LCC$*ie=jlia%wMm`xB{)nIWHugO3MHC`DP!;<+=ur z3MJ#*$y$!1=E(fY;$J`FGYTPXlZZ28+g#Eckz%+_4qoY%`8drN#I7u2B$@ZA>Z7q0 zCB0dBtYHz~1fNU&zTbZzLMfeA)vQEWR180Ib-)QonCCEwag06|Mp_u+cPnn%$=KEA z%7{n2YH0{I+RWu}J<5p??KbaD_#OJ;WejPFJHFXo=!+j@9uKV*+{J?rdb%6&KC>RI zb?}L*+xpWE@puq@oOo8#-bKeUM1Z~IPE;y;YBRrmCH-iy_h;XVmpv-O&)#6uSj>`g zu^<61^`kyQ$)C+m09i$`w*f4&m82M=H9e$me)9_Lwe7jq(5;Rhac8NMFpsmruWX@0 zCYQn5LA;4mO48UACMVkRXXH^3(&vzE!oEL|L7IT zRxr4y^W&>Wxgt@PHekf>Z2)eZBIpY2#u&kp&nu5iQTE1pW|V$vFg6&OYl)m#tun?Z znHfnE7rXCw#Gth~ov7RMD=kl>S<4K;<75WOE77u00{(32DCx77Ij@j!Z=i2opK0*Q16U7bosCEVr|pSwmBPI;@o-_ou;F zvzELheuxS4Y(-AHb(G@lZ(bgPkDeeUfu5#hYo|F@sB<3ew{1zT9BOYIr6iLIs*G}{&XR}^HprUtX#+&;!dV*%tv3?k+` zo^GJZMhm65-LI}es{@{Z0#Yd_T9_H8hXDPcnWV>fDVs7~(q#Uce zRvD&_k7va+LR@1t&v4|wA8-kY0~e1YiGjd<#U%H`2bY`=+fnRn#Tc+GxnJE@csRCaTXLjCylr8|P7U7ED$; zB^@$9`>vVdsX`%JAyLG;o`=}VRKQ_#mJ>Ij=x6=M!@BAFW-;Uk=c;&qcVVaik~loi@JSBhw(tss##rC4r!D zttj;-iLtD=S8{(-cl~QCe-=sK+ZqX}El=vr*vv4xy^_VB2{s3v*jJCfd+3dM<&iBu zpMN=CDwc#hs`j7I+dqb3S?J0StTj4RAk5i`uIl?@xlJSc6!c3sAlq-{sapXhhRlTK zG?{klq3PbTj^XNaGRZ^u{v+kLSq$<9?VGm7i`s+QZ|zk084Isr$y1Oa@_~xx>B=*Op>hKAB*h=DS`MUHPt-piyXb99)Z zIBK5@+s-y#e46x-;>)DgZZAD#uYKW=(p506vG{nObM0!|9bV=WojzOjnWs${sa9%5 zpTZRx>pw(q6H?00!c~}+gsqo8&FKls_7RZB(Ec*Xa@&##ws4^MeOkvnRdvjD$fFeX ziTn2YK|F(1O{i(!ZZ6KeO0@4uTt|d%To^1qIz#WY?yi+bKen1*--qtphS7;0~sh*2DHl z9{RyCwvGMGn-0=d$F|gma^HVoxZKH%c!2E^G4hd(Ub7q+ndl@lgmYO}POw{~h|k`s zh?12N5B+FM22>Gw{!V+^_@X$<-o{ya>?IY~h#rRe+wG(&7AT*8jAzAYCwprgh)9dp zqscb2e-+wh_Mx=yy1tKIdXv3Y!gcX+*%2KxwxqT9Jsp7)^31D7sRnJ7(08qD{zIe2 zJoOqMHXgIOLFzPZY$4O3rkEMjLJgW~M5ZiCwlYi_?_5X2al(TeRGvgHsxvq?LVP7V zBY%{vJ+ z?oyF!!Zv1ya%K`Z%ifD1D^Pi&bCAI7%(sX#mJF{n z&#l9mN^UzKfn$Aek?tq`W*o2kf+YVg4p%(9++1zkNr2tdj%Z&Hk4~7@TK@TB7|-DG z#&O{hlIa1x1vs$Y?OD4Y=W+*i@ZRR8PSD;25-4{7Wo=@{7^-SeGss9_BtD0QLVt6w z;%BGTR%*^edy7HQLm43d1uegSaX9O8>RP%p|E`8anc(*^zd#^77{c$To{?dU$6iG> z3r|qG{}JEr{YPqXl!n!WzLEvr*ITKT)m6LM?dnKN@xnd)knlIzF4dB zFRoMCDQ9+^@ksV+`0)?F`enx7?O?H{c^W=`qALESa3hEDdZxFOqvMWQO1AZ9j)bz2 zz4nOAD*ipf(I&Fe&MkT?+IkL5=aoGbY^8&@Dbf4ypNi$dLHZ$^DQvque2eVsH)AMRR!rMc+`ZWWu%=>iI;{0WLeZ1L60a`+Ly*))xlpxC?1ow) zY79VjgT!Jx-1pAYF<~#_nYr-XUo*hbj?qNn!qZL_cLqgix^Q>%Z{6`=)s+M4PQT>< zr$q{uAjEHO*8OOnQKB3{Po-*E5fj`jq9moI6Z4bq(L_y=RN4Kpa4l?ITwbTm6bSS$fN&52s|%y-E#(86dw;NCBd-t-akMZ5 z5}31HZCNErJprAA4n@cSiS4y-8ZOj!?ZB##Tm_1X|1|zQf0>Io1>hn8kxU+A;$2bW z!J3?^tLyq8s%T2Zj<@1c+a-D(hehY}`_hO`mu;4_0K@%dcm+IDJeyb@N}x8H zsh^@|g{3NxvTfZR0FX2!w;}j%{(p3gBG-}u_?+%zflbzEf+=d1hPxnpoQ+fS-7MH< z&sMiPkX}ocqbaiUs{GpB01AT(#=w1Pxfn34g3C{4H4CH)rTvE5K;XU5?K#id&nHTN z;-}~h)+hoe!(2?bM_~s@DJ@5DRY24nW<;>*=ynth$kZE3?yzh4cp}}x>FMI*Qw`+S zD+m|w!b&5d6M#t4*pjb+&m&Ncn5sER?0!XG2uK&un_SJ zLj=zI_NA|;d#u=7Vk&Rw#14dC+m$XVSvlp$kIoO#(Z=!AOz0UEy_@6W&mEgoihE=r zRE_ku4^0cEyy0qw@)}xcvg-Qw^rvsE+eV#u5;?7Z0kk+bsvH0<0zj+OvQYXaIIB+s zn7B%8s4g%g1+KbKXH}H*m$R0!8`vlG*oB%%hQ`xGn_tVN*wuJ$M#ZivC!(acPK)T( zkT%XFS&-0-01nBw2LsIrpfJV=t0pK>GZ~ zOU_B8WZ^CiyTO-NfZZ_g>jLCZs@hG$l#ci3Fs}PjwSq-{r?_xwluJGQbz)}4uW|}< z0@+SrxCK+)n+*d3f(aoJNk}kNc(2L~Vqc)KR{yccduTzg?o>Xl3pdzx z>(eeo>jTw2ss4O--qZBdBC>Ro8{VAr!7eiD8vsD;dFp@;K$wShb#>t1A(dAWet;XT z1D)oPC?r-CAa-T*=l?~vHT=q`8-3UunjWqcam^!?G3Op{M@e~rL!j=NTun2dt^-afOYbf&>jdyAq8))G%~8qSmFKn(#U!V6OSKCuo#}+Wa&*{_GyZGtEeu(35@}`+6C`}{$rbYG30~-aGYd7v(6mL z@EPD6a2TnenoL$&A2+xFI*Wp7C*U7jbVX8?`=kDOy)nOZWdPrdXjdK4PKvIF#+=2W zHTU`q7pd2>2YxpK)K7fFWDt?V^~(PJE@W0K5zc2q14s*g=Q1uLt1ZE~J+y=d}_OzPSJ;A4LipC}n0N->x|E%4W1qsl_3^5gnfV{K_au@snP;Z?LtV)pB{_Af_+UarA8Ls7b zu+@H=0`)YYR03h|V$2pOa~J}~sWq%FbPSMe$j7+;t+wYFnDK2eoz<2p`9${((wt}a zOhe5b;+1XonMmk}9rrr7Fq?yekp}JWm)zJZ%uQ`3)o#VvsLXZB9X@YQyuqg11Q6-5 zQBjUa9#+eqT?eV_hT{PMom%kW&`Cs)!S(46_g1wt#4+i0xB&bIDeC`%ukf7d5j|HM z!=)=41njA*AY@qV97a6j4k9aeAbJs098nyj)711EJFGZNO=NYXUs>P9Pe!gg*fOUm zHUtMkDPcCS!Fyf#AU&0SEcqGj;b#Q>66@k`IW0x9$J5;sT~RxR048f*Lne6mGXvm4 z0W(Vdeq$i8DcNeQ2W$j>$Hm*$=)wkIf&=P4;A>uw$spqA0V#pNmS>5U=X`;W$9rj8 z4Iz((-VSJ6E+sST@L2cpH*C^LfZ+~7Xg|!m9x5yUBrwW4Ucm|;12o!NRa^DZua{%^JR#M7ws>=biB@h}w z90t2dHIHZMe>Qd6xW21)nk2NgU(j@c4HbUL;p`6a@a-B_U0WoT)Uf$25aAF>N_Y+6 z0cXHovah#Sx54WGaA*-6HD&;12fD`->^6#Pe$msT=to9Hk$bL`K&4}OljS^Qo6iUG z8dN!Y-{5T#n8fdBkKO`g{Il4s6N8WuRQnC0sJGmPE$t3K1%4Z_6=s3qg>B=B z+ReNwz{FlFKGj#m^#Sf4^9&mASF)^*yZTI1u7j^?A|0ux4n**o?67>({=Rh6b zX|DCwL`eI0u2?`?b73CAch=%Drl+0UpHr$R&S-7m#j%P5}q- zngC~tbPeH5a41mMv2SClnu>%#z5!eixS1PZ@c?8XcOyn9NCg47G}vFo0V@@-6Xplr z26~I&41iA~w;;E-1AKS@y9&un==VKkxBLDD?BGCL&IP9lB>WL5rnd-$g5MnjuTobz zxjw**t-enVj2`6rpZ{4F16qp7Y1g~aZafS&wEoFU+QQigY%WycK)e>&Edb62=sO%m zd0?O#0PU|@BDge|k4V=Ex9S(5Qep!13m724($cDRDKw*D$~XEG>{?7Zz*39s1cs!e zpMX3B@HzHiCtN$-N09gF)8nU4eZj61Kw;Kr8+~J$bpfNny9JXA%&i9l_y!b8KnNDt zT-1Pt4XC^H18&yuF+d*P?qLVa&t$MJaoMiTg8VcMDW<|XJUC(A7%H4upYC5BQO*?!of2Bfox;V2wBzR`n$(k-9|`~IwbVbLP9)r6qD{Ryzz zK5srV9^Gtp>m$v=aQXv;v!UWPaxJvXsI#WT2^yF$W8Sfl0vJjm^FM>27Cs-Sn>w3c z=MFwrn42`*;gQ1m`?2_&K{{ru>u}xsenK~bRSR`%XQ2`Da{!;id)|1dF*5w9Nl)11}`?6WY*FXsP%?7t8MV^!sDokL-Vj%=k-D zm9GAtL%p)Ld$z=bP-61RXz?_`9a3*0JOGy+ApWyYWu0%;XFk#%+_v-|iQ)2jD`yOQ z1o6R098iwudcjydRk3L?+ElUg?m1v0|J%Oy6!vT#(|ny;Mx=`o*QqU;y|+*5@_a20 zZ^$%6#U9g6iT~d98jftG?JRSVNgtrp&|x@%NcHpfAKdSibQwLp zl+SU1gXSZ3RrhaC621C~qzk@VK0q3M+_~I{YdGE@F4Gh(0fggkFh`l7<{=R<%i@3T|8b1(C{am%d6BXjK1~`)JiWN!l(^tEmwpzbU8#?7NOk2xcq4+ z5-#q40$yDpsBZ|MJs16Oo5y}m)WMw=>9jk~Y5(&HFbo5fbgenw*-C2?(`T2aE77YwqzlCyFLK`u75dL7 z^iU8}=af=`UIkR>_2q$~Hw6pcoDvX*?*nf#9R}XjOHEVZ01H*!=Lx`lq(k}7C#`^M+KX!7#! z0Jj|=xl8pMym+9mkZhfcj+z3_HF&Y!z{7(U<;mrdp6dcOKCLlNky7j^ivYx&?(64v zxp8g|z_1AfY8{4T21w3YsGA`5axuR4W?=gsQ$QZF&SN2p z5JOM|Qc0opsR{xlUaT?m?q3?I7yeUIQ-Jp;?S*p<2`8Mmr^53$2we?&O2YXPD4x9e zji%BG5@!h0kCvCaI4S0%_Q}_7^=H(kb?7T~X z>q=`NE~R*om7tCcHU8=~6{bsf&9MZ|Ni(1AjqRHx+L1au`XA9R-}QcHe?JkCmkH9d z#^pvH&ZYLW27;N#&%>jsv-e{VFncC2a)d8;Ef-jQhU-0D?csrKBkKCHwO3Rxzj<@u zWi)PWBPVpJO$N`4e)-C&Aa3Z|HKEK=;yeUri)^Adc^6?XInIXeN=aZ*<=7|d$x0BI zLgb&(-X`OZ)4TQeBEYl`WYtpKc&@OfTbfZL6M~2n`Z*w7pE>&K@3rEXuTUtOI&K6A zULmeIQ~{FBf7XPby-qY&aI9OQs(>qguu|49O#EsEm=r>ZUeSY1@}C)5AJ2St6M7{G z6{%z!tEQU3j1+DMo9|S&Wg(<_XC*2yB>WrnUgHAD#SQXT|JXQ^=HOlKtI*#G%rw8R zm;f=Zo7Ojxkp7EU`5M=C2o91usP|sI1g@gxxIrLq^6%1h6)k{JM*So2m2u#IYQgb8 zAGyfH0zCmI`^RK!NDcD;xo#TW5S$kE0SHwSNT~|apo`{zak`2o?w`e1&`r$mw!tx~ zLm3_%1)L}tggWm7yUkxLAgTZN6$(X{bp`X*==ytLjB)iln3q7r;~LKGxFTjSlfHm| z0B{xM!QHgPH9xlvFcVIe_0Q5#6mTPuj!=wX48U80)a1GH8n*@k_PFnrRIyqaL=9@L zxSh*#2tCGYc6t^BF2C`ooK5r!sIu>ZY{M0xz zYVBneJf?)y5O-94BX(JS^(dDSB)*A2&&~YlZF<_WooH7RsebMCk4b^MW(3ntV%$y; zzG1TmTXP!8sna7TCNo#coD+Z6tSmmyH5IL?)AF!kadnV86s_*uzXLCi(q9lWZFuzG z!HWT~Vdb_4;9q}B`d#Cy`;RHL_c1S}QZ;m_t>PTj6>Wmutbz3SjCL*7oux{4C>FFg z>82F=;jI=JP8P$DPx0{i&A6f)zCW0h21n0;Zs?)yJH zrF)qfW6F0jGep+}dX#!Nt5$$!eS-QC?=o{ponqd>vV(jDkr}l!2N6tSVTw|!CtCAG ztUCINn256M3Hr}rm*4BHdIMe2JhiP9q=m4OrpUsd`5P5eg5jLH-@yjv-?^x2SI{rb zVoN9kbf9D4eGvj`HU@~YdV*f~#QRYl57kFmuv@!@$y7RqvXsHz6Fk@&Wtr;k@kO=C zBOOlDVI6u|%Z%}Hd1KmLCI|ee8q@3gW~=5~0-Ax&ugWF2yQqxDIcvXEF)pM%`kWqK z<-W%_UPqTbmQOXNYow{h=7}(f>*ws7!yP0nNR5#*t{-3UOz2+f2d^wBeEd6OP`@Zz z+GDR%MMbk}mb-e`5k@R73-~OWMNo-M2nr9y~t z9cQ7idc^sA_5E2wUAebe%Q{raQdeXacLzE5#x0^1nyg#&xEmp!Mzn`dgM=|&S)@Pe zF%4lmX!OP&i?}W}uJrZB+C11pgXI6ek)z7!ej-y!V3JG<`9PcPngCOGOBe*vozqy6 zL8~qGyHV_ImbMx#X{l|MF4dNg*WUe5$~KGfl;+S0*ty3ad zp#qUC9w!hGrY_w={aL(eLmnAVutZ39+mtuzTWRP0Qwu$6hat>^LIjh;No7$qg1&H= z$o;8>PnlLu`k+88jzrNCw}&nczQ7h|{1-pydF5XQvvscY8Hc7|m**+BRW;zt#L z9{Wddj{A3F6WC$w_zp@NR5>lJf7e^Cl!vdc{JHHRqI9uaeCT=V^9*A-4u%ep;lpw! zSsLBCZ4>NMb>IA*`&)A_)Li92seK)vR)$~`iL5F)()E#Lke*fKOr8#Z?PP)9cHMmI zu6YXAVU3XC!QS{_NVhzG7RKU=~Aw#D-rjy=mr0cTNdI*x0JQk=9B{K_<*g2PvetqkNpPdRaYoeogS4PVSm;e*P=+zjV;%E z?`b*J21vix^MCEHshd9NRe9zXZ!$%fX)@{-7>pK*pB>PI~vvWRj z*vKH1r$s{TO+Lh!_r7=BDj?2sIe?RrO<}emhOw(*)YVI zX_-u4ACB@ooASJi=Av+-k%IO+O6a%E_(21MEt9&z;mo z)AUc}4|@8|YedD;_0C#ZPEG1O;vrZ6mAON4AU5fx9Z_#~$N{wQaUIj>>^8Q0rDD9f z=%^TMo!w1&)H<*GqBOg{58jIR5vx9E#D5ysH5aykcwIm9e9p@8kaC=uwAHIQa@qVK z%teQnE2Pmc*TU=+)!!`;Kj9B4`wR!)rKEM&$?0m><(bu@wS$!Yqh0S4pXv4LeVf|-+LVZ)>Ls_*n&HFb zmrOnPk{=O?CCzYsOTFk;g zges*@70rwjH0Y17P=lG%aOC9eUyWhq!@&tR-x;{h#q6G5yAk4GYyIKpwv5bUz2pQb zS7o0GO=5qFYs31LkC}XQ3ypX#iGTo%o66Y8_> zsWf#dh|uI!H#}RuGq`V zNWVygvRwDp%u0a|PEar1Sw174V_@m)C;o7UMoq=taPRT9`vkRVPZT-Bj`xRDf+y9p ziZwzHq=@8%5S*cd%_|6EHVp=;(W;4L4i3sLej`qIJ;|<1zzx6FaYJj0!mqPA+xutle-%wd5tTZLaV?Wl+N#qfSlZ;CCg05WMA$ea zvdDTo8_?4|NDo7sDfJy&{^|c%J@()s4II*)n8oHAygBx^>uPr624NPshSuHM7)aEv=i#Ichpv~Mr6q+vxU=hMr+{nT;dnGZ@Xr0B%IJjWDSU z-d1CBbLx~lt?mOHC&{T?(%F<7p_#fzpJZtV%;g=F5#Bx-Zf! zMb-o`KRhs9J8g&5Q@@;+u9`FNLdQbgEDV?!s?#6KcWTkpb#Q$dRKG~J9T)ipg?c7S zVI2jEF-CrVhwCvmcBI{JFhylv_Q|h)Fcn}I$(g{>f1>iblf~qK`prnAlL)^W#3m;r zaR276lzI{5==FAm&iyYqcFb5(RrjQbg43LS+=#Zh9_)f3j3>t*VuNT#xJB!ChQ3Sh zT1>X2q>S{g`JwR1sSW+sW&>;xll!;dU}3WkLf_vl)erR0{>Mb!%~c@i7hG~I40V$B#**dXeb>LXLt)YLs9jl|0&pBbxb+$Um_Y_0H@Ac#G&aeJRtv^pX-U}me2I(-hb`&m3v<}`E5l_56tnW7>Uy&%l0cP*=xH~k5^P{eiSlq;|=h_eA$SJdK^{H>N04jan7FMJsVc4 z5h_Ukr-NHI2}9YjgZOF6KC#_#iGe2s3R}`uo7OHQ&8kNZO~yHf<^Q{_q%0<*08Yfr88 zp)ZR{paGN*7bIfyavPQ>s(AUn!h^}<5*E+}LteXov9Iu@DmKu|`{L{>qhl)cNfseq zS>rSlDW+*5FxPwbFrlsNsf3DJm+d zP+b@4Tr*)-X`KqpKE34Cpz19s#*Kfb0JF?Y`pS=Z|1`myi@M3lF8W`V!kd-%Jk-AR ze7qjS1vd|cG2M!<4q*dkcCA{Azi4%1F2Y%QWR;0w`0kMH>#9e-9?t|K=xOyKQQID! z5GIzSc;yYT+iFFAm+z%Oa`}0MtS?{e8sI#u3L~!h{qqWA3ZVq0G=RpA@J*Jf_W27V?)|4B zW-QRg;f3P~V?y6#8L=%UC=|L4gxJ$*`wn>_kagbTv=l=Kvx052P{p9^)pS`#CMxh~FOrNHWzK)1 z!l3)nFGsxwVEOQ#-lNe&!#}^dv()|r?k`tw-v7vgcXHM}dGy``ntx`S_!|nonvf)1 zNUKW=)?0j9r5d}}vWc7k(;-t31uC1}n!_<5KU|J?o6pnvimA|m|SuWZ14 zZ&S*nUXRCLOR+}b_CNatL-tmi@*sScp9|J2m=|r|6Il`Wcf;h_6#jjFD}|a*&@W@F zGxdCDr+PQ%Mo8F9g6hKR)Bf$p=AvuH`$>*{6fTtMXV{5u2<|qGOTT5Ux8Za8`9KY; zgyaLj!HCz{9JA&60LA^?+iUY&3TS_=RGa0-fN@D>Gzi|4tY#^&PeDNykZ*kq*7A{FVkDlzN#CamTH zF^2$v&KKzFW0MQ=^YM9tN;@c5zL$S{<)Z!EV({Dwr;66Ka*Lsbw7UU7v$A6@pr*%4m+uND4%7k()C zm1x%j+!?H6tQ(0~@O|~X4k+quZK4TZw8PP#yw;^|p%n;IFSwFdE*I6x3EoZndkq~J zgo*F_tLv@J`O`rCZ^8H_6zFem-9+mLlVX7q7yV-Fuy#Ssed)aj7H&TCeK9eytq*&w z00nz69cTz@%o`e1|1N~$p`dM4x1^Bf^yy?pcV~h zpJ>u`ou+DBzsG;FGqhDSM)G$oO#$ZqP=C$&NHoE2wPLh;sq_ud8ilR%MFT+=i5YCL zu%Cza=66W3MWDjW-p4IcQ1jdyRE4+R8O?xVJDiKO8vv;nPH@1o^5?k2e7|?qMc?g* zztO6{?Uz?qw>DiAhEUkRpi!^;ZCs4+)iWS#eYAt1R5OSZJl+sF_#QV3Xee*+* zTV$7B&G*Q+80ekq6|~gXEGAejv>43*pbNM)Qa#pzz4>A!Q1a-Qn2;L008i2Sy1-x8 z0if@f=5YDnRlA8F*BI@*SBb9{VRoZEpnW8o#k1poRmzmzwr)TLQ%y`mI2EYI_t>aE2 z-Ju(eWl1-DgSuRCn^E_U!$-WY0yTnm6}W*zPS)5}oMx zZWQM$Br6OF9ztxquR=&#=cz^o{mmxE9nudg3yEM8f$G0^A%pg`!9EF4jkq`{_+mF zXiLZMQ>L?8^+SFBVJ*47yVWg)?0aM^q^oD&dCw|Yj(yiZtlPF+>}r~?o*s@g9;%&5 z_ZC*PemZj7H90fKUMZV*rTepU01uM9&wfU-`Ayg;=gCa{;QHjl*|mVbdxpW&{eUxQ zMuB0gf-R$9%b9kr*ANNVGMsOYercV3+O#-v-{q~n*1|atH$I+aj<)q~xfoetF;5p_ zqE~UFWA{XES9H7ehss!1++Kt_Ar^)MJ3*OiO6Rb{K{tn8?8y2b9h+os)_Jb9NGuE53GM7xlPvRhZf zB9gqhWq+#wt=k_LWOx504|H>kV0_{Ar^J*yZtzEEmBkBNL4Tt_^mm5+M*A!dr@Nt5 zc*mYD-C|0%7r&(a5#_6iAJ{a|aZG(WsOj)wnB<~I_V!N_+n>clS`8n!y44q+6P;vRQeRyM72#Q8`Koi=146MB~rAI_E0vB_|(7e%h zuW_(Df#bILhFh(~Nim+xMG#N+799WlfH08xaWnDL;t#N_qKh8>RNQ!iVyvELS#HAYmQCTqjza`7*E_vP~TP2=Ow{s2!k5uX6l8M~ES zvEo#Ctl8@t3*l(;kZ*6eyN$gS+sH&DpZxbhU!u|R8Q#WzVEPKLqh_k0%Zm7gqY2v%!+ePGPGBv;WFJN2TK8CshvPIsB^~pu1 zCzR2V)m*-^Z#jcMd2m`@ME@}h$G-1Tx{mONzb{Mp>0WU?4diL~;@Zoi#YL^XN6XY_ znF`RvZ1=&6^2#ls+VZ6T2GhJuuRzP#uj+{;E~;!;;0-y{r)isppS#U#MYs7)>))Ib zU=)u=Y9;Jjq(OBHV|sEY?)_d+>#cY_uDZV?m~QGkG}ptlxn#*RpU>Ggr!g?x_(QTf z=%K{LSYn=lsPaa|sp{C(_Zox)kIw#!Gj#_ zFPUJBquqlX0jFQSwP!+|rim;hua|G}&-cS!<7Ootdr#cHcrV`Vhols_?>=3!5L^7( zqNUxdj9cr0Ff&D+(kG9w%3Y)&_S4U2)D7WNXYgwNU1l2qvG57Z^_w;92@^kc#$TWh zzJD=ZoI4$7$F?ctcUVg}>Y~aYv19r#qYDET&u{`W~6}$w9 zo%0Ag3CFAs!|gHrmDB>~Cu>mnsNUX@Ye9o-QTnCXATrsw$HLpB^_Z30%xW!|J>*u` zb$HRYZbg?UR&p=VV(vk%xj_zmo*@UcCbxOP8ZXDC{GVSfYu$Ejjo-RQENUr!F@fV` zM-K3+!}KY`kJP8{P!81@|Ds*?x*dj%+NFAj=kM$Ab9A>|B7_kIzij>%(@}HYjB|3X z`!$!f#T089LP2c~O7Ur_QP)`eGHT7mXDy~&-H8-k_pt6Z+)4}ozuJ4RfGE1AYji*{ zp`Zeyh$1SIL6j&_qJkiZFk}QpavE|*0YOkvqLOpYdB`d`>X0ReAqUAB{%U>R=YPNN z;+%_fd%TmG>E6A2@7h(Xs@8I|S8hH_dZL!=+DKF`xAI14Is@ikgVOtPQ;|P7i?7`O zV)Sh?4N612u4_?e(hr5pz7oVwWw62)!P!%4&RrjKGKV|v2 z<&R7}nW649Df@{nPgT`Y0eUf?Xo>mvMb6hVbtk&d?r_|OI38SFz`AwBaE-%^+RX#d*G+&#wesFeNl{Niw*Y>5gJ0zC& zSy5GX4eg0<*&{7%s>NCqgwpDw1&s!+l z|M9scKpYf*;Dc{(OcY8>FQ|2^WR{N&Z#L5#%s2`ab)8C&HdCf1Ha3W*i^`-c;IxfD zw$!oyxGSKgv7}9HLxi<3Gnct|g*Z`LVY$FDLD_jNGvI1`SDXKCNl3S-Sy~WpOj@kq z{`)T@&|LdPO;jYw>DW_#@32^}5c57IYIP4BQQBXWf}4Cdk-x=2d}%z)DJFy^t0~=Z zS>H%?rh9e9cND87bEo5OY`-FQu<;631|KIP?1eKhuAqDm0pA_9O44S=X6Pr`UW2|2>fvRJ= z3J&6l!2k6Dz(?m~cA)L;fB&Mfu~GNoA#|f)PRWxb5?_w9zGOok#3r8z?U3*RNWTB? z4`f2wb{x%q6{MD*NrB|@87}!5p+x*zk2Fi*yM2!G@xpn9Fdl=-KT}s=+#{#mUVqgo z<3C>(nQ>ggJvn|fv5QQtjZRIJ__Q&4lJ ziNA4#-i55-sH_ZCOmHiQepHwA;*S7`@$T_h%m`&Aa`&8P>+%(!!W-(56}-FK%D#}| zcVsU6`xmvZXpjj2nsj(h5H0&>=n{q2@8e9fC{#AC2qh-+^l04QPH)#QGX0Hf z?XD6(p4oCoSgN_AVe5*e{LJj(xjUjAzl)u!g|AtMEzhV}JKmT{szLKbhfCb})8drYo#D9K<{W@c}c6^7hyxTo3! zwhafk2V3{5Ua01+WE6!$X-@2PuvwpMU0QRLNhJGYo=MG6&jb0V-4bV1OUd4`o6a{B zw1j^j?6XW%=lo$WIB+*()rKmF?V2eN5oJt4(={K_pVugnZJn=vIh{?MU+%VDhEd(u zJUxf*%I#{U0QzBG!AG^8A;?r4f8S2Ei>a? z97X%BhWcZ2$CjeGg}u?+CaTXpNJf8DjonaF=3#Afe!XQ?7_m&Cp6({$=Q(#|@k=s= zRyLiI^DNtfQKmTcc}L3ksfQOo8tCeK~@^@0UsQQ$829??>~i z@q!LZC$6LZ!Zad7qC$6c*5+z_Bl3DD^JRIkI1 z`SXgO%v?Bgr$v>!J;U5A3wxw3F6Pi%9QR&{=lvJ6x=^FKda(!^HfV+7LmxO)oBy3bQc?Q~swffa1%96bY<cxaCH|(S%&FpoH&MS*#~vsv>Q>`fuW#W-gUuHECLcW*6SHqO zy0l_@BAAPscYxO4t>mY($eESa@#}=%W%(R5ZTy9s%U^9U6cQhBUec58(d^x;WRC@C z|Jg&sJiCd1{Z$lvroc9^U$Z}FbayVdAm30h}?;_244KNR_Y4%|&kAF!eN8cLFF z{hl&mN2jctHi~|pM*4#35A3U2owCtrx%)w8)>$mZx1KPKkZBL3LL24L|mfpKYCC3(pFkEkJc8xn_lQJoj5dkts#L#0s&xk zA7xCTwHze=nOwL1N%qPP;U)K&4{uF;#;wH;!ppQ?ab2Nc+Ei3N8drXa+-;q4aE*#B z=o$(7$J=M`M`XOwrDGAsv_}d?Zmp)bbibSW>lPj3E_N^?(rgG-_~R(ao;H7@e0FVO zEXVgpJ%JSGB+s%Nm)Kw5w($t5z_spGVD(`AJJreANj+}SSpu7)Kc>>(#iF16ps$#? zXfBkd8*QfNqISSqDU=wlk=xUKuV%Z5sYvoSEVRKit5-_G**tw{;2C*o4l zc^rFNMNkOD8ft%MBS2^3V9lmUhM*IkGl7V{XzECeIOH{{_qZ)Ht2u zscX>eNmj)O?pU+L%S?3Zv~6;7AxoDD4Y-G7{fI`_Ejfv9|T!TAm!WD1|^-B(-+c$jwghTwRfIhw=7h? z-u6YQ2iH|=)4f7(ER^fCuG(r9xU6(OD@D0_)NSS)b4%l-nwlHy>Uq}KaOGRJa^b?y zCr@mX1;0<*>5S|YxHi{1%AxIPq-L-k?4-}kwsy9>syox$1g|KVl}HrPCi;L{-i0RT z;^gKfFXMz({^~kb3VSE3^3HPHCge_^e>4^IPS!v$vIpx;v1iLn6(?^qrTb zBy;klO`5k)y?W$6zgv)BvK#;cYap`t;se zzEt6&)w{<%N>>Op+7*IE51ooN_-p|0zvK`7Dk?3|rXQsEXWmsC<0s7qccRQ5(F{GT zlPY}u*iL@x{!F)BOE(5w>F85o+5Mm0J{~6f%hFUbb8GePsj!Xjq{4qkKX0;0 zbHMXUH9V>9c~OX)OVGyrq#<^Uuqd)HZ-k7WbGp}G`<~^(!A3LoV9FQI0)|`ok z$+Gz{!(SY!4p8@u>3u+ZzC(*0o$^Y;!}WXLIiC>+vCW$o zw;$e5D>T1&+NH&KUi;8TgMJ<0?K6l{IrBC&(7_az!t2oK4?XDlrNZp})qYd@P`iKP zdMb}Jr4e+7y-{fdXO-Znm%SnT~KF z^?`n(p!})FhiN+9YdXC88Ey}^il($}Y^`|^`3-;9rs!=kVOW4F0V z<#yj`)$_FTBiY(qfUcc-wu;Sz%b3ylvFyxKkwOmOguJ2vWJlrP$EM3*(txi2h z)!W0sT!nI+ImCIrgJPtDH@kmB$`Te2q~7FKHm@Fr*zK9y6ul(-E-7|s9&DGrc34Dn zU>640USAwk*RH0Iuy|d5h8LAc4B~(thpmK%-aI?n#)_8r3ni&*FUQBP_}V#+RJw-c z(%8+Vx!9JxhC;V8{`s8mYeW%{05Ly8A~N$W^(RJm2AN&kkkuqTBL`!`EWz zn^_Iob(Pe??3E4|7hj9*vD1ueV?J08mL0OK93E(oPgw54LQr3|7tkdD^^L<2MfW2R zRF^}VqKbT%r()y~Q|h)Lc;GL!Bj~ZuoV{|GoSj+UOx9M0_w8mJeC%xa%^^gVLr{5R zGn;CbooY&(r_o;OU^maLpK>Mg*D%b=FX-jJbs;CJ9COM7uxoimeDKF8$5}=`ubo(H z9b=#^=s+*gdQ;~a_Z;jMpL<3VZ%x1j5`Ow)cC#QQ9;BE;BTZRk3+cF-^kRrHSGY$Y zI~|j>RJkj+C#Uz~MFS&(Bk!gOJlF*Ky--E>yYi*(Va`I)Tai>C*d8GY`!y#U_`Q6) z%IueQ`G|OrBoZ^sK~}lPEg5K0n9bu|VVr1bX7+`XBb`*yS6h}naR z6@+1jr$dY84z6P4HSvULH3w8_s9xc>?Q~0)111ZH%V46uB{7lWDnP*#Pd;)H#+V_) zh7~`Iya)0xcvEDM;m9CC={s1_{W)@rW2Z019LimR9Oeo)Vt!rWW6Iq1>q@Dp0s7cK}`r^qv~zt?$dlUWKPnz~BF)lMz7}0ky?J zz&^pbY`i||P1D5K36IV12olgWrCpRc|9l}hG?d;o^VF$R6XdgyHM(LmgOkYU$Kjs9 zG{M=!AVaFLY@A4+{rCQnATQ`A(O+g~4j)4%gkcWsI*bsSwF4DXj>CJpYlyplGEDG;*Y|6G0nnB0_D)JQblpDV!QgCeH>vN0DlO7nR+0$Ron8DQO7 z%rs#HRMr7oa3&jeA^=?eaiB{Jzh`6g^5rB#o&oa*R9dx+Zrhp z;1ywkQLNS%|AUAZ>R)>MpStVp+ z0aQ9o`d^3O?R-Mw5*9Q(-FJAp1tqQZViFkQt!N@ASR0b1c1vs_%-(F(qYRqWz!W)r z=&K$5fXfi~0_fFORFQkk>q0S)ZYJDAWQLHb1Q=p#&E>_}%I5EE+k3+6(G-Au>c5}c zLuC%~dD5dV*J=Er2rS1hO3=x-uql=p%n3u>N>2IjOlW%-x(Eqj9b%}T(2a#l$G`@Y zXbI) z6;i>qk6VJL>Ld89CcdAA@{ZgkXlN%-4-%t9y}}W7BEsrCkj>s>TId3*pZ|OZ3Mwf7 z^PQkOAlAE_{uFpPel;GyW)j4vJB(ly%8cG@C_)&2B9t8MF?B5#^_vBfiu5vQ)|5c^ zwFZj!W%^^p}Dj{HeyQe_E@J2EU_HqCacsWdblJXr+eTMi; zf3K0z)E;NEG^8m&K8g*}|RjK<(ORn1z;=Rn|-d@rCs^yr{v7_?%%4M&$40BOBdq0>Tj^+iiAOa_6by}Xa_pWz3$4=&K@*!s3HNRL-EBF4)HwBo4vn0}#IcY3| z+56hfjJqVK!hVT3fKLN+^`X+1gqfR z|NG(!Z%9+NeT4`64G%_--NQsAwWo4$8U$-q=-s>sL_1 z@Xys5{I8UXT!HoS_uid#B$_d1Fw(!bFnaFICJ%PtnZ(R50EE19#VAw$@6EpT#u9ge zaZ*tzwS_FT65MANP@2EDjb*Z(CR_)MF{d?QDa9hEDAGz~8-1gw#+s^91 zZQcI8ZEcPirO&B9&v;p_Z$QI;?|CPZ;Djrt{CSXKooxZtsTa`h!3skXC;l^}Gie4! zuan?!rvLS~WD^~=h-E*y8A2{yZ?>}%aE~sF#JSyXkw1Bc64}>^PBz}SlV5*JPc^Vn zJ&%Do(@)hSYb)U#UsNW>qo9z%P6H#V{=Cf9yGBQKTYGEbR-k!)y&wnIkY3WnMu78p zd!3+SGNn?P_KTLKH1F+5W6P7@^9It>Tpu@2(S;AcH;iNHt!|uf7;D25Jtu3`KOt{< zxC0LAJEujgR0bz9Wl}I@XOv14u@AF42XwQSr!T`-+hj@N~|&yVtLXpgE?Nd#;*2)yV%@HM_GG3{g&Lu)5O)RJDbn$>K`f;21Scr^ifymX&K-VykRozc@e#*Ba=SY1m2|N|o{Ws;@eK25ecgI@ z9BX^e^ISW#FXgGs%?o;aPfBi4zU`M#x6qK62Sr9< zl|vKc?r>5Qq#t9_ys=AY_hS~Qrq0VLhXa|#~JG1o`9XZHHWfO8fO5l1m=_o_-pP; zM}o69*R5Bty{>uZATi`yks16>urkEsk`bPjwISG>&ht|MM#{X#MU|Ub!A+$;^}ZJi zPSr{;GlUGe9HP-^JP%S%0g5PatcWtG=E)2z0~{``+*Qr0DS#S$rHVP$i#m_I6-)8y zn7fYk^`x9a7nBXIR{VID$Xt~^QSL6rbaPoLjU)onB#xM}LETrHF6&(|yb9PxN*=*k zjp_ly9DM@CuagR8n?sM^4Voh5zO}Ito16E#|9k=4%%aVT*=W*>D>7(TbcicgBd*o? z@lcZWtDW9_{dQ2^ymp}%I3C%-c0dJ%3YW>usiA-YotKNYpS(+b<0%Vtw!fyT7u~L^ z**dI@Ydjg&%~94H!Np}UqwI{{AH2_RQ;U9R+uR`pX{d@vAZC$~1;lm`fFYF}IYudu z6IxFRWWUa0DK4&@&|}$Aei5ed=J72IiJvMRT&@pjBka z1nrb&Mw9|n=_{X6F5Mw~{2&@V!L|FNe9?p1(ouxS409R%I^`Da{(71<%S9bC+uGz0 z&27}5f^M_$xcOR|9nF+ctMfT+*^q4~k%XlzdOqo!Hh!%hO?woZnwYU8O?7#Y;E2AYDprGN@&3=;Y3o8X?W`al7$5gLC2+`~>^L(#<6| zN+t@`d@`oqtc#`LV#qLWa16x0N4DZ5No5QJjDd{4*rYoAk{`d=R>5 z_{rCQe_}EUm=Jk7NZH?qZ4LOqKWB7JB@TxJs9!-_6j}SgTFnR}xcQ&UNz5aK-zuUz zYG7QbF8=4quGa|yL6h!2km*Uu5g6G2=Q#CKNjiMoZ~e|*bRGl0zC;S^=&B1USTn#y!?AD&O)%kXtt{-o0Y8#C8rq^R5 z?kWf$Fm$;~T->A%=j8R56;Ot)i-#c220|9hg=!(h-+&Y<#!k90j@3!<#!rgS1A{lY zW0wp5^bU&&RkYvb<8V-JXjy-;{-BLDxR_~MS&yz%BQ?L}N^a%K>oT$eHXU1|FHRP^ zJ9Dv>-M8$F{MTNOBxHf>n`;Qd>cA~^o>?EbXW@30$i7!#8*DWj=UgTeTzQ$Oq*~`| z-hg9`4vi3CujvSFm9epDWD6)JoKc#U{!CS%XXTnE)A0t6UI<^Ocv*@x0;jm2$$RKV zbUsgcccK-g=g_DYnh|$a{W>_R+=C8Nk-B|I4YeFWuQwdDOihW*zP7`mi`}oSLodWx zx0=YRlaKklUh#sasktaM?a@KUDWglxX+(z5ZWA^s?Izgy>E!)){N84Bar0IMXHBm_ z2@Qt(<^iLVVEF*u68Fs-xHDU)Vurs6*?n7`Uq z2!F$T6)Q1WWwpai`EtF>xI37)kIZ7)J=*~DI*IND z8wU-O5KYo>l*WR6&x(VjkD6_~Z{ykFcU!#?L;3elqB#rWb7@N&YM1N7n2ie~Qk1@w zRXI$Jh?CdOpJJLhp-}3>xO3eQDhVppK(vTZUDHE^UFL@JuEz-{T*{bgs%uS`s~>}2|huM ztgi(1txeMzMe3YSXD;Y{@#a?%Qh%w!TC7rKOD=t|t^rIspCopufY%IfiLnNEiUZwLL>y>6# zzg~9!b=09+%zjEMg++5?M${0C#p-_2048#Pu8Sjxat^@;cO5_{nnB)|543Uwx1@{5 z#&y4qE^Xeg^I4~&9*rx9&CQO%lK&aU*48%g= zO*Le`Wl9>KVr!lYX$Wmp#>{8&dd94v-WqXZ6*ar6(igmTHYcd4OLqIFz66jbW6#%L zIBHWMC*YO>#Y>N-oJ(`CjEYG#VEOCz6DttK&`-HVjM&>0| zGI@SmbNW_T{YaSR3~QULO>H9U>6H30lX1~=VHbM%1+AndjY1PzB(RfLRnfTKcWaC* zRAiMDb+s;Bth5W2Ed@a={=8DzyyUT=1A^2KBAQt&l~W#d6$3THkE)*XJL8ST7x1&f zPeWV2RX1-ndr+S?^n2}ZaGUfTiOqXXOP||44?Hq6RdVCtbk=oX(XP~JBlIDFgN`B- z7d$nU);5sJD6FGQ_q+FvCf=)~Be3i-KmSJ0>D|EE zg%2HKoI*B^3e}nwUSxr5`fukV733*8#^gQaX>P1iP#IE>ITr`mMe@gIJM51ZJ)G%a z7CMbi6mV0CsAOcNqpgXhntXNXp_SN+0tYkim?PW(g6O*SRk}D$wMEwcvIkrr9NJ!h z*_0K8CO8EQT)bg|e#S}IdUxiFnVMxt*@CHuJs1m_!} zYu2m7PdPJ@7KT$IP{~igTq}bnH=SPOBq7hzw%a*XPp5A8gX;MDtrwZ6mNV*T8#I|d zWFR>9ZY{J3F)P=eq>li)Vn;sWYcCNGXeBY7^5xIYgqpW2K0`)DlzN-8r&SY28JcC% zZ6D-*ox-wn7~l#91N&~U3fJ^qoqdg~t=t5hxJ+Ym`FGex%iVUcgSm(yp_VUwlP7WM zQogk4&4sMIz1&M!ZSjj`LvaId`?lv;j!W44B_->#65C8?7c=OKD;KsGbde_dxnzU> zw(i4s7db@1nd5-hhX07;3Zk5BTld^Mp|P2Uy3z_=!~v56G%S$-QNUO(MA{};&J|gNpjow z0D%XJWlBFpYXx&b9b??6 zhSNgfSY{O*9(Zz8_#eOo!Id~aA{%*gANrDZp?VI_^536%*7{NU@o4e@?KvNDwa~ve zQYeP)r1)4+seA^h)Qcc`#{92enNkJ^NDcoGs4jtyd9y022yu`8fKEPHP5Dz%*0_fV zGMJB=fAgPNKPLh7=n?kWHo)Q>kdHF)`S$}>@N^N`2-cxF1jcS7%%O+*1KdsPK);0^!Gg@M9*K&~swrbh7H_I4>E=m!|lSe)smX+lHjcK#^%U zj8`4Tn;5!ic`$x4_wENKp|qd0?+cC;em`_6(Np{#W38YMu>ytpGDki2D!{|=7`Bo# zgg!PP#zO8UJpG2`3X}dOpETMOXh@Un-E80`mBVGNg4j@(e=twvd?z7ulK~p+8F>9c zmp5po(K^GfyQa?0<@RpChp7eA2|yQW<2Pi%iwrv?_arR8ouo(DV9k2b44Ph=HeE_% zt+Edv!XY07<{;-iWLM~J-SVws4&yY;d-jzOs}l`9!!~d+U%Mh$l1QFxf(1n+ z)el!4`RY+Sd=)5gktEQ>n`0s~X)j>ZV}m9gNv%lMk_Cl_hcYpT4Rrj9^~Wv#1Ynir zx1IlZ8NxxOIw;A}#6F|ii6%tNYr{n;rq><7?eM9RgbdgkydOz3pR3a7GdsU{cjxnK z)xShWvg#8+#RE}{5AeqqLuu!pEUw&PHJBOQbOwTWrV%KsDMgs|cX^$+Ecl3#v_8sa z57q&AU1RKl06QQ_Wix;@&;-DGJ0?5dmEHpk&#-%J6iQJ~?>ekZ93kNq*$ch{nj?@X zQ#1oJ+p}2bPs2X+b!%%2Bqf?uQIL}+V3>xzFR;>Zqjs1HEY%L=3h{3M%Q7|X&QKj& z=%B0vNhXL@Q^22%NkRq&UBzSUA>$wT&7g}FBc`cPx}hX5FOLl$Ua*CA3VByT`EE$Y zohKu!9cGmc<5bOg_G&<$J}66;f`S4#eWlCofGqh_J}LroA~AVTzc64~e`PG2au9lE z?V!@2Si%Po>3ugvCiVm#qfQ1`tQ@*o*5=3E4PfVL!^*1Fz5@bKCnH&nk?AO9aoNFLD}uJaG*f2;RhVVpzjR3Bvgm!R`XEGcxeKC>2IcRIfgW{x%{> z^A|q=Oe;0*=f)D`ngX->%Hu*c5GXh&MNK<2+?THp^YKA^hR`Ao5nllMb_x#bCdTC7 z+5JTp_+a#+jrR`Xq}E;1_p!4DGI$uFBD_CpY4aq|VSRh^9^|t@ZG(m9+O3(B=~ar5 zYTF(2W1o2XQQLJp5h@hM1h0TRNd(czGgu*0JCI&FPI{H)TMBed0{!gpu-hSPF>T$` zs0Xn?hfm|O_9-GWYh@5tOK1e)2;`Pf$<2+8sr@=-ATDNJzWk>A)tds?%aE~uS}s)z zJ;JQuv6!XHJG}h}s|Ic=0doMS1Zoc|hVMbHNSq#a^duy?f-W0vz|l)%S#u6fuh8P3 zAK?+0h9X6F+zcC9T@Ml89`Zm}w`Y>0Vd;GUeA%5g?z`>b&RjMeHIE0jUmd^tG2h_! z^78UnSCReX7yY$Nq&2sg>wU#M3h1Cv*Y@ECAUbjOYbQJ)FxB>0_nJX>YDxHSK!h95 z7@!b|G%pf8_?RtJW^WsCj39pk6!gxcx`1PNy!r?!Ii7qaXAp1!pUhz3-@Z{COMqx0 ztaxR0A+_K%C=>-4kQ_BE1`nq9#!e%q*K9;M=y({vT~_(=I3rl{D|A_102DF8cM0IT zCv>3n1^}(I6iN=gW(o=lkoB!ifSfoh=$@gMRm6WHhpYBj1)}ja@f;NcNjy~S0`q{(gY}(LZA^+}^A@WWCFD`YXMIgv+BS z0oBSpojG9XhqH;6hO~MaM$zjB*5fLsfi^F}$OfF^=K5r?8$#_lXII|~SDX!c9ZLf92h zTqSe`l?)zCj3iVv1P+qZ7WcD86WW06$l2{Dl8b->y3Oa-PmmG(SyiH!dCVdnB{xLE z8680o)5dK+#G+k)0@C$sV3#(xosrQ&<_blNkZ^Y^YQ6_dLWuG@r|X{m_H{{JI0B%x z-2f`lzVSi^y@o$jRMc)^91Mnxul7!X~chAhIJ z_}N%M+E*T)D}rQTLn!-c=(q>Powv?6QAZ)ryXFc0M0>PwmD}OLozz()$PBMsdfpCu ztOCv;I9(S266*t56>YbDI-YrSB)?rqwq_<<#TpJ|1*L?Jp^U&ZikP}W_cOGR=YarC zUe5JS*ciweA%yn$8Lo851a6D#TxRi_h#TY`0Z`zyo$m#rfi>7vFgVy8Ae>>?m8Jyi z>%+a-l;wIp=2}3$uldb0=ePk!tVQ{U>K=A>N4y0z_irv66 ze5E~%YYvWntz6yOUxmi;PF#7#!>;PC%v>@L9@xV&z#l=y0DlF_6AyjLuG6vJAmhg1 z!EExUWm~}`$%1sd6f+GYLo6N(9 z*u~O0IKclf4}hi%&!BY&@Nz3EDuUR+EOS=L7=XHvIW3=tCg+k7!V31sfSy2JzDfU` zICx42o}2J;zj+bEnZxduLj}~^P2NO5-8>^As2zk+Cjh-g*{ngbdG3on)ZsKzr}u%w zQmWs6bvq{vBsJ_D$@E{o)N;TF?}$x!7~?V6HP5bmt{XST!J&lJv53ffVDL@gyX}D| zK8&?WcAhl=359y(U1(@%sHmj4MS~#B8rb6mGzO_(3m}=Y;e~Aby3?%SKEN84P*wx# zSK-b^9q?vpmndUk0-@vswo4U05_JuL5L9Om*k=K71b&gymVm9rSEQ@MECZgGRkm0P z{rC-KZ@_YRBkGsX-*QRV)!EE!5oAxQp#NT_O#B<1`mkoB*|f^H-qJ_?aMP-&Y8P02 zw0)#=1b98Lpt2AJhHQPVhsS(Kqb+Q$XB94ti&hlkM~Q>m;6B|8PVZ1msFq_YOW`%- z7%G_p>VZ3V??Uw903KSSX%{s4mw+XN^%Vh(6(AR}mO?>V5D#}wFDcpbqSPwcT`bEk z8T0++QHcZ18pPdfN=CVtf^oUlgy^!&&(9-ejLPa>uqG>DlX7*LLKOvo)-T(mDnJLV z@G#dE&giTDQ@Qn5VR^td3f|jJH}1>BpWzwE0iQOzhneOW3aY*dVq%@)kwFGy5OMxj zGw%sHz*xV0`GU1y9s}*4rh4ATs;Ye@mg8%bHJCNgMDRf$M+rC}=K}%iEOJD~ErH>G zBqPHQ&XryUQ3(`C-I9v{RRNl&vTsXO5T%5$X{H^Hg3is4!`raUk&2HVW)WZ`fi%Ek zT-^0#22?Lpi=yHrmwMcQEB))?^(71zwAUPq`}aZhfE1DpLzj~P=xG#ID%Sb_CL$sN z)CP}5kIaK&j0^1A%7Yza*jfl*1ZsU9kt>McUCzjF3!qhoLLm}R)d7)(;Bsv*6@Gou zcm1Gn6F?^;>3`J)j87A~NYwwT;LEm%ZCF1IWrRo}A*wS{>lxI-C$?mrg3Z1H6Ceco z8xHj1kMH0A3cHIndX^v7r_={(RSgZ6rK5k&Q=eX#9{simL+qX33Q|{WI%Ihe$71N@ z8A-SOsn+nGDmS-n9QuK@v?I7h6OkyyHv2e(ZYUziig$bl!boZ*7O~~WH)5mR4jka* z4jGqh6IgA5aAF_U`idNi8cGe^b4-2}%!H_$ed16W1&aR(1}~N0zl^ijA^A|roCIVh zD^NQ(P~lWuejFtQE9V(|k=Y=V+?Lu*LZDnzQ_~K-oe9CSmmxnB9u+l|@Q`d<$nX~| z`YlB8%Uu~iWl#zjw$oEWQ~U__rdXE=ecofADkJ;%vg+^yOkVi`wJ-p zdWn*Q@I}PHUvQ%r9C&v1haGZ!_@LxyS2Z)SmVS6$AP(sCQUNf(n9Ha{#E+_g-7*#Bl$Aj%Di@1pm+_@yn_oWfiF*O1V{dL zOKYBC=llq-@9_2+e?K_JcAEf5kcdXa4|jU1Ui>nE6SHH2CBmr2=NyL1;1?)BL7=n? zyPQ`JZ+HQoWoI(nyh)X}zv^(`JeSEj!jw`T8cM5G5wQe?yAbsK0BeY68QDa_u8cF9 zAe;liHl!R-$FVRNiF#`Lkyqvw>md>@mEw-ND}W7`=?FEWlXvP^uG?fs~<2Ha0YTRPo2B{>{fEBhBTI+ucbp4!Bb&# zEkU@h7*X7SF35KvOg*zNfHDVj>WtJ3uFGcn$Nj_64}u#i4wv0!Vf}P+>U=x;NsEYx z2>Hf3uC=9wTzL3zlMMo1iO7z{832b(Lizh+iH~3+H7fkuy|hI*)MCWCTMNmfyj#)({a448866Et|x=J`-ZO0FOYhg^is&D zGErA57L{!#yY8GC&!WjlB<-#Ht7PC?E7yRF@x!lcYY9ba-AsI=nxWBLEMK=KHI^UV z!1$G8D!HK{HVaJtPJ2(0sX9Cdlue5&B{C_ zKEpU^MbY5Jq`h}n*&xy178iFXyGLE?$!)%eVArn2S-x0hE9v(6C_lxes~ z1d_e$v&!?EcxRZ;-&dYL*~F4DkXzOlSe1}tp_R0a66RdM*kVvT_n=F6}RVGjQ3yN_jP{+E@M)}0xZsgAK!3+aGijjfo46W?*H zHwk>2xxl>lu7XL*y^F<9@JVZQUyrXiG7)me8|hm-fad`7rMTx~HQ54a zn=ZsRBKLv*NqGG_-)u_!#6qtEw!bSlsPJ1rydHL(khqXxmwNR>;HK?_NHAvnnkHIT za%8zA$wHdPKJdVEkvx~9uR>MMYtkLJ59OOR{=9#ZRo5oL^pjfya5g7_0WYMvpQA?o zcHAtPwq?a#3|ck)DN`%OafX}i>7yj0%dQF^?_N(`vCort?$o*#g&h#ty1EgR!TbiW>oALLUxs0`R7EQ+*83s$$%k6Ww z&aJMjKyvRhME~+XUxXVDGlp;3 zSlQ^9t5mMmH2)-C-vpG{kLFWz0s%K1tnwcP;}Qk}`Pov{mP(-tZK5vC_O#Ht3M)6o zMf%~cG2CRx(Z!od)^qILhpQ(=7{KBZDIJKdLRAD~vaYg&j;*xPVyRg=udrFIWocMQ zka}|}PI^!^8y(@IV|>5Vfyj5HWB5twIuFS5E#IgcqviM$#>xKx6jCDSPgO&$PwV2E zJX(jL`7Ya8ou;7FY37!U&$GNg6~FUUI(v<$-77Kh<=r*awYa2Mx^9P*>dw363xqCFwY@1Ybdx~Sv4`p zY)vWZksX2?aeXTG^Qpsv zmaLyl)DxA-^$<5$P68ZRQZJJP=EAy{E5EJC5`AG*1!6NL^pTKr8nUWp?718@u46__ zGCXnZ?R`=j5z7J5Kq!Rl=mxCvTKB5pR^9zdPyJRP&0Wa^+j&!>kyx zTPO7VaIKLQ7qSvkCA#8NiU$Lg_8(Q$gG48J+` zcZ$~?CH45}2Q<%GH)%H(LuI8!*Q*O(WSX4ICT?+ol(*EL3Dm?xMT-ek7RO{zYT4gE z+60I(zSFh1FY*`94JTz@4QEFuNN+XR6o#V1Ax^D(DR z^h3ov(6=2t))qbv#orLeLP1?G)E0|ahH7Bm`V2zI%}3;^52%>#@{5-*!L}i368Qn} zb0jh#bf^83BB>q{K#F(p1x8&);5q<2nTZ^lp}};*yf!~2C>t4Ll%akeafDU4kwf%? z`yk?##X|HIrv`BZYqEP7BDJmi6#@?hI5nXf+x^BKtStIS>i~mLh@|=aToN*#N!!|L zN@gX$=}6^DK_to&5goCAJRrGT>zOB?*h@jQtx0vBI4Z2YOuei^)2oEw9W~z*=LK*Z z&7cYYZ*IojLtwpA@w%?{PBflHeiZU+6wUDv6GOLuCAeIWcR}epfDa1%EO>yaE4;yfzI$$e;` z(sXnzjqti*I#iBbS-B2VF^v?z^PqBs;Zeni7W-r$J=z110Dp|P5rST|A``}tR|nn{ z%*qXC{-jnc*aVG%vXqr+s)p>bGq5DWH@|~nsV&Q%16UuL&FOhi^3dK?{j;gLnH{*I@zA4> z1pxr)t^>%v6t$(N35_h0<-;EXPAv0EEJ8%*lwk;t;MA(V`!85<9|3<8uow}!bUzFN zqO#Yo39)B35ItpxESv1m9T33dkD+D}crOfC9+X@!YV!ch0Ff(vZw=xuO(4x%xzbet za*_4`)BF>EJX;3`YY@}SddKe(z;OtBWVjBEmzNiE=;6x@904bu;Qo_H5HWy0!I|#N z@cA3i6(0{wl}^}8xeK5kC@m)!vJ^2>I+w-F$oMIo0O?*r6{?n4=p@TUzF;=|1#q%G z#P3iEpp}h0iHWIR8i*a z{Rs97^3ITmObmj!lBCDcL(o6iC!gf$>4{K+n$@}HA>nr*eSK|hqg`z$_z3|0Haej! znBYxZ4-5^p15~Ox>Q1En*jofnLMR3WbFvW+91FD>z@&F`_@tt=^r>A-f)lVfJ8PSQ znU9~Ly4jy5Dk}O7A{W2{_#Nmu-$7OkP?jn1euJ~iQejC0O2nVsfC6O-BKOi=SYnWi z8-m1H!>=@pKE2RZ*e{UvfWC}3knRCS!tbwW&XV7=p8gmZxC&{9rdO|BvXP7O{7)S$6;PGduIHw_l9N2h4BznHTcxj?B5S^g{@wQt-J?~r4ss|(sns(5e zi-Vw}`)s@PDDOf+hgsP|5oDOjN>H;1;NUU=)u|T&{Q@!G7d1qDVEWe6fPWN5HVDlX z&Y#r)O`Tk}hc>-b@ByPN>n^R(XJ+Zc{3Hvm^3IOgC=~{c;ZQ-T`e=msy{M zpvbP3tLM?@+oO;!%A1KL5sEg435KNe0HAi3;FBGvu~wgHxZ#hii`2u=YM_#|TO44d zAw@w09bxss8p7iUPBkDH4-OCZHQ7SvE9D{D#{pUh-7y%jVSRml(A~8MbxvR=@MH^U zs4Ch7&och-8iFvwv)O}jff?w>>`y#H@Co336b7erY)p4s9CcO+Mqcv~ipR>Kn!;OqL)F17%ZBgrp-?6QGC7A!#tY?zSrAg=_sXh=BXI+WXY0GPfZNpkWCn1gG09k`5*is1Ke6?Ob(xw|6s{TY7 z$_ixwRtI2K08N|G7G>`tgOa$Zn!UX}NMSrV zc!0_wMREprVUO;oDWz+{sU-zaC0Kqq3Zt*4!hZ@(T3jBG8dSVpSFQwW3Gm1U2qYsY zDKsdr5BQx%{e(l*{koBGOAMTVu=pK7#2FggTp<9D0w@b?tMK@{y9fXTvapajyCo!4 z2{;^*1O!yvX>HmPN-H3S--K)?qB0J7E~Ix$@C8CY3IKsKIPUAh41}l+0@%4Ilo8lo o6e{M{|No!=4|Bk@vU^A~-AI1LF?QO`O1oW0jx>lbVLT2oz#^d{p?EG#Ti7j(jhoBhP{Z_TDOuTrrE36LXK97 z^TUKnf~S&_{DYV(>j(FKXx)qAH+?{(sEIlr7(h6PWzJ5|zE!Ilz805DOIMw8y7I9~ zHWXv0#$62PgNN?oz+n$gZotynFR;>8DFTC4A%t@B5bAI?b_7uec@heDk>LN^W$18} z0TDqvdA3Vxmf%B?(M2HVGAO znS|00SA{^sREId>Rf8{{w#&0?Qjxt2@6fYUt;e0hE4X>@1!Fr;#RG^nC2Z2T}JT5TE1|91sv!2`lRJ*-F3_+`A63LMO#!6ZYsHQ~Wy z4n)$t+Bb*qh`kngo{syQQ0(3d9%Whyi<$lG;D4iU5zdFdy!9b|k%~^4Lc=We7Ono1 z-zsertYk>5C6fnwJn`WwK1fcuSh01?A*8~1EB&owA3SP-OevpgtQmLL$#Wo&aWjI@ zkdg}Gu5#*T?mGg79Mo?KAeh8crM!;h8AS-iCH#_5M^yJnU$a9y9G23U$;`O36sIgR z(n=F*4k5i=@Ea!4!>U&Y8Uu}~LKfihedVYJbbQF^(gegZoLEKCJZ3Fw*@gZC5_T){ zjJY^LJEv}`nxWO~2A5{`9EfVdbX7h4^en|bcn7Qx z4OM^`5E^^hF&)GynOBCz$DfX7ob(Zd%_qcIo$y%(3a59*5}l(ZD%@PBKbK6F!z9C8LOi)3iN<^Lq1(3yv0#%zRpwPc-fJf^T?zP!@!G*2+8fm*@-cSRhI8&% zd@K*HdQ8h|N0Gfn7RNMe3%ias&k0;MN!p)GMzL}l2&E@#z5h*j^J94rg!guh4Z(yU z#QueLJWu!(L(9rMN>?9ce6|5J?eQNOs0@sNQkN@33*28>yiwtOvqE_ z+I7N&&RIyT(;L!VJ}`FDyR`;DU2|+^Jt|)4msT z5U{kNAt=OOzHgu@{V4hi@1jE7{L-|0%`|)TZ1RNSVlgQWr85It$w%Fh8q%6rTAz=D z?%&~iA%zzbEO|T2(Zhplh{fpP2P`REOgTxYgg0IA!zb}lpT^`a$0?eJm|xrQ-(DOc zkBV;Bz9V5ivG_rXM&80Qz5)M@i{cEd=22b9bB?`-3<(Y`Nl(S@`Ce#;Q$u;%$DFM` z7mbL0mofW<c?AuN*d_s9=f^zUT)_ej#H30}gv9y8DOGg#*`U#%K ztNxm#KfkT#vb_~l{w)5uj`XI`aY+|(^TZJsq0}D*^X(_qYTCn3@xeu#8BkfyITKCQ z9*V)%M+2SKgbh`3t1QqzlIo}RB=f!br_5L$*3_G#>6^^8X{GD%$86ZI<)Cp1IG)n& zT73zUuF}Xbjm8@zYxIdo)Su#r=8Ux)8}U2$Zf(^|)bA0BUBF+IP&1EC#-L7I+jSOy zqvUXF$Gx4jHQcpAsB`@-WCPgt zJ2)HRQ*d^MZu^OXJNS9{Cw0*33XNo>*I_tbyF|7ZRg zy5XEclcJK;Z3eDxFN9zI=U(7IN z#e@fu8U)%s2ARhE46hRYms5wL>*pj|LkpOckoG*^b`6?sx~7ZXeO_wx`F^bOf~^d# zVqrgSoG#pEQ{oVgsVQ!c*UFR3gtj|g`}P*WnCb5CF|{tmoIfBf$%cnp)FsG%_&aW}t!JnZ z7CE2ec3$H2BznShB`cpPs%%EN*Z54}&%M383>GkV`%&}N?4gkw&-!75he^4CuC6W5 z8`MzJr7Mj4$eEQk2~Hmf+ISUk%tg&%iaW*X5T~K^@4YmS+I|f z<9Y7gYY#PHM+hre!-SI}RO3?jG`wKuVU^p?tS@SsEw`3ON3LqEjJ!#uwSMb2YI9C4 z*P7KQ1V~Rycc5FZubBEhtDNUsi23VHCo0YK)T3TgbCWT^aESv$REG$gQMeD&W{?)W z|EK1j71d)7N6C9OXQqgku^SMU0v=2+iFTie81YQj^ve`|)zUU1B)QjA6=kKTFbmNcMcQpTCi$i%wK$4~@_;_Fjz8 z>x-TbJl~4GozRB)Dx&`A)}_PMOy^o~bACR2I7dd$l1$GukyR$}bcG&j{dD73S;hwjKwoadI-^_ zI-+*2TKJeK8>;8KQC7D-fy62Kf$kS*br|{3-OCmfXX)P76UQpUNZi7%tLn5No8ex$ zaC`TmxcS=m_u2P7GGqhpr#^>Jz0Pv!AX@HC?5(mbZ@oHGMnKHk=>c5ZhJVj{u+x0Z zO6L~obFiB2_U#_I%wLMs=O2TvuR5d1dqZiZ&(=OxH0|G{^S_av#lqfKV>kAvqV=l7 zjsHgH`m}RvvqrA0NzkRJ&Kg;-|?{AH8b(oZ@MM{9#gi3cV5RYFp(tI zIZj7fZk&LrkBNz?Tp$LJrP8!Lw0h`1G4aT)nBF0~7_WJ%^L;4=tLHqZIvZWe5HpHA}B%lWI*Q^I_9dL;qF)o_@ zgM;H~XSoyle%wO!?E8+>wc5_~CIN?Id3Al$U5im{+l9SN%&@Ou_WTHkz#{lB$BUbX zXxjWN8bPBFnPQIjEqda>0#cn#d;gv2L2>6^42sX2wx3H_!|2>{jGKM=4C^U~!?DvY z#!Z4AsN$y^)Hz*_YRE0(PD@;09CVrun#B1r!zQ&r)Kx#%kf9|G$2Z$=@Yo=eL??BU zNtM>r&~Q&YZVkNLZay}Kdb5E`@ktq9wd2v7hJ0sNmHnONZVX-+tWU%9LiJ>>MG?Q< zNeg2Ps$R-m>%8;+w9%{1Zj6WO{W$e4EXaJo@hk|o#=GTu$PZv~v}MK3@pENX6(Q5a zEjJO(7F>V({B|Cw_7FXSdTjiXx%@kNefpaX%;Ap$mBjmh?w%VZ*oA|-A@Uh6Pfb$K+s6D!ctR@G6M8#q;E*)Qq7 zYC8GM|42*#HtE?#T;lBO?DLM_=#{{m-v~X@ej4RBYs4b%(O~-=4+sy3r#YM)bL9z< zD7Reft3VB+hFK3d5iCo^+m8G`yJlOlpSv?lx{2*MhE4QJn!d`utKTSZ#7vcIH*tx1 zx}cfD@1bmQ=ean`*#D-bxkwJz_0ed9m+DO3Yt$c=Kkid#ulayFubsIhWpmi>4;6vK zt?BW7aoh8CM=*`e!9@?QI*URcs+*1I8Gc(`HnYrPCY(9FTLMkyGPc;7M$ZoHuk@44 zo|vA{@1xZ4S^an0+rtPf+Bq-&V!n}xzvX1<-t$kUA)0HWh2&lEHx)nNrftDDG^F=% z=}Ox)FgvHV+G)1IB0&V;Uk8-l#`&e-=JoitgxgA~XN zKdx5z_GZ22zc|gFfe7q|aVsK1z&2Gfihp}`3N>J5RgvI zhdOU`<@wT&KF_-s6Bdk2bW4Lk0^3}aP9)!QU+(&1wAy-@h#F!l`TW0GlD2Cji82rT zCgD%t_?bpR8x6lZoBA$J52RkLReEgbK<%IO z!xyTDWu=y)Z;Lrjg|Eczj$)$Umgi%DEx+&;Bk#%N`LExs-U~OfsKQR--HzbVtB=?h zK&bV2uDqw*2&$hkTM(2uMI)#$u@oC1I08+(9lsd_mb&%CIE4`*Z*D6Y`>dqsJz<-d2GL`HRaCTb>@?rftQbqz=(*KV z<Uo zy3&MBK^)1frdccnb#v0P2K*g)XA;OjY#gat8<8X9ceYMi%5X5IYx2cz?56Lyo_ZVu zA|#5{cdea(txMs4n`B*m{jGmj=YUtqcI{!%>vP1EUmBA)wm;y6eOuFY8nQIvXz+;U ztFsN{f@L4N#*Drw`N*SUBS0^2bPOWZsa+aznO$AavSrK~bg}m{H$ub~whKywi5u#U z-V=uTCJ>=R!SIhdgk*!6Y!}~}E7Dkh6^oaUxLkW|MiA3?Oo$%lCJQ>lu#AFV&3CV7 zVhRZf7UCht3J2c`YjSkbn~el=TrysglqaKa>4SII?z`oCyGIAp?~zfrVtq0o3>x6Y zCBA{^E>V0n-ecsn-x6@HE_3*^#<<+{HOhT;Kys^Qv@(gY#Rr7b5SD)n<6q)bb>hWA zWB;nEF;|(KfRieA;;E$&KK*jRhNEt_5*=3J5WF5}xXcagOY5}o#OyV{8~oy@pdPh= z>;@i{nnq&iF{Ul}w_CyPWLZc6CW%8+Azw|ena5kDlLsv_?1qPE6%Q?>E(}kRC4)`qgU+?63p<69w~}ZwX6NX(a8W zT_+1+pJ>{Z%*2acOap_p!kb(?SU6&_2t)K`7iuV(QeapSedqamn znAvJEs3uy*PA4QpSG7AL+Ofdc5kNj<)Mrz!Qr=u_NXvUJQxO$h$%O!4j+yJ77)dx)pg~zZ!a6m~o`%_X; znS>RD$qR+SXL+=GU4iY)JC|;gd9a@wRmShKyqRhiY`R07I|m!=5RN2Nsd6|Y7AUGTnH*vLfHo-6fYL(l20vIM7s^tUr` z+1H0L>0{W;8wx)aLrMhGE5>PChXD|8&DhZ;c7QQRY}YRrUvPM^{d*mG7~*#Jl^43( z%Mc3LbIJaIvVmYTz6&=?pRpTGbJ+L&LU*njS!uXp!hXI0w!qh?k9J99eYsNn#{XzUc?q~{yMY_!o%a~k+C77suKI7a^&JoB0BmuP%4+bK(DFY zcd6a!idigKBK{z%7~}}35(6MOy{vwl0Qn$IyR>zUM9sShq2AOm$%l^|08@Zu4Iz2n zoov<3t4r^7eRcnBJ}FY0aRCs-tpCBFo?r)oGbpOU3gM%i>p$$v@A2_R25P?!;&^*@y4l43I^2wmuT+)r}o)><^9m1W<@7v<^pg&v6Yso*$s-(- zx2oQtnm4My%xz-rV3}OZOfqM4)Ef1g10Op&AqN!JN^QohDiJ8@mrK>m$h(kU=U|&$ z)9I7vyqv*GNDfHQ)#>l=*B7V1bCC|BLm+PhSQfc+ws(e^GP;Kb;eis$Y5qM75#z=d z7wk8vpD`iEBGLr)&nb!+%e%SMIW5XWByW*SdHl{=PD71;QWI`c-8#e|1Fg4t#aD3T zt%NG^<--8JwQ9$q&U@^FU5C%vzdx}cuL(n8tk8O%r@ zQ4iMZd9;-xU*dhN8Og8l+3Th!AY+ck^K%)1W8I(FYh_>gH-xSG?`*G@zduy;DpS(h zGDUUDUV7}d6FAS*Q}!P%Fn% zN{uY4Q^xNod?1pPDepOH=PH1iZ9}1RfaL<_8}>+;`~gIFup@#PaCG7*##gA9H8yX) z3HjZpa@MvqA&1=TBvQRKGWTy^bEK)|1 z;DD$P^yIp*uXd}0nFJ=nOf~q}yK29G|MuFN@;LjuE-`5$&rO{lDihJ!;JGzbxg*hT z#VpJ))c+kB+T7pYPd&eWgMgVG#)E@T()MdEM72C4UD&#Kz9A2wLBZfI)Rj3P)MLfk zotGrI@|7wA&3{47>PFQrrlA0LD|=o@w#`~-HJCwxK(;L(`0X@Dl{Mp}8;VkMtJ4Gs z-st+BFYx(TEa3d8`|?GR*2kXYJ1QjHPDAmCT{5yIY++a|q=en(Z1!WtqVPV7u`MvZ z*=-HXq=HmV=1SDEe0e;79Va~0_EHWTNf<{(AQH!}F11zjC@8~Ok}UHAq%t}>x)>R} zf_d-dctCPhX=Kj6-2l^Q?7!E2eRbTr%fbf0q?G3-cm&L9;1_>ioE$`yu;zc0MP{c_$`fy`QW!%W$HU7`o@8;v2 za@hhn{_KZMjceY!5?})H3JMy|iV?O|fG7fp`H-0`LSY^-Rld3z?9L*P#7m7DJulA= znZ%tgK#1niAAwwMVNbtIk9f}uBnh#C0>>$m%lYe9DfbHInq3cIK)X65u91v)__cZd zG4YqSPWD2lQ2Z<;EJ-C5nhx};xRh<6hXDbr0mMId_N!;pwT|s`J@HIlYu^_q01*U5 zQ++YZd+9Y0hy2fvw#+VOntTpECfvWtGc0pF13q2mVz*PE^>^r@PEb-XV zH@%}OZ^~XxU?YG;(`0{o>E)Ht1PaJxwY40EtQmRG+Drnz4uHI|-`1ChQDgxi_nJ{f zH(U0n0tQEI{p1^1$rb?1Fb@Z79u@Kz=vKZfDZ0S$YX;)aY=cMRcEbix(D2quHm-O& zmHXh+?1<5JDumbej3kgw2-y60|Bh-n&o&%_!X)i5Q3*DH(_9mS!9*L7DWojF(MUYV z0y&iLWetWsqXC=W*p2;;`tLb$C<>z9I0l>tR1NK>J79tG=3_^htpW7`%F+b@pag}< zAOZq^rz|IT9`uz^U9W%V_-)%3X5)Xg% z!ua0+PiE0z7LPZ^$r~zCuRp}k0s?=yIZ5qbyL=}HUghYGg8Q#qLnLfqxOE23&4+KF ztN#-QeJGTZ@4s5x?Yo@__cv;zfJx$+&zWSy_(k=bTBh%3!F-0YBtd0U|J$iPXAVFm z5Z}fbth4r3-kF%E*t$)^k688n@9+f))Ode!tQ7eGpU7W`y?t78@WX$Rb)gisc^EvE zTih;w^|@lbJGX;@&v>5coYkSWRg?O_KuqGZ9kUR_k{(Em-Nb4BofAk4Owt4^Vm`_I zl?8E%?4PqI4N$ol6xpgvwlWn}uKbzMNZpVN5+di$7t;M`kme zHbAQ#=!@@#Ki^917lq9&p1$Mv@*`2E+SO95z=EE}bp_*xoxk+lUtVa<84m=?3{ROo z!8ZN*y(Pf!pn{qkG1)%cQk^z06k);0$;3>W(U;0UDZ*d@!)81%@k2{(2qD7S7t(|* zDwAM>#uSBu6ZR`Vvu=)ziZi+1^4F~0opB#F3c5Vh>w9Y`#$ZA^tQ5I~9{i>$9G@t; zNAR4Zv*XKjFUoxRqn!4~u2j=?k%IBTn5C$fq&=vRw&pua?wUcW+@hNzKA-@kH#;&N zJ&|E3phk2v7^0-85O@G}zpQm&po4MNSyYutupp|#vmx`ZM%9wJ%sBW?^pwL_?tYt8 z^Wq%su<-s1Wv58V@ZD^PJ;;W|7O7sx4f}x1t=htz0K!r{vy^p|`I0i1&NBm`vf}}h zgeZF17hl>$-slSV+@0fjW7yo42hR&R>H66>EG25=Va2uzc*rXt)0=bL2756tEsHD? z0+PZ;Yv;aN+ZXg)2a4DAZh{hTA^#)y6ZVJ*a~;i;p$RmfL2aDqvC;WF{K^O)dn@_y z^Q$5gR;x$2u|wsC3^7)5m80;$EM)f6%X&($!Ve6cSB2Qy^=B4ulmkV7s3SB$Mj5y& zR+bWr9FB36_J7S;19>PVW>wx1Wy;QdzBXZ9bkkuD4Mens7~b+Wg~umz@WaN>f2`Zb z?8nbK={>X$t%{-)lb1UmBT@ohf1Ps*ccXnd>wPCow^F;=V!@h!ME1&b%%Tu=Jm-hW z1C>c+HUc;i@W#^3w`a(%_hR^O)g=z-hKZMODUZ`1{r)zARsu{2pt4>t*bMxnRB{(@ zVT3&Q%d?NGi^SCLco)v)olnPa!+NH=zEVaRL8_M{Z}9MBhBv-;+n3(VTYDPa+&LPb zJd#zrB$ktX-Z7FLlc^h(4H0-*Q-rY;oU z%)1Egv#9sL3qHP2ATw7LoA{Z(jGo9IbW+8qgX(?%phz5RX~mX7ww&0Zg8_QW2~tDJ zO5fZqvJAP@mKh}OFWy?wzRNwV#1;c%3n1${l;NDwnsW~3o(i>X7IOUEDnEAl9x=|Hhvl@VGgSVAnn745BB9>SI>7^ zdwGVjADmnq46}*Yd~3N281H|N^RGMFje3e*Qa|EL5OsA0WJUlK>z%P+ow zBDS8cGX3AGGY)T0#ag)09D69V} zF^4ds&T01Lvu8g_^{X-8jbVVmHyQc-oO{jhT}q1N5!3d|7O?u5jRm;6Gr-xovH`%* z(eDi;4f{XE5tV$WySp3I&4~KK<_yx0YjtzJ8<9*-!KN4|#u$iI7%&33tT+&9`Ji~d z{Xnc?6LXlb1j6G_Ev0P#eN^47=QOA}jxUXhq0HS51;`6;1rXz$-?>4718G8RfhUWA#En;&kRvYv zmY4?Q0Vs1_O-UhJlYp!PwHC`d3g85Pz?dnVbA49Rfs=9YA(sFCUKh1l2#B55qbd9M zZmR?7DJcnZ`XAbrAmR>bt{K{O+ry`)4Nr-meS&mSY=Ek?J>7KL(7InN6hE521Hj&bPsTQXs7W zBhK~QX^{|MJyPC|(8#K)Du7ZM|9gzcH3&L$4i$OaNC}dt*s|71(l(coY%h~#HGJpy z;(ilmQSG09Z5v2n-wFW_+SB`v_0VR`E%Vy_jsXY4OmNz-Bb-Rjh543i`z?~k-GX90 zJW_j)REmtpAxRcIIXYyrk(X-`2npak!Rk~!+|8>wcB zk>Zi2MNP!EH(Y1wmmlf5@etdby4x4}!R7}msZHUM1)Sbovc>h1)Gf5sY5 zS2F=5JO=>6a^u|~Ab8GAdyk=ARgTkQYqv44R)qTQGY$9x@y(ldze?|wN2}+`&Zgt6 z0sMv!NbHH3bx$C+o`tb7YnKegW4PT|F&+mYw{=PiuP)e5WdzJFSY4f)35vj?)RyQp z=SX0c@x|mhJr_1fJ2zWfTY$K`+0Z!Gr^7*ZKZ}F{FTArH7}sigk0$_B58Pu9Igpd` z?j#joDCc2z>*J8%^Ygzv9H)rtHW3wdm=NT(VCDn1R2Q*8Q*eaD;{;}+O9^OwBD1K^ z_RLq8wS@?HWM$CU(No@|`!oa{5iBhv7|7pj%Mch#udis(P)^Oyxa^lYru-`bl zB3XRUZ}jYiA4ZZ;<_brj0c)R(Rk|J+s({*5oq1QsFsVk4s_pxQxMVdQ{xZ;WVUVwxgN$#D*w0R$NHUg~(W*g^dAe&ak6W)jR=*8mq>LQ?5J1{Ge}|1O zxrCuWvWggDq>D=vu}!vTJNjmXP;{my+y4B#&^$#FqMlX6DZo>dyei$~l0gXt;Y9@ezv$%7G6TK{{~TYA%`)1u1Rl9B7ekAvj2fghaWkdLSZr z57LG!&#kPZ9D<1H`YK`qWnWZB9MyA`r-`uTZ0**oeJh>M_VP0 zzQ3v5#?jlejp8=n+mgrMDil%f6@ubOfi^>e(i2+oPbma4X*KZI9o{*bs(^4l1&Ppn zl2v}WsY&|16YzW%MuUfO8z!-`WRO~52p_OFdYRox=%@}rFUHq{@~U&yE02gK^CIbU zT%V=(eFE-&OH#+DofXAK{$^@MmJr)*TLDtL?+v!`2K;jCTe^{;jlf#UYr88z`3>K5 z^zO=jjhLJM?aPibb=@4_g+KI+Ydn0ae!Zfl&zI#KGcJ!yJ091OtO>cDAfA6xJ>BmY zD7kQ7D$OJ*{0eMSQsUPk)t=6Xiclnhc=|;zou_=~4XuV&C|=XMo(o<6{&#ZCI#;|a z5+;)A#50<$cLfRqcJ7_-d7$?W(J2~BLy$WgkcYWni;4(|l`(0d)@e2dZ+w9f5tMdM z!{GxfDXV$`K_4Qs-$1V8@@dDrI@Ai@+2rNBp7D3qxp%Lx__jT;tDN?KeturjxP2eW zwE{F3K3?9=P4H!Zif-hWz3a^2vAP%$3mg$AsbXH@)Lyb;ki2p5aYWi#C!Q5IK^~D~!Kvw0rPcL3ctk_U7BTjTS+;1LK5_s(CsjWH=^?K$*?NhzC zp$|TBp*`$*5N{ul)bIlwp!H#;85g})J_1}%Ai0*l{Bo068wna^FqS*YOdbEg}w`7S%ND=?lg{DeQ2 zzh?Kjh-M>~f;W^W_wfWIUh$>VZe(_8CPo*kxjLbV{049i8J(fLj#`d((RIR>^2A_* zu)g2j=`>jhqGPXfgUtwoi5; zF9ILOxMnqu({#k2g{%hBGMyTTi21=JY1n_fPAF6BeoC(;QJMxNvk|>%)>SJoc%STe zoB{Ekn3Of<0~JE?X*Ya?i2gTrWAeFb4(QS#;?4?=_;~`PHDGy!;?Xa-Bf=s^qH#qX zu!yp^6lgNdjx)7Cb#7L(nIH5KB=2PDyapQ_A@rb2j+ z?ff}7ozc^eVo4L0`5nbW4rTHb_kQ;Lr82d7UZ-W}=vFvr&$CG!|Jzn*qo zos&pTW6XP0Rz9Fc#CsU&AnPl|2<3Li1O+gSqMf+Sx ze=B$)VVBc@-a&eEL372SL!9WyE1fX^TdI6y02J9z^E!YmgA*!zMe{j>+2}USX%vaP z--MUz`AyjyhaV#kAIEMUHf~aD%H_f_?-{wc4YW%j1ArRtfnuthwSh6??)jP;OHRg$ zsmk3+y~(hf7k*o|JFS%K`$L1Su3=GrzaA-2#Fc#UlEn52BBBQ>04? zJs)fmUJ%O=QoeZ4$~+nXm8gKMtBTF2U3@nmCrtUXzi6^r_ycQL>YG*UhcseY3hC!u zr}8t_Za}Z(<>f^tkpaQWbD^yBbCNc~nz`UQ!KVW!*7_&bht-HDf8La3MD`-6GP)?C z84n)kzUtLC9DkYFHNxy%HNpBqk@r3Spe*I_ac`xe6TE7QK#T3^D0)oB(x7h`x-`Qf zCVz}vXv(0msL~o|qzYU>+J~daAA_|8qge06nIwr8SQbYPblcxQ@uutth=zZf-@s3`LN(=EhUFhk}G#20&-++v%& zKuTp*D~2t3$Or+fJfX>nCXk)6%4Q!po@8lKsq*tnqz6%#5X&r< zjX=joH;R(p23fTtfeJsOWBMn%#kaNbca_+ocoSlGgPfx~Ej)N~d96y!>3NrmUpl++ z1v`Ibw~DWX?#Ru;ohyO7<(dtT8 zM7)FQo3C2_1^lDoV!t1mBvWMv6TEh~6fesUk5V{OCVho1>a>>3n2q~6_vu0t!Y1AA z1UB=$y@HpjxsX-t2O7nDxYe7J5FGWq`t`Bqhk&d{Y=oQV4R8K?Jcu*<$BMVO8W%4) zCdC*!V1MO2=IflPHm967Tr{*(WzfHk)N`yLpn1<`|7363`RfWD6#fYK1=Uw2G~kYt zRnedpQ94h%u<-Mu<<9c6p07zg<0(N`*m5BfV97x9qv5k>p%HQQYae0(uZ@a;E@C4> zXcHOcbM~AU@xEwy8yRC4Pm?YB_>EexPG;)FY^mxZ5h#H}Y z0KTS#pC>JM1{$hg#kCPMw+2DrG4=Mj`&BJBmQpra+=vQwT%WHBt;qQ{s{JYZ6AMbr zLjlLfSir~qEJ)+vyl%fLYEl7$My{;UIT^$>&1vdd=!fnJW{p5txda+s%%%QW6N-fZ zX~hfEa@9vD(E(I*O(>k`I6ZC?;pO<&s~jmB!P;=Re%+{~o31=%+1;OYw3!s7|G}h6 zW<9Y?#h})vxTFReAcrs_e1d>x_ya*B;Mi!aITDTO;$5XT#~;tw#06Z$4TIN{EHHM@ ze4s(MGozF;oto3=r$Hd@Ad~Lt8oEZH_#2enrr-m$R-xqm(b8q95M;{&8j?Lcu7e5!TZs` zSiO87#o7@n>3g(!dN__RN4iDw)qa2Gt2>j3O>i%)5O}`)KFgoWfL_DvGSm^Ungtrk z0fPoNV_LX@^3)kBS>AH?he6oNwB6OM!RxY%TEoUN(lq6^SN>FuZP@vM4-ms#6pp~e z0X?lkTQ1Dd&=8~p0Sj<_K$bu7E|(UW3ESFhJhxtEE!cni_RZ9~+qm2?FR8-|T z(EF+B2KIpWR%<(o|L>Ng&B;;F9RPTPg`~J1(j0FSlNo|nN|X5z7{49RRC?l1s3ks2 zjSNhib&i=RKyD7eFR5H$2i3Eu`DuD4I&&y#xRPrnHz%lXD+bvhH(IPk)F@~{T zWhmgdru)Y;!+AKoFPlOwj|D=G11h-&#(L{$&+r06ScLy3& zy#S!jfboG8lkM?KU(baOCYd?Ggy5C$(Y%#~OAerq@EWP;fSsoJGfFFDJtuBMjO2^F z8zu0gVwK=;6ok;#GAfO<86a-31wJ{ybFQohgcG3TRW3l`Zeqx3c)86Z9!g5>;6)lv zzMBqknCePH=;`;lItb7!F*YX{9PEUS4h5)0?68`ZjI#7i1WQmU4yrLv`lcee7v8V2 z7i3<)>{o6b3hLj&Raktgyi{VLqCP|KF$9l85z=exWe7qJ$ZF3PgxI zOqhN}fo6u_9U9`HC1J#uUN-U^#rG6Pd_q@k$h#>|M^D~=4rl3;udbxhS|+!TN8Ma} z)||Cxed7znu;X#z9+>Rvm~%DbCU4jImH?zFB?!B7N+sqH@XjfysRXvw6p7Dqdt3$c zOrsWXvcg-4HSsH}5i$V-Vi3i24n8-?!ZSW5#jGubchz?$idof93Omfn{49W6iLL zJ28Wo1qYfY&(<|ylUS~!itXOf{S6P0SRhJI?e6>2)Aon1Gn>fsmxkT}UAf!gXfI_rz|2cGe=Ut+=Hk>b}&^qp8&; z%s?NQUL|1+YW)GEkP&hEJ(%2*16J)HhuRa8Tycdh9Hz~HGbCW8%-T6YZx%ZO5s!#{ zobZA_jVT^+>fzxT!pL|l&f|fbP3Vuu0^>Lh8@u=icCKGhc{890#^7y$E{!l%bqd7j z<5yq5Fp%ErA#E=zN3FShMa9w!M7qN0N^yEnz$NGjtxiHsa&16xP*IINTrGm1mfR5U z_zbLiw>ba3o`Fr3TKxo(s&!dv1WzC(jAg9jU}KtxHHrPm}NI06$(YQ>Y-9f?zGQz(hSeHR92|8OZM9SXSHIO0-r9 zz#PUSpQy4V2Evar0-Tnz8a-bD>d}otu$PI`vm`O?IH1$5QvTK+#=QiE5rj*pn|YQ9 zLj#GJOzQ)o$&xJK5hA!XB5&igv;sC}&iCG|Jrs{S$J<<4nJASS^!5Qr^IFZhjSJEI z4d{e3;39yUQ@rQ_ zUfuF8R5+!H|DGZ^QwL7eJb>DR0jzU;dOr2u=!5U<0XyJmeUxBH&=)j8+k7_#B~l{& zNgHTZa%$YF1u6Z{kgs|M$qm4S=%5j*M?5$LWCO#)N+~!Y;3KiQDU^M{>ADPYXXBSI zV=S%JSf#yp-`;y(X?pMP5gWNXI3QR2aRp>ZHv8AbdR1@~=(4$yn{)TRt(%tC2!MN` zcq;GHHXVWQ>%-v=mIoR0Bm^LsV{rKB{zhpvaPL<0QGz<)=mYJ)@ui(xFU4L}gQmGd zU^$x7zF`DI7=v}$`1u+1bpl;{0DvPfJiJQr;%|D>6|K*D^Zq+8iQG@pUQXFB$5P{) z6?AtNYyQgmT-ux#4(~00iN_rJdilk!-Sm6Q!9s?peZu`eMazkDtPNP0<_2K09|K7~ zfdSDoZ}bT}-GNbJ;S($WJNiD0lKyAx2KQjkOqNeYYVprs6xOD;`q0YR`327%#*5Dp z-nQOy7L8|>IRghtxLwcFI~HAlSA-!&B{_9p(a9>B=0hO_ta>ZL`Rf2TJw@Yu zg_;BnXjKWMh`y~G>L#Cm_LJ8b!~kx(bHnh_uuN_pJgknHChDf)RUuk}Xdoh}5%sE~1G0%S4AY-}T)~A1$00CXM2m%dVq)cKo z{kXs}j;6+Bom{z~ByQ1hHu&2tWrRUrBVWr=iKq{#8=8+6CCUuy=Blj|)9mo|67<-i zK^Sgl&jNQl3m15&ay-svrjy@Cv1jCd6Xy*^&6^j3{?dg!UA%HId*;e6@<iY zqPjV3MX~|;$`OBBVd#1%kdp3w#s#np!60D)#*BH10l<=aooaPU+Pm`K`O(U6&-&0J zr|ZIBn~+z#Rs10XKL}Ib+@#*P`SOzu%w{sz($u#7rz(EPBtA{G#ZWxtVd(I+HLpd6 z-F^a#6gYVG3v-Ihw*y0Y=7jRVNgIE5M(csD*VG65`=(mJbnoR+;XsVS_Xf{!`hVn$ z>ca&!-8mvm-(L;DkA%A(p1fdf9`hzUdo2O|8DV8EnWQHpvZ_VTqwm1hv0o{oXY@+a z+=z2ewBtjEt9>{p4r44hx=vaBMXcb*CAwm}t)lbh3xxJB;Ox1f`=~Xfh2-Q+E(*SZ zZb+x9#Aa}0OYa5fcBym_bz%H7Ucp9})tw2^kWmdX)z5grp?%qAa6+$&qxjH}gCe;b zZT?F8&v5e+Vgj01Q%48$ zEU;Bt7&pV)gW%}8V@}?IyoEWD2l_eWNM0#g_PhWczMywjPjCG@2zs+~NOMR|^jeZ(p}?Nfx` z@YyQlTpaJ56r}ZhM+4mN+!P7UxY4?=-UY4xS&|r625b_!0K4Ns?!{a1pOugg%qk17 zwZduKiYaepD3<*iD0l3Es+m?b+yLFT^^dbr4jkVMg&_7i*3O~Mi|^JB^dQP`%&Fwc z1=&8h`vp>;-#%ZixKxEbF<75Nf!F)Px^t}b1fD#T(6e}kynU;Me5|Qp3+zvv2eJW}*kF;zCkP*gwMWF*K_krRP}KSw*LdK&!uiA$6rGCI;olh8(@k486- z?I5y6R%EYa9OD>SQTA45*&HJwB9hsV3PqH3e~y0l>viA%JkNbU&wrfny1v)<8lUlg zzds0yVrSCQ=VQtqqr>l|TZ z%cYgg!a2^kQEq|U;n$^a-M$FFLXb}(y_~V^ z*Rt+GJi;m1p6h$uoZFai6=|Z;qXefqW^={sER0O`(|h*peWJZ>vzc*s?pd83Id}AV z&&n#@BkN&23W3o}9YXhO4yg}fT&+KiRm7HyQ{4}+OUDZSFv#r}cx!j&U2{yy1wc(07Hx}tr;cbx@FmpcXI&EVN)RmJB4-%v;W-H&%b$C|}FoZkRV~5t{@N`kT zDH{uu_;(ISj;vvCv5rYW^1!OtXQlQLa;`>F7H33mV!SAkuI7oNdQUGUnAu0JIy8vZ zx+7$!U!wg^^%)lliE6cLI@X-VNLMoGJRO-(M;M`GG%l>7y{k^Gq?S({@``gT*oO?= zcTrS~a`NhEU|XE#z_EXh$(iV{coxp*-K}dSJx~A4%Ur+sOSnu6zjtz+%lA4P0)&Ve zh3#eSFLV+Mly|?S;s>GruQ=@9Aw8zS(WjIzh^mPETvWd(Wn1Q7Gd>Npyy5WU#t?PE z2+MP$P7DXib&NSfKGE1@J@TxrD*wfv)j+0#t5b#~*>3my5RGaw`bxU4lSK2z^yVV^ zT*T)SC0Y&1$#?>-?R6tluFmz8mhm^985IjZpGq3+VkXB&Ddy;26iz#TV=<+qV@=3V zu>U4y3huqOz0b~Bv9~3x5(K=~tOzuuW9l2>BQ;ksSDa^#7v&P znyH3(9ECv``G~f@MSs*6LL}b%DA2fHWk9!LbqgskJ%kQJhz+cjEh>!vo;{JBN00Uq zB&(G?&r(`T|o zDO#q{yKQFXt;UFLady*h?B4Rrap6JnN;g z?}ltf22682%3lcGAb)rhsU{%u_6h=H6p4D2?sB}uq~Jt#&^~#1$NA%QB@1(2H0p(T z{LO1^!^UYB9D1Jr*30R8C)mC+UOj|Pz&(PLIhmM&Np`v{m1Bxmsc0*)CIW+dwEszR zPbj?Y;qW$J34K=iD=o2pNgh|u6@hbu9v8{B_tz^^_CyLgh`Bb_WXsZ&Cdho_Z&W?Y z5z2K@J@qSE7MAa@Z$61Q+w&|mJ1Z_JA0@35N-rvQp=_h= zsjDlB`AUq7kjA-~AhW@>pUK^@fl}&)4pS^W@xfaHZR(uY)xvD$bEATJXOUO?8)sZY zn8z3~-OA$fLkfL0Oyu!j$BPwfZo?8I&*)&a1QFJC(Fozl+e>JKgKNZR$4 z$LlVp&WCufPA1-2k2v(Yz!i^jrJ-|k4<0o&Jh2hs#^|j$V#00Aa8vWc4o8(rZ_dr7 z4j4P!rt+`p2Fq^$&vs(rb*<6#$6d3MrfuSHLp9Nh455)j~gX0yR(-Wqgdljsd?bX56 zHOmN;3|V%b9&f64Br%nNy}pV{H}Da?eMJA=8VR9O;Xb<`a?5MK z5S-p`J?>hhd<=+we6)!01%4R@2ey7Yk$51WhJfY`6l-qGTY~H=c=Hq4o6JLKCLl4x zGn5gPV9zVVsYnZ$@gU13l7bUN`w*w#Ji<|?JQ&HlD1Qoc19^7dNmxy|F(&`ptvdN% z|98FtoK~L80bxzu9tDPUL5&9|T_s^tXi!p~k*1C%Jnv6|EXB)+{sDR&L8`{M!ACwO_Bj%&Of47l~wkzX;IW|y?l4{D5CsxH7Tnb6R6q1%W9*M zaTT0jPxmS7KN4ch<`Yga)wZmgHGeEsiA0=jNin}gQBK-TMkGt7?pS%MvbECkvOAV3 zSAcEMjoyazW|5Qqf87_4VpjBewJ7YLYhuFv#;%XqwO{7)1bVfmnU|e7q>wP;`7B>O zZ#C}CkxcS@p6#awtX?tJ?G(NVJcvkkg$5PzS6E?)`mN16m1lF1nUw`nCE(sf-<=M@ zyqaDn>la7(-pW;9MSteOyj9IuMbC&aogFXcX3u>#Txk)7{`-AoNP&_~X3D08*sDGu zr!1W>a_ih1QXc8t@B>>~&54f2F>-r35mS+0q}ze@I!vhGD%psCUxiOb{sRaL;YWdg zF2|XV$l|v?XLzp(p8gs7CFb8L2LJElfA02vFCHOt>)YZPd^kQh29$-|Amd`bb%qf?>w{w;XabHh3&k^7 zFv+qv`SmY+D&fDJY1?g+frnx~ z42bR2ZT|ah@4HrMhhKijrb^(5F~yXNK(kVd&a;}y)+Sy)&+F&DEyn!50qiPh@PjHY zk?~DF2xCB&*t>a+j7k_4mOTAhE@xmmbtHEPwgMXcG}(^W-!~yd0T*0SS4@BW)Q+T$ zaVsq1`>U!KJ68%;pE(mAuYJTb&v~WbA z3v(ClALr3KZhWSp{CTsWSMl_eMNh^0uADt{{Oip2>*Srjzn_vfIZ#vc%Oq*k$9?tL zJV>csx%y(2uQ=qnZ`$;Y^En=$&%8bE;ZOMazRTw{Hf{IcozPx%s*VDZ|O)~6EwceJ~yEJnH+Qb z`eYRGPR#9Xj1JXtZ)q4)zTz4)6|h6@WddxjgH*AX&+FpoUn=UWY>bTxs0iM&N@f+8 z8?5K`*~Z5ngVVpByd;2OJAU89{llO?6K;O=2sJq(rlO&s*GrN#PfmOk3?X*GU@04W zU)6m%`{D0UWkLTH1lY$g7?_s}aVzAR=fIFM6ht)IzGyc4&>)bC(^o@Wj~YWX)=PAC zGDTW{w2I8*uCQHG-V=KjS8nG~by}dYa-n$W%1IMT!A8V?=l_wIt3ZbV3MP9KoLP;4 zZZuujmv7h9crgz^lZ5nJA6m=seohDY|3KEThb$X9QQNg4NXR_bJZfz>h3>3*FJtG= zO3U&YB9qgnSEq&jOztkvzA)h*Uds5dDk5B z*y#o)82EpbCr&<=!>DMxU*)TpdRw8<#8Md<*6%Jd!`=Pm#*MQHvyQYsYou>p#dj2> zydi``hhMs`_pRiv-gnA;RYt_x+*|ctgT<^ly2#}6rxTCgv@_$%Zc^#E>5Nungv*}5 z^yyxGL5NVJLr(lQ&*9W^8@`IcbIc^i6^=8d*ZU@a<8NjBhx=iS*h`(YevPx$#k{q| zI;J$wN*#QPohwW*&6Ug_NK*|ZSm}Q>O`qu=;H}kjsK9-;9)N*!Xtx6)SVLpusN=U? znSZ)Q7WnHFn_<-{O3~z@#t%)zTV3X5l*tF{j$27o>Kda_GTh_M_Qm0&a*s^zAvdo@ zrVyA1+?r;+#QR9oFI85tHC;9}3>#g5a)kvO7_C4u@cZ|1{cD_JYro!oPo{R$vQx{b z38V{|De-eA|I9wFBt{7p3JF_0xE^m0x}D+jf6et$S2yUhbZ1{rGqJS71}5$Ux$E;T z`Z#BZ=KBz*BH2+=fgArs82?7H1?I6DcWDDdR@zH^i z^(_~Vq-;B9mUj{D1$2tyfsp;RIMlca9449j8IPYjeoF*bEo|X8BB^}9-)m-j=clAs ztz6_0RV~UB$|A+|e!AW;n7a}$=M&fW|IG{YT$ zP}p}@?{}?1z%vv=?)c})3$#F7**ffxpKuF(Z7yCb1_)-<6my{QONNjCL8|36MNQvl z#Fc!713b4~RfZem8x5jwfM+MC_}UA=Xgp22^9Mb11IXS9QGyu6gcjFZm{u@ueZlgz zt-iU{hJ(%T0Jtr6DO5vimUas*-diHCD$QyEV*rTvUtn^3M_k-$!a1 z!mNM40~7~O!XhFfI;7!X4Gau?{%4O}-=8F32G6ZvvGh90+&~C#2*xyJ(P{qoQf{+b z0MzKheeIe;#HtoK7wCQKbcB$@2SUJ`1_CVN<*M?DmG8fHKYQokp9smgBcM{(r|?6= zia?4PLY~iEckq;dXi*Igm~lz~Qm;$7gl{b&ngYuN6w?A)G>$TiDzj_tJB695&-g5$5e{OyO@y@?0 z#I3FT)0E!4r)OO4%sFIZUg#d&8Ekdf(Ghsi z1d=ZQZUL&9Xo$-{e&El>+(4DR30{RT5|Bf%IM7qf#h~_N&_2xa%TWv<;}ak4riKV+ zN6KFh0YM+<2Y$@Vmf(26C2roYt!^jbHAO`49$#X*4_yb;sV^1J18ho4 zql2jwh-5CMBj6YIMO=h8)$_reE$vp0=gWxU;SyaAQ@F~fA;`ZyHQYVvXh?S`6kTZ0 z^QMp9X}8KEN<@H`{Ckc8(M+sh-Glp<$8e1tP=jP%I0>yt14ku;mnqE{SA0W!z{KP1 zD`Rol>|$Fvebp{&FYMS5n|cZFzFEv3YAJy>H~qxB+zuFVFaZF(pMqO_w1FLs6l9CL zJFsBwfxM06&G@JnbNYqz$=FCk{%nr*zjEeYROp7MI-nLNN*3KL!Fs8x2fZj6YZ0|~ z18JOwgRx)^Y%pfaQ_z@-k5-iq4$?H&1XtIT3xnx?%kq^MT_on9NerU=rV-LLnpH@) zBe6ff29z&=1#(b$L_oil_n~9ApW^=f{t5iR3(N~JnB9e?EYz)tPP04QY1^M&!{;SS zF#0_aBhYscDiUagISP^w8p180Ks?VU!Rp|4u{XGXV2MaX`mE-wiMb4+*@=56L!|?O zk`64$cl8b=nO4MbWBS}G>-+CtmthJO>q@OtO~r29tTm(Pa`*mVG1j+ihA9A#9?x;1 z_!4F9EuU})Pcg0e?`4%Eq{j6O3U|I-#^7ZuKuU?4q6Tbhnev<}{}>h9bAo?0 z^<=hYkz%F2Fo5I3Qe1Z>#vC#b#Z=LiBdnXY@bMEphJscQffNF;g+ z9z<@Rg!~=#tn0ZGg#ndE)9Kv>SI9z~c1mOD!|VV@Fi0Wdrs?k)l9bpMvmShxh;aOW1N zn?SPI-0KoFm*=O;KjAh_`QOctIH$!DuI_o3k5D;-4dALJNj)$wn z7$xAlfi$CC_f2^I7d%-G!Y~&=xznurTC|BbdSVT{Z~0;)$q ziWzSLGxbw>E3mT+PW#%jAJf~8?-@JsP)pYJ73rB3 zWp7Z!k(;B#dy${12@5ew{+s4`ODaQXUJ7BGXB~hg{Fo9W`9kj0_1ST7O#92%lz^cK z-|lG6xnP;lQx(B5xBO(@Lr7S7bd@wI90XaQasc!7S=2vDBS^A^f$9|y9h1#JgVD~L z>jv_LRaI`FqKT&*+quHt+f`$n|5a)nWJD-Gw?Q(le$yMq#4%G5^1g)4)N)IA*`!@L z&`Ncg7yLGCd5AbZ3!|ZTS_s4s!1I0EKUeP%JJb>jxpSyZ5LZEmaOl1R<3-kOC{t529ewG< z@(&xLFRQq`Oq;$tJ9?mdlxxhyW9JriaE}T*+T*WjzH<3~dQ8-J1+nA22E4<2lJRxREA)xX4+838o0pZuac%BM4!yYG-7!BQ>;!3M!&1Y*1Zw_Zits391RmSH_1^xKxs zCVy?|OQpXj4XqXuZC#_6!cGA%6>v|>Quq^}*+bydRt(@O$Hd$2Z?K-cyRpzB0lxt1 z=aTv7+t2_{N=^my8rNc5) zEbD9S?SU~enE(%(nRqW&G%}TcCMkV{{nMpw-JD6?#ck=YabOn3yKQ~>%sFFTaz1vr zU1s@f%L$k!d0Z@|(t^}KS!jZ|CrR%_R+Y*>Ugm5)K4cLc#Gm$PD@8f#H%E+peJ(1- zqyJZy_hT^WP;w*h<9GTHLHvf=HyRk@D9H%PPeHMeTO_|(NFz2n3|A)DbL#E?-j(9x zRv9_!un2RszAtya&SKw%;B#Q^kIA+7EjC~YBJvznlsYNP`KQs^JVnK6EX1k^Yg=1E#$gtS4T$k8a9PsDH)ciKPyhs4J&o%$&XvG1tJO-qq0o|UpZ z(R}v8+`k4++$w3V|1qJs`HuD+`ClgDNAV+Xm!6Pz@G1_94vH7(pr~!`2WLU1qDQ4` zZHJ(f>v7y7?e!&$(KunFAfi4u1?LLf1=0%sX8Ej6RWvizW3 zZR|QVyrWXf!JtSS~_dDD?bhmtI#U|Em-4-*CTGH&(7P5@v6ev`w!in zwD+%Kq~9?R)+cv$h#M`wH1d6`a=b#jbNQJsae4erGIJrV|A_ZaT9g)^hNKnxv}bWR0zVN zm=OO}fI(dtr?+La719C*&ku%pn_Feo|h%(|`Zgw-;B zR2o34ep1z&9y47WbV}YG^Z03h=ZGMzSKMKmlZm($9^D$LeZ=Au{w7B0V$wY%uE;T> z@|_OK)RVC?yGhUbCC$G9+K6(kk>$}bY!r3X4OTz9~Bos@sA{4uLBQ+KbOfZ;RlEeB)t@w&;}u z*>mY1D>R+Dhr%~)HjTgE3-BP6a{B92+I-%%p+VWiO@hf_SATrX;(_<`f0A5b##AD<*ud?&dU@%JKQGP8Bg?X z1b+q95@};L%zk-&dVIJtU#cMUO6#AFe63G{l2<-!4!?UY_U(XGb~*z(g2v)mNFe0? z>QdP8|G~1;LzAAB_4CteO}n~3Y$J+M_WxCP>KFqhGXDt;mhb`hH%;NA(Ja;h!2e%8 z-ffQX0)>7Xuk6OnZ+?7L&hX#91S*6H#J=F^58OYKI28ix>wW?#bUp-7-$!q51Hj~i zn%$Yl2W-gfEhY#Uc)lz=3Hy_!S2vN|za8u2+xDj~Tj1_^Na?lb(4tTW2!enCLOP;w;L5D5M(ei3WC{q27H@xmq_g)`qG0^lq^ewu>|0-fUvu89hQ;e`sm zSO;8{?ZyR<`M)`sI=dEU+-F057s^f*NJ9=#Ih8Tova~Ohv$HQC5oQ#Nj+4*%GWt|q ztU{{c`EcB97hPZ55@WBv5VwCBfxrzOOuS`D#bUFds~jUKVPbk&8LjB@$#!%Tlwl~d zlgJyR7j)c~G$)r-kpPtcB%|pW`>_Z^{_Di)7U9iyXHH_FtW!eJM9gj41Y1$_U^%{G z?@~tl#*g^>@;PEGDAyW3k;b;nzbUU`AEoL$62BGG-Ko1}6#u`33T3M7pn!y)rrNK?C1`52rWKb?#RaZ={iN0AuI(tA4IhnSvSgwPbqBm+w;ksNb$=F)+KR zFEb>Rf2!M-u7L2ZPn9f`BlAhEux~hHYtW3JJQ|ZS7KF}3oC;Hsevc({B%n$aFAGi+ z^@Ig~a0vYy3?A62zQ9g}UmgJ|?nRB9129~V{Qt>8TEzGd7MQ!fcX2FoJwQMU1hFBz z|2fw7Q`Seaer`-dVF{S_`?UC?Y080t2jG-NM=wBALrDAotEd(L;fIj#fH)6^@DWT6 z8rGBg-|@yC7kIdd0KT1>F@VA+fGb}*HVmH?gWk_dsSN?`Z*WQaT%Lgu64GC1*IQvy z83QSKAZxM9xOM&aJNyW9APRm5zpa-n6uhv*0g!$Ykg74OUCi}-7T^5uH(*l$hL{Ke zvPNOxr2c0$y1i*4iy5@_0XMYnKNAtcf#Bub<5J>0o^1h9VLykS*Q7tcqjvPEQE;m4 z@z-+GFVDk_MPE*{KmL-qW|iwDk`IQt<@kGK;G)IOevl97H-wyN?YNrpY4GXJ(~g|l zN)&dO91NMAwf8%qW^pldRkstrx?u|xLD6D}Fq1&PWb)5?gXT_J`}&=~E5rdL@VSK8 z`2tn%bh=TOZ2PTSMDNrjwXkxA!;^4NJsY;ld`EFoUr0T1CS89Q^tT%AMmOo;!#fIx8FQLxiJ3g1|FLz+M>-dUSm;vh*&%?>gm_kv+I>lY)|`zlC2`J1k59s4gDb8N1UC_~ zgYLev_)|<3MU&-~zZsgGM@A8$-StTGL@BlVLkxKmXCw6;@;~R4rVq0RukgyiJI5yo z?edmpV(Y_|%dqPb>S|fJ!4#&6%4Y;e&mppyhLl8m#(X94cNHZ5sG%!B7CUWyOB-87 znpciXS}RcO|L-Ug$x48v=utRjXAu!9EtUrUCO=YI-h3#wyT>H-Q(||m)2ZUuR58v- zeelS5-A8w8{B-)zjbNF@HPydByGzMrfODhGe%YZx`uIm3*M=4}|5v8ij7m!~qYE^l z{OzYT5e({Y_+%^yf-@?odM4Kpt5)lbT!#@(@x;?mF0e$87n69Zn+MxX@k-a;tfYIx z{*Hm;ctYQB_F=b}l+6?;7%#Lgm@c%JSOqIhqpBO_1XEN#0gbq~Y{EP^IT8Un)?1ObmUthCkR7;g-AO7TOvBO`yFnAd1 z^C|3yp?twoRD-v$M;ZVc$bQxb3k7?3yby|I_)(d)5IMg}1P#wPY%|P%qwAUFP+nDZ zc(B{Tk00afEXBhYUFBjhH$Mht^d?cD!<5MKyV23T1jeAEiOZiu#t z51jxZH@UIbu*HT!OxOxyPy0e@#Z5O~Qw#h)HVlBzLclg!4tOJg{QOTd!pu5QgFW-} zbmdP72`~L~c~>BgK0Y#{mV7E4JX{um{WHk9*v%b8%kCq2APq+kz#<4z#eH}?;&%%8 zjt17|DbttF!~TL^FfGL+I`&UKLujrQ7fd1) z7}EO=5^J_EqyQqt{O>^8%Qf=7#v{S*Nz$1?I?-;Kyh7vn-5?0;8^srqJ7%Vd>+6y{ zz!gtYsq@moCdC%7mx;wN@zBz^Wu##(1-qNPbm9Wfzt>eSWuzbJ?ne-2IRV%J#AjoT zC))1|vM@_rxaxJk?Qwi(>ZEPqIYvQyewmDHj?BWc_<_>}iA%RmRFdkJP7bjg5SS+c z$SDw7n^ufl;`7B?xN?Y?V7gH<6t`ZO$O|<$p331@PIC1tNKkGdC&+4XLEH2I`O1I+^BC1@8fef-rHYt*Xx|j*YM;J z=|h{26U+7#)RW^)tUI7g^vcmx%;u;I?=!WnB~A!$D(9t z;M)$NEAi%dS{iQK4C}Y0buq)b$=?m~%!uM(VZIub=@L)m^9@tJ=*wz4#00aHwz-2( zjW+4tN%wvZxd1jpE}9cFeZ701mbAEavucrfW|50J_Lcw<*6LMRc1g)iu9#^_35wXhth&XcNI~L{1 z-t9o5~W=riTw7jAZi)(;H(zTcqQ!U~w*1Eb3=&Hmf9 zK$*)6n~u*k9IWCQm!C#w_l_D$$rt}{nx-7=S}PMcPm00GSxl)7U^G^?37#+3{NTQK zMRNdrQW>3bbW$;p`s`>1gK(YPCk^TKXY6G^Z?rVE3+VMXeqrJiZ7>f!Z zYDB2xnSJ-)Xu}3m}-Qldy{Zn1?Tn&|GAws2a1H>t}JW zJ@&gMC($I}hp+GK--&xZdRUhFT57(y#R2GM$~$wq)2SWj_*&Xx%<86VPUaOg_46G! z?L5MnwEgtJ7*y3OOsgy%=<0JY6Jh;X&lB^T4J}B% z#fibbf%+R2mp&N(8&tFxQKkm9P?7E{B%F9;a;g+fRS@lvJBpokM())IW}^I=U>T^# zp~ueJKsrMdMz{S4lUV4}%(EVoY5y+sx861cR2mVCxhm=GXuAhZx0L3)u0tm`Tq6jFDOZW;e6Io{~iPxP5hG3mCEXvVP{H+aYZ9@Uc^Lf* zI%{MQoT|e@61)>c2rdQvC*~$)X6Lu{vNN zgQ(GDp~|Vf+brQ-XOk5o7)I;CC)7-0>n4Dt1-InQ8R0zTbI@8xj$1EPdSo+%YFY>$ zXFn`1R&8Gc9w9*9dM!=~FhIL`Ae({w0p&Yi05>pUYE&X{2QB@A0`deu0g$$ehK{Hz zg^F{qimT5PghM!YiO{#J>o>?T^+gjC&=x>8jw(;{_*LPyw;@pgC;=-IY60vjNI|er z5ZYz~15{9-n-O}${s?{pmS6%cxz!42i%=)O zDODe&fzQn3EP+Bd9a}PoJbh)HIh?@^1#-rLSu|5&@)-zs1A+`3paKZH*;&ndT}V3c zN6owM*RD68g`jG>vtQn=W6C(Lbn z8$sb_4H=9{=iNph3$;!S9gOk0uOw%efybs1qt3p?^_(;+Vb{Gv+vwUb$gq7Gg%CZj zUg)0-k^0S2QGe%^XqVd%NLG4OhGreAC9J`TY-)A`bl{dAy{89P3sGQQnkBH#hl;r; zv!~iZy+>2Wm9vqIDX}+2U1X~)uS)IkrSfop0deb3b+-lD>q;aA)=P3IFFAXiF7}|c z_k&2p(1waj+`u{crx$vqYBD=^ycd@8z?zy|j-A_Nu6dD{&XA+POS`s#jO{j$>+1L> z@@smwof#wb{ZW>p9G$G3Gwda|%w=^seeYW5relDCN6P=49&eE&({hA2b$7c*={7Ay_htFs2y}idRctL^9IRO3WCDvDr|pK+uDtWi z&_wioj8TXoDyF1UkBOX6wWDm&mHK(b8@SC3i>(H|#35s}a-a z^=`3`dA+RR2jglr7usF@q?iOkWx!jCL-Wb7Qc8(i{Iq@a=p>w2dQwi9la4Mkfgx(& z4AsMWAUxnq0%&6uzOxUp!uj1*7Z5_rW_@V8204{7iewAwel!)nSIUzxDhkBO0|sI* zgK7}l=5Z-gUZUoYo>(`rOt5db;y;S)fjwe3-rm7~Hg;}ipA_Hr<9@ZUb=)tca@ z;n@mqoy&_;%qr$eIy$$MEMV@GsQF;tV(?B4r20)^?42=`$X4pHo4#w2+vi(&UIf@- zap)3e%yh11cIn5n**Q_KyDHcc9C*P&V*P}@cRQIoVdvZ#Vf5|XSGr%U_`Q4OUlv@t zlSBpcI=S4LL~VQ*?Z!@QBwxpa6m>yWLn;FnHn=sF#tN7Q~&DcD? z=xOK1qSZ4Oal;VnRY#(7ggJZHjJM63asU1wM90$O+XLu%SOrQV{~%)NZ*?$O5liah zzgMpyMp|rT>T#bZ0a+?89;}gD<_9g7Z3Wi3`^KYR1#}KKngZ5N(cZ>J9*Vx=wmvnm zVqA3Y`5mG*)jL0&mQG@D@2RaVz=56i>?{2Vsf0SC%JyWoRlf{VgGz*vHu1tYEnM zkz4kA-s};IwJpaMJ9)^CVV7QK>?};J?;wI|^H8gw>#i?+)?7xcL+IQtN5lxi#&ZrP z47qh-BWx;fE@u9@>Hh0aXUv3sphM=PGd#5cP}iMgA&mq4N#T`umv-*2@3gV3f1%*H zB>Ush_1bpjwQ5}!l3?A3oINX73WmdsS1RD42)8BRxGQef_0H7VBs(USm6e?n2AN|z zz?hqlzbsM%-B})CbG#L@57J0b3>4-R^QqMDO;h^wZGJ+TB90JB0_P1cW?P5C@5TKK zKnaJqaBl>aLQgkA5&nV`MTTfVN6o=z2KGaI@tH%&^ox_4y}og9W8ooOWC5FDEku%l zvEl@QUeHfZq1TZ+QJCUtZqB3sBpm=Bdr&+>pS7hy zxe!#K7!ar~oIE9b3CI1dWxNVJVa*^Xgz~Q+K*mYIP*|P>f@>QAZk=ULkFhG zEdVF>&~_FU+TuAfx%k}1d7!@o#_;~duPt1mLp=q|0QUna@_c|N*#-v6o5C<^c-OLE z0s2^mfEvV?$sPIR5ZVZK6evO@6Oqte|4KiEett4?0Jm5K8p`&Kxl%3>CY&cVKiQ91 z5oB-5cCA5)AG_|iAqVSO*Dz4BV;#^N+6aa{Rm4Szw9kU?3?X>Ci|B?^(4a3VIDF3d ziQ50gK{CCX^oW)5Kq7v-s^qhlJ?xA+bm!t)Uky~9PB%Qdgm_jK^~FenRhC$%VZ%$i z>73*BYOT3VZ5MIla6c1+lOvZ~E`X)^fDdumixxOE{#9BflTnDqa*dO_amnd1O(ab< z*n0`IKLt7H1)}8)k*(=t?4o4#W7R2X)oi!)LV|Q18ky2YVLN$Zr16bS&#A9((VW}A zyj#Rl;#$;2x@_=h2)L#0I@$YYb`u{88w|f-t04Pm$5H;X=HX?7AZFqsCOxNlV)heb zeWF|L6fwH0lOJ4TRoS~V@?WcQh^Fy{eqEAG!>R9SPow3&W=BaZ%aZF8_1MW|NE?-` z56TYA#~Bl3EYn_~d=_0g8Aq>Of4joz7TOk>cV~9SXMO*@i-A2d-$tEI$ZmdCCN3$K z_!lS1y9e6kjLerJv`q7N^f~w&yj;^Tp#XH|GYiMg%%8edPdtAB2@q*nU1`NbGz9j6Zqq4*5az(6= zl-*btT7sBa!h+&jiHn1>lSH-%gUops)J%@>DB~Z3{tqqKDwXPExAYbI%f11b7MT=7 z(p9oWT2j5`WwY54!H8~N$F9*td(trQbwH|+$35XA>%BLl72$8p8sARCVJPYhOq zNyhp-h!Z2i8)JXG$D}8*EX(F3TcpN~Z>JVOqu{%s{29sz(NDpBdV-0^=@v4?#eF$} z(7R9V%{tIMg}`^gx&MnX_y0IrZFeFURc<0p17Ulu%UZVXof-T3y7nEQCX$IN|IL`5 zalr77x)0A0LTH~AAKLH2=g8(9v+TO@(XE0aEQ&T1g8Za3-vN>jX@Hx6CxR7oW1&?N z4JvT0E>noSO)Y45Y?NvdI3%mV;d0=H{J8A}p;#_Zkjez9>vSw05CS6hbOOI29JD0> zew1>&JOG$v^taAK%`U@^)%qALwLjih7N-r5(g)lf%#>WY3O5eP^>;3{KD z2Nbae&*;vPjBX{C@{U`SY?sa-yQZt~bf>^zqEz7Ly zoDbc2yPg-(&Yy}?{|e*`d^ZB^6WFu7U;t&p0JGpN$EEUwzii&T#%Mh@EF6E+R`AcZ z3+=tlX}BbIVq}t91^gGp?M5!b@SJ}cZHA7m_?h|riO=| ziVjcBi4~gr^H|lN>on#$_rvkxEPqD_5I6<;-jjQp~EnVqcA5i;Jry(~e8n5Tn zjdAYcA9|QqPh%EgyCN!qBC6M*B&qDbH&K#m%J`6n&9(VTkPFutx(x4RU&Ot67%uH0 zft3nRVywxum+}YDCw@28C!@*Q4HU3M3EWTGXr);P&@~c9QZLYZC*qYa>I9Oa^5kZH zZn;>B1kOda*jPZUpaHm20>7@Gb4O8AfJ6`YEQLrj)Fk2ZjjI;M9k>kw3nX$*4pI)8 zPha%@8OA%T{Gr!Jj-s?56JDt>KGDL*`<;7_M$vN4s?YVKac@|{2jZ-U={3z4bcA); z5_aznCEgFpfCfXz^ZcjevHzYr*mQPRH;(TJt7Dd}!0KNuqUYr#A!iZy<4B@K3n29{ zfHywOHuFAUQsAFLh%Pq^&_YeNVv4jUcg}a@@hOE<;EL=+S_PI|VAAX^MOqw<*nv+@ru4r9>y+SWQL#`u;TGuP0%>x- z#d{SDI^{=r{>EVPATAW*6d|O2%Mct}4px(ci~vBdWxX1>@M(`7NH4suzRtq`e(l4D zZ*SXK{#$?{pK(sYd8sR0uKO{p^Gewl{9M88jUE7yKR%%56>%kAf^{#GJ8u~zLVJ2& za31~nX+vv{1)d0rLv>iZ#KB87no9-Tn~x#<9DL zvA@77-2Q4wIsz*(6k6>O{EMGm1Ff?@B&7ol!XUl#KY0>-is#ovTAsmc;@6uC&oq~DaV80Bma31XSt|cJE_a1 zjJlt0VSa5Cmt9-|D{{p9IZh!x<6cadTB#vzgu^E!rvtw=O=bfrONBN^uMl53Ju@%0 zBWFRPt;A2ogzT3$?sXP<5hdqvU+1riBeLUBE-|KOi}S1BdckJPsdoG7?I(oKnv2N& zM9d#X-I9{G9Dagv7dPNlaj1_}{%EqVbUJFzhNHYj)k=DENh-BF34QK2C;D?`+>6s6PAXnhN8(3-9eT zuuSzy4HqdLne1!n7Ea8{SkeipUS1fwF`}yz;j4R|F~i*#s1fTe^JxE9t5AS0$dpX0 zBIxD!&_^%DEbRiS1b5DL>t)O4&^12vu_cHr%44**M_#-`T4;zpYhP85vtJ%=Rrb!m ziLh@t!Pe)I?E1io$v#_q^Wd`k!0yabv`|bry1^<1SKgKU;)d7d8x3-3X|1}LVK4JO z*`jmkABfefHTR>kgF})OYUz|O-tP~jLJfQ|o}s_krHMItb$qpA1--e~SH-J>a-#Bj z&WLfi<}>Tm8xVzhLIfo<)ukv*wz_=wQWm#7Q~9=s^EV4p`UbOAPp=AH;7!;$%TDNU zP^|UM@vn-5wDoer>7JAPRd$`?Q4gCaqg76)I$9BXC65W2cE=7z64&e4WBL*bPu`EB zGZ&s$7|YypT)f`H!prEmTg8|6K{?p+!J89LOb@E=$e-AL0T}jMY`Wu9W(c-z>Kszfb*G5M*y*nKuDqrvl7#sZizVY z71}Etw64TDK80~Qx~JANFfwMIpFYYjly{AI|Ok29G=l0MuncX?cI^)0;qfv^$qt1$} zRxFE{!_-JDV;H%CvktO0OnVHC8Q&NPZQO{IeKXF*%w$Cx^A1~U{`x}#u(J>nPza*? zxzb)~!o#S)>^G5~UDOXUgbEA+B}65l5ldy1Z zW*+0)Rle#frRy*-UV=sj@bkU8vs91x?rG5KMyEC^TRB>)xpNwG@bf+*(Sf^7Q_oQW z?X!0`GO}Qks}u8w483>_Mwnyh3_o_U8%rhGGlf3wAZ+s{s7*aHxRV*C)ArY)kJ=M}_;_Vb% z=*efEJoz~jg`_sp8|RfKTHiUBfMzB4cV|E8KUJgLm_LyfH#ZhH#Gez6+hqfqNsuci zu8rK}w!3}11}hk4@=47S@`(YX=s2bozQ;~p1OyM&O_RTA zy2rV1`V%v6LcPGN7L)I(S4PEgW#1~S5>M3nlKv>7e_cGMQ_d*3@4&YOEF^k(x7rTR6IB$yZiCYXNNYNba+ zZI81A=VaNYkWWrh6D@Na(X8zE3MKeh$jwO_63CTEVP+GgV$GPsD$82;N1g9^&}XT% zzaB0_b(jn6y>Uuw?(UzT&w0sao6w|qm3u%!xr+a5sIZ#+%L6@x z|07$ma)zQvt=uvJ?;A5c9xCNp;@6+;sGA9Iide9@?+ zUan2({GqkY9&QhY(n3!oUm8wHi^tUnUg+%?wEsyV@oVd0ps4?5p<3ed&oqd&teZow zuz|6dq%`8qU9R+i^t#A+F4~;EO>rh8$lAwUWwt+!X|?kDc(bjrs`PyxJ;RTCGrZI` z;a}rxund@w*KRgV2wP-*XcY3$dTJw_l6cJz#e9cjougZgf3xt2fw`~W)n75E_^sFi zGiW#bx+-q3pI#m!b{@?1`x-DQqZ0OR5U;SQb2Y7KIj{Iey^os^CY&D>HYJl=X(5cB zEx|fmz2oc0Q9oGP_CGrN>ZqvNwqK-MNrte9 zna@^lZTGfs`&u}~hqWBZXN6`N9UiFUb7&b}_G#e@oPo#$1)b+I{{~jC%tvcr?7FS zFsP(dI?!cs zb#1b4{rl_n?;|B+V7*Wo-3~4Fdg93{(Xr9YFPViFkJYVX6r@9i24a#D| z8w&`@C9n>74GrP3QfqWt8WSlYaEv2{rbs-l0m&JHU(arwF(TA7b7O2|aXPl3(hp`g zDdO)?L5`7(a_rsVW(h_d8G8v@|usVg1 z#HXdN&ep2zLK`?KJ3C>LiBxrun@++nh1?M|Z}ymFu9>N+oe<1@4}8L`HEQrjeqfV7 z`$dkt9t2GMAH1XqJ*l*zY8Faxe+?(aZiC8%%Bt$*6&g@_ybqURJfs4K3|(qf`57eo zp$a?iKza!(#!tyt4#7v!F=kET=8V#~V5JV;c4f=8bH8KRpJnhhG4JP4?G-}bej@NqwkTw=dV52hkCOZN5`Ddy ze%9^mb1y5Qs`@?Hc+(bF6V!Dr@~{!z0Mxwr7^s}yn{CL!@m=X1@t*^n8w3MTET0^2 zGJoPzfs@&~|0_Z1lbvA)iU_FwDsSyMOhvxK+Zs?@y4knmYS4}HAFojpbuTE*vr00cGll$+nM=Hur!1dM0wnaF#lh zoB*gl_eb0`xuglVp7J5uHcf}@6PdHfE^=b^spGi-xrfhHEAYuw6#So9d2Hd8C~-5 z`utX_tx>41l6g8(Zyg>H0^I0tjux(mo~eN*izhGw#p0DLOZGvc=0P{&#YJQh8dAZL zFM$q^s16fa##eobQWuo4F2z6H?qOK@`MFWXUS3CSl=ZE<(Hlk#HaxX09=|?cIM}!i z%3d7#wWFM_Uj*)$7&Ay-3O1Idl?@ZI6$>0C5sk-2Ce-ZYxb@Pe)kTbQs<$9|h!DTl z-QtCf-5tU63A6bwSx`gpD~-gm@X5q3J3$Y3)pH6(OFRdE=&dq0i?mStRT-SB6-t5q)ZAFAH zQ|TR9)sL_e8N6!gU2BF)n#vPR1G8~-%I}IwFHK+CL7#BIuEjHDQ*2g@hg$C7PMi)y zNs>#d--pr{4&!B}zU+y~b>kwFqe8R@?^C+i(tT8A^$ciJEANeLG-48R)M~W9%4jrM!+RY}cv-|Sq*)2ji`tWqL`PRM-StYNXrUpe2Fq(*ty_#=kW zT`lK|dBW}Dl!{v?w%|^TA*6DW?OX9Q@Kk8F)+87gd!tap@-9!7#@7yK2zPwXsVHc` zPM&krs1rYBVL)Sq@0{x2)`=?fX)kFlWZ|zU;dP`vm=OND=}T9_=qphk=)BL^y6Z0F z%r{Kwsns;DNNeZm&v3HCD~Gh#IIS~jP<30N00Y|1w*2!O= z6egU`tqmp4*r!~=<4P$+&6{7Lm_VF;95Vah>iUBg0!8;{YhR=>${*&lGR$L$TJwvk zaw;)`vaE(DdSovSwLe_APVu8{0H8Mh`#<%3o#+ogbkeh>ea=BDTklim*TLAoVkUO; zqBvN%4ePY*(#|O7bsm0sw)rGdede%ayfg2^hmz?M$~x^#6Zk%~jhR7}7hDiq7AOP$4vDJ6U65Zl&twm&JpVoJHj&cl1tQPCI07Zbxk9>MgwB zoV%zCZRB^=hBGzBGYy?#ZbQ)t1}js>mS0l050St4>u?ysWFTZehTzDGONl5QsR;@m zQ0Ds23t2=P#1{xgQXI(H_zc1yNFI@(Vemg*9Qqb;mg7KdqV>fqptXkv!vat>fPt%Q zr~!f1Nre@zEF=!O?Ep?@kKRV{^+0bsVu(}>DvGht<#`H`;TZrD85vbT4>#+=-%SI6 z9|c@+O~4~Vt?@HqD~JjZdVz&6MI7jh?*cQ}No0=pz-<$(e|=nTY(Q6NUL|u0UdnCi zL@`s)&W2FO8MX;<-`tpMYYSWkho6`47XUnr>ue)cfTCU*y+IuaRx(E&m(an0$~G^z zKg>(BhgJnUgmup{8vA%43?qKHfWs(HusQLPL19!m_L`TLRtMGsMo!?bpW@P)GiPX6 zbq1#23rD_vHO~)e_;FrBerB+{z6wony$weqsv&yT&l}1XUqumjF??Ib%e_a z1G=|{2!#tmI=n5e+%bXbZpxHY10kK1Hlg`b#;wpVRODeANV|@~=!&bwT0f{{ESwP8 zm2^`v+Ib#`8^R^V7F~}~zBd#t*+)4<)gm;wu5XvpraG9M6+?W#2z6WCcjnvSZL z9H)fTTT!A&(A_mPt|sSpq~t-|-t(~*Mzh`q4k zP!g)fA4+buH_q@i%ySFQl{|e7=D9i;6#)L=sXBs{wuvn30p`2h9ZZUUZ-!EDfM}VE zRzcMp1?nUbW|0C=6Nz^5)9T?@KJofvm;1}#r$D#`$}T;TndVCL^TgGs6WkWxp_7VY zD_mYpYQMM7LJTAG0W#H}E&1}^(nhZGr)Qyh2=wC!WPF5jO7LE2G+z`E=gux2Gll6x zwH<4{a~TtMXB?|M8{?976)rn}_715?CI5oyE1$$wk4Id!oxKzHB*)Xkz7_qjIJ-CJ zn9$CCw@{+qm;{yx4Fu~`l?P;9K19FrL#Q^jNXjQMUQCZfdIJuabOgn5$(duoEaH7>gnS%{j)_qYb{-plur@A_W)yArVo+*;Ppc)W% z3OZHawol9d(MNIuQ13#)kph0b*Dh#mi-}fLvwvDOKo8^YhI=oD!vCJ1Q*m_Nulno6 zdiqOhtOLSI6V#u-Z9U^(^iWDzW5W z%^%hk<*4o(aXREFx9~Fca%qYNq$bz6LoY#JAdau zV?&{Wi+z7KA}xpILXi9eLzzzf*7|;2x8Avvv9aFpjk>3jP#i$2f!b$mm>YeD; zUS}VaKx>&D@Y3R!(3yEiFgaiHxZuGzM~SM6FxEGCOv`)07P3eyt^kk@avc0C`g0N7 zO8|LPf!rtklyOim)Bvt8;;uwrGN$Toegb?gSsU|>-mk% z)-VcikgbA6QJDUyH5>6d6}mTz05EP1sa91L6%*^ipHu8cU7V=8q6fVjfbv9KDZsgM zm4VW*Fb%YwyfYs0Uc{l0DB>&NSpxVd(aUexdPAY}RNV}2AfSSIM|gYyGDvpvunN_E z21$8}0I;9K@%KW6|G+Z?QymPtYV6={@4vg&A~|Q;y|8u*x>q>#bt1sFU8w=70}y2W zVHXq-XT+UjP3H|FOz&haPW=GZ*JbrX+;bD$ zCZo+YZ$R48x&EogCcg(-rf+?ldhT5LgU*z7%PHyO(r3SKT62x0acf)VmEZjKwBqJy z0|@)ow}hsFt`Qz(`rSs`87on-#IDiguF+eP_e}zF$`GE#qG>Y80>f!cS-dwdN3t6F z&_nAz(3)pJC0JgVNGTNzrn66`{+W(ZRQCZ@O@$mW)}j8vCkH#Z9EG|L8Hi~rTp@5P3K2TyX}WOjRff)M7Lk9UeT5h`yI1< z?%d&y>9Zn?{>fFr>$>!t3o--qWwbyvlkH!UY-o(K55Gg#EzCeggg+RG(fP=CO5utm z--^>(<x)ISeHW-X`r6THnHxIIMQXXKr=@ zezNr<;A<}2-M+|V4;s`Ygk@C$O~h`9nul@R_cXcq*hDEcRfnAr;tc7P8koYTPBXE4 zb=Dk`?~~;4DmA7YDe%uxvfxf9)A=_K+Rh!a&9=AWHA~UtIraxaV;~jQwqx!}(BAIJ`cG-$MNiN3g`M{gnku6@sB;PsvQMob*`R`{>9 zqvRebkoXtBUPYu6g=G|woG^ud-U`NkRlQ_+Q}z+|37lsHt5cPJ6mpjVWa*PK*aRtK z7wY2g?6~pYo$94YYp3HQeYb{hWqZFhxzWY|;ZF{7_h1_#CX>n8UNxzAuz5;YDyrnY zurm7RPHkv4Xx+kZUgSZW2WwqW>hKi#ZF%2-;$^6gNAe`6n8iagVf-#4`xjUq-j6Zn zg(rh0&=D2x)1~$eD^$Lf$Q3TUwS`<{bi>Xm(0724Ig=N1RR6w(BjG4NM={g*Pb7>A zWNGgQogTVdBDJVRrCKx`TCEC=Ed>ndd3entYTB&vEP^L<8ZuuPqN)mD#2z)H-fPID zKW$fjpW;UUTQg!fTVJa`_SY+M9iK>L$3#+b_&jXeSxpNr=6C`yb{ z^sdUoLyDbTFPcA=l$l!28b!^9!RLs?bF8N{9G)ur5SuMS_!M0Iz>3$KN?PNpl}a4? z5X6<@H4$6;Zd301S(L9wQ=eQIn3y zd^yMb@3MldYaM<1+F6j^hvz3?9{L{@VN1Ixrdll8SSNOTZu?Q^tK$JDKmwCOF_Eii4;^ZtfbK9CCC&5+*1i8dDPrhD&vgNrBmzej z%L*gkmob*%?Y!BipFiz6boJy^^e&;xw|;5;M&oPU%Bos&;pu$%S?>T1CpNszb1P_G@yA(_4j~9oOta-Vozbhua6W zqH%JY!7dKvbj&q~XA#kDxup0~cT)r^YpxUKlMadD@*m{Uj@^^hR9x+Eps*s@e1 zu1toKCGddZUg8Ems>;p#rp%vBzp=cZhqu;A6L5=K-OcCsrh6B`SHXAr^pp%P_hS6f z5vSriUQ+tBv4QO#uMoa?r&%Mb!EbY6Bb2n2zNHgR}y6EJtMp&x@d) zjjz%mVa;!2H6aiT=qc%ma9pARQet171=<gds+tm#CH|Y8Tom?!uuGaBh?ih*)b>_7E)r^7tW^ChP6bC2(aU zg~~0f2lP9{#7-hrzC1@)L~}{QY2egA-`4~}cvFF-0fv*QCL5|fk{0UZvpAD=N6J0ROjgS8z1SGK^} zBL@aG8LA3qTPy#6_hB?4;^JSVO9;OWX#{?-5fS4ck1ZW(a;OA7scuT{B`Ac8z!CGs z0xLCBA0*RS`j9TJC&XFm#MuOsDCLroyMQS;TP)sykMGL0El81V;+9CFN?XP+`;@Qm z84tyaoGYrX44HDw3zd)c7r*(Qq1Mf#BQ$aGDr#pW7b-@v;IT4=scRXLS9ZsseV$z0 zRMjCxf_R#yG<8~}j$b$GCQM7LPT)XAWvRmAuFJ+^nxr#rkE?6Uv*ye9vB_CkWaa19T zp_I8Hq>SHBb9zflZC{SC=?-IA-0M;3as0a^UmS*OWHnmFGvN~8_TV^*;&_=-Rzs&p z^orwdPvKY_PU9Yg>lxsAaqr)s01^Wv5blc zz(7FkRydqLK7h(eHH%U!&z`dd4lT4$Zh`JpICP-yt&UPr5CU~#jRpKVp5dHikLP&~ z2FVwU+{ID6VDAJ01J*CEejpRb;@ZSJo~7OnJ~(d+0N^ezE{?pgu|&lyZE9yBakHI3 z`sPg5+mAv$RB}@fhDbv|{_+9XQc_SYsao0Nm<-4B92f70gy4f-{#BTo23&t&6( z%=_=48~tzED(#Lt|2bR_QgbGK2Ns*1m;TQBmgrML4nCBKI$+9*-yntP$Xhr?fHPOFg~ZMFy00V6-Xq+)Pj*z8T)+mmW4(P6xCS8K|;bvhB7zB#Sv{vb1$Uwr>wP)M=*F%+u&oOK)bfgT#An~z6y;~kRm}~NNAlmUm z{`uSX!x7{%J8a$6MklLPS!><|t(_jycXsdY=ch-4%Bb~9XQqQY~PUE zqwmb?=R?8!Nb2*}>UM;`%h{O)r;d?xXf3J8roP+y&esDNaSA+y6ehmG2W=ixo~=M3 zWT{-CJDhu>X3`2p`IuPQRF#hrhphNob8lOlrDupGXT`w+OskZ*`Bq!w1m+)(_}rXD z_d;w%JBpsnxk?H`S$ve&1=PsGRlI~ulpYVd37n{CC*ra32+0X}&f`!pq5vw2e2!@2 z@w1M_C%WX(RE<~^R8w6wozOpYC>LE!maubO@*Ur<+U04ZsvxwG;(3JYn6%Ab zTe-9=R@i;S-eMM^vnFfuXQf8SyQ^DVuvjkZXD+DpJ=kQ6e1)o|dDbOgQ}^1zcAATW^mEJKuYjM`^fC*R#ZF>=Y0N{~j(r`k7zxSzqr!Q=G6jA?=5 zXEef`8-w6dO84MfQ2DlNZNtwZE$g3pjCjYy;XA_YFqz z6tI;5!8=*ZAiKcUb5Twc1Y%${%wNBrSyJ9qg;N;xeuCZ$7~~{^BNqOTNPe^E(-mpT zoVf@3xksK7aabPPRlnKK-2GR_E^WPxx|&{#rO$l7C~623qo<6@Eo>1P;us*wrc{6) zoQd^yojalf-5A-H@O!50A~EBIUlv_3(tGF~w23zJuBjYk=m(Q>_0^%T!3C{yGwP1m zeP2XexzNqWF|T#%q2eiBlAQvAnlc-Hf;W+;aZgTi!2&>pKk6F!@O!p{&@~TY+8~>V z*B5-?dBt-yAyUpFYN)6#>du$8Se?zzad|+ohEyVoM$6mKg$Xf3vW%dMqR zy?tdmW9xGL(%X%W8+ISR3ck=;z1Ta+*d25j`dZb81_IpD%WvNR)D$>*Cd0(%-`q`9W?jk2D6LC31lWKL27~%iLU_%WG&%Tjg)WVE1Q2lfa!ilMT=6u3%}5fD@Q$h+2m zEiUOs?ZKl5AEV^?SyLD%UPBL0D;SjW3J<4DpL>_W}~a zW8gu#f?Ps9m83j9#KqVD=U4f}<#|pKh2+r+H$1te!(~M0(hzOLm3XTci^APs>a7@C zTFL8Pnu3qA6j(wv`W9{#M!t}+rArp{41sh6cQ8g_Sf%!98#+n`9lG`=)iac}%|Ur4 ziD7T9(7c)pdXG4L>Xa3gk>u%*Bry(T2{XMyR#LU&iDo^iEy)jggx{J)Z~UUkG%oBO z?I2&vYJEjkVPB;FV(jJF;D%N&P_J>#jy8&v8G^tDAI>5v-4L+`c}B^r-_n zyla~iH#qa1_7}ftQP~)VP=gQ$8H|{=lO%<|tyUZw$^DeGWZ{X%8kM!}}0n)*xM zT1+Yo?lf=@97NRQ)CreW(c_P9g>CV>y9{>*sbQt_2nOWBovOs(I&k~0U-CQ)P*0c5 z=0br{RO}pB9qzZL_{}i8EciK(FIS{EAzetZBVqHW-Llr)aC>&BVvY=NRl4N8{Fap) z8puF!3lH)fFKtK83)v_|&-^@-8AAqsoDKvxWq+l#U&^mMA=Mk&A&eop#W-Uw+buOP z;Y*mx85-p1j5!$Z1nV%|rAZV=6a+htDM*0|&Pi02032M)bF31vpq(`>$0<6ZPCmYb z!X*Ggf!hR`MLTY~{T=21^S$!l+-7))w-_l%`@s7m^mkPYz}+(CZpP!RQl=rB`oCwo zU~SQYuhtG>I&m~s6JqrLtN_jY02k{OsRQrv9LNe2PXJ4e<6p+>G1J%pT92QnRRts1 zy_rG!6qnl?DDRj_W5D9+F{p&kONY!RVmW3X-0QtTv!BoxWm13akv$NWFJ_2 zARH9|>UX$g`(K2-y*g$HQ`|a>ET^9+7q8DB$L@K%|1`0_U?Jrg*85PXbw!A4YcDnyjEP- zI|mia##yB)j#hB^ir?51_3bqFE3Zm&*f?0cUT+?n08Ns)7}mu~nH>}J6aYCHS>Dp- zQsSeD%FOV^VX@MqSugBErXL6u=QO63yTyWGze&_Pgy~sY1E{-Xn$2SS2lduDA5I=` zl_O{KDtjic*rYJXqd!tLk)>Z-L@!qfRK86ow9IbM7d2`QFqTu!cZXzu%W1#WQ6QFNm+Cz{MfZ$7Up>>;GJVnR1` zzVBkJ%sAU0lb(6dqEjCh3bXP0(r*^f^P9 z8=#)vfAEUuVXqxel=~PSE3wc#N|l{MDk8>L{}n)#Ol{^E>E1j**0RteCNsfM71TbV znRO0o4pOkxLLSo3CnG<_8BzcFl60(pM^hf(**KE7$uTI9F5YFreJpF;v_|Pgo~^c~ z>L_{M^!|ok1N)06yfO&%-ohawDzV($*AFzfDidzIp^sX`_%YR8TKg;~82L|KHmos4Rq`=CaQ| z2wyE#d3Q?x!pXp>Ck?~dnSId0e8%Bk^<&!QcstW8NDB`J-%%C|rPy^n_pT;!ODj(s zjPTb9TL$#)ld!*HHh~zWn0!i?$Ge_o^|u7^%N2AU(^4||E{R{pk3-*v+OrE?CL4>> z`|?^@et*Ht|C}(3Nvm3E(r{vLKdpqR_2(($QCSLPA+C_v59&vO;FR(9$=L!VO$x6!U_VK&;ho66;Ax-D>5&DA(X2Lrk- z43%}5xCoJWEX5bsD*>PdcK)`(h>h8m!FH6JW|gb9q_i(o38&9YDYY^|EpKbu#_jU9 z2}j*Us@HWb0qm~BfDJe1_i2$7epL#mo@TV zCIe6Hc)pyqP~irvH@(Gx45Q04vYc{J8vrnUDBK``zMNVq9^9E?ddry0P4^Cks6uM`Wt)gW{-XU}@s=uESe@g{8NCpVqW#B6>F$JL+%(IaFK+r0H4w&C81k zJ)GVt1st8ZSIFHC3-g`VJ)VPRrpg+^@FwrTxj{I~vE)m%F@9s{fIMRu+7 zS)OgEQq2McnEo`U&us93(iY{vdgFz7Hn0{H5pRG%&3&w_<6nsacx_u-TP+JvRVn2m z1um8A3iw9=*Umoh=M`c4P-4jm(ZIUXx0s2s?`t}%?#?pO+k+ch3>?JafC+d;;gqC= zt-6|n2|Q!S!~nlUggF3hV8@mhAbpI+^%80U$Z-qsj3DUMs?&{n4dE3IGzrzmEMO7* zHy>(yKs@ENPDz@2p*7EhQ0SXRsVGCI6ZD{Og2*gnv~x+Sg&|i!3|KS7TwhCDm`{SX zDD*MHL8SCb9)qxi)v5F~RyYD}2d{I_@);6HPRIe`cupI%3Xiw{|G#Me z+c(7<^75pwf85S_NZlF&!-=~=VzNt@aA5Wy?evunXX`s~4P50m0It8jyBhv4bLmYL zQ3QvWN>!Rlh>n7t{?_~;P{M^eCpY9=!Q25fP)^|hXOMb$D}r;HJK*@i0rM-k-aflP z)jj;eW)Cp?Jn3X2FdwfDJ|_#Rwq$rk2CX`@a&scL#(M_?NLFz6Isft+kTj{t$-#qK zAGON9Lze2_Tm@GL#MwSTOl5VPy}(2xIoitp^PrO^3tX~7YmU>Lju@sG>RrbO_Hc7t zAhSye0DQ12IRwcO+T)oerL;2)E$op?ukM`vTMzW0_4n@n)FdMI3i-shv~kwf_akj; zLBmrQWrg8k^Gm4+xwP}=bdEN?e@3L&U&M7-^s8@XgTfXpxgfIx?dxm=2b2JXZe~32 zzlh5+*%VWI);W0%3(NOfR*M=d55A1tN^PC~+b#(C#@V|Vw@tkdW^jmsZ`^+lNxlou z-GHciqS97W>^V^MdnXBhtqM_@ zYAQv(xNh4g7@{F9lgh{ZN37ttJqLuaks|}$^qbox_cwv@|2ac~jDGdBZa+%9BFn*X z`#B9qNN=4=icC__5&m(>;MYNQrX9^7`*(9h9F|!=MU{PtNvsIGHWEtO4v_f{m;w~4o8euUV;+}>NC$m-M)1q1nH>FL6uv9Cx5y?&?~ zcT2T2%QxJf3ZZjULw&MUdr*eq3Xtx}=TP)v=}f4#i^ zcc>6IdT2Q&bDP9ZPzOW&-E)XbqKly}@5olWL*oS&^!sh`RF$RDv#Fbrq_rOwo#|D? z)&pbsb0#}j(v>Qs*nZO7%f7{r^7;jB`=FeHBvE+NoH!?mwol`g00Fxr=H)0ZhMU?JHH!MQ!Bz5$njLaB!91r~S`yd@Viy z{(d&-RNja{8gRm@%n3&a$;odn^bm+vPm#brb&wpiq`NH?X81c*)s80R+9#q!&sGar z9rUD@&2XI^>wL-da$L|PlU?ivD-q z7il|wJYvIt6j~wCLW5Pq5AH}T;S2WJ%Mo6pAynpPI*9`caeS_qU#7v&y_g=CbdoBH z+!&uHU0-~uZKpOmLu=W|h?|x#anroXa-^n4`L(H{9s28&KM7qXc6QTJR9mZ-=5|uO z*sM)H7K7`eJC|F(^e8-gBB^)oMdI|mZ+6Y@yzD3Z8u<0j#psdZBV11+;|mh79l9q= z;(Fss-QD(THhMgg^ezeZk zJLl@QCygc^tIbr~>@UvTrV*F-ckogoSU`=vcF}lFVaHf@ub(VSYtinGy8Kix)=Y*J?g#DTq^g)$}-TD{0?29Q~;n2-(9W^9h&xQeSn=HK+k`1mLxiB{JiUQ`07H$ zzjOJ78oT$6DS0~}gy9gS;rB7``2FfwtUeVcOIci=14FqDlshvxkoe;ik&Bht7U@Lk@rry zeYW7TeF}f#WLC6Y3*8cGz#F4af$``XC!8iKabYekcx~}3{KDJW;iinVrQ@+> z;Stzl9W7(dqaPy(k4?AqXL^4;l2R|R3K|diw*%S&A=`c=pJ)#=SNUazPIs^+By z7_%xwVYig{@(C0#_@r;U*bAk`Y^;YGaVs3}w``5#{0ad{QMZAK^IMXw9gO7@FbQ-<@O1ou99{c67fXAdw%=@;mDs$^BWb zG*W}7M`+YrB;qx(-ERt;+*eW)Wds!4i0+ZEw>N)^T${_aRq3X~unib;7FW~Hx!$jM zZ~bVn^oP-c^p0Bkw~@I1Vllyp!PNB9HnGV>dTJP8H0X|}ovb26bZu?O-MwMDZ_UZo zeI~@OH7h-4!R_~;?UGAq_N$)}`v#c=32#gPBvty^G2I>0zp|`sz0`4$Xm0SX(fO+Z zHJ8iS7s!$Jr{t(tb#i9s?TzcRvh-|yEJOkjOw?XPi1O%&z2Tqza{_OYa#X~x7!LVT z-&0Qa(Y?P_Y%9~vjA^{|{jufpV0V_eZGipZ@7K1H@TL;s7N@lB5{1pk;rgPJ+D!Ap z-2Ni(U2o?zssx|W1tZI1g>M5BdkJA~L@XWY@9gL)+$WM8mVQ?{!XB4-v*hB*MSL~( z2iQ|Ty3)RaUFy*CaD(eG)9V^a^R9scfiodJJt2OKt^&i0I(PTjclHvzjB`|)7m9zb zzwxV2ZJyTUG~2D`I#7XS-H0P4ic~hm%zg-53n+`oNeDG>&^$YU=hM9K@z&-7t|-*s-;b5W;}DlWJ{ z#L{81$T5@`$Y2X3@iRh1LGC5=>u*KxLXxSbzgW2YS;!8T&V22;Kq%fI0l)kF9kwaG zs1@}>0m`^QrF{NJSZ~lyA;iUazmu6!d;vsJ)>m zGxTgI65BCBjE|5b#$-je+?JHb%I>--bgI6%v#MH9KrvY4eKzehy3wqN0`>a4OZ7s> zCoVfN0^|)!I8xdscircmli!7Z_U-N}Tn8r&@93nhrgS&i^t(H3vUPX_pH5p49{)7P zv#Ro5?PT7zp2aHC=3149sBDAci*@`r1GLQxop@+dTABI0*F6`sujfA>8Lh14*!sFw zOw#AK;L(M3VGt3~-nlp|2SH2v62NcShV>=F8Yr%eYJrf z&1CiaO)ua066uRvB-abZ#+`sMEyq?5S%@EsyRoeC^83D9Aa}|idYv$ z!sZvhE%(2}_d+)nXpW+EK6EPoCM}*G8OVzyc>G$=q(mkwTRyGwefm;?k)!0v<6m}a z&JzA<%6j8aIQ81s({Raj7^n7a5+)tue36KUI4-oez4;D4Q<4wNpz%Qq} zY1xAv?2=7p`%ZBD`K)}yK2PX&@O^aJ66$`Wibgi|*5{j&yaEXnF!a3%=mN2fkndV) zap9`8!)Qx;pC&R7h&$_QR2G)#M?vN;;T+5D^_#@lZDQaP+aUbHEUBiYe~>#wjXn>+J5q zDW=3JDr({4B4}-AfpxQRbQg5Ocz}<<_m0lib~bhx>%D74g+)aLgk=Oo#SBGcIK>pi z#la5|VL=&55%k{m7Fdka{(=&M!e9X&l%12cv$s3=RmTwg09OlhiowBe;0sA`KlH1Q zq&P45Pr=2-0fWL=YTAL_RV2iu1jQu5$2^*9hT2-3qKe>m2RnNV_)isMW$z5#qG;>p z>Ev>C$X&@nH zB4aD;=VLDArMY*x6xv0{O2XY+SsgG-1!n?&=wP)CjB(zwyI=WhpD#psTS_n@O7a0w>wzRLGr41ZuZ;udj zP!{%f^mTR@RyGzE*3)x9s9}-db``XUhoq$f*iFjCzzU_M0?2`Tp%A*@69+X14-I`W z9|KKEOBDknCrylj7TQhMUe`)O31zILFK&pH6txh6ONfhNL`3x*;jX$EBU=N&WED>> zeM5B@FB26vDI4f-O?7ueF-si}SG11(6p>W$0{6o;(Jm+@w6vC;5f=K}L0dw?R{NH8Kk(rs;HB{aN25??n>@@qLNruA8Bt_DIXD(s1d{fcdWak zilG`#+e$(UW2GgbrvN^MyEv$LAnh!??JdC{>MBaA8d7c|l49Co;>JGCC^tU~4;?QX zxT~wApRu^Lj+KkAFhWb&LD^8($U{ZX8*8MZ=Im#qVd5p`rX-;yj8RtdGq%Mlp=_`Q zD$aV=diJg=`kKPRz{=o8uFAfiC>Jj+oEysBK|&htZK$J&GC&v_+W_ip#XXFT5LzZ) za6hz#v$U|Kr?jOnTuRwa%}2+{$wwMvC<1p7QH8tPN@+<-dl<>+TcEr({49-CWpHjj z#=6$hO2U3JSPM&gH*07;b$eAcV-XPvYeQ`}B{u^}xSoxJu92#hCfUBwxx;i>25t?wuyfexv2{q+KB3kB5Z9$wbX3%RK(oTa0MwpLmMM!OSqS(x2=qq zp^vAyn7f)MxL3zVLR4DRR$A1aD>}A>RO|)m;8Mn<)}zrwR}P`{{T6L;4rV*Qjk z%F@humL4CH7%Iz3C-dy8x)~e+Yp_&OD!F+x{(9DyrXprbQU4Qfxyqfhmo0g%Z<(F- zojJ#mJM`;B0zG~6{CdgUEBB7e*pAB+MYXGF|Jv2XHS6{v`_ko&64&`fi`hWGnf`KY ztfe9mF&Q=N-+nZ(hq8pBsf?ma&V>+>khA^y!z(w*OAG3voxRZpTI}&eGo&WO& z!u`}Zkt6@H5!g5V6kE4jiM7Y+f1o1-5iRo%OR;@TC(2H8=gfKY@1#LHg`wO3VJRZw z2{LkgIzJ<`%D+7<^e0=lCDp$@CE-tLZdl%!Y0ISWKRhi_Hof~lECp+j0sNWJP>DSG z?`#EEljBoJ|B2`h3Tm7TN>ujtf3lkmmeT{HFkk014y&u+;xV zM3C73L3CNc8S6*4<~78khW*x_0ZZoe!LLwStW2>iQj%+yk-~6WjxAhT&GN ztoGiUrfcID}Tj&i{(rQXkE={9s>eSt5<-Z@J-q&1G~cyC`Yd}28vgdN7= zaBtQd<-{V~aE%yZV0||=*ft+<6}Wc^9+=tJ_>fcCz4%XN(+1k7`S(A4q;;*g$RrBE zk5R@*?;FwG7ev{}Z=6{NMs?SV7+zd~O8LdUhZ`^&fCuV%?mGK6@>2uYOre{Y@&4a8 z)f=i1nog99!-?_RL+SAOw|q;BmX5S;C*1310yTOBwfqyG>rK6w$xODQe*DGlSQH*x zt!uF{8>e3NomV1L6VEVaDqu7%JR$i-h4(~I3}9@+`McX2Wbk6Q(LRO)BqHbjYMs}3 zk@q!Sa_v3}vM~}@q1d}M-aaVbN7)d0KP&_Xze{*7}H;Ob*HJ}@0H{STaZ$iUVe zHIyRu`|6^;D-2hm8Km=)W7o|1|m!)cGHD{12lK{07_qF#3Pa@c)-l z$CE?WW~l>yOOtwK`Ir1x5!RQ-0|2JK^^)H;=(CvF>eXOY^+HJGp1 zWr{W>;_BlvQ3FpjIvPdD*{-b&oC@d8M$#5k+}bxNv!m2F->W;#hco3%&s{uud@nE0 zO1Lj!pY*;Saf>AJkNB|^1UWT=!PYmpyQ6L=MG*2=N^)u(V+#Fe+B2|66bEsqQOIVp zaaFAhV|J2Mu1yvQibCi>sCu}yK|7mR>HdDQ8ftv#Mewb}!L)d~`>6hTLB7@D#gW!a zm)m!^q77yON-}a@?hHu3)!3l5aY}fp%ju+7H`q2>lOsUpK=Wc{z<$*$643_3yx+R6{xao^;GK0`SC#_q)%8PDA*56;4~O-JP2qWoxs%_tEtKO! zZdlS?Giz*p*8=T5Pe$0=S4x|4znIW-bnAz1u)dg0JEC&Puh8i>g8x%D*mMeeNI*%T zp(YBQvI2>qevj()OIXHBdsJhNItTYJR_7fl$8)YeL6eermEYR-_~w19*|pRm!DD}^jUYZge{ zSR#0Ac%@BFW-Z2S-Nakv4$!2(IQNz*>3`+Bm%F%_eyyG^G5BQaen)onu3~o&%Dh;x zZp~66cqd>RTh!euP!cT_9;2%(RNYy#@UZ+k>cM{g17^)u6bx5Ul zzHT##^L>u$2|TuwB~Yts*R1zeUzG!Kev`J;7sD0T9u`=hk#QGbDsFB`3HPJW9# z9^2N~1+j5rPn0rJ?+A4vL>)*$*KJCe74k5sMxb^Tnsy_uz^Ud;Ng-UC!a%SIJxap0QEn<-l0ZGK-GV(s*nj)$R>9}w#w(&6 z)%Um?qQA}Fr`f;rcLB;3a9|gnN4I)mBmgV~BOz&o7t8PbiY@S-AEH>gY4BJ%grSk4 zhuFnNmwo>Uq-?5a?#3$ke_%#j`?=*Yd3QHmGI=wTl^q&UYt!(@OsrT39kmyg0aDn=9 zo*ZQ&VqE}9asmOz-f(rF>yEOx?;5px%H`La!;OTxrHd*-rIIc#Mn8bIq62tmq}3G4 z2{Jo!{2J#i-G;^!XMg>0;ewDW6t`zdMpryzQ~}*gARTtgQj77rhlobBz0aSvuDrVS zbc91Z11b^9(GkQp_QM*}2ayGcHeZf$uYibl?T{Mo90NoT+rM3dR3dp#C91^VOhVNP zL^e=~a2KWsSU3sbJ5t(^Dq1hZm+rje6F9W=o<@v~C}Enx(wSiCUNFtt1;`qRpyjq6>T93S{(eMprWC8o$eM(?n zaTxdUC(sjq15X&GnCi~J6IKDFP7gRHnRYq;ES-1uuX(kMG}8Vy-w23&vh73 z5CTisTlQi4$qLAFjnBdIFzW1+5mFE*{0x{PkF?gfz3}mX_H=>m$&d6@v>>!_4<(Qv zfAcT$1a3fsh3|Rmc@DA^)PDV@ft8q{R18!|+I3Z;02PpPGYyD%a1)~97R>cS8MqeX z`b>;KN9rCOtrtXJLTDohmaa0E*qAAUt`_v){NO#FD3oZE4$&;Sug|Ic>2#1j8v%Vr zI_X4FG5_0KQ>jf>bWSfU2s!Pw3dv^jb`8a$m3D9(;4++%!b1U()4kQ$bf5?JPG zg5>_gGJ%{q!?Vr~kpnqTYP&MA@GQ2De0%Q+p#t&Vvr%bXsJ;mnHslA$Rm|3O9`OL8 zckIsH+IZ?~W}Mj_GLXIn@9EpRm^>pdpyC3c!u1!UAMk`jv|J>O>TLji5*r&k2y|ff zG>EUfc?K#^z2E^?TlP$r0|gkOt)7|H98N{m8Qyn=B+vqA19H}|DRDtU4~qm972_%y zMv{O{c6T-!Vg`8!B@%@k!n*hNu8p$h>*4_{zXqAe)#WUyAAnzvg}04dgTvbSSM8h0 zAs9^C_q}pClV_oe)j%DGn&TZCwSEJD=evqlm1KufP!!+XaS%61_qf5Lr4<2jgHVdO ziGr^5375XY3G`B^G{ywcG5~*Y@xS;3 z1*w{o5EW_Pq~#FfiEkQnrXeaw_o!e>G4?nKsK`SCD$oiss#$Un6>VqP{khozSd{Jn ze^s#9aZ&fuLi2gt%1l2FIQWqhC2pPCX_EQw9WttppI_(|U#ZkNepa%mmHrLuPICR0 zvUaMdvVeJA&5Y}?|9tJ#)?(+0;IiGp#l_6snmDTymzKK9Vm*I^I1KvFRjGCgoqax2 zWu_i1ST#z6zxIm6vw`@Cq1%g7O&*?}KZYsRYvtx%d{3Wr*U#{PJYHhA!$=Ky2-t0? zjA`l4N0y)uG)`Z76G(c*UI{o0x}fJ_gYgLd^ms@AOPe$0xTVSWxZUkl)bRG4ch9dT zR_7}W#Z$2ak5KrSIzs_~+QX9o)ZQi{Px659Mw`8-!W+A*94CTr%k#F3GowfOQxl4+uqthC*hxN z2hIpB@&)uz;Q3DA*M6G&Jm^1z^IW~6kd;Gwcx3k_uYWaUReYsyXCO?J22eX@-9K6J z2C_ie_F3&usi+Z6?SEs)Nn9f8BmwcpdQxg3v*!lere8Rl;bjbM8(CHRLl$hCuN{uw z-K^g&92gwzQ9NY$~HZ$-!h{Cigg;2>&twCGgSqnGk>&=Tz%2hJ2uqh0olkl1`Uf$gq-sSh1c{yKK81(Z7t8Jy$JB=#LnE||g zvftLPre)2>#x3XI&9sdftb8Gt?t>zo%j?#jjdZBVOgewP0K7!J5PhF>{q~aPPB4&N zKD%N>-s8tyo+IH5$R!_a9}M&?F8MiK{@Fk}^tj;tbl{`-|sP$HjR=&7=$On}XKc%#p)e<5E2o-04qOQv|L&CmX+h8yJwQUr<~0 z=Mxuao?bVXV+#zJtD5MJqT^p;fHj-SZcOH^o@tp^UOE4?zdmyPRO4fY?a19ywA`NX zqp(gc^bn-#0Z3I^m#K<`gzsm}H4)KXLup=yQxsSy+6FObLB7G0sS}%xk`$G7iy8By zH371lZz=+&ACetfz^{8##_7CSY(F8mwgmJu*T`P}MVW_WyuWu#g^-YCvar)wYodQJ zS7D-(8r`VY>EK zsxBKf#pL?e$!gMtY*XZ;X55*oI-N90?G`?|b%gokn@RwP@Cs)@|UJ z`_y>23RTKb7g}1mOvX40?~L}zxR2+3)Rn(lxw|sxJdIG^(L$b7Kh4fxCmI^5o!uKL zzh%L>z5b1%K94j$l=Mi!>Tt06t>P37Hqn4J7v?agptTYI=*`XHT{Csquky_83)rpA zxtFDWi8=_5gIR#}v1EKk7$*&7>#hxwq($>R#^@v1v-oN**FwG9XV#scN0^(f!M@Vx zLNViGlQc8U(cPRp2@P5S7|tR2tvB+MR?byp^tJG5ki#O%{DgvYWApD*T<;e980$Fr z{D*sQPfvK}mWh7^1>X9>tx5+de>0Qld7{gDl|MhcNVMk1-FOZPRki3z7g0vRhv&$~t< zyZWIqas|BXiDbDna;>M!Q9J56{hQV-yop)b&u0+)2GLv2 zdI@aEn0Qw@%8W-DGhO_)o)0#c7fh9BEetH`Ra0QH{R^^) zr2iv)9_cBh!0)Uz$xoeF>y%r(z0(jp}RG6_IRwNUozj zeZ44%24#Xd@3kPv*dfw`knR>u{M7>y>q} z%QUB-$ z)~cWS6`BjJI5mpis6)!YO4!vx9-#u)0?!sAYi&|Dh}Htq;Re<++IN!oH&Pf%SGWu0luw_9nYJ(Z2b zv+leh;{~+{;U@YrUixa<0o2e4=e&1o7HM1R8WRmZzf9r|d zW8T#}3o(U|&))c@zfP~SzPlcKIKs}H<$I#H=U8c0>%9Nt;u-g5na)eArrQEZ7L{A; zXCKE}46aznGt8Tl;{#GsZ-W5426Vj_f+&fo%b~1Ghd_EY=O%{aKCtYT z`-%6tEl8M6YPl4Oi=6z{Mk1FP7-L1QfNmk$zsjeE7{DkGz$UDlgBQMiuBR#T_E765w*n8f@Emn>TKJi=U3p3;w zQ#(&z&FAEQsS?Y=?4yh7Gp7Yry&!ip_DT&D4kKTk!i?iIDc4_oXqXkholELgn$kbo zxV^wLlK2x4Y_Zr+jDsR?;Sh7yf@qi7A(yF z8`FIKt3k-bW0Dx}oBs4=qVc-gg5VinGQskbIOS6~^*h!*u>q%(7$Vzp3>Jp+ww%M; zP+YVmK}nF!wD?!JgGV`aavq*FXMWXHMOUgOVyv0J0eCU-{>&DYfuP@W|> z?)|Hgp)e_l9z><`cp%qN@eE(R0M2aWcm{f1D7ACmKieDEBE{X>wum`olJAY9E9b!= zTDvMUCx+{rcaw*Yrb)pC(h3!ap(H>Nq%{$2c85dnf?wFP<@$?jYXhC{a)zqsrUiU_ zKxP?+&OKDQuEJ(%Yv(eC&u(%Gm6GyKD5i30 zd+DQ$r#5)Zpi>%$WK&(+D%D!QN2)S=8fLXv4r}pTOv1^Rccx+=`k4^u#(#gzT_5PCU1cD1r1y#v<0*D_Ha^u) zg=p;5r)aVzv34~d?HZdvnfd1K*vOeMdoQ$bPd_*J89CGg0?$W+mV!btojZqL&sbr0 zXP$}z6WlQ3p_^>xt1K+3*2oCj__0xIt;*!8l+`k$>dFs7m%?I>bkkZtJje-=#l^)_ z4WuX9WpuqKjLU8An2GSNmMskHnN%g#$7xNSA131Syw}7zHatBwwIrj#m>{||J0Un! zPmNob)MOL_B^fBi!}K$HTql%dn#X$b$C%MHPG167zhLfMsVmlIIU3h+zWmpkh3o4X zgeHqYyfcCyzdS!|9?9HK99h(Ua?vVAMkSKTxpp~eCjw5vFvi(i=E;C-=^S%?JzZLF zwha=v@;GO`t^UrS$D}DBujerdQ=J8w>SN5F6KH2=y<}W!AT-R5_OJ8hNFvYH(^@ zsj$6wU(UB|2Wjg&>H%`60^Q8ZQ=n5|bd*8xv%f=7UgGeCv^}x_mP?W`(=d5uZ*911V$EE7u zv$l_QTDJvQrER?=!wqlrRwmSx^909dX~ZO-SY02~6smnxx3gS2{k5QvE424RuBPYL z_2KDCxo-E38H%{%5roX0@Kc*?zc6Xfg~N4P#j!qdXeeSEC_F|?QX z2-Ds2=)M*WK@)Yuyt%SC&FjkRwZ)G*-9$f1@EaBC`K%)7E`< z-la)RdoLB4*Vhe>1cSOluTT=!EIgC5Ck*XIZ=$#S*D2k}rs2QL;Jg%>qap&n;Fce*Z99wynNMGG93cJ^U1;sCoAyrzhcq5sG>suO@*Gd0VD4pzu=Lx$<3QbvkNAK10zZ^|F6M2Gl$kB1EM!zy9hU!wRH9mQ(W3GO8Czrf-vx-m-n}5tp zs*?3IWb25TQbzdSzo8C^W*5l$z4ZBI zIj4Jy?-4=v6x#GU(IJi`jmU+*Ufm5VP>^gZ;x1;t3so=gPBn6~327z4G3wz~XOkzsLNwreYOYyTG{L)O0-zvJ86JrKNLX$?)H^`f562?A$x=cZ zAl@$bHY5a^G9}%cJ2LS*(pL{j(FOK`wAMpq?xxUweD_`$|iI@Iaw6zEL6W$ zw$rQFpMr|fC8~}Mi6mL*5cFlNt$CwimLRlSyxESuoP)9)3r>5owW>RGWHtLlV6!6c z#fS^$D(61kZy$eMIYWc%7xA7vL~;vsF*wbxE6Kw063H)eFT+B9zPfQPSw_jXrK0Ai zY;b&{$%3O}S&LOf;Kow_%jg<4)9WSeZ)^BYWqbjhP)cIo_cZ}m^5p!kI&4Dq2lKyz zKf5-WyvLef9D$m@ZJ+vEUz}_{#Y2y4Ih2>q&`4_kgh(i=vRrV9uNHdB$CYPVVu}h2 zHRX}#hxw9fTTmTsm3LaCsxZ#n86I&#WU9|Iw`Z^$gDarnSImx8B?_H7WZjswC8b@r z_U*@$%%ROY4;N;m7G~oXjK2YZLTul%&E_K}M2H1^GDpteWeK&+rswzlpl3Ce79%T4 z=?zuIJwT2Afrd0^Xi|+3m;dx&Whqrf#PPHAr?qV6m;#sHNLtQf*?x!whtDWQnm6vHfoO*EpQu?0T0 z{+csXgXoa{ehk#DuQwbWOGSy$-!*`Gtt@-pkp-8BqSj=T6x4o?{LK)<9-orm_d`vn z?7*nG72C!rIvj9ep~TIacR9NX^7DUqxq!K71gJhfFDNd4*B|RUHkY0?0hR2J?v?CK zQ%sglhMp!Pzhnv1`AOR?r4XPqn+w^U89m>1h0atl)pzMKVqI?T?&atjAJ;swA~Bm9003*K)(O{>)!}mR=Zx9tC@UQR!|{1AU6t^nzRpX&9P?HkBO>h61#+jY6nAZ5GGh$#$XZ zg93F*#{!KPc8WGO{H$^;%ouYXv4aGgD3oN6BPp%qP`m6E6Uec$nvcB)I^s-OOycoK zY`V7?IJf7HgKmCuE?;5Hw#QlI`X<#z5gs z1GWC%I?ez)OGt~Wt7X6ETr*?j&Q%}~v5zTQ7SL&rg=d1kNEZX z_2nT;Oc_n)Q&IQce;|x&gy^>t2rvMBq<8M?CU&97|V|b6x~u^mN6bHpwDk$oB1&C(iz}varX8{J+zQc+V{nI z7pTNa&vB2FK??P0k9{={%8s#wo`6)~(%C%c>gl%08PE%V+;v&m->#lOk~kdztRS%6GsCw0)~u7{l#sa~|J25o)x4Y> z#XBsQ(6fEd&OZ@Tj5zi_6uG;zZ8giW$OpAmp=E$=LFn|-t~KN&&jr}zy}49{z>inW zRj`?mjjZr;4d&4CBlBFpzda-?g6rmCw(aTe{%c5K+-PPr9qPpX)F9|Y63QUq)j5PI z0E92)H#Q!>th?QD8Pcp-1FaHO_wjZ;F>J>%kF2UH69Kat>1Qz_r8GGce?bgk*k?$} z{bMwtAAn$T@0bqV*r+zWv@!P(qC!ws^*xrNBuVh<8C&%A5|Dl=xwdvWik89+o}CA_ z5du53w*()LcqH3h6~I$*y+82d*NW{#N5)3@&)M18&{J(WH?_63e|hCK0RM3- zIfQWTC?9knp^n)UZ_8I|av|8{$)VQ9OZlbIMme2=ptWc)S>}xo9Io*#r>f{JHP8HL z=;&A+6pRcE48%VXc9UCdB|m?MF!MtT{4iLl(F4pd2taJT1^a`oZ$IXEPWS$R*$+N< zl;LKQcFCfuIDCSINmmt~9_~v=a*NR4BIDa5$jG%yhC0*(LdC)d>p9tVKT3ph%7HPa z!8d7h=HE~C6nbXi{a5GG^rDO^5uE zIV5d~$A6MDt*5COjdOH7o(|gy7^?AopYftKDHUZz;wDX4VPF4Ky2;tH4nbCQ@((L& zwDl7QZK!B3B z-U0PNSFxaJ#`R>`tky3dq({0E;^Im!$Z~Db(ssWOX9$Fp?2WkCh(MifU*RJFPDPFjyq zlq8&hP(pw5cFIp2+rcCO%n^3}-Me=mKCHjsz{FsEq`!#Iue(uNg8A{MW zkl4vP@p7cg+WML9n;>-1wm36Y;Xa|W4PRSXS!r+?)%y$d3ZF!_)VNg#c(4FGc?tqc zsG7d~F^cLObu6Gju0uT|e1nU#(>D+d13m6qMmvJJyCZ1ONbF?PIIzIlX+@8PEBWcF zSRx2iRDoXwbpT&+1x7xSFNAm^=udzg@vm_DBFk(GVs3$nEYL2Rg4zH4`fB8CxsUER znA*s;hxDBiXcZV%%)Rzagp5p!oZ2r1QBiq(%HG*=(nJ>Uo7V$RQ-~_<4eg!#^3pv* z7-}@x_Vn~r$yUJXQ&IW6oTN~&9xw4UervtPchU5$PTmfHN|9%$2wD$Dp`f8Lz6+9(>~|%BfGUi{z2*MI zfnO|lm@gvnc>E!+yNOpoL(B(b9dl-Of-FAM(R z#zgvZN=gK`cZwwJdq=mPf-dmJw#o71#|4k&e7h`K%J=v?i~vm3djt`B`acyQ)i?SR zQQZA2KMnwJC;#K2 zz8$$VH93jMQH@1j^qOh@F@Bd_(Icy;J~%k-rE`&UJHJ(eIOJK`5PQ>5^alSNX_;i0 zJhKcxBjse8<)5Q+=R+t`Tl-G&RjDDN@kCesGlWB6%`|(UKnu|SKSAN^uBS~$x(qZL z5?Uf5B~{oM%`n(ErnYHlYlhM30~FRy41q(ze9OC-iLJo0U9AD0ir61kqfF})eBHak z+wbPCJpMAbaki)G=pF8Qom3HB8|tp6p@^bQ__9#ofS4sDXA4$Oo|CxN5R%@*n&c_}d=9ka zWR^QJE-oJ(85|rWz0*u^&52TTYxly?F${O-AeD;&RSrg;(^BDwsqIKYsR6z0cbSI5 zH)KuaG^+(;1LZ*J0S6{VaUh62?HJL;s5w#>Yp+L7SugKZ z__atAQw|EVJ#r!#QTf>|x*9ZZ6UeTBtm)(CX#Mcpxa_|^9G^39#hO(*Wjv4q?!HaN z=W9bK&|x_rm-Rc3XUNxU+k%G^j=k^B9IMadeK|Wfgf;vT1VEwh9w<|FK@4`d!}5PU1HTY?8!7`9_=)QsD$A%MQFM94=FRcG_zynqrKgGNCMBm%Cr3{U<@r;52T2d;|#AL9m@qCBNmOsX`qr z^&*(JGvIeL-w`UB%+Xa)SjbS-go*{vu&N$lXnex(_uQ1_sXbnGhn9kuw%h%Mp5vpHSgs@Pn8O891~FVv+NC z6;Fub4`u;CbZ_4_IXX6UUNxqBgn=Kwqgd~Kp}gwn!=4ut_vBg!K#(F`T=W@~MfE>_ zAq?wWN3r}%SgnSSA#*8J4ehxtx)xs7&* z4w!c8p6|rj+Nw>LQhbL-Qz7EsKz$7bJ1S5BBEMu>Z2BUIp>Y1MTc>*JHs{Wk+$(a< zk0TWwHl|w{R?G}z=BK&q*Iapm3lACF)b*o+UBfXv!(lx9j%zE5p!d|N>ML$UQrywe zkGF?{PpVaz@18uCZ*SlyLg|f~c-NM4q!I5!I`pFkeXb|y)=!jjS|M;Fj!So`S5N&J+vISg7+_}Xj zG}kfQ>Z=~G5&oIQ!c-_K7&^u9xypT6zN{trxM}eAGBgG5M0#efc0Riog2>HsFkMDk zw!mWvT09o6_YdA1!3qctD3Y~O$pSGk{YMgFG7So~ED6T}@!ACy_yvk`=v=^1q#;qq z1j7Dzr65KNik{d5K>m`S|F#p#k|EU#MF7>yPU)ivT>@pX-1@h-sY7vOWuyYWC$}o= zmIao3N`s7M%R4Xkh^Gx$1s=!u7GHTnS_}fIv2L&5*oL$3XxL{cTM!twT6-sRo&*Wz ze@#s2Wy8Tql?DB^i*qj&OIzdU&=6j}3YS}GMjAIXA?yM0_?gWN)m}V;S^GI63n7nH zH9>)(A1dHsZ@YasO)>{f6JNRL7X#}iNt1NcpGbF)8F=BEvoY|g?(y4L)>1#P(5&6& z>U^oSa&l>!b9V3R@S{DWE%BsdP7V%+{HT&8&ojbK`(SVroNDH`ymnu;fet7qiceYo za+A&P>R-@$nZ9kiF|}J&x9B6V@h5oAXfwVCfce5{rFFiFVV3@HyrK@_k0&*J^xfnq@}$E3!f5l-X|l}81OR(WxsLd zJB%Frrph4%lr-S@BrypRy0?`r(uixBb`OgLy1_y`(`3V6pGLP(nX>h4O*mB6<+g083wmU-b7Br3PG@aN!b!AG5xnKPvZnk zUm&y}$jKlYfUCv3jcHIqBlLv3jKOVi`ZOVQ9N(cJoRu^`Rh*%6LyA6}wA@Bt?aWcI4nN?H4^XJcDBi3WRkJRINpy_l1 zZP$tcJMg*LR^)$y`FgriDXbzj|H~_S-#g8`f3Gy8A^{YPL?kvq zMcSo5P=YVmNS;Tc&q_IUha$09tj?>AmDxhlVh{l6sU66!0i>b=)0#?0PTo`nM$Urm z0I~}r?_X2mZbx2CSSNxX5!I3OJ5A3&i->%Fk7Q3+C3upRl|q8bl}gQ(_d71c+YePsAE&d;L@=W{$45>iUi6M_C$ zMTq=&8^qj>QR?8%rzr5EmKGomAj#-kkRGL+%H#T%K8x+^^SG?K=7AlAdJXCPX4x`%nUusTsIpOIhG*fW=y#gDxK@g1?U9djc1C?~ibYL4@OOGef8_X? zzdkufR@hWp1JjRlohSmpq7w?HUJs1n2CIzQBQzeD2%3%R+{bc$^XVASF{4}Se>=Tn z`tHG{E1;KH-bydS6#B=4bMRu2YS7l0?^}otH~GI-ItLnI1vH6>LJb&Ir3#Sjef_SJ z08m4r6Qr-hC~z%toS`Q*x+JXL-ln~s&JKx@pcB79m}z8pRVQzGG+wvBK=+c-Rks$- zDdPBOvH3+W?a>`8v`gVYQbA$}rIz9o}_0&^hB}Fcgkvc57yzGRWLW z2-=O93H};-gno7#fK|79=cOm3zamswe3cvrD%6=lD_CRbF0UNHc{ z_vb;g6-^88*@4{A#wCL%I!0j13i*m92#nzxKodl=VBVQ>2wdq`9sX(tj&M|C7lQ79akzr> zG172o_`S3*&f9x?QdRy#A%66#8S8qp>PifpM1-QSq-H^*wrqN;uoMX z9w@!~61m$Cn&}>cGy$HK-wIE6LIY#DOp$|JKfy7VNtu-T?O(CMhPD2y&x>C!9RU{m z6wM~z-8w%5V!`F*AW)7*#@^`ot=5tNdW}pnG(ha0y;WpVZU~ZEcR#jRrU&5YyatO= zx|H*S>Tgl}hNWh;ZY%d->MRypuTBJIkzVlC3yfvCv~;@&Uu4TJi-c zU`Y91M~BXK${Q%9|2;8-;Wzc_$5aPC|D6H?R6OdzfcxEG8rRi|K9Dddw}+vP=wE(7 zOhZR}&a2!ePro{p?fG7rfwyWTvFQ{zk@WNWrG%zaoLh0K^ZMm-KB(t+h(kuxBNaf$nX+4Z-i?IA(p`=H<6A#Q=mw-UQR6Pj`Mb= z8yMyMS$0m7h0!0(9(8kH{zZYeYNBGk8947hY_K@d`N@%Xki~ySgmZo@j-hiH^l&#^ zo82VAtHdfk4<#eoW;fpZ0|QmbLgX#$?k?&2aZ2)6#N|_94}!QQ$TCenf|26bz!|Hl zm&4m7vY;&E-tmB0Bk^M~==|k_=rkV${{_I#P}^G*@+Sg)p4BDbK;0q%;TNQnm-qlJ zN(0VH>cWsZnokq!m#07J25t@x3%)>KtQ&GAoTP?g9X4i&^S$q_^Opa9_7JnocM1m? zZU#7Xhy&q7b}T4_X^+jugrSF$9)z9`fzgrVn$ZfGcQ3qi4l0ilbNL;fg9-vN(Y*nz z2u|7X`+j=+=7fCFct=LzW8F+MfpH#{Vl-akh^d}qz@~*^Sqr|cl9dbF znF_LoK$2Xz0OPZO)aarC*~zRMAC5HK+=~N&I|1Vu_yQ$KM%?V@Dudv`5fBAv)EN{a zD$5n0>Cbl;xb-nTfQQO&|InQRC$Xq*-%dpHUXWCT6NNS(T40rzTkbYUEGhz%RNMDy z95TEH?C!1JBAovusDd4Mm1RNMLoP*uAFT^xZ9TckW(CkFYyI3y;3u;Ed4_{{hHZV` z9GA+|%TeK7=W4thxDj^w0rNI+lM63*Zi6;dJLPC;H}rlESug_#ox&;t?uJFW|Dkep zH>knA^Sa*q`)LvgR(b7Xb}s~fXU?241Vk2>b571jwF4zLmV3`w%NO2TCe3P4feIO;(^5t32yzR_oq%xAI$W-M#*X z0X61}?gOx^oh^=ptBl zBD#M#LfUhxyY|z!8L>={59+dgvSSGb?Or$k#Wz~Wg_;P;qx_sIxPIqwdD+2!~vx*Tkq$ z(x}k0(Tw}+X=GQ*9!CGJxxuoacM~R-pN~RlR&x{1`qwcQ)q>!C_2f>6G8-J=JbdTV9HVjwnHO^3jt=e9am7D+t$_%AI(|bK$8}G@<4zJQ*!I z<5^y_*hwArBnAfxk9)v&M5ln+p>7;L{{Ja1dkqeMO2+@>Thw_sBR?FxyS?oHe6-CR zOc(AYl*DA0L408;`DYN=dNn^$W)X5tPdQ7en}?yg)Ocs6b4mQ);%PyUYWom_((Pkl zu3G9JsK8TX8FKRDMV%JHMQ9va{3RAEfz!AuAL9dD=j-293m+&+-=-xZ3atmGA#w_$ z#hnBUpQf5!Kv$?;4O$-y)d3Cusiu=>&zx!ZOiB7p?GcdlGF>=xRfP>ga~?yxQ^aHe zba_j^-UO6#Kk$K<4)}q?1EdMy;2R2@9$rpNas8dzL?NQpfqndqwZhNz`9G$CU`X}I z6>t_XIO_($tjA_}&u%@{)M+_^-;Ol~#EUZeJk_^AAu$ZCcl0BuSFdqg6YXw&0j7xx zl9G~$zFR~xKfY01-XYt@ZTS`&g(HY`|3y*3Ai1&Rf<751dq|FdqkxVO8DayWsQD7ra%?BE+~ytS%m;9^g4yBD zX_42TLu`Mi;7<`CynOwzi<2x4D22xV)83oMQ@MqIpyiOAGF4J!D3#14bJ!G$$hfg( zh*PvpDRYP@MQGAMrexg4*yaog4TS7%7E$Igvkdon_s*%#{hf1v_k2G0{&7F|kJIPi zUF%)zS6Y2*Vi{XsYM2H zTsB2G7*V*7x8M9WRk>{D(2=JZ!lh^iffVXZqUUROy~K?`E7)Nv>7QPOAhSgR8FT@) z9|?Us(?~0V>wJ>?gSJ4QE6Knxh25cGZ@IKZ`34TMJioB-yx@;3CeOj#rL>u?Bw6ITdT? zp}4F`m#SJ%ua};Y|9T71i=Ah>o4rMne0UG0S zTXB`JY2dEMS5N+Mw8|x5kuChHSD9}D7=*w+qX@8KL}=`=FnEKA}}1w@*5VV+ny1I{o2V@p>XYFI0B;P<`r~i|g-pg=KdL9A9I)3A-0K zXrl!Q12a@XuC72aH~2TH2?+YIc`rk5I?q_Or-{ z@xrNxQJ*8*>d`Cpa=i|pWW3w+^J`7_E(Gf6~gtVjj zLzpWR{S``&;tlONoEwz*GWjdlBTgN#Qae0qMTr{<6w)bWP0w|??c*?sCk#zp!i%6W zm`Z|O)WVk`PfKnrSHJnb(hoi03uSRw*XHYHdzp&CW}p-}>I(KO1@{&U)IpKmti~6& z(N^YG`^IM~H~@Fz%Md)a{@zIBy{VhLmAIAQz0^}O5+LEUxa;B`?ic|J^kWzg@U^6N zG2|>?;vha`mVy`fV?CTu41-5gXw&L4&|Pi1o0t61uVDF`S$&(pxKE|_+Dv$E13otJ;U@qULDL^-y{E?VGsSrb zUinNrL6dq9a*MNUkgAR}mPtP1y?sZaeO0#A^EYvmmd?VlzSaL(+3T* zkXd$z^!~af^Udo#2vUlY7loFrbGrBQm@Xn1!W7({EUoeuZ4J7;VSCIVg5ei23N02( z)(``0CIvIBA5J9f&Y?NPjK>19yYHp>=poovLscs8^VobCx~>Lmm}9_sFBg1_^eI6% zZB34U+z;Ep7Xc~IyO-<#Gls@=zdX_r(8kKvd}d!xc%LSx&F>sz&_Sx~Nsq;>h-E7s zpVdD`t=9sSr9?zTJodDlzu;w-5}r~Er#Yk1I|0!abheh_iZiI9>>5s6x4lE_xe6mt zb?00>2KP`Qk(w&&x+hE*4SJB02qZIg8pc%Yz0D?}E@YI2M&c_O42?!*-XMwS3vC+r zxd0FWtsTUvXZvnK)l_w7)2BND35j`}@}3XuaNzhKm`^!RJLLZ~zId(9Q888F`jAFq zrp|22Q-wc^4fOnVYcTCewN44`nO_FN=Th!ncG^mXI&|*aPjo_Gf1oJC&JbZ9rug!Q zLMFt;ftbu_C~rCNnM5zu85=4$ERJefBpFgg}wO z@Ou~4Yv~4f42_QwG-h(C9UV@N_llN0P@|P>aPKAuna_~;-0t%I#KQ<^VE|Y=bbhae zQSSVcNaws`+6a__fy5M8sB`_+&E#=Jhf2iQg$M8G^$A4?GAtty3?QgV?J7N*RXNjl z6OqRCdPyDG+i_wNuB9lGiIQoCFfRTCqZEUsM+r!;NoXqs(_4){ zbsWU!4G#~iOMPkN>}0*7Z+8PRaQ6eGhszDtDqc|?=>t@fh*UOW$?-<2U1BSsgEBwu z{)6xNsmt^ij7tp01e;$3@qf_#M3{vb4juw(#qi1(!{?y{c&W?erRQu=jzDwNd*~W^ z&IHAP>0cbLP3QZpEDc%MzP`zP(xt9w`fz5Nz-A7FV|0Ij9UAD}so>UzL^9^9tMDG< zv{K=jD1{r_AIrL=`WzR{7z4r|zxx<*A79HyT>rJh**tNx$sS)3;tYEW6ApZ;!% z%`F+pk-AjSuzqo#Tj&dG5pVz(zr;L6CD~y0s1G>H;6s_8xKoDo0cVn99q$~aRn!U= z&hn@%Z$+}9Ctgy(v52QNHr+%^gM`1Rl`#$&eYmf$=H6hE@}R z3g~)C3jyhJr%>)ap6{In^C7Sn!pGZ`;Gd;;gS`xdZz@Q%l}izFHi@#I1GW=EtC4Q; zMnCph*vI1x#kGl+NZ`zq^w0}`el-u|5CR_*S;TH){@5^E=kh+p4%VPlT9W$@(cn@D zbmcrV^8V_7wooMabpC_lcnv@$TEvQ!OkVSu$;!#m>w<8`aLM$C%dENYqNf+mhWo0g zJTmhlz<`6&wG;ElW1)uKkr;pf&TwL+gYK2NQJwC^4{k&VW|~eu zFf}zD?z~)YfCQvS%Zt;b2YhA$_M6yv+&JF z#-PDR!d*r7+t0|ts8idyM=1w19=0k3q<*g{NsYL(i7b|iezYAt__gpf^x{pVL8|D> zJ&l?7-Cv`O-#-#Dgf_OfGV!%y9;Ze@&TE?ke#nf~SwOc5?gAH4YfwuC?O_8;UozGEVnx*DkZwXyge*#aF3g+)irBF2ZS z#fxgVuAVWySV4o-Wa4Tci>Y&F%0bwP2#{8ptALbcQg&>lf|iRIcJLEOli{YARp+A| zIWvMseRkQO{c`u%X6$tS%6zwZ8M-t(TdbX7Kd~s7OLE(@-dwU-%t8UaOO0*Va}CV~ z{PEF<#8`bC`YgJoRe2|x?z!1Qor5t%K~sA8!W38CUET}jC)ZzHomMXmX&k6SAj5_U zXDc^>>W0W0`(RN?z2>Kphj}3$;dw!aIT2PDrm(te-CjyH59h7MnoE|$>|1hJdGF5S z-m0f@E%(qab*Um@vbAp_YtOn94hR%5$j@kt>`|w`7miH`s9h%GXEEfBBgTS%QjAZL z&z`APt1yYnhqqvz0MwAi&YA3YazCdCiaS;3q^Lg%BYh)udcQTXK+*cTZNPxkuZHow zEAIK2C6@V7TcsND%3CQl;_p8#9|Zf4F)`54(1;BVXX~WeosPsY>Yz z5N`?&-Nyv=BWk&&C-b$u?`LcnnAh|T$G$aOhRaIA=jDMe9lT8(h_KB#S$hB690EOh zj^QqJX@XkWMk&J&?+N%!1f_`{JgX;BJAZvCLMlOpx(I3G5)EM)gVM|jq-esz9Vzf8 zYsRO$WvAag(QVX0nBpXbDYUdCWJ7#CHYzhtCrT1dc-w>ZKcxg~aW(Kjp#PvK%TEi5 z+tZprLL>`csp8oOV5PhkL*x=GUrKLJ)fe!$e{T^2U(Z`&5h{w44eeh73Yg!AF896= zTzl??ITzVXHAIqehBJ^0O-3p3s8!UzUiiP-Pb`i{d5{$pRBSOzSAtaw(IkHc*+b8A z47&xVQ$3Gi5vnj95+5bgdTRKPJEo#W@GsF-(5R&SZ43*t6Q zcJF)1S>e97MX@Y2XYNDf{HNNI;U^j6_4}z?TTD%7Q@Z1=OZ8hyCcH_h>0pI>R$RBL zYWDd)Ce^_ZQV>xT@qNeOhK*FzvO?Ou-m1x@p?gQ9srt5{LE-}Z>ed(`iSl^I}utc6aNh%$-%5TwS?03|lWjK7X?m!9$Ro~_Z3yBv??#@mmgBhOKj`kUACv4dSQO02$so8YGv*k&7 zGX9BZF$$*g6rd2hzh}kJ2i(FT}vsI!G2i#qwpUEeadM zxbhseF3k)=CvqCb9!Wc68#)1O;a;e8@6A1rjB`V_DcT*Od*%ciV|v3+UQalI`iLr2 zPQx-)0V-I`Fl9=rf(3LHlKou2Z!gw5B&Nhc|C*xI( zB-DN3@Yh(#XRljIjnYR>p7q-L$)^!<-A})aH-eOnGFyCSC~V^j>NJg0tyr85 zaIkmfu5H@P#Ep&>qwK5tol+(ByKPLK4s+8PeG=V=GVle7lba2z9&cs#gC+@V@TmwVjW=TZ6}E@8&px>`f2#z&(Z; zrzzH`wYNb9h{S3=d)4Z2Sy!}qylsqc4P%ECujags`v=;VU|;@g>1v8gjaDw7W$Hb& zu*I(F+O~p1w8+BHS{vgqT6JrA2K#V-Y>n98Pyi?A)LUp)zWgD)yXwlJo)ry7Uik!P zz@^0C^T4E16x?uAzJLMppa;%=P+VKi1*Ezp1ZRoI&^86{(;Tj|of~a|p*rvhP}?I6 zTnZTc9>NXADkuqzMlYJ~jcwX|>X1c3bASbh>@wi`rCm#}_v#nlq6*&T6$5;pgU@ti z*q$l41N}3A19wFJds{tcAr}9dk)0y~du%VPd4O_@WX&ILI$?2iV_c29Rs!caAB`xP z6=4H}K@za}gzmsSmrfwH?5xfcobv^0EOzpvXBkZHuHJ2}kI%~}8cwCLj=A`1 z^AI5VboSIk1T0Fppy%3fy=rmLS|!J^Cjh##YKQ`x(j)-ve~l+-d~`>(~*F)1qUx<)g1~k*~_Px; zI8wWm%u1L=7i((j7bukc3(>qh=IM_1l*a5(I=J^)@p_FBXRFMNv#pyaZNUmT!5ias ziN&sKKw-YEr=}85AR5>v>Zu`@_9TV5wH`iGVt#xEJt$UaeA4=H_r3EP=Ed9=2^iM% zVr#oT&6my_gz=bp&yu#Vg{|ZQ?m=-+{Rp0*4g%6#Cm8N5cf$k9s`N;)rc~EAqvYu# zvo;so`tVX;miA*omUZ#^G1sNXRER5*YBBOGSeR|(_7a?c#Ll+kv?~Dns0EQdHw2h9 z(rj(1C8_1*xXbpavU~n(kl8focJ%-+h{a`$5pZ3Ho&d+ZK6Z7~iYv0W6YxU4dM|6p zowdk?@^fc^c}>!kY*bCFDliUM$f9xviLBx)X+|T0J1k>6QLWd!I-tR~jmpjRFK> zr^WT%*jL%aPmHjNYY~>OTONaO+EBWq zw{20^n8PD`Ccv0F7vxQYbcQ@fFkOcBGJ4$97Kj|5p${kDYsKyl`y2S+UMD49uCQ1a z7Tk#3%FR~GR^E%*D>yF;5glExhC{$`m#iY^>#t**0uiBc~w z5FF2J7x1YbF5k+?e>S1{(zF|dsea2^7L9#5oq zB>*d)R`*rd0MW&S&a@>Nzi8mMBKgMO)kAStJS24#RTg$%H6M3*Z}RB{Hl2mNu13bc zg?kh`?xkpc1{9=svI~@?$aybRa%lnnXIFUZ9rWh7X6b3x2`+Px!HI4qeH=sZ9SLt) zNVGKl!L?8>I4wkc_v-P>yW6_UP=ptc{5EZV0GjU5EmtMgz1H>--uMY|0kN6I`DTt)3IPVjSf71}Ol+i)_2eqJ7%^)%K6m(eowFVOKhNpc)1D z#n%zAKX;g;C*cm}w@)g(x+d2(q&1|(=&4PxZiKq4lAgO|;dZ);D7$Lop&a5?dgTzI z{LUBcXfU-iPfvK%U-<}aiaIoWVmC}(+YWHJ_xN3a&tsT7FN-}q4RA_BGfk&6=d9_z zyVp+fo%T@oY|e=eA9t^j3msLp)AfM}lzvz%eZSy`EtyWWRlZfDDa#IPho%(cS;zPg$<)=7PoE}V6U)-#N_D)>OaqRW zD#6lDv?k2`HBV}mP`^81OqniT-I}WnHWS;SfNeVL77hEsNDJT$*l{1}=VuJH zH)`Ex6?+Mc^}P?C@fDIUkkX&WAe$r>{?r%bNdb||e#wQ{%| z*^##9)S;Y?(&&vGx4fG0+3>bvss34L55zLrRSn0rjtoK6h+Sq5RlN_@2DmIz(({my zJSpPY82r(&Y>9Mtph}_};Ds^W`m%ersodONp#_`T)s`( zDjc)S^4$h0*2-AL@}qV;$Ya!&5zyha8VpXeUvYRa{00`{!k|rFKCDy-xm2=>_hu8aV zh%Xw|89!EWcZp~3vRQHAK?lo(#k348g9&6!YAJA7SQ~dMGhxDgjmQ~172w6}s(NuA zic^aYcUlGuJzPn6e1u5%z5wH1HEXHW>oQ9+Olm3VmoT8eF$EQ$PdI_$FtZ$eVQZx- zk6FwbYfy7mp|qJ!T#C2*;#d2y*>u9M@I@}m<5e!bWr4FE*qyI7^3Vie>368!N~RaJ zxrgopQ@8rofak_(IYS@Ti8$YR5&5SxEL#gXFR2U~n1>MbbClCrV%O%00LBU6+orQ8 zE(GkGLg@$WjkM2e+Z2_`wLQ#IB{plZKhm&<=$}x9;i(!L`i@=vq#nlB|xc1TZou5B*=d$H;;cxa%m7yXw#i<3T`*R=U9NU_o z4(+}EaYX+l8}STwfz007ma9Y=iK=h%x4nTXkFpxF1aWwI}@u-{Jqjnuf;sckrt z&@)x5&M;46e5wY0|3NB(c>Pgv`F^o5a5jZ3?Ot}cZTq?4IyjtHvvoBNZ081X1KttpclGzZE51TK zW9ePS=Ni`QSb@yDK-a5CT_Gb8LG4Rb>lx;19g&xVhG9GlW*!PCQ+>Z=xV?mAm}j!Y z9wSnHW+1JDQdXgg+miiG&G$#FB|S6- z7CPn!W|9Q_h8;v)Vs+X}An{(eIaxL+sd0zmQ`Fs^!GHRu_vGj2uinT9> zwnAsLdnX>{2u69*{nH8_U-+{Z95Dep%$}qHk)@;K>glwK3l&315i_CPb^!s(PjnZ& zlJLD)6W98+PXke^!{rX{C6aOzY3@LRlDp}FQ}%*V;YH-etY7YyT)Df`46#~ znG%c?^Ew=g>92)(CNqa6%P%#CnOz*VT)7=G)yG`k4XBl*K6_}->@#W-%Q6zCz9lNU zD#v+M<u*by3~;6KIVeY25yzCVUxUbXnY{J^@5#!nP2DX*{&b1Q!lQPLmJMb-YX;K zr=gxpp`4MHZ-Z(OhRcczual|dj$tJ9%UI7UXl=t)DL69otHhSpNp=WIC6K(T z=&H$R1<_dZ3wA0an8ZtB^*Nm9 zZgEMTw?8s8dkJuU;TomZ&?(NQoE9qN!iH)Klqv~JehIBBrVVQhkvk9V&2vYt^oz$7 zj;iBDIb24il{B8pp-hJ3F!HaT%svgFBRs@jU}euhJcf)?my45Co8p)DL{`=C_?~6C z@`vzj6{~W(leKqg4cTO_%Rhf63T;cha|;36q{Lx-7d<-0^xb}ISpZhR(&ez93XQs# zo{emWNNmQp`bPC*>8(UXPBr}1$7S{2j{E!bKT!7lu zw7Kb2$5@4oJK54q*Sh~lDwq$MF1c9+(b4yVhMU1OT@qF3Z3SP~j z?`{+7tW`bOo&{x5)AAYQSWT%!-!L_n2dwxy{UaZl<1jpFYzMNE!MhBK-|UBkKtA8e zsRQwmed}u4(k%m-?8fg<%7*p6*$pvMsli^0_=N=J^wv9aRCjbTm&{K2_w|}A^k!!v z@q!B}D+N1q_~nUgcC8?7*Sz1#6M72({!pYU^|{o`A7#9d-t(W|?PNJ3k)@uRm zhB+fgYGmT)i0u$?O09eE2&T;K{hZX6v)Z=NmmkUzgLuV}j*H>!*Qgi0?@^XtdnWT4 zn{5q~!>7y6)Da}SVs2dxTj^+D-i9N1xXy%>Zv}-U{yMrcApme#Hn;el=H0| z07Bp!tQR^MX_ea21_a%b4H)s3X>h24g-&(ur(D4(Q)e}Gz8j8RT$?7H2!>ci05;^o zHPy`zc)5xEjvo#VLBGm2p_*+*h3H$3bcPBzp(8@TI9oFJ6N9(~I5*zb9)Y^3Q{c3( zTw{PT+^-h^AS7LZC_>O3L425?y52sqswMpD&6b?#*)-hMTDXsCvil`_LND%@{A^b2 zYEQUi3$|{1&oW{PdoByKht^c|#^kqJ2CGO}aH+XxwGc1iqaNDY7KXg)Tdu-9AXUDq zDt#g%ITD5zz~pdT8O`RKFZyIRfI?2HxFd$)Fcr1M>HRl+E%zl+g*6$Zd<~mpYs9Cry zKzR5)#HOZR0||S!>&SU3S&UOpG_JAPA`|1gMl-N;^L%}e6;J3y00)aBL$N9=6; z=D?@+&BxMFe7)>6kFHd0YG-!!0;BXatW_oU8OEM|S72EDH5QWd~{>(!s5E{Y9VMyNM3QN8d9^ zUlc`ieL{=84maI1V0zDQz6h#_*9q;fG$&*mn{ zr>6$%goS^w`Y^nLh}Zs!LXf-nZqRZ#YqMmjmP^~|n9oWMnp5X6-%v>wAV%~{*?k=# zN6*r|J`s>$S~JKItfGo+s~G4TS;;t|-CCe;u*&&1QkZQxm3S!o_eSa`o@&W_ec-&h zIADl(v;6uFssivmBane|XYrxuG-f}#NvZ_P|0=phe%xsYV&R6e@Fk*;P}YX^_P4Y< z&K1V-X)4t1jPusUKu56SNa`6VlTAwRAvy5a^4PTdvKZzB>5uG;0Ic7(&nE)>ve^Wb zQ{C&g3$$BG7^FRK_nNL#YY&kXnNta_*{1e{Uh0KISJ93uuA~8h4N^XpmH+&oE^i=D)l2`Dp#H(JjHV(wGp1bd_bd8vi-W4WSRS0l0hkQS` zt9rP4GT7rb{lInX^kSwp5~FJN!c`eSR8Uk8f{CXJO>d)T56XYq#b7pdHDzQi^k(@H zsslc)kIz5-K#oT)Qo_b{NtiR?5JeVeVZzlt=%_;JZ^TfWC<|5H(LE%F4~lDD-y4_(WQh=-v%{E!Z_?c&~I;#52`v+ zgTif%e9Pw^JDl3{w#fw^-odTiJGWUmor+F_6;_?43#(SAi8^g@*l%0+4lPM#wfbWa zD`IqMfpiJ_Dm6LjvZNQIs_u&p1Fa$-?R6{$yi;4@(Wo*TIDJ_~zD*Di_hX<0T3wPi z&|im*1w&kbxXG`L9C*xA&X5f8Es9XBlR-k2q8a2+WphE!h%~Y!L6G?F-<)5bK$fs` zb61WQ9v`h)@Oc}Av8$0VHYhXUZI5^$ihqr@ONik~@u0Z#+aY z1>E_R5a>9611CaGv8^5 z2HyuxFe1_67nDW3Wl5TVp}X{eng<_LqW;{5NRGrf+eD>}E+p=ZL}bOWZ`SyDiMj>w z;GiG4{IvJ&Zl-Soj|iU_FCN8=b2-mX9{|)$rzFvE=lF(pz|OwWp$OkRaUtcGkW}0m zhf_WZS2wyytjRR|0rYe>rm(|%mJ4<;@j|Arwgw;5<8-ica%RWhlOSl^_^41!3wczD z^5~Xu>#gvpkt#Yl;W(-mgM{#NSS}GRP!mZ}8d0Hx z0c{gCeW-uw@e_+BNO8jeBiCV&|Hk<-626!xS!?fa3cj+oE7}uLM~PM z$)orh$cI%{yYpN|*aVgP<9O-A;{jrp$;Jp|h-g;-=B=>q2gkP~g30EU64V;QeYj0w zTX90H0Wwk@9!(u;c=F@$F7L90yWU^^;lM0@_qiX{=eg8e&KGY6E!>?#YE-n~hX;6yYeL zw*`9uD1iq`q(?AB4NIM|Bm?>+oEIf)!c820C79>9 z71>?PGfwhZSW?8u@Dq58JQEXc*5V$0wgC*mXsTJu>cJX5^stk2u^=Cj#JB&eBgbKD zAjCHO-xOk?8iStZHMk?mK&Uas+; zZVBFzrHg-a!PYwo5saSt9OSI`EM-=@$G>kMPRouZ_QUNR+J zyN9rU>|y4jMt0bgr#MNR1CjXW>AqMZT#&Voa?MyY3-2R>?*<64+a4d^kHe@U@kjP- zC*k4tFG(WghjYm+^Dq@*aYi^;`Nw}AvkOOdW5wIb_H9-z>dRQ_zj#CN9N9YTK5uT_XNTR`hlOq44wuKaU@L zpJRC@5=bk)4fq~ar1q!~>b5>Q)o?swfUkVd#_z;!CV0*Wh(3x#cw1Oon4CD*qbO5N z36d5RKcJ1_)i3(-0iVpE{%r^%t*-6RKlU*l0}l9h(X5C)n<>1**WWtMUw-3W=-Mjn&CkmtwZJ>?&e5)nV_tddY|qEJQJ<&sXXmkHs9a6%Jnda2 z$meTz%tuQ?h`>IY!y94bgRsnlC+ePqPbOK#{uJ8f_aWu%LrAh_wq7}%KDaU(>th_j zw2OfdQE)>Nwuamj0W10k1sY7}vM8DgaXdOF(Yxl}vc@>Pf9>d; z${y)(bZs2~EsB#0?>}`0gZw#E`tu_G>=3kIrEfN% zy z!myi5T4OY?ysTKoYKG-(#v9MdTg%h?sP!Af%yu_cy!+;{{Hf@m-$$C!6vC_BGmsee z)j}9X{bVv}<*?0RVi4mWNFgG=hJjJK#}_A(4qsk(mQU5TJ?m|pLR#sT;X$Md$cugv z2#74!`*9+o3kk;cy}n$Az8}{NJojTiPJ@%?hlYC(Wc^UZ<;b|zEDgFp4szP6t zaVk+_zgSjBApTL#|t1)d{jm?1V2 zyy6`AQ}3b`Ujp&rr^~ENAPoA=0%XZQ0!eg13BSJ}Wka6u&S^-RHGGW5E)HvGtS_Mo zX2p?3H@xzWG1O=8*1x>|M`kKzp>t#t1Ag-ctn7DNQ)MWNqi7b&DySn^vvji=>&9B1`APeokUN-pnClCg&X4@vi+D6bwmV|+ox8R^+z$d z6bd1M&&~s*oY$@+(*=I13WR<@^W(4FIll70zL1-!`^Wlzexo-xx?EjIVtvVP1R+CQ z8yf)q~wD_v2s43B?(|Y~p`C z=Wi7A``Z!y+nnJ)dF}6e|9#ls>8ULsp4`8_DdnFuvoQmf{)>PAKId=TQjUVNeX&lY zKX-zM@?Z82J=;nWU=)(ide>e!+nwuYTT3!_3iBW5{6%X0{nRZ^r~kKGRKrNF&7qtTD9OKW z*#Cu4`u*A>=JpR7#d?60_(!4SzAX9+Q}cine0S}~$iKbte=n4-{6}(vQ~8fa1fP-o zPeMta5P_6B{ZHg9SO1T5{#QZwneSq<9{`zU7xQUAIcX36KUY=A@TcNuk5f7jFdH z{4eYE=exC^0^JJun$G^8R{HC8`D=|#hoJMpp!2~HIXe0C*D&F)f_mNf15XFZt~$JP zMgl>k{|M{1o?GR>_IwuP;%)brz#Z39-FI5m(42nF=frLPQ`87=Vi>37QFaJOJ wffj|~oUb-)*wAYsFs6N^f9nYe3A^aoM`~auvOz+^ zFKwW$ansNK51J8ib8@H83U=u>rAK@c7q>2p%9M4Q19qHSvS9EYRsBsf-odLoH|kOn zMz!xbj}~!iL0wBw@G|s_5bvoXCJ&5rJo;P|6x6D z&!tx8cTMYbaNzp4bvLnroV}@|wJS^8L(PnSt1F^W_adN7c?<;YSVbI*;8zSh>R#mk z-Nca(^zl!96BCog3vX#=W@avD*Zvp@+*?cY*FLBF_S$73dD<+W2k!0z!^1>%6ZhGU z8H3`#aDKVk7v!=!@0h^YP zaGt)yZJ1x#%pGAxQGqdj>w_i`2rnhI96Fwn6ngCtCLUie$0@SKT|EQQyVL1LR-anFAiArn#A$CVh zh2+d53_Rus6*^V3#C$7tg8FYrB(pCX;i!}=gZ%9Eb(U-8CmX*1g6(5TGE^*8<%3W* z?#`i;ESHCV;1pXjd_ByLLkO*Yb_g??BzbIMQ@i?gVkMlE?g0OshTuXYGio5M?vAHL zP7uXhV4PxShvlWT@R;tf)1{gW=B!`304q=u!KDZ`X+Vg9i}%un+FSZeAqv|h#Qdf# zO*DO102)vEJtOpigwv<^$3&ZBHM~b!YzT?Yf0ho3BMrp#ksc!mQgFF*DsKKc49t6? z9|`su3SShqcpWpl)oTP%I3TqOAh#^PfPiQ{>MxJYz0l`+g99LgV?hc5Rr(JLkm+e@ zgKyLw60mG;J`cRS`b^vGD|yp;7{`&1f82q^amdO@NF*)X*HNuy#$MEW&JdKGr2MMF zyzEUaR_2d9A+@EoRo+i3$6@ux@7EU05=E;N=~B1M5D2-RzGLeAygahzsnJnE>Tw3K zHyIfm`TgWCUTix5&4w{curC+Tv2NL4))y!-dA=gVpQ6`NY10TlU(9m`;;Vd-=}tHt zR=$!1<(AzZRy&enW}47SV7dYj-0hG2{P}a5q?MD)_b54P6(2HvZWr~b<|j)X!AE(4 z1DBZ(--eT4Fc5VzFhE^N-2HsnpYnSglUxG#OBt?c1t{rN_yrPfdi25eQvP^~Hy`QA ztBQU}Iqtyg<5dO_fsK-iit~Jqp*eo@sDXij!1Z>k`PHuaw&rH#4gp0br!B|s7=~BG z_whWUYt-Oe7bz!BO9E>|t=n+G^&D(YfT7bjznuvyd7}F1W+F37DF*JUfi8=IW>O++ zxCt)o`2GjDlv@Z$Y3~2Og{e|=J-xmp+{?XIGJxvm75wMVmbY)e&d7kgNnv3-$1a6? zvFHYaQXwy&YxOB9E)J)E@@t?NhnuQ(yWxEE_f&O$SsDF;lwAvcclawH@Juog#5M#E7))2^T5+nmyll-%K!~am!nfKOt>S6@0ThYDQ<1$DMioE zJ8i-QaHduG@5nQFNguPq`|ID^g~BaIRKMC_z$04cvXoH!B>$ac;MS*?(W_wqI9o2? zw6{;tNhs*_yxIATOq{xw?mx420XqaXNlX31)fgxj1@bWoftCJ@8z?AO4+e_vQsxNU zRK>iD99I`niD2oq$#X@Z+pqzuIYJ7VS~l+vXE}OpG_}wXNspj{jRA~-6Ay>ewdM+} zSiMlMyO1T}Gr?*L>Qgg;O`0Bz8mwNlgBYlE(64<3|7a8b8*Xmxzj?rQ7nl?R*`5Wq z_wIAR8lsKB-q_gJs!hQ~MfwpNvGdX=w0e-x=<9cP=6Ha*3kji*jZ9`Ijvmc&=)jhR z*CzV)V8zN&%mE+0PA8u|4{OvmDF7C7qIsmYYP17(=1Md>_~q-u-mYZ)|; zDjpA0<&nCy#p=%(xM|_Pn3%43gDDB#H~O2+k}uRX_p5?9zW?G7T15#5CqPvlhgr(_ z=r-Bg#nI>auZwTOU|DoD<4Rv(sir4(berkov41VKEd=oqmUlGJ^z+9D+xx`ECrfY? zwa_ab!IBhbX#3Vj$ydZ&t=uiO|IBPd50nCFID}kk-S{|UY++E(*$%t;DpMJWJI4+x z13=g1h@HoavdquoL+!Rg`OXPV(E5EOtbVGx`v{k!VcX`%w6edykZ8mkb*i*nb#-fx zT9@zZV>GrwstWePUU^D zb+=%_IZ;*2`$FEMZZ34l(YINrQ?HJLq8@UlXddrL)I`5wKtRCDgv9Gxu{{l&$J@is zr@)mn^UzMhyB=l1NmlkZx*3fedCqW^C$d1``tBlg%Tcu(Oo(D2PXkW-8*_K>-esP8 zUt!n!GL(3{q%NwMtbydF1G+FaHr7ErSqIxKEG$M=9EgV*J-V9ch&fHPZX$ns$VmXL z`vO+|M8YRvZ^l;7h}AGN$x#=5n?;b63}2B^IS3 zRq^FEjha}a3>gD=5&7Wc(Hzx4IX^wUXS@TK02CZPG~XL1S`}YYQ&UhNLTjwU_Ohr* zOdci)JiEWY|6`GFk8O%rh{$nHtmhh|s9#$*dh+?y)QywN;ES7@_Mf~t%BE;A zwYI+YiI5_ zC9%rjS2#NNqx%+=19!`MER)jHciY#4Z;&<0fv|B8jSp)=U^5&-xS(JBx&nzOd-B^O zp_yL7_gO-Umlai*cO9UwwP{n1brpZ|~~%-l*z za8X)WSy4Q7IdM}z;_`v|u4VUlhC{y$mN$Ms9P|n?Ckx!~VhnkioGcU`hk-veM<6zO zWkXEK0-KK3TCv8N|4eU>Xfeif%I9IRdQL_#Q|U`%+tbeZ0DgPSNcIlX7#=TD&2W4) zOMIW5@AsqV00C(=y>TgY9t9C;2ybbD4jgREbu+hu2uHPX`thecoj8#yQW0B2!;iob zi#%n*_XniQI;eS0U!!&Zan51?-S028qK`np5rFUR@Hm)>1%mI`VAJTCHf4u^tOAm_ z)I)+aG{Spaw|_)3iiwK0?|lzsZaqi%Nu}sM6~?O?d|2006tOe*O5WMlKh9PIJ^fZ? zy)E)QA)K23<C zfb3z+o_;R7MRiIyJL5a6>ONJ~I2&T)lmX6*4hW$5dI>f7L#gwC!8MJ>Ul%?79>}=4 znc0AUSkKBhw(kYklhyJ@`;MSnip?B^pNc~{swyGJ`v4K*k~-1UCMRgiVCS9u)AjG= z_O|T-3bL}Y3JUe*&ro!q!GxCoSUuLb46L6X0m6mli9S6 z(&@s$*IX9f$A55kxneKejFv(Wd}%f{gAb+4EWfdk5`MxF86BrPmBLvT@O)T*!c?vM zm&?4>fSJn52itFm$Ik>p#8sF{2tRMj*tJ}Beoz1NYBG-x64{@?nUa%}&@EW)*|z!q zirdhK&plz#=J$@=HuF;9P_xql)vk-va>^-YiNO_-+%Z}g*kiDA+$X&Zyvlj?+KG^ z``_f>aSUGI&Z%nWiRuxz`Q=xUzjm58T~l-O$oYIVsSlK_Y-~XFKQ`YwkE}th zt*rsCevuL|{&i+%W>QkFVR)hA>9siU(hD!xrEmX?aMD4Sk;()wN;>op4z|z5@VMB` zvt!s_Yw6!@HO2RHe=fJ7&e%{lemr6)0)gDfQvAa+yC0KXjz&OK?zy}7^nMFQ_6y*} zv-2lb#R)Fdl2#Qx0R~$`?JjO^!)O(&p)Z=a6u#-F-#l*Jx|Qs+M!CpANmr_^?C}i) zXAk@PI2emXoaD3%c)`(9BrDM(S|tW!Kqg$y;rq#ODM0VfVc-ogwA9-fC|ZM^J#Zq^ z^6yFXqTm;%Z1L~&^mqx+t7UsWJ5T*TE9vv{`mX*|o%j|wRcasDzzeOryOI53UV7+< zE4t*U!baNeT7S4%18ZO_cf_9^?A;738QD z$Coiea;fq6m%GidB+nlV?N4&0@?{6E&(XET=vs^JOp5)G_J%wPVP`6WGgRi@tfRQe zoc58SbtEk~d&I6X?c+#$`&9UQ6QpMitqwKOew9avo87r27q5gPooL0SsFUxyJB`lK z1zFf1IN!{|KI(*jm;L6-;)TH&;eo4G+?-Atc)Ba_@B|SRIaoS31zI15#T=z4t=_b>Ns?AEHt_P|2 zMV`YurWw*l{Nj6)WOurY;eM*iRd`wvQBl?7jc#U&Tp|NhD*pH+-1J$4EjGn;?=P$C zpqP-*n8Bj_B~&NRDeWHzH`psEfI^a8iyq8UGP)2YK8=AxQTh~fTr72Sr)@3ZPW6)4nIFsT2P=pG!Azi z&iOL{DS92o<3qL_{c8vv_mljCW?krRo)lT*%u}8RmD#!IzkmN06c)~e5>J?0`7)L% zsuIr9n8J69K^6nzGBUdF7fO5NNw)iL!9)VzY_2Z=TZXIaJe;T$d8)ED+RycJ@dM+x zxESY~78Y}FRsn|K7y>I7UP8#S!dBUJfRh%OWL7S{ME+=aeSJN1$`9Zkknf^sjnbu( zu90HkOyuZ~pFW9R>{9mIxOiG_AGE21VgXy^@e;cs06DwUn5Ej=hwQGwG9Y7`4qkxmat3x%u__*E~-H|H>_?O(id<%*J$)x35D zRL9uLYW7L2pOO$~sZ}LE@K(UnxS?2R39Jm`*c)%;ont6^5@TXIma;{M*4wCH=)Adp z$Q$jk?Cbt!PLJa|LAblnm)N;dlm)z!NES0a*tonmHa?yqZ6hYX@+Wu+?JEY)jtx!k zJo)2QBe98ro04T%V}i2Pf}xU<4}X40N>^()=D@T+W@ss zx~E?EhY|UlzdpinV^&5TgvxGe+nIPg-pV1^LQzx>e_#LS;^JKw7ZLYQOFIGTo2VPW zD|8-jeX5bD&V_4f)(2g29att1l1)%ui=Q$&k7m~MuHASgH}|6}vPNMJfashAuWc=U z5`~itWh-9+acuSu5b%(-Y2`15xDMJF-v z0L0#)@A`M?O$FQCz+jO0Sl7Cmn3*y2v^gRm9u%C>6RBAtL#sOhXMXZdcgze>UsMBv zUVu=L^6s7H8&hj@fIIGw<#uA|u`{^;TIF!aZY}B{u6#}l?Gy27_wUMhnkj)@2X^Gf zArP`Z#~8Zb4al#ns}msZ_gPigCh>{T0Gn9S7pc#q=>FC!kVt=;W@X<$H1u8~Xs6V{ z!^SnW^vwJV1UwWylT>4_mPx z$Zaz+B-(35Bwq8L*Y@__th&<;YADav)IVj2 zJm_D!y?vve1H^Ws#6m(s;?)7jIAm>`E~@B@Ah6b^R#wl_y$JZ?@gfs*0l5W{*~8C` z4PLYM=H|k+&HQD{L5$AO(9<9TgZ@}1VNCF?TU=fdjFbdc7CfYhbUBJrm0j@smAfBP zGq1Qgk?H8@kP1)`j`$MxWh@e{NI^ZP>sln@9e{eF)=$i z?5wyB27$Plg%?JSy<`l-{oUZu zr7iLyv<7aP-Zy*n$K@g*d2YGju#qHHj`ZETY)({!St2}g;kw~OQEkRZLKM7`=%u-zfgxR=DYhf=+G7*H z%e`v`0Xf!JnsW14r+J5Z$5eW1i>LMb#^(F!_WP%VotN>0IiM2g5%ok;DB+(s1zis4 zebDA06fkrnKlb+p?&6e z0+Q;LrXNKbI%(;{;W>&PUqO=Qk}l#3gt5%7RZ&UF7nsK6PgF0_Xoh}rwe*-CGD1Nn zh@btpe+dRd;Mw9Uyy}if;19(jYb3UXtD~SR)Ve1>A?g~;H{Pi68A zMH-3qVc^wy)z#H)Z9ywzg&*!Oh3kAum)0_k_;y*4IS`kUYuo&oa^`D(qS__-YEps= zC>QC1Spih%Z}G&D4{2ku#(?*SP`^`no2gK13ydI`L93&$pr z(BANbgoLp%vxBm@+-lGfdi|Q4E2D>p@bl`2)3&xYY-WnP-i|{Y!7Z=kC zdwpv#0UaVx@k}T*%E)_hugvHagN!mgo*vrQMDvoNImh_oGYdlL-|nZSr9HmR%!o_T z!gyYs;o;#C-E)5X9Hdv3_R@XAn_*C83Idf+a!=w}g9kX)e}S)IQf{mDZ|Yyiz|U)- znL*}x0WZPyKZ9K4*SgN}g-xHeS&>Zzf=adb_@*1X;dM^D1VHu~P5HR@D|2fLi{<8B zvoL4UvnuQSdF10*jV#pz%Ag7_^rUoE=FhI4+jeQYMf;^@y+w=5qjr zmL@GUuTZuPlUj71_uy2jl<% literal 0 HcmV?d00001 diff --git a/docs/docsgen/source/intro/images/linreg2.png b/docs/docsgen/source/intro/images/linreg2.png new file mode 100644 index 0000000000000000000000000000000000000000..5eee0d5cbb0e7165276218eff435c04d4e849310 GIT binary patch literal 27462 zcmcG$2UJthwk{kSii!;d0V&dyF1~#r3fOOgx-Wuq)7`U6cwfS z79a#f2t6QO0)&4@&%NjV@7??EJ!8C=F<>XVuD$kJbAI#t=G+nbx@uGx=`VsnASw-Y zWdjiC3>*YHZFPYHShG>GRRb(ec^IfY1eNvOSq6TbwNunl1c54IE*(8P2mGddp>FB{ z0?}}g|4wNbu>S&q9I+b8ibgLj2{@`(x9}PJCsvN|$)Eguyj$5|%JHx!C8gIerI2f% z-yYK7ErX$PPdJdVv?IG#K7|5z^;NP3FhS@N9&wh3d=XQ8KV1^1Yh*&R@8T|UXnicw z{ATPk`K^*?^t2-hvyqx&@|U11Y(}TaUrztuT`Ihh@>prr}@*@jM>U`?e@m#-#ryJ`p0 zi9BoI@}Dg8Hr5QDMO^=Ryg!p9KK)Y3(g%Ob=Xe)G@D(w4`+jQ+Dd4Zs_MVOnW9e6F zq4uh%yxd!BQ1~R&x9ExBcZ6Lse!44a!7IeV#J7!WAO7x+`7Idw-nQFz(wDJ`^;GM* z4BLJ`GuX5HvTT)(E3XVczTkei#@iF$TwCIXNnNrMuM)0u~3W_G={J@8~wYW->06PiLDQc=oaFYfyC?*$=bPUB zy0f7`g3<$=5$95_A9Zi;DKTAA1V=aJY64NK5)Xwn1OC>y) zVtbMWl@@wZ*cil|V`olC`|g`RzZdfQ^2+8jf%|h7ANGHER7?u(?b8eqk%8Ez{<_I| zt!LWtt4cm4m!0-~LJ$`Z_vxl`VG$xk*8(EfLh)tK=THkC_HjXkr8M>9Y>Ch9-D!`V z&As5Xtm*a|UwP;Ek(rOr@7B>Deej!#9uT&3d;sNOIRgs%CemTkIzZDcEZ`7{R}-X6 ze23_IFjiv9#=*f6pPI_;jq^$}2mY-imV6KJN%-Z0EX3rb83X@ff6SVeSc9cWt%%1Jm&Kxd7xk!q<$J-MiPzq ziP%Z~V|%BPljLdt%T3Ir{g;al<=?Aw+jpyX{q}#$8-n@Z&>At~hDn`uNqs{y0vJ{| zp4dl1qA6n`?PY6Q3Ptusmvf#OBr&))J{l{xnt{AFSgtTBw}kd17hjPM7IAz_lp;(B zl6q@c+dEV#v4{j25YqoBTALi}RX}n5o~p{w>5mo8i`JZXqD1d@UKnzxSGElEaqIUn zx9V$zbeDD>G*}k3)h~#IbQv%DDz)afReVSK2Bz=dDjzspID$;l`!Bvv(RsG{&27=r z8_`7UwAlAn&s)p-WZl;^h8Fsn*0=H5`sDKU`=>za_BnhANO`h=w%~=tclyg7Kfh68 z`F44%4&L#aRdx$F-@RhZX)#u%jw+Lg~!?S^m(lQRx$4|EHQ^U@i%%HIuqv2NwGW6lBIF^|4J2YuzLTK=%z!FOLV>BXGPakWa# z^U&=WioF$Aw>K^(oxJxQVE)r}qXU5o0~NUqO}W;(r$7p?hWS6CwYru0Mf?k}{y*6y zD>kfjI-S>8Lo}H)W*?*T99N%}Soq-K@o{l?@9AZFI>nZVr0>Z3nA6B-TWwW2YfTQe zgR^Atb;oS8XN&E(-wu}jyo#KDsYw-V$ULBrYYctl^V;s6X(MQ5W6KF^fGbPSz=u9K zsrPJrW`};`+rG~{P6m0DVSnYrj2g}S3@xSFZf&TDMEE~16B zSQpi@`x?jO8YOAzL_HnyPPHR4F$^=H_Tr_;1rr3nOXJlBWnYVZ$NSNW*(0m_R#r=S z3R18k%UWp)Ol}p(yFE)D1iHqU^Lk5qj@TU@;jMObr@N2UsG+gl*WT%w_hw%&o2qX# zhm_(w3>7RAWr|wMU*7o8px0`bTviYl()!}4G+QZb#i<+7+INu(X~;Z0Fh=wdQ6>h* zP`CtYrTQ-p+(ng`Se|Sw2n;-EiA%kPa_lGxMTk@nuPw+PB$+uML$fuo1 z!YfvM>2;U9P~yX`CWF$#R#~BRLuM<1Mmi=nMu6ch+dPb zdnOjoMJpmt>Jd3P&IQxLGh!R`$5~Lj9Lk@ME_Ty5J+uqtYtriT0(TLocyMNA}wYk!q=eW@9*5Qa|VDV5rBicGIXK-8J-}t zIO75r?&o*sw(xit$R{B8>s7YGAD9ynu21)w5UlKi`(1csv$t*Ao6_LD%agyZCf1d>v!H(BurI-IioE?TixKQ=ZF_jh(G0` zJ=l|31o>z7B6;TzcKzdOky8zu-)$52$g36f6bmn_XK5T9t*N*l?jt&KhnsEDUr5Bk zw5d|^4g6SjWGXwQCw;mAlCW3LgM#IzK<(QDNOapVV@M1XD{(!`416jyaJx2PrES?xYA>hrOp4 z)f_7Y+nup2S*j1%O2``bSpCNd)Eyp73B9Mjrz=8%uJ>5%iV~l;@3v3@tB8L4xR@1a zSyuP+B3O0*BN>yD&pG+9@Ul`cKKQEK{kK<{5jje4ySy+Kc7ZGGQiCtnrzEjb2V30& z*sRHiMT^#NGJh-nLEfxNh0aPE8WD+b_{58eTO!&kw>29Y+n67cLxO8`4Gpx^*xG)Qa}>-Cg24P%{kN06|B#q?^DqIN`uhSLI8wnT`S=#ORo8Ck z-1vXI6zXz!cc1!m6?XZDkeg(j0|M=c{EYwg!m2nL7{V-r z#|P$3uZ%Apmw#FvR0ev&I5G2jXHU97r5CDa`XgSOt4Y9c%y_i>l_5eHxt82b2I>hao}Ciu?ja2CVZ;?k=S$*UNeRG1gYJ# z#B)nXY2~EJ=f|Yo*V)Tl*E^H#F)?Wnynh;U>(pYCbZ@05P1;k*@6K*-@JF;O_?L?L zMhJe?AuoSvV$bRAK9`FKH22DUzUd);?-U#4q@>=!q@o))G{yGT^D@M2?nYU03CbXn{}{xrvHwJNUkct*y+9hA+P6G$Wk3 z;Ei8-4J-YUe{faWE{HA&{3?Pqm7CJYKX`1+C~|0Y3ybJ|ck6ql^32?NuA*4+i-<%Q z!C8?!m8#<95y?YZ*Z8=EErKs8FRs#IGr(Jmn$g2FEs0YU!jqSuuWBjOy|b&-OGDsa z9q@jy<)@6Y0Nd{m;-4r+Ns>J6&I4vvrQ?U3u;{=#A-7 z?{$Ox*wH;<28t;3D&3U|N5n`J)5j7m|< znAo@?;vy~Gos`$b+Y?N&>3~W3(=vq?p3FOY2oVvHaTQHn&CVkkL%%MU7-f;@Q!Elx zZ*S3G!SR2BoLCA-TJtCqt?V%HsD$IUbo17e1VR7T4G8uCD>{zm>U8xoWJdVhN5MuifH8;{b)I(KyLT62i&BzfiQ z>OyM|xfmGFU#`NV%tfn<@IWiRSx#(te#m6p$|^saDqW@z>5}-h_S#Tro