chore: import upstream snapshot with attribution

tests/functional_tests/ASR_dev_run_MultiTalker_Speech_to_Text.sh

@@ -0,0 +1,32 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+#!/bin/bash
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_transducer/speech_to_text_mt_rnnt_bpe.py \
+    --config-path="../conf/conformer" --config-name="conformer_transducer_bpe" \
+    ++model.train_ds.input_cfg=/home/TestData/an4_tsasr/simulated_train/msasr_train_tiny.yaml \
+    model.train_ds.manifest_filepath=null \
+    ++model.train_ds.use_lhotse=true \
+    ++model.validation_ds.input_cfg=/home/TestData/an4_tsasr/simulated_valid/msasr_valid_tiny.yaml \
+    model.validation_ds.manifest_filepath=null \
+    ++model.validation_ds.use_lhotse=true \
+    ++model.test_ds.manifest_filepath=null \
+    model.tokenizer.dir=/home/TestData/an4_tsasr/tokenizer_bpe_asr_phase1_en_v1024_beep \
+    model.tokenizer.type=bpe \
+    ++model.spk_kernel_type="ff" \
+    ++model.spk_kernel_layers=[0] \
+    ++model.add_bg_spk_kernel=true \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_to_text_results

tests/functional_tests/ASR_dev_run_Speech_to_Text.sh

@@ -0,0 +1,21 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+#!/bin/bash
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_ctc/speech_to_text_ctc.py \
+    model.train_ds.manifest_filepath=/home/TestData/an4_dataset/an4_train.json \
+    model.validation_ds.manifest_filepath=/home/TestData/an4_dataset/an4_val.json \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_to_text_results

tests/functional_tests/ASR_dev_run_Speech_to_Text_Hybrid_RNNT_CTC_Prompt.sh

@@ -0,0 +1,30 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+# Functional test for Hybrid RNNT-CTC BPE Model with Prompt
+# This test validates the model can train end-to-end with prompt conditioning
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_hybrid_transducer_ctc/speech_to_text_hybrid_rnnt_ctc_bpe_prompt.py \
+    --config-path="../conf/fastconformer/hybrid_transducer_ctc" --config-name="fastconformer_hybrid_transducer_ctc_bpe_prompt" \
+    model.train_ds.manifest_filepath=/home/TestData/asr/prompt_parakeet/multilingual_train_target_lang_field.json \
+    model.validation_ds.manifest_filepath=/home/TestData/asr/prompt_parakeet/multilingual_dev_target_lang_field.json \
+    model.test_ds.manifest_filepath=/home/TestData/asr/prompt_parakeet/multilingual_dev_target_lang_field.json \
+    model.tokenizer.dir="/home/TestData/asr/prompt_parakeet/merged_universal_tokenizer/" \
+    model.train_ds.num_workers=4 \
+    model.validation_ds.batch_size=4 \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_to_text_hybrid_rnnt_ctc_prompt_results
+

tests/functional_tests/ASR_dev_run_Speech_to_Text_RNNT_Prompt.sh

@@ -0,0 +1,28 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+# Functional test for RNNT BPE Model with Prompt
+# This test validates the model can train end-to-end with prompt conditioning
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_transducer/speech_to_text_rnnt_bpe_prompt.py \
+    --config-path="../conf/fastconformer/cache_aware_streaming" --config-name="fastconformer_transducer_bpe_streaming_prompt" \
+    model.train_ds.manifest_filepath=/home/TestData/asr/prompt_parakeet/multilingual_train_target_lang_field.json \
+    model.validation_ds.manifest_filepath=/home/TestData/asr/prompt_parakeet/multilingual_dev_target_lang_field.json \
+    model.test_ds.manifest_filepath=/home/TestData/asr/prompt_parakeet/multilingual_dev_target_lang_field.json \
+    model.tokenizer.dir="/home/TestData/asr/prompt_parakeet/merged_universal_tokenizer/" \
+    model.validation_ds.batch_size=4 \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_to_text_rnnt_prompt_results

tests/functional_tests/ASR_dev_run_Speech_to_Text_WPE_-_Conformer.sh

@@ -0,0 +1,25 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_ctc/speech_to_text_ctc_bpe.py \
+    --config-path="../conf/conformer" --config-name="conformer_ctc_bpe" \
+    model.train_ds.manifest_filepath=/home/TestData/an4_dataset/an4_train.json \
+    model.validation_ds.manifest_filepath=/home/TestData/an4_dataset/an4_val.json \
+    model.tokenizer.dir="/home/TestData/asr_tokenizers/an4_wpe_128/" \
+    model.tokenizer.type="wpe" \
+    model.train_ds.batch_size=4 \
+    model.validation_ds.batch_size=4 \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_to_text_wpe_conformer_results

tests/functional_tests/L0_Setup_Test_Data_And_Models.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo -m tests.setup --save_dir /home/TestData/nlp

tests/functional_tests/L0_Unit_Tests_CPU_ASR_1.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=0 --num-shards=6 \
+    -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_ASR_2.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=1 --num-shards=6 \
+    -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_ASR_3.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=2 --num-shards=6 \
+    -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_ASR_4.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=3 --num-shards=6 \
+    -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_ASR_5.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=4 --num-shards=6 \
+    -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_ASR_6.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=5 --num-shards=6 \
+    -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_Audio.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/collections/audio -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_Common.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/collections/common -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_Core.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/core tests/core_ptl -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_Hydra.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/hydra -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_Lightning.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/lightning -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_Others.sh

@@ -0,0 +1,29 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat \
+    --ignore=tests/collections/asr \
+    --ignore=tests/collections/audio \
+    --ignore=tests/collections/common \
+    --ignore=tests/collections/llm \
+    --ignore=tests/collections/multimodal \
+    --ignore=tests/collections/nlp \
+    --ignore=tests/collections/speechlm2 \
+    --ignore=tests/collections/tts \
+    --ignore=tests/core \
+    --ignore=tests/core_ptl \
+    --ignore=tests/hydra \
+    --ignore=tests/lightning \
+    --ignore=tests/export \
+    --ignore=tests/deploy \
+    --ignore=tests/functional_tests

tests/functional_tests/L0_Unit_Tests_CPU_SpeechLM2_1.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/speechlm2 \
+    --shard-id=0 --num-shards=4 \
+    -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_SpeechLM2_2.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/speechlm2 \
+    --shard-id=1 --num-shards=4 \
+    -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_SpeechLM2_3.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/speechlm2 \
+    --shard-id=2 --num-shards=4 \
+    -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_SpeechLM2_4.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/speechlm2 \
+    --shard-id=3 --num-shards=4 \
+    -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_CPU_TTS.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES="" NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/collections/tts -m "not pleasefixme" --cpu --with_downloads --relax_numba_compat

tests/functional_tests/L0_Unit_Tests_GPU_ASR_1.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+python -c "from nemo.collections.asr.models import ASRModel" && NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=0 --num-shards=6 \
+    -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_ASR_2.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+python -c "from nemo.collections.asr.models import ASRModel" && NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=1 --num-shards=6 \
+    -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_ASR_3.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+python -c "from nemo.collections.asr.models import ASRModel" && NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=2 --num-shards=6 \
+    -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_ASR_4.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+python -c "from nemo.collections.asr.models import ASRModel" && NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=3 --num-shards=6 \
+    -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_ASR_5.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+python -c "from nemo.collections.asr.models import ASRModel" && NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=4 --num-shards=6 \
+    -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_ASR_6.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+python -c "from nemo.collections.asr.models import ASRModel" && NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/asr \
+    --shard-id=5 --num-shards=6 \
+    -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_Audio_1.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/audio \
+    --shard-id=0 --num-shards=2 \
+    -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_Audio_2.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest \
+    tests/collections/audio \
+    --shard-id=1 --num-shards=2 \
+    -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_Common.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/collections/common -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_Core.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/core -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_Hydra.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/hydra -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_Lightning.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0,1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/lightning -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_Others.sh

@@ -0,0 +1,29 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest -m "not pleasefixme" --with_downloads \
+    --ignore=tests/collections/asr \
+    --ignore=tests/collections/audio \
+    --ignore=tests/collections/common \
+    --ignore=tests/collections/llm \
+    --ignore=tests/collections/multimodal \
+    --ignore=tests/collections/nlp \
+    --ignore=tests/collections/speechlm2 \
+    --ignore=tests/collections/tts \
+    --ignore=tests/core \
+    --ignore=tests/core_ptl \
+    --ignore=tests/hydra \
+    --ignore=tests/lightning \
+    --ignore=tests/export \
+    --ignore=tests/deploy \
+    --ignore=tests/functional_tests

tests/functional_tests/L0_Unit_Tests_GPU_SpeechLM2.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUDA_VISIBLE_DEVICES=0 NEMO_NUMBA_MINVER=0.53 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/collections/speechlm2 -m "not pleasefixme" --with_downloads

tests/functional_tests/L0_Unit_Tests_GPU_TTS.sh

@@ -0,0 +1,14 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 NEMO_NUMBA_MINVER=0.53 CUDA_VISIBLE_DEVICES=0 coverage run -a --data-file=/workspace/.coverage --source=/workspace/ -m pytest tests/collections/tts -m "not pleasefixme" --with_downloads

tests/functional_tests/L2_ASR_Adapters_Linear_Adapters.sh

@@ -0,0 +1,24 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_adapters/train_asr_adapter.py \
+    model.pretrained_model="stt_en_conformer_ctc_small" \
+    model.adapter.adapter_name="an4" \
+    model.adapter.linear.in_features=176 \
+    model.train_ds.manifest_filepath=/home/TestData/an4_dataset/an4_train.json \
+    model.validation_ds.manifest_filepath=/home/TestData/an4_dataset/an4_val.json \
+    trainer.max_steps=5 \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_to_text_adapters_results

tests/functional_tests/L2_ASR_Adapters_RelPos_MHA_Adapters.sh

@@ -0,0 +1,25 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_adapters/train_asr_adapter.py \
+    model.pretrained_model="stt_en_conformer_ctc_small" \
+    model.adapter.adapter_name="encoder:an4" \
+    model.adapter.adapter_type="tiny_attn" \
+    model.adapter.tiny_attn.n_feat=176 \
+    model.train_ds.manifest_filepath=/home/TestData/an4_dataset/an4_train.json \
+    model.validation_ds.manifest_filepath=/home/TestData/an4_dataset/an4_val.json \
+    trainer.max_steps=5 \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_to_text_adapters_mha_results

tests/functional_tests/L2_ASR_Multi-dataloader_dev_run_Speech_to_Label_multi-dataloader.sh

@@ -0,0 +1,29 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/speech_classification/speech_to_label.py \
+    model.train_ds.manifest_filepath=/home/TestData/speech_commands/train_manifest.json \
+    model.validation_ds.manifest_filepath=[/home/TestData/speech_commands/test_manifest.json,/home/TestData/speech_commands/test_manifest.json] \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    trainer.max_epochs=1 \
+    trainer.max_steps=1 \
+    +trainer.num_sanity_val_steps=1 \
+    model.preprocessor._target_=nemo.collections.asr.modules.AudioToMelSpectrogramPreprocessor \
+    ~model.preprocessor.window_size \
+    ~model.preprocessor.window_stride \
+    ~model.preprocessor.window \
+    ~model.preprocessor.n_mels \
+    ~model.preprocessor.n_mfcc \
+    ~model.preprocessor.n_fft \
+    exp_manager.exp_dir=/tmp/speech_to_label_results

tests/functional_tests/L2_ASR_Multi-dataloader_dev_run_Speech_to_Text_multi-dataloader.sh

@@ -0,0 +1,22 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_ctc/speech_to_text_ctc.py \
+    model.train_ds.manifest_filepath=/home/TestData/an4_dataset/an4_train.json \
+    model.validation_ds.manifest_filepath=[/home/TestData/an4_dataset/an4_val.json,/home/TestData/an4_dataset/an4_val.json] \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    trainer.max_epochs=1 \
+    trainer.max_steps=1 \
+    ++trainer.num_sanity_val_steps=1 \
+    exp_manager.exp_dir=/tmp/speech_to_text_results

tests/functional_tests/L2_G2P_Models_G2P_Conformer_training_evaluation_and_inference.sh

@@ -0,0 +1,34 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+TIME=$(date +"%Y-%m-%d-%T")
+OUTPUT_DIR_CONFORMER=output_ctc_${TIME}
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/g2p/g2p_train_and_evaluate.py \
+    train_manifest=/home/TestData/g2p/g2p.json \
+    validation_manifest=/home/TestData/g2p/g2p.json \
+    model.test_ds.manifest_filepath=/home/TestData/g2p/g2p.json \
+    model.tokenizer.dir=/home/TestData/g2p/tokenizer_spe_unigram_v512 \
+    trainer.max_epochs=1 \
+    model.max_source_len=64 \
+    trainer.devices=1 \
+    do_training=True \
+    do_testing=True \
+    exp_manager.exp_dir=${OUTPUT_DIR_CONFORMER} \
+    +exp_manager.use_datetime_version=False +exp_manager.version=test \
+    --config-name=g2p_conformer_ctc
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/g2p/g2p_inference.py \
+    pretrained_model=${OUTPUT_DIR_CONFORMER}/G2P-Conformer-CTC/test/checkpoints/G2P-Conformer-CTC.nemo \
+    manifest_filepath=/home/TestData/g2p/g2p.json \
+    phoneme_field=text

tests/functional_tests/L2_Longform_Speech_Transcription_Canary_Chunked_Infer_from_Audio_Dir.sh

@@ -0,0 +1,4 @@
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_chunked_inference/aed/speech_to_text_aed_chunked_infer.py \
+    model_path=/home/TestData/asr/canary/models/canary-1b-flash_HF_20250318.nemo \
+    audio_dir=/home/TestData/asr/longform/earnings22 \
+    output_filename=preds.json

tests/functional_tests/L2_Longform_Speech_Transcription_with_TimeStamps_Canary_Chunked_Infer_from_Audio_Dir.sh

@@ -0,0 +1,5 @@
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_chunked_inference/aed/speech_to_text_aed_chunked_infer.py \
+    model_path=/home/TestData/asr/canary/models/canary-1b-flash_HF_20250318.nemo \
+    audio_dir=/home/TestData/asr/longform/earnings22 \
+    output_filename=preds.json \
+    timestamps=True

tests/functional_tests/L2_Longform_Speech_Transcription_with_TimeStamps_Canary_Chunked_Infer_from_Manifest.sh

@@ -0,0 +1,4 @@
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_chunked_inference/aed/speech_to_text_aed_chunked_infer.py \
+    model_path=/home/TestData/asr/canary/models/canary-1b-flash_HF_20250318.nemo \
+    dataset_manifest=/home/TestData/asr/longform/earnings22_manifest_1sample.json \
+    output_filename=preds.json

tests/functional_tests/L2_Speaker_dev_run_Clustering_Diarizer_Inference.sh

@@ -0,0 +1,22 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/speaker_tasks/diarization/clustering_diarizer/offline_diar_infer.py \
+    diarizer.manifest_filepath=/home/TestData/an4_diarizer/an4_manifest.json \
+    diarizer.speaker_embeddings.model_path=/home/TestData/an4_diarizer/spkr.nemo \
+    diarizer.speaker_embeddings.parameters.save_embeddings=True \
+    diarizer.speaker_embeddings.parameters.window_length_in_sec=1.5 \
+    diarizer.speaker_embeddings.parameters.shift_length_in_sec=0.75 \
+    diarizer.speaker_embeddings.parameters.multiscale_weights=null \
+    diarizer.vad.model_path=/home/TestData/an4_diarizer/MatchboxNet_VAD_3x2.nemo \
+    diarizer.out_dir=/tmp/clustering_diarizer_results

tests/functional_tests/L2_Speaker_dev_run_EndtoEnd_Diarizer_Inference.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/speaker_tasks/diarization/neural_diarizer/e2e_diarize_speech.py \
+    model_path=/home/TestData/an4_diarizer/diar_sortformer_4spk-v1-tiny.nemo \
+    dataset_manifest=/home/TestData/an4_diarizer/simulated_valid/eesd_valid_tiny.json \
+    batch_size=1

tests/functional_tests/L2_Speaker_dev_run_EndtoEnd_Speaker_Diarization_Sortformer.sh

@@ -0,0 +1,20 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/speaker_tasks/diarization/neural_diarizer/sortformer_diar_train.py \
+    trainer.devices="[0]" \
+    batch_size=3 \
+    model.train_ds.manifest_filepath=/home/TestData/an4_diarizer/simulated_train/eesd_train_tiny.json \
+    model.validation_ds.manifest_filepath=/home/TestData/an4_diarizer/simulated_valid/eesd_valid_tiny.json \
+    exp_manager.exp_dir=/tmp/speaker_diarization_results \
+    +trainer.fast_dev_run=True

tests/functional_tests/L2_Speaker_dev_run_EndtoEnd_Streaming_Diarizer_Inference.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/speaker_tasks/diarization/neural_diarizer/e2e_diarize_speech.py \
+    model_path=/home/TestData/an4_diarizer/diar_streaming_sortformer_4spk-v2-tiny.nemo \
+    dataset_manifest=/home/TestData/an4_diarizer/simulated_valid/eesd_valid_tiny.json \
+    batch_size=1

tests/functional_tests/L2_Speaker_dev_run_EndtoEnd_Streaming_Speaker_Diarization_Sortformer.sh

@@ -0,0 +1,21 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/speaker_tasks/diarization/neural_diarizer/streaming_sortformer_diar_train.py \
+    trainer.devices="[0]" \
+    batch_size=3 \
+    model.train_ds.manifest_filepath=/home/TestData/an4_diarizer/simulated_train/eesd_train_tiny.json \
+    model.test_ds.manifest_filepath=/home/TestData/an4_diarizer/simulated_valid/eesd_valid_tiny.json \
+    model.validation_ds.manifest_filepath=/home/TestData/an4_diarizer/simulated_valid/eesd_valid_tiny.json \
+    exp_manager.exp_dir=/tmp/speaker_diarization_results \
+    +trainer.fast_dev_run=True

tests/functional_tests/L2_Speaker_dev_run_Multispeaker_ASR_Data_Simulation.sh

@@ -0,0 +1,20 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo tools/speech_data_simulator/multispeaker_simulator.py \
+    --config-path=conf --config-name=data_simulator.yaml \
+    data_simulator.random_seed=42 \
+    data_simulator.manifest_filepath=/home/TestData/LibriSpeechShort/dev-clean-align-short.json \
+    data_simulator.outputs.output_dir=/tmp/test_simulator \
+    data_simulator.session_config.num_sessions=2 \
+    data_simulator.session_config.session_length=60

tests/functional_tests/L2_Speaker_dev_run_Speaker_Diarization_with_ASR_Inference.sh

@@ -0,0 +1,23 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/speaker_tasks/diarization/clustering_diarizer/offline_diar_with_asr_infer.py \
+    diarizer.manifest_filepath=/home/TestData/an4_diarizer/an4_manifest.json \
+    diarizer.speaker_embeddings.model_path=/home/TestData/an4_diarizer/spkr.nemo \
+    diarizer.speaker_embeddings.parameters.save_embeddings=True \
+    diarizer.speaker_embeddings.parameters.window_length_in_sec=[1.5] \
+    diarizer.speaker_embeddings.parameters.shift_length_in_sec=[0.75] \
+    diarizer.speaker_embeddings.parameters.multiscale_weights=[1.0] \
+    diarizer.asr.model_path=nvidia/stt_en_fastconformer_ctc_large \
+    diarizer.asr.parameters.asr_based_vad=True \
+    diarizer.out_dir=/tmp/speaker_diarization_asr_results

tests/functional_tests/L2_Speaker_dev_run_Speaker_Recognition.sh

@@ -0,0 +1,24 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/speaker_tasks/recognition/speaker_reco.py \
+    model.train_ds.batch_size=10 \
+    model.validation_ds.batch_size=2 \
+    model.train_ds.manifest_filepath=/home/TestData/an4_speaker/train.json \
+    model.validation_ds.manifest_filepath=/home/TestData/an4_speaker/dev.json \
+    model.decoder.num_classes=2 \
+    trainer.max_epochs=10 \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speaker_recognition_results

tests/functional_tests/L2_Speaker_dev_run_Speech_to_Label.sh

@@ -0,0 +1,22 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/speech_classification/speech_to_label.py \
+    model.train_ds.manifest_filepath=/home/TestData/speech_commands/train_manifest.json \
+    model.validation_ds.manifest_filepath=/home/TestData/speech_commands/test_manifest.json \
+    model.test_ds.manifest_filepath=/home/TestData/speech_commands/test_manifest.json \
+    model.labels=['cat','dog'] \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_to_label_results

tests/functional_tests/L2_Speech_Batch_Size_OOMptimizer.sh

@@ -0,0 +1,23 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+# 1D bucketing
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/speech_recognition/oomptimizer.py \
+  -c /home/TestData/oomptimizer/fast-conformer_ctc_bpe.yaml \
+  -m nemo.collections.asr.models.EncDecCTCModelBPE \
+  -b "[5.0,10.0]"
+# 2D bucketing
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/speech_recognition/oomptimizer.py \
+  -c /home/TestData/oomptimizer/fast-conformer_ctc_bpe.yaml \
+  -m nemo.collections.asr.models.EncDecCTCModelBPE \
+  -b "[[5.0,30],[5.0,45],[10.0,57],[10.0,71]]"

tests/functional_tests/L2_Speech_Distributed_Batch_Size_OOMptimizer.sh

@@ -0,0 +1,94 @@
+# Copyright (c) 2025, NVIDIA CORPORATION.
+#
+# 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.
+
+set -euo pipefail
+
+export CUDA_VISIBLE_DEVICES=0,1
+export PYTHONPATH=/workspace:${PYTHONPATH:-}
+
+python - <<'PY'
+import torch
+
+if torch.cuda.device_count() < 2:
+    raise SystemExit("Distributed OOMptimizer functional test requires at least 2 visible CUDA devices.")
+PY
+
+CONFIG_PATH=/tmp/distributed_oomptimizer_transformer.yaml
+PROBE_LOG_DIR=/tmp/distributed_oomptimizer_probes_${RUN_ID:-manual}
+rm -rf "${PROBE_LOG_DIR}"
+
+python - <<'PY'
+from pathlib import Path
+
+Path("/tmp/distributed_oomptimizer_transformer.yaml").write_text(
+    """
+model:
+  vocab_size: 64
+  sample_rate: 32
+  frame_stride: 4
+  hidden_size: 128
+  num_heads: 4
+  ffn_hidden_size: 256
+  dropout: 0.0
+  activation_reserve_mb_per_frame: 7
+  max_activation_reserve_frames: 160
+trainer:
+  devices: 2
+  accelerator: gpu
+  num_nodes: 1
+  logger: false
+  enable_checkpointing: false
+  use_distributed_sampler: false
+  max_steps: 1
+  limit_train_batches: 1
+  limit_val_batches: 0
+  num_sanity_val_steps: 0
+""".lstrip()
+)
+PY
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/speechlm2/distributed_oomptimizer.py \
+  -c "${CONFIG_PATH}" \
+  -m tests.collections.speechlm2.distributed_oomptimizer_model.SingleBlockDistributedOOMptimizerModel \
+  -b "[12.0,14.0,16.0,18.0,20.0]" \
+  -r 2.0 \
+  -s 8 \
+  -t 0.25 \
+  --memory-fraction 0.5 \
+  --nproc-per-node 2 \
+  --probe-log-dir "${PROBE_LOG_DIR}" \
+  --probe-timeout-seconds 180 \
+  --probe-memory-reclaim-timeout-seconds 0
+
+PROBE_LOG_DIR="${PROBE_LOG_DIR}" python - <<'PY'
+import json
+import os
+from pathlib import Path
+
+log_dir = Path(os.environ["PROBE_LOG_DIR"])
+records = []
+for path in sorted(log_dir.glob("probe_*.jsonl")):
+    with path.open() as f:
+        records.extend(json.loads(line) for line in f if line.strip())
+
+if not records:
+    raise SystemExit(f"No distributed OOMptimizer probe records found in {log_dir}.")
+
+buckets = {record["bucket"] for record in records}
+if len(buckets) != 5:
+    raise SystemExit(f"Expected probe records for 5 buckets, found {len(buckets)}: {sorted(buckets)}")
+
+if not any(record["status"] == "memory_target" for record in records):
+    raise SystemExit("Expected at least one probe to stop at the requested memory target.")
+PY

tests/functional_tests/L2_Speech_Estimate_Duration_Bins.sh

@@ -0,0 +1,36 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+set -x
+# 1D buckets [SSL, CTC]
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/speech_recognition/estimate_duration_bins.py \
+  /home/TestData/an4_dataset/an4_train.json \
+  --buckets 5
+# 2D buckets [CTC, RNNT, TDT] / with tokenizer
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/speech_recognition/estimate_duration_bins_2d.py \
+  /home/TestData/an4_dataset/an4_train_lang.json \
+  --tokenizer /home/TestData/asr_tokenizers/canary/en/tokenizer_spe_bpe_v1024_max_4/tokenizer.model \
+  --buckets 5 \
+  --sub-buckets 2
+# TODO(pzelasko): Figure out how to quote the value in the test properly for CI to accept it...
+# 2D buckets with prompt [AED/Canary, SpeechLM] / with aggregate tokenizer + prompt format
+# coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/speech_recognition/estimate_duration_bins_2d.py \
+#   /home/TestData/an4_dataset/an4_train_lang.json \
+#   --tokenizer /home/TestData/asr_tokenizers/canary/canary_spl_tokenizer_v32/tokenizer.model \
+#      /home/TestData/asr_tokenizers/canary/en/tokenizer_spe_bpe_v1024_max_4/tokenizer.model \
+#      /home/TestData/asr_tokenizers/canary/es/tokenizer_spe_bpe_v1024_max_4/tokenizer.model \
+#   --langs spl_tokens en es \
+#   --prompt-format canary \
+#   --prompt '[{"role":"user","slots":{"source_lang":"en","target_lang":"en","task":"asr","pnc":"yes"}}]' \
+#   --buckets 5 \
+#   --sub-buckets 2

tests/functional_tests/L2_Speech_Estimate_Token_Bins.sh

@@ -0,0 +1,40 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+set -x
+# scripts/speechlm2/estimate_token_bins.py expects a YAML input_cfg file rather than a
+# raw NeMo manifest, so write a minimal one that wraps the existing an4 test manifest.
+INPUT_CFG=$(mktemp --suffix=.yaml)
+trap 'rm -f "$INPUT_CFG"' EXIT
+cat > "$INPUT_CFG" <<EOF
+- type: lhotse_as_conversation
+  manifest_filepath: /home/TestData/an4_dataset/an4_train_lang.json
+EOF
+# Conversation-style inputs require a prompt formatter to populate input_ids/answer_ids
+# (NeMoMultimodalConversation has no plain ``tokenize()`` path) and an audio-locator-tag
+# so AudioTurns get a non-null message slot.
+# 1D buckets [SALM multimodal sampling]
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/speechlm2/estimate_token_bins.py \
+  "$INPUT_CFG" \
+  --tokenizer /home/TestData/asr_tokenizers/canary/en/tokenizer_spe_bpe_v1024_max_4/tokenizer.model \
+  --prompt-format plain \
+  --audio-locator-tag '<|audioplaceholder|>' \
+  --buckets 5
+# 2D buckets [SALM multimodal sampling, input_tokens x output_tokens]
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/speechlm2/estimate_token_bins.py \
+  "$INPUT_CFG" \
+  --tokenizer /home/TestData/asr_tokenizers/canary/en/tokenizer_spe_bpe_v1024_max_4/tokenizer.model \
+  --prompt-format plain \
+  --audio-locator-tag '<|audioplaceholder|>' \
+  --buckets 5 \
+  --sub-buckets 2

tests/functional_tests/L2_Speech_Transcription_Canary_Streaming_Full_Manifest.sh

@@ -0,0 +1,40 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+
+
+# Test with waitk policy
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_chunked_inference/aed/speech_to_text_aed_streaming_infer.py \
+    model_path=/home/TestData/asr/canary/models/canary-1b-flash_HF_20250318.nemo \
+    dataset_manifest=/home/TestData/asr/canary/dev-other-wav-10-canary-fields.json \
+    output_filename=/tmp/streaming_preds.json \
+    batch_size=10 \
+    decoding.streaming_policy=waitk \
+    +prompt.pnc=yes \
+    +prompt.task=asr \
+    +prompt.source_lang=en \
+    +prompt.target_lang=en
+
+
+# Test with alignatt policy
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_chunked_inference/aed/speech_to_text_aed_streaming_infer.py \
+    model_path=/home/TestData/asr/canary/models/canary-1b-flash_HF_20250318.nemo \
+    dataset_manifest=/home/TestData/asr/canary/dev-other-wav-10-canary-fields.json \
+    output_filename=/tmp/streaming_preds.json \
+    batch_size=10 \
+    decoding.streaming_policy=alignatt \
+    +prompt.pnc=yes \
+    +prompt.task=asr \
+    +prompt.source_lang=en \
+    +prompt.target_lang=en

tests/functional_tests/L2_Speech_Transcription_Canary_Transcribe_Audio_Dir.sh

@@ -0,0 +1,22 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/transcribe_speech.py \
+    audio_dir=/home/TestData/asr/canary/dev-other-wav \
+    output_filename=preds.json \
+    batch_size=10 \
+    model_path=/home/TestData/asr/canary/models/canary-1b-flash_HF_20250318.nemo \
+    num_workers=0 \
+    amp=false \
+    compute_dtype=bfloat16 \
+    matmul_precision=medium

tests/functional_tests/L2_Speech_Transcription_Canary_Transcribe_Full_Manifest.sh

@@ -0,0 +1,22 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/transcribe_speech.py \
+    dataset_manifest=/home/TestData/asr/canary/dev-other-wav-10-canary-fields.json \
+    output_filename=/tmp/preds.json \
+    batch_size=10 \
+    model_path=/home/TestData/asr/canary/models/canary-1b-flash_HF_20250318.nemo \
+    num_workers=0 \
+    amp=false \
+    compute_dtype=bfloat16 \
+    matmul_precision=medium

tests/functional_tests/L2_Speech_Transcription_Canary_Transcribe_With_Prompt.sh

@@ -0,0 +1,25 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/transcribe_speech.py \
+    dataset_manifest=/home/TestData/asr/canary/dev-other-wav-10.json \
+    output_filename=preds.json \
+    batch_size=10 \
+    model_path=/home/TestData/asr/canary/models/canary-1b-flash_HF_20250318.nemo \
+    num_workers=0 \
+    amp=false \
+    compute_dtype=bfloat16 \
+    matmul_precision=medium \
+    +prompt.source_lang="en" \
+    +prompt.target_lang="en" \
+    +prompt.pnc="no"

tests/functional_tests/L2_Speech_Transcription_Speech_to_Text_Cache_Aware_Infer.sh

@@ -0,0 +1,19 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_cache_aware_streaming/speech_to_text_cache_aware_streaming_infer.py \
+    model_path="/home/TestData/asr/stt_en_fastconformer_hybrid_large_streaming_multi.nemo" \
+    audio_dir="/home/TestData/an4_transcribe/test_subset/" \
+    output_path="/tmp/stt_cache_aware_streaming_test_res"

tests/functional_tests/L2_Speech_Transcription_Speech_to_Text_Cache_Aware_with_Prompt_Infer.sh

@@ -0,0 +1,24 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_cache_aware_streaming/speech_to_text_cache_aware_streaming_infer.py \
+    model_path="/home/TestData/asr/stt_ml_fastconformer_rnnt_xl_streaming_prompt.nemo" \
+    dataset_manifest="/home/TestData/asr/prompt_parakeet/multilingual_dev_target_lang_field.json" \
+    output_path="/tmp/stt_cache_aware_streaming_prompt_test_res" \
+    target_lang=auto \
+    att_context_size="[56,0]" \
+    decoder_type=rnnt \
+    pad_and_drop_preencoded=true \
+    batch_size=8 \
+    strip_lang_tags=false

tests/functional_tests/L2_Speech_Transcription_Speech_to_Text_Inference_Boost_GT.sh

@@ -0,0 +1,56 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+# Boosting ground truth - sanity check for per-utterance boosting
+# RNN-T model
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_streaming_inference/asr_streaming_infer.py \
+    --config-path="../conf/asr_streaming_inference/" \
+    --config-name=buffered_rnnt.yaml \
+    audio_file="/home/TestData/asr/canary/dev-other-wav-10-boost-gt.json" \
+    output_filename="/tmp/stt_inference_boost_gt_res_rnnt.json" \
+    asr.model_name="stt_en_fastconformer_transducer_large" \
+    streaming.batch_size=5 \
+    lang=en \
+    enable_itn=False \
+    enable_nmt=False \
+    asr_output_granularity=segment
+
+# TDT model
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_streaming_inference/asr_streaming_infer.py \
+    --config-path="../conf/asr_streaming_inference/" \
+    --config-name=buffered_rnnt.yaml \
+    audio_file="/home/TestData/asr/canary/dev-other-wav-10-boost-gt.json" \
+    output_filename="/tmp/stt_inference_boost_gt_res_tdt.json" \
+    asr.model_name="nvidia/stt_en_fastconformer_tdt_large" \
+    streaming.batch_size=5 \
+    lang=en \
+    enable_itn=False \
+    enable_nmt=False \
+    asr_output_granularity=segment
+
+# Cache-Aware RNN-T model
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_streaming_inference/asr_streaming_infer.py \
+    --config-path="../conf/asr_streaming_inference/" \
+    --config-name=cache_aware_rnnt.yaml \
+    audio_file="/home/TestData/asr/canary/dev-other-wav-10-boost-gt.json" \
+    output_filename="/tmp/stt_inference_boost_gt_res_ca_rnnt.json" \
+    asr.model_name="nvidia/nemotron-speech-streaming-en-0.6b" \
+    streaming.batch_size=5 \
+    lang=en \
+    enable_itn=False \
+    enable_nmt=False \
+    asr_output_granularity=segment

tests/functional_tests/L2_Speech_Transcription_Speech_to_Text_MultiTalker_Cache_Aware_Infer.sh

@@ -0,0 +1,33 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+# TODO(vbataev): fix decoding with CUDA graphs on CI for this test
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_cache_aware_streaming/speech_to_text_multitalker_streaming_infer.py \
+    asr_model=/home/TestData/an4_tsasr/multitalker_parakeet_v1_tiny.nemo \
+    diar_model=/home/TestData/an4_diarizer/diar_streaming_sortformer_4spk-v2-tiny.nemo \
+    audio_file=/home/TestData/an4_diarizer/simulated_valid/multispeaker_session_1.wav \
+    max_num_of_spks=4 \
+    masked_asr=False \
+    mask_preencode=False \
+    single_speaker_mode=False \
+    parallel_speaker_strategy=True \
+    sent_break_sec=2.0 \
+    cache_gating=True \
+    batch_size=2 \
+    att_context_size=[70,13] \
+    spk_supervision=diar \
+    binary_diar_preds=False \
+    output_path=/tmp/mt_parakeet.seglst.json \
+    log=False generate_realtime_scripts=False 

tests/functional_tests/L2_Speech_Transcription_Speech_to_Text_Streaming_Infer.sh

@@ -0,0 +1,113 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_chunked_inference/rnnt/speech_to_text_streaming_infer_rnnt.py \
+    model_path="/home/TestData/asr/stt_en_fastconformer_transducer_large.nemo" \
+    audio_dir="/home/TestData/an4_transcribe/test_subset/" \
+    chunk_secs=2.0 \
+    left_context_secs=10.0 \
+    right_context_secs=2.0 \
+    timestamps=true \
+    output_filename="/tmp/stt_streaming_test_res.json"
+
+# Boosting ground truth - sanity check for per-utterance boosting
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_chunked_inference/rnnt/speech_to_text_streaming_infer_rnnt.py \
+    model_path="/home/TestData/asr/stt_en_fastconformer_transducer_large.nemo" \
+    dataset_manifest="/home/TestData/asr/canary/dev-other-wav-10-boost-gt.json" \
+    use_per_stream_biasing=true \
+    chunk_secs=2.0 \
+    left_context_secs=10.0 \
+    right_context_secs=2.0 \
+    batch_size=5 \
+    timestamps=true \
+    output_filename="/tmp/stt_streaming_test_res.json"
+
+# Boosting ground truth - MALSD beam search with per-stream biasing (RNNT)
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_chunked_inference/rnnt/speech_to_text_streaming_infer_rnnt.py \
+    model_path="/home/TestData/asr/stt_en_fastconformer_transducer_large.nemo" \
+    dataset_manifest="/home/TestData/asr/canary/dev-other-wav-10-boost-gt.json" \
+    use_per_stream_biasing=true \
+    decoding.strategy=malsd_batch \
+    decoding.beam.beam_size=4 \
+    decoding.beam.max_symbols_per_step=10 \
+    chunk_secs=2.0 \
+    left_context_secs=10.0 \
+    right_context_secs=2.0 \
+    batch_size=5 \
+    timestamps=true \
+    output_filename="/tmp/stt_streaming_malsd_boost_gt_rnnt_test_res.json"
+
+# Boosting ground truth - MALSD beam search with per-stream biasing (TDT)
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_chunked_inference/rnnt/speech_to_text_streaming_infer_rnnt.py \
+    model_path="/home/TestData/asr/stt_en_fastconformer_tdt_large.nemo" \
+    dataset_manifest="/home/TestData/asr/canary/dev-other-wav-10-boost-gt.json" \
+    use_per_stream_biasing=true \
+    decoding.strategy=malsd_batch \
+    decoding.beam.beam_size=4 \
+    decoding.beam.max_symbols_per_step=10 \
+    chunk_secs=2.0 \
+    left_context_secs=10.0 \
+    right_context_secs=2.0 \
+    batch_size=5 \
+    timestamps=true \
+    output_filename="/tmp/stt_streaming_malsd_boost_gt_tdt_test_res.json"
+
+# Streaming MALSD beam search - RNNT model
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_chunked_inference/rnnt/speech_to_text_streaming_infer_rnnt.py \
+    model_path="/home/TestData/asr/stt_en_fastconformer_transducer_large.nemo" \
+    audio_dir="/home/TestData/an4_transcribe/test_subset/" \
+    decoding.strategy=malsd_batch \
+    decoding.beam.beam_size=4 \
+    decoding.beam.max_symbols_per_step=10 \
+    chunk_secs=2.0 \
+    left_context_secs=10.0 \
+    right_context_secs=2.0 \
+    timestamps=true \
+    output_filename="/tmp/stt_streaming_malsd_test_res.json"
+
+# Streaming MAES beam search - RNNT model (MAES is RNN-T only)
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_chunked_inference/rnnt/speech_to_text_streaming_infer_rnnt.py \
+    model_path="/home/TestData/asr/stt_en_fastconformer_transducer_large.nemo" \
+    audio_dir="/home/TestData/an4_transcribe/test_subset/" \
+    decoding.strategy=maes_batch \
+    decoding.beam.beam_size=4 \
+    decoding.beam.maes_num_steps=2 \
+    decoding.beam.maes_expansion_beta=2 \
+    decoding.beam.maes_expansion_gamma=2.3 \
+    decoding.beam.allow_cuda_graphs=false \
+    chunk_secs=2.0 \
+    left_context_secs=10.0 \
+    right_context_secs=2.0 \
+    timestamps=true \
+    output_filename="/tmp/stt_streaming_maes_test_res.json"
+
+# Streaming MALSD beam search - TDT model
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/asr/asr_chunked_inference/rnnt/speech_to_text_streaming_infer_rnnt.py \
+    model_path="/home/TestData/asr/stt_en_fastconformer_tdt_large.nemo" \
+    audio_dir="/home/TestData/an4_transcribe/test_subset/" \
+    decoding.strategy=malsd_batch \
+    decoding.beam.beam_size=4 \
+    decoding.beam.max_symbols_per_step=10 \
+    chunk_secs=2.0 \
+    left_context_secs=10.0 \
+    right_context_secs=2.0 \
+    timestamps=true \
+    output_filename="/tmp/stt_streaming_malsd_tdt_test_res.json"

tests/functional_tests/L2_Speech_Transcription_Speech_to_Text_Transcribe.sh

@@ -0,0 +1,18 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/transcribe_speech.py \
+    pretrained_name="nvidia/stt_en_fastconformer_ctc_large" \
+    audio_dir="/home/TestData/an4_transcribe/test_subset/" \
+    output_filename="/tmp/stt_test_res.json" \
+    amp=true

tests/functional_tests/L2_Speech_Transcription_Speech_to_Text_Transcribe_Boost_GT.sh

@@ -0,0 +1,31 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+# Test RNN-T
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    tests/functional_tests/asr_transcribe_boost_ground_truth.py \
+    dataset_manifest="/home/TestData/asr/canary/dev-other-wav-10.json" \
+    output_filename="/tmp/stt_transcribe_boost_gt_res_rnnt.json" \
+    model_path="/home/TestData/asr/stt_en_fastconformer_transducer_large.nemo" \
+    batch_size=5 \
+    device='cuda:0'
+
+# Test TDT
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    tests/functional_tests/asr_transcribe_boost_ground_truth.py \
+    dataset_manifest="/home/TestData/asr/canary/dev-other-wav-10.json" \
+    output_filename="/tmp/stt_transcribe_boost_gt_res_tdt.json" \
+    model_path="/home/TestData/asr/stt_en_fastconformer_tdt_large.nemo" \
+    batch_size=5 \
+    device='cuda:0'

tests/functional_tests/L2_Speech_Transcription_Streaming_Inference.sh

@@ -0,0 +1,56 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_streaming_inference/asr_streaming_infer.py \
+    --config-path="../conf/asr_streaming_inference/" \
+    --config-name=buffered_ctc.yaml \
+    audio_file="/home/TestData/an4_transcribe/test_subset/" \
+    output_filename="/tmp/buffered_ctc_test_res.json" \
+    output_dir="/tmp/buffered_ctc_test_dir" \
+    lang=en \
+    enable_itn=False \
+    enable_nmt=False \
+    asr_output_granularity=segment
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_streaming_inference/asr_streaming_infer.py \
+    --config-path="../conf/asr_streaming_inference/" \
+    --config-name=buffered_rnnt.yaml \
+    audio_file="/home/TestData/an4_transcribe/test_subset/" \
+    output_filename="/tmp/buffered_rnnt_test_res.json" \
+    output_dir="/tmp/buffered_rnnt_test_dir" \
+    lang=en \
+    enable_itn=False \
+    enable_nmt=False \
+    asr_output_granularity=segment
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_streaming_inference/asr_streaming_infer.py \
+    --config-path="../conf/asr_streaming_inference/" \
+    --config-name=cache_aware_ctc.yaml \
+    audio_file="/home/TestData/an4_transcribe/test_subset/" \
+    output_filename="/tmp/cache_aware_ctc_test_res.json" \
+    output_dir="/tmp/cache_aware_ctc_test_dir" \
+    lang=en \
+    enable_itn=False \
+    enable_nmt=False \
+    asr_output_granularity=segment
+
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_streaming_inference/asr_streaming_infer.py \
+    --config-path="../conf/asr_streaming_inference/" \
+    --config-name=cache_aware_rnnt.yaml \
+    audio_file="/home/TestData/an4_transcribe/test_subset/" \
+    output_filename="/tmp/cache_aware_rnnt_test_res.json" \
+    output_dir="/tmp/cache_aware_rnnt_test_dir" \
+    lang=en \
+    enable_itn=False \
+    enable_nmt=False \
+    asr_output_granularity=segment

tests/functional_tests/L2_Speech_to_Text_AED.sh

@@ -0,0 +1,44 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/speech_multitask/speech_to_text_aed.py \
+    model.prompt_format=canary \
+    model.model_defaults.asr_enc_hidden=256 \
+    model.model_defaults.lm_dec_hidden=256 \
+    model.encoder.n_layers=12 \
+    model.transf_encoder.num_layers=0 \
+    model.transf_decoder.config_dict.num_layers=12 \
+    model.train_ds.manifest_filepath=/home/TestData/asr/manifests/canary/an4_canary_train.json \
+    model.train_ds.batch_duration=60 \
+    model.train_ds.use_bucketing=false \
+    model.train_ds.shuffle_buffer_size=100 \
+    model.train_ds.num_workers=0 \
+    ++model.train_ds.text_field="answer" \
+    ++model.train_ds.lang_field="target_lang" \
+    model.validation_ds.manifest_filepath=/home/TestData/asr/manifests/canary/an4_canary_val.json \
+    ++model.validation_ds.text_field="answer" \
+    ++model.validation_ds.lang_field="target_lang" \
+    model.validation_ds.num_workers=0 \
+    model.test_ds.manifest_filepath=/home/TestData/asr/manifests/canary/an4_canary_val.json \
+    ++model.test_ds.text_field="answer" \
+    ++model.test_ds.lang_field="target_lang" \
+    model.test_ds.num_workers=0 \
+    ++spl_tokens.model_dir=/home/TestData/asr_tokenizers/canary/canary_spl_tokenizer_v32 \
+    model.tokenizer.langs.en.dir=/home/TestData/asr_tokenizers/canary/en/tokenizer_spe_bpe_v1024_max_4 \
+    model.tokenizer.langs.en.type=bpe \
+    ++model.tokenizer.langs.es.dir=/home/TestData/asr_tokenizers/canary/es/tokenizer_spe_bpe_v1024_max_4 \
+    ++model.tokenizer.langs.es.type=bpe \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_to_text_aed_results

tests/functional_tests/L2_Speech_to_Text_EMA.sh

@@ -0,0 +1,21 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/asr_ctc/speech_to_text_ctc.py \
+    model.train_ds.manifest_filepath=/home/TestData/an4_dataset/an4_train.json \
+    model.validation_ds.manifest_filepath=/home/TestData/an4_dataset/an4_val.json \
+    trainer.devices=2 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    +exp_manager.ema.enable=True \
+    exp_manager.exp_dir=/tmp/speech_to_text_results

tests/functional_tests/L2_TTS_Fast_dev_runs_1_FastPitch.sh

@@ -0,0 +1,37 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/fastpitch.py \
+    --config-name fastpitch_align_v1.05 \
+    train_dataset=/home/TestData/an4_dataset/an4_train.json \
+    validation_datasets=/home/TestData/an4_dataset/an4_val.json \
+    sup_data_path=/home/TestData/an4_dataset/beta_priors \
+    trainer.devices="[0]" \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    trainer.max_epochs=1 \
+    trainer.strategy=auto \
+    model.pitch_mean=212.35873413085938 \
+    model.pitch_std=68.52806091308594 \
+    model.train_ds.dataloader_params.batch_size=4 \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.batch_size=4 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    model.symbols_embedding_dim=64 \
+    model.input_fft.d_inner=384 \
+    model.input_fft.n_layer=2 \
+    model.output_fft.d_inner=384 \
+    model.output_fft.n_layer=2 \
+    ~trainer.check_val_every_n_epoch \
+    ~model.text_normalizer \
+    ~model.text_normalizer_call_kwargs

tests/functional_tests/L2_TTS_Fast_dev_runs_1_Hifigan.sh

@@ -0,0 +1,28 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/hifigan.py \
+    train_dataset=/home/TestData/an4_dataset/an4_train.json \
+    validation_datasets=/home/TestData/an4_dataset/an4_val.json \
+    trainer.devices="[0]" \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    +trainer.max_epochs=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.batch_size=4 \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.batch_size=4 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    model.generator.upsample_initial_channel=64 \
+    +model.debug=true \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_Fast_dev_runs_AcousticCodec.sh

@@ -0,0 +1,33 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/audio_codec.py \
+    --config-name acoustic_codec_16000.yaml \
+    semantic_codec_path="/home/TestData/tts/TestSemanticCodec.nemo" \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.sample_weight=1.0 \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +log_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +log_ds_meta.an4.audio_dir="/" \
+    log_dir="/tmp/audio_codec_training_output" \
+    max_epochs=1 \
+    batch_size=4 \
+    weighted_sampling_steps_per_epoch=10 \
+    +trainer.limit_val_batches=1 \
+    trainer.devices="[0]" \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_Fast_dev_runs_AudioCodec.sh

@@ -0,0 +1,32 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/audio_codec.py \
+    --config-name audio_codec_22050.yaml \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.sample_weight=1.0 \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +log_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +log_ds_meta.an4.audio_dir="/" \
+    log_dir="/tmp/audio_codec_training_output" \
+    max_epochs=1 \
+    batch_size=4 \
+    weighted_sampling_steps_per_epoch=10 \
+    +trainer.limit_val_batches=1 \
+    trainer.devices="[0]" \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_Fast_dev_runs_EasyMagpietts_Nemotron.sh

@@ -0,0 +1,38 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/easy_magpietts.py \
+    --config-name easy_magpietts \
+    name="EasyMagpieTTS-Nemotron-FastDev" \
+    ~model.phoneme_tokenizer \
+    model.decoder_type="nemotron_h" \
+    model.nemotron_h_config.num_hidden_layers=8 \
+    model.nemotron_h_config.hybrid_override_pattern="M*M*M*M*" \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.tokenizer_names="[nemotron_nano_30b]" \
+    +train_ds_meta.an4.feature_dir=null \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +val_ds_meta.an4.tokenizer_names="[nemotron_nano_30b]" \
+    +val_ds_meta.an4.feature_dir=null \
+    max_epochs=1 \
+    batch_size=4 \
+    model.codecmodel_path="/home/TestData/tts/21fps_causal_codecmodel.nemo" \
+    trainer.devices="[0]" \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_Fast_dev_runs_EasyMagpietts_OnlinePO.sh

@@ -0,0 +1,78 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/easy_magpietts.py \
+    --config-name easy_magpietts \
+    name="EasyMagpieTTS-OnlinePO-FastDev" \
+    +mode="onlinepo_train" \
+    +init_from_ptl_ckpt="/home/TestData/tts/2603_EasyMagpieTTS/EMTTS_Pretraining_Qwen_WithCrossLingual_3_5_Delay_126.ckpt" \
+    model.phoneme_tokenizer.tokenizer_path="/home/TestData/tts/2603_EasyMagpieTTS/bpe_ipa_tokenizer_2048_en_de_es_fr_hi_it_vi_zh.json" \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/ipa_manifests/an4_train_context_v1_ipa.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.tokenizer_names="[nemotron_nano_30b]" \
+    +train_ds_meta.an4.feature_dir=null \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/ipa_manifests/an4_val_context_v1_ipa.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +val_ds_meta.an4.tokenizer_names="[nemotron_nano_30b]" \
+    +val_ds_meta.an4.feature_dir=null \
+    max_epochs=1 \
+    batch_size=2 \
+    model.codecmodel_path="/home/TestData/tts/25fps_spectral_codec_with_bandwidth_extension.nemo" \
+    +model.vector_quantizer._target_="nemo.collections.tts.modules.audio_codec_modules.GroupFiniteScalarQuantizer" \
+    +model.vector_quantizer.num_groups=8 \
+    +model.vector_quantizer.num_levels_per_group="[4, 4, 4, 4, 4]" \
+    ++model.add_language_to_context_text=true \
+    '+model.ignore_phoneme_languages=[vi,zh]' \
+    '+model.training_modes=[{text_input_mode:streaming,streaming_phonemes_delay:3,streaming_speech_delay:5}]' \
+    +model.reference_free=true \
+    +model.loss_type=grpo \
+    +model.scale_rewards=true \
+    +model.grpo_beta=0.0 \
+    ++model.normalize_whisper_transcript=true \
+    ++model.reward_asr_model=whisper \
+    ++model.speaker_verification_model_name=titanet_large \
+    ++model.use_pesq=false \
+    +model.n_generations_per_item=2 \
+    +model.batch_size_for_chunked_tf=2 \
+    +model.max_decoder_steps=300 \
+    +model.min_valid_codes_len=4 \
+    +model.max_valid_codes_len=490 \
+    ++model.aux_phoneme_loss_weight=0.1 \
+    ++model.best_cer_threshold=1.0 \
+    ++model.worst_cer_threshold=1.0 \
+    +model.inference_cfg_prob=0.5 \
+    +model.inference_cfg_scale=2.5 \
+    +model.gt_phoneme_input_prob=1.0 \
+    +model.inference_temperature=0.7 \
+    +model.inference_topk=80 \
+    +model.inference_phoneme_sampling_method=argmax \
+    +model.use_local_transformer_prob=1.0 \
+    +model.cer_reward_weight=0.5 \
+    +model.ssim_reward_weight=0.5 \
+    +model.use_utmos=false \
+    +model.utmos_reward_weight=0.0 \
+    ++model.pesq_reward_weight=0.0 \
+    ++model.val_n_generations_per_item=1 \
+    model.optim.lr=5e-6 \
+    ~model.optim.sched \
+    trainer.log_every_n_steps=1 \
+    trainer.precision=32 \
+    trainer.gradient_clip_val=0.0 \
+    trainer.devices="[0]" \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    +trainer.val_check_interval=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_Fast_dev_runs_EasyMagpietts_Qwen.sh

@@ -0,0 +1,37 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/easy_magpietts.py \
+    --config-name easy_magpietts \
+    name="EasyMagpieTTS-Qwen-FastDev" \
+    ~model.phoneme_tokenizer \
+    model.decoder_type="huggingface" \
+    model.transformer_hf_backend="Qwen/Qwen2.5-1.5B" \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.tokenizer_names="[nemotron_nano_30b]" \
+    +train_ds_meta.an4.feature_dir=null \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +val_ds_meta.an4.tokenizer_names="[nemotron_nano_30b]" \
+    +val_ds_meta.an4.feature_dir=null \
+    max_epochs=1 \
+    batch_size=4 \
+    model.codecmodel_path="/home/TestData/tts/21fps_causal_codecmodel.nemo" \
+    trainer.devices="[0]" \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_Fast_dev_runs_Magpietts_DecoderContext.sh

@@ -0,0 +1,33 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts.py \
+    --config-name magpietts \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +train_ds_meta.an4.feature_dir=null \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +val_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +val_ds_meta.an4.feature_dir=null \
+    max_epochs=1 \
+    batch_size=4 \
+    model.codecmodel_path="/home/TestData/tts/21fps_causal_codecmodel.nemo" \
+    trainer.devices="[0]" \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_Fast_dev_runs_Magpietts_FrameStacking_OnlineCFGDistillation.sh

@@ -0,0 +1,91 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts.py \
+    --config-name magpietts \
+    name="MagpieTTS-FrameStacking-OnlineCFGDistillation" \
+    +mode="online_cfg_distillation_train" \
+    +init_from_ptl_ckpt="/home/TestData/tts/2602_FrameStacking4x/frame-stacking-4x-english-nanocodec.ckpt" \
+    +model.teacher_model_path="/home/TestData/tts/2602_FrameStacking4x/frame-stacking-4x-english-nanocodec.ckpt" \
+    model.codecmodel_path="/home/TestData/tts/21fps_causal_codecmodel.nemo" \
+    +model.frame_stacking_factor=4 \
+    +model.distill_local_transformer=true \
+    +model.lt_distillation_start_step=0 \
+    model.alignment_loss_scale=0.0 \
+    model.use_text_conditioning_encoder=false \
+    model.prior_scaling_factor=null \
+    model.local_transformer_type="autoregressive" \
+    model.local_transformer_n_layers=4 \
+    model.local_transformer_n_heads=12 \
+    model.local_transformer_hidden_dim=768 \
+    +model.text_tokenizers.spanish_phoneme._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.IPATokenizer \
+    +model.text_tokenizers.spanish_phoneme.locale=es-ES \
+    +model.text_tokenizers.spanish_phoneme.punct=true \
+    +model.text_tokenizers.spanish_phoneme.apostrophe=true \
+    +model.text_tokenizers.spanish_phoneme.pad_with_space=true \
+    +model.text_tokenizers.spanish_phoneme.g2p._target_=nemo.collections.tts.g2p.models.i18n_ipa.IpaG2p \
+    +model.text_tokenizers.spanish_phoneme.g2p.locale=es-ES \
+    +model.text_tokenizers.spanish_phoneme.g2p.phoneme_dict="scripts/tts_dataset_files/es_ES/es_ES_nv230301.dict" \
+    +model.text_tokenizers.spanish_phoneme.g2p.phoneme_probability=0.8 \
+    +model.text_tokenizers.spanish_phoneme.g2p.use_chars=true \
+    +model.text_tokenizers.spanish_phoneme.g2p.use_stresses=true \
+    +model.text_tokenizers.german_phoneme._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.IPATokenizer \
+    +model.text_tokenizers.german_phoneme.locale=de-DE \
+    +model.text_tokenizers.german_phoneme.punct=true \
+    +model.text_tokenizers.german_phoneme.apostrophe=true \
+    +model.text_tokenizers.german_phoneme.pad_with_space=true \
+    +model.text_tokenizers.german_phoneme.g2p._target_=nemo.collections.tts.g2p.models.i18n_ipa.IpaG2p \
+    +model.text_tokenizers.german_phoneme.g2p.locale=de-DE \
+    +model.text_tokenizers.german_phoneme.g2p.phoneme_dict="scripts/tts_dataset_files/de/de_nv230119.dict" \
+    +model.text_tokenizers.german_phoneme.g2p.phoneme_probability=0.8 \
+    +model.text_tokenizers.german_phoneme.g2p.use_chars=true \
+    +model.text_tokenizers.german_phoneme.g2p.use_stresses=true \
+    +model.text_tokenizers.german_phoneme.g2p.grapheme_case="mixed" \
+    +model.text_tokenizers.german_phoneme.g2p.grapheme_prefix="'#'" \
+    +model.text_tokenizers.mandarin_phoneme._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.ChinesePhonemesTokenizer \
+    +model.text_tokenizers.mandarin_phoneme.punct=true \
+    +model.text_tokenizers.mandarin_phoneme.apostrophe=true \
+    +model.text_tokenizers.mandarin_phoneme.pad_with_space=true \
+    +model.text_tokenizers.mandarin_phoneme.g2p._target_=nemo.collections.tts.g2p.models.zh_cn_pinyin.ChineseG2p \
+    +model.text_tokenizers.mandarin_phoneme.g2p.phoneme_dict="scripts/tts_dataset_files/zh/36finals/ipa_dict_nv23.05.txt" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.word_segmenter="jieba" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.phoneme_prefix="" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.phoneme_case="lower" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.tone_prefix="'#'" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.ascii_letter_prefix="" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.ascii_letter_case="upper" \
+    +model.text_tokenizers.french_chartokenizer._target_=AutoTokenizer \
+    +model.text_tokenizers.french_chartokenizer.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.hindi_phoneme._target_=AutoTokenizer \
+    +model.text_tokenizers.hindi_phoneme.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.italian_phoneme._target_=AutoTokenizer \
+    +model.text_tokenizers.italian_phoneme.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.vietnamese_phoneme._target_=AutoTokenizer \
+    +model.text_tokenizers.vietnamese_phoneme.pretrained_model="google/byt5-small" \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +train_ds_meta.an4.feature_dir=null \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +val_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +val_ds_meta.an4.feature_dir=null \
+    trainer.devices="[0]" \
+    max_epochs=1 \
+    batch_size=4 \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_Fast_dev_runs_Magpietts_MoE.sh

@@ -0,0 +1,41 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts.py \
+    --config-name magpietts \
+    name="MagpieTTS-MoE" \
+    +model.use_moe=true \
+    +model.router_load_balancing_loss_coeff=0.01 \
+    +model.router_z_loss_coeff=0.001 \
+    model.decoder.d_ffn=1536 \
+    +model.decoder.num_experts=8 \
+    +model.decoder.top_k_experts=2 \
+    +model.decoder.routing_strategy="top_k" \
+    +model.decoder.router_jitter_noise=0.01 \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +train_ds_meta.an4.feature_dir=null \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +val_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +val_ds_meta.an4.feature_dir=null \
+    max_epochs=1 \
+    batch_size=4 \
+    model.codecmodel_path="/home/TestData/tts/21fps_causal_codecmodel.nemo" \
+    trainer.devices="[0]" \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0

tests/functional_tests/L2_TTS_Fast_dev_runs_Magpietts_MoE_OnlineCFGDistillation.sh

@@ -0,0 +1,50 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts.py \
+    --config-name magpietts \
+    name="MagpieTTS-MoE-OnlineCFGDistillation" \
+    +mode="online_cfg_distillation_train" \
+    +init_from_nemo_model="/home/TestData/tts/2602_MoE/moe16_sinkhorn_top1_valLoss5.0469_step2625132_epoch524.nemo" \
+    +model.teacher_model_path="/home/TestData/tts/2602_MoE/moe16_sinkhorn_top1_valLoss5.0469_step2625132_epoch524.nemo" \
+    model.codecmodel_path="/home/TestData/tts/21fps_causal_codecmodel.nemo" \
+    +model.use_moe=true \
+    +model.decoder.use_moe=true \
+    model.prior_scaling_factor=null \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +train_ds_meta.an4.feature_dir=null \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +val_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +val_ds_meta.an4.feature_dir=null \
+    trainer.devices="[0]" \
+    max_epochs=1 \
+    batch_size=4 \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch \
+    +model.router_load_balancing_loss_coeff=0.0 \
+    +model.router_z_loss_coeff=0.001 \
+    model.decoder.d_ffn=3072 \
+    +model.decoder.num_experts=16 \
+    +model.decoder.top_k_experts=1 \
+    +model.decoder.routing_strategy="sinkhorn" \
+    +model.decoder.router_jitter_noise=0.0 \
+    model.local_transformer_type="none" \
+    model.encoder.is_causal=false \
+    model.model_type="decoder_context_tts"

tests/functional_tests/L2_TTS_Fast_dev_runs_Magpietts_MultiEncoder.sh

@@ -0,0 +1,36 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts.py \
+    --config-name magpietts \
+    model.model_type=multi_encoder_context_tts \
+    +model.transcript_decoder_layers=[3,4,5,6,7] \
+    +model.context_decoder_layers=[8,9] \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +train_ds_meta.an4.feature_dir=null \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +val_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +val_ds_meta.an4.feature_dir=null \
+    max_epochs=1 \
+    batch_size=4 \
+    model.codecmodel_path="/home/TestData/tts/21fps_causal_codecmodel.nemo" \
+    trainer.devices="[0]" \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_Fast_dev_runs_Magpietts_OnlineCFGDistillation.sh

@@ -0,0 +1,97 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts.py \
+    --config-name magpietts \
+    name="MagpieTTS-OnlineCFGDistillation" \
+    +mode="online_cfg_distillation_train" \
+    +init_from_nemo_model="/home/TestData/tts/2602_MagpieTTS/feb26_Magpie-TTS-ML-V1--val_cer_gt\=0.3258-step\=1000.nemo" \
+    +model.teacher_model_path="/home/TestData/tts/2602_MagpieTTS/feb26_Magpie-TTS-ML-V1--val_cer_gt\=0.3258-step\=1000.nemo" \
+    model.codecmodel_path="/home/TestData/tts/21fps_causal_codecmodel.nemo" \
+    model.prior_scaling_factor=null \
+    model.local_transformer_type="autoregressive" \
+    model.local_transformer_n_layers=1 \
+    model.local_transformer_n_heads=1 \
+    model.local_transformer_hidden_dim=256 \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +train_ds_meta.an4.feature_dir=null \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +val_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +val_ds_meta.an4.feature_dir=null \
+    +model.text_tokenizers.spanish_phoneme._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.IPATokenizer \
+    +model.text_tokenizers.spanish_phoneme.locale=es-ES \
+    +model.text_tokenizers.spanish_phoneme.punct=true \
+    +model.text_tokenizers.spanish_phoneme.apostrophe=true \
+    +model.text_tokenizers.spanish_phoneme.pad_with_space=true \
+    +model.text_tokenizers.spanish_phoneme.g2p._target_=nemo.collections.tts.g2p.models.i18n_ipa.IpaG2p \
+    +model.text_tokenizers.spanish_phoneme.g2p.locale=es-ES \
+    +model.text_tokenizers.spanish_phoneme.g2p.phoneme_dict="scripts/tts_dataset_files/es_ES/es_ES_nv230301.dict" \
+    +model.text_tokenizers.spanish_phoneme.g2p.phoneme_probability=0.8 \
+    +model.text_tokenizers.spanish_phoneme.g2p.use_chars=true \
+    +model.text_tokenizers.spanish_phoneme.g2p.use_stresses=true \
+    +model.text_tokenizers.german_phoneme._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.IPATokenizer \
+    +model.text_tokenizers.german_phoneme.locale=de-DE \
+    +model.text_tokenizers.german_phoneme.punct=true \
+    +model.text_tokenizers.german_phoneme.apostrophe=true \
+    +model.text_tokenizers.german_phoneme.pad_with_space=true \
+    +model.text_tokenizers.german_phoneme.g2p._target_=nemo.collections.tts.g2p.models.i18n_ipa.IpaG2p \
+    +model.text_tokenizers.german_phoneme.g2p.locale=de-DE \
+    +model.text_tokenizers.german_phoneme.g2p.phoneme_dict="scripts/tts_dataset_files/de/de_nv230119.dict" \
+    +model.text_tokenizers.german_phoneme.g2p.phoneme_probability=0.8 \
+    +model.text_tokenizers.german_phoneme.g2p.use_chars=true \
+    +model.text_tokenizers.german_phoneme.g2p.use_stresses=true \
+    +model.text_tokenizers.german_phoneme.g2p.grapheme_case="mixed" \
+    +model.text_tokenizers.german_phoneme.g2p.grapheme_prefix="'#'" \
+    +model.text_tokenizers.mandarin_phoneme._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.ChinesePhonemesTokenizer \
+    +model.text_tokenizers.mandarin_phoneme.punct=true \
+    +model.text_tokenizers.mandarin_phoneme.apostrophe=true \
+    +model.text_tokenizers.mandarin_phoneme.pad_with_space=true \
+    +model.text_tokenizers.mandarin_phoneme.g2p._target_=nemo.collections.tts.g2p.models.zh_cn_pinyin.ChineseG2p \
+    +model.text_tokenizers.mandarin_phoneme.g2p.phoneme_dict="scripts/tts_dataset_files/zh/36finals/ipa_dict_nv23.05.txt" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.word_segmenter="jieba" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.phoneme_prefix="" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.phoneme_case="lower" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.tone_prefix="'#'" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.ascii_letter_prefix="" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.ascii_letter_case="upper" \
+    +model.text_tokenizers.japanese_phoneme._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.JapanesePhonemeTokenizer \
+    +model.text_tokenizers.japanese_phoneme.punct=true \
+    +model.text_tokenizers.japanese_phoneme.apostrophe=false \
+    +model.text_tokenizers.japanese_phoneme.pad_with_space=true \
+    +model.text_tokenizers.japanese_phoneme.g2p._target_=nemo.collections.tts.g2p.models.ja_jp_ipa.JapaneseKatakanaAccentG2p \
+    +model.text_tokenizers.japanese_phoneme.g2p.ascii_letter_prefix="" \
+    +model.text_tokenizers.japanese_phoneme.g2p.ascii_letter_case="upper" \
+    +model.text_tokenizers.french_chartokenizer._target_=AutoTokenizer \
+    +model.text_tokenizers.french_chartokenizer.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.hindi_chartokenizer._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.HindiCharsTokenizer \
+    +model.text_tokenizers.hindi_chartokenizer.charset_version=1 \
+    +model.text_tokenizers.hindi_chartokenizer.punct=true \
+    +model.text_tokenizers.hindi_chartokenizer.punct_version=1 \
+    +model.text_tokenizers.hindi_chartokenizer.apostrophe=true \
+    +model.text_tokenizers.hindi_chartokenizer.pad_with_space=true \
+    +model.text_tokenizers.italian_phoneme._target_=AutoTokenizer \
+    +model.text_tokenizers.italian_phoneme.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.vietnamese_phoneme._target_=AutoTokenizer \
+    +model.text_tokenizers.vietnamese_phoneme.pretrained_model="google/byt5-small" \
+    trainer.devices="[0]" \
+    max_epochs=1 \
+    batch_size=4 \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_Fast_dev_runs_Magpietts_OnlinePO.sh

@@ -0,0 +1,121 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts.py \
+    --config-name magpietts \
+    +mode="onlinepo_train" \
+    ~model.text_tokenizers.text_ce_tokenizer \
+    +model.text_tokenizers.spanish_phoneme._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.IPATokenizer \
+    +model.text_tokenizers.spanish_phoneme.locale=es-ES \
+    +model.text_tokenizers.spanish_phoneme.apostrophe=true \
+    +model.text_tokenizers.spanish_phoneme.pad_with_space=true \
+    +model.text_tokenizers.spanish_phoneme.g2p._target_=nemo.collections.tts.g2p.models.i18n_ipa.IpaG2p \
+    +model.text_tokenizers.spanish_phoneme.g2p.phoneme_dict="scripts/tts_dataset_files/es_ES/es_ES_nv230301.dict" \
+    +model.text_tokenizers.spanish_phoneme.g2p.locale=es-ES \
+    +model.text_tokenizers.spanish_phoneme.g2p.use_chars=true \
+    +model.text_tokenizers.german_phoneme._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.IPATokenizer \
+    +model.text_tokenizers.german_phoneme.locale=de-DE \
+    +model.text_tokenizers.german_phoneme.apostrophe=true \
+    +model.text_tokenizers.german_phoneme.pad_with_space=true \
+    +model.text_tokenizers.german_phoneme.g2p._target_=nemo.collections.tts.g2p.models.i18n_ipa.IpaG2p \
+    +model.text_tokenizers.german_phoneme.g2p.phoneme_dict="scripts/tts_dataset_files/de/de_nv230119.dict" \
+    +model.text_tokenizers.german_phoneme.g2p.heteronyms="scripts/tts_dataset_files/de/de_nv230119.heteronym" \
+    +model.text_tokenizers.german_phoneme.g2p.locale=de-DE \
+    +model.text_tokenizers.german_phoneme.g2p.use_chars=true \
+    +model.text_tokenizers.german_phoneme.g2p.grapheme_case=mixed \
+    +model.text_tokenizers.german_phoneme.g2p.grapheme_prefix="'#'" \
+    +model.text_tokenizers.mandarin_phoneme._target_=nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.ChinesePhonemesTokenizer \
+    +model.text_tokenizers.mandarin_phoneme.pad_with_space=true \
+    +model.text_tokenizers.mandarin_phoneme.g2p._target_=nemo.collections.tts.g2p.models.zh_cn_pinyin.ChineseG2p \
+    +model.text_tokenizers.mandarin_phoneme.g2p.phoneme_dict="scripts/tts_dataset_files/zh/36finals/ipa_dict_nv23.05.txt" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.word_segmenter="jieba" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.phoneme_prefix="" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.phoneme_case="lower" \
+    +model.text_tokenizers.mandarin_phoneme.g2p.ascii_letter_case="upper" \
+    +model.text_tokenizers.english_chartokenizer._target_=AutoTokenizer \
+    +model.text_tokenizers.english_chartokenizer.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.spanish_chartokenizer._target_=AutoTokenizer \
+    +model.text_tokenizers.spanish_chartokenizer.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.french_chartokenizer._target_=AutoTokenizer \
+    +model.text_tokenizers.french_chartokenizer.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.dutch_chartokenizer._target_=AutoTokenizer \
+    +model.text_tokenizers.dutch_chartokenizer.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.italian_chartokenizer._target_=AutoTokenizer \
+    +model.text_tokenizers.italian_chartokenizer.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.german_chartokenizer._target_=AutoTokenizer \
+    +model.text_tokenizers.german_chartokenizer.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.portugese_chartokenizer._target_=AutoTokenizer \
+    +model.text_tokenizers.portugese_chartokenizer.pretrained_model="google/byt5-small" \
+    +model.text_tokenizers.polish_chartokenizer._target_=AutoTokenizer \
+    +model.text_tokenizers.polish_chartokenizer.pretrained_model="google/byt5-small" \
+    +train_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_train_context_v1.json" \
+    +train_ds_meta.an4.audio_dir="/" \
+    +train_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +train_ds_meta.an4.feature_dir=null \
+    +val_ds_meta.an4.manifest_path="/home/TestData/an4_dataset/an4_val_context_v1.json" \
+    +val_ds_meta.an4.audio_dir="/" \
+    +val_ds_meta.an4.tokenizer_names="[english_phoneme]" \
+    +val_ds_meta.an4.feature_dir=null \
+    +init_from_ptl_ckpt="/home/TestData/tts/2506_SeenSpeaker/T5TTS--val_loss\=0.3125-epoch\=8.ckpt" \
+    max_epochs=1 \
+    batch_size=2 \
+    +model.grpo_beta=0.0 \
+    +model.num_generations_per_item=6 \
+    +model.reference_free=true \
+    +model.inference_cfg_prob=0.0 \
+    +model.inference_cfg_scale=2.5 \
+    +model.cer_reward_weight=0.5 \
+    +model.ssim_reward_weight=0.5 \
+    +model.reward_asr_model="whisper" \
+    model.local_transformer_type="none" \
+    model.cfg_unconditional_prob=0.0 \
+    model.model_type="multi_encoder_context_tts" \
+    +model.transcript_decoder_layers="[0,2,4,6,8,10]" \
+    +model.context_decoder_layers="[1,3,5,7,9,11]" \
+    model.context_duration_min=3.0 \
+    model.context_duration_max=8.0 \
+    model.decoder.p_dropout=0.0 \
+    model.context_encoder.p_dropout=0.0 \
+    model.encoder.p_dropout=0.0 \
+    model.decoder.kernel_size=1 \
+    model.decoder.xa_n_heads=1 \
+    ~model.decoder.xa_d_head \
+    model.context_encoder.n_layers=6 \
+    model.encoder.is_causal=false \
+    model.use_text_conditioning_encoder=true \
+    +model.legacy_text_conditioning=True \
+    ~model.text_conditioning_tokenizer_name \
+    +model.forced_num_all_tokens_per_codebook=2048 \
+    +model.forced_audio_eos_id=2047 \
+    +model.forced_audio_bos_id=2046 \
+    +model.forced_context_audio_eos_id=2045 \
+    +model.forced_context_audio_bos_id=2044 \
+    model.local_transformer_n_layers=3 \
+    model.codecmodel_path="/home/TestData/tts/AudioCodec_21Hz_no_eliz_without_wavlm_disc.nemo" \
+    model.alignment_loss_scale=0.0 \
+    model.prior_scaling_factor=null \
+    trainer.log_every_n_steps=10 \
+    +model.inference_topk=2016 \
+    model.optim.lr=2e-7 \
+    ~model.optim.sched \
+    +model.use_kv_cache_during_online_po=true \
+    exp_manager.checkpoint_callback_params.monitor="val_cer_gt" \
+    exp_manager.checkpoint_callback_params.mode="min" \
+    trainer.precision=32 \
+    trainer.devices="[0]" \
+    +trainer.limit_train_batches=1 \
+    +trainer.limit_val_batches=1 \
+    trainer.strategy=auto \
+    model.train_ds.dataloader_params.num_workers=0 \
+    model.validation_ds.dataloader_params.num_workers=0 \
+    ~trainer.check_val_every_n_epoch

tests/functional_tests/L2_TTS_InferEvaluate_EasyMagpietts.sh

@@ -0,0 +1,43 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+
+# Not making this deterministic because the HuggingFace Qwen2 rotary embedding uses a CuBLAS matmul
+# that is incompatible with torch.use_deterministic_algorithms(True) on CUDA >= 10.2.
+# Setting CUBLAS_WORKSPACE_CONFIG is not sufficient since the error originates inside the
+# transformers library (modeling_qwen2.py Qwen2RotaryEmbedding.forward).
+# This legacy EasyMagpieTTS checkpoint trained context text without adding CAS embeddings.
+CUBLAS_WORKSPACE_CONFIG=:4096:8 HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts_inference.py \
+    --deterministic \
+    --codecmodel_path /home/TestData/tts/25fps_spectral_codec_with_bandwidth_extension.nemo \
+    --nemo_files /home/TestData/tts/2603_EasyMagpieTTS/EMTTS_Pretraining_Qwen_WithCrossLingual_3_5_Delay.nemo \
+    --out_dir ./emp_zs_0 \
+    --model_type easy_magpie \
+    --batch_size 4 \
+    --datasets_json_path examples/tts/evalset_config.json \
+    --datasets an4_val_ci_nemotronTokenizer \
+    --use_local_transformer \
+    --topk 80 \
+    --use_cfg \
+    --cfg_scale 2.5 \
+    --phoneme_input_type predicted \
+    --disable_cas_for_context_text \
+    --run_evaluation \
+    --disable_fcd \
+    --phoneme_tokenizer_path /home/TestData/tts/2603_EasyMagpieTTS/bpe_ipa_tokenizer_2048_en_de_es_fr_hi_it_vi_zh.json \
+    --temperature 0.6 \
+    --clean_up_disk \
+    --cer_target 0.10 \
+    --ssim_target 0.70 \
+    --asr_model_name /home/TestData/tts/pretrained_models/parakeet-tdt-1.1b/parakeet-tdt-1.1b.nemo \
+    --eou_model_name /home/TestData/tts/pretrained_models/wav2vec2-base-960h

tests/functional_tests/L2_TTS_InferEvaluate_Magpietts_FrameStacking.sh

@@ -0,0 +1,36 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+# Tests a 4x-stacked model with local transformer inference.
+
+CUBLAS_WORKSPACE_CONFIG=:4096:8 HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts_inference.py \
+    --deterministic \
+    --codecmodel_path /home/TestData/tts/21fps_causal_codecmodel.nemo \
+    --datasets_json_path examples/tts/evalset_config.json \
+    --datasets an4_val_ci \
+    --out_dir ./mp_fs_4x_0 \
+    --batch_size 4 \
+    --use_cfg \
+    --cfg_scale 2.5 \
+    --num_repeats 1 \
+    --temperature 0.6 \
+    --hparams_files /home/TestData/tts/2602_FrameStacking4x/hparams.yaml \
+    --checkpoint_files /home/TestData/tts/2602_FrameStacking4x/frame-stacking-4x-english-nanocodec.ckpt \
+    --run_evaluation \
+    --clean_up_disk \
+    --cer_target 0.07 \
+    --ssim_target 0.66 \
+    --use_local_transformer \
+    --asr_model_name /home/TestData/tts/pretrained_models/parakeet-tdt-1.1b/parakeet-tdt-1.1b.nemo \
+    --eou_model_name /home/TestData/tts/pretrained_models/wav2vec2-base-960h

tests/functional_tests/L2_TTS_InferEvaluate_Magpietts_MoE_ZeroShot.sh

@@ -0,0 +1,31 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+CUBLAS_WORKSPACE_CONFIG=:4096:8 HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts_inference.py \
+    --deterministic \
+    --nemo_files "/home/TestData/tts/2602_MoE/moe16_sinkhorn_top1_valLoss5.0469_step2625132_epoch524.nemo" \
+    --codecmodel_path "/home/TestData/tts/21fps_causal_codecmodel.nemo" \
+    --datasets_json_path "examples/tts/evalset_config.json" \
+    --datasets "an4_val_ci" \
+    --out_dir "./mp_moe_zs_0" \
+    --batch_size 4 \
+    --use_cfg \
+    --cfg_scale 2.5 \
+    --temperature 0.6 \
+    --apply_attention_prior \
+    --run_evaluation \
+    --clean_up_disk \
+    --cer_target 0.07 \
+    --ssim_target 0.73 \
+    --asr_model_name /home/TestData/tts/pretrained_models/parakeet-tdt-1.1b/parakeet-tdt-1.1b.nemo \
+    --eou_model_name /home/TestData/tts/pretrained_models/wav2vec2-base-960h

tests/functional_tests/L2_TTS_InferEvaluate_Magpietts_SeenSpeakers.sh

@@ -0,0 +1,34 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUBLAS_WORKSPACE_CONFIG=:4096:8 HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts_inference.py \
+    --deterministic \
+    --codecmodel_path /home/TestData/tts/21fps_causal_codecmodel.nemo \
+    --datasets_json_path examples/tts/evalset_config.json \
+    --datasets an4_val_ci \
+    --out_dir ./mp_ss_0 \
+    --batch_size 4 \
+    --use_cfg \
+    --cfg_scale 2.5 \
+    --num_repeats 1 \
+    --temperature 0.6 \
+    --nemo_files /home/TestData/tts/2602_MagpieTTS/magpie_tts_multilingual_357m.nemo \
+    --apply_attention_prior \
+    --apply_prior_to_layers "3,4,5,6,7,8,9" \
+    --estimate_alignment_from_layers "3,4,5,6" \
+    --use_local_transformer \
+    --run_evaluation \
+    --clean_up_disk \
+    --cer_target 0.1 \
+    --asr_model_name /home/TestData/tts/pretrained_models/parakeet-tdt-1.1b/parakeet-tdt-1.1b.nemo \
+    --eou_model_name /home/TestData/tts/pretrained_models/wav2vec2-base-960h

tests/functional_tests/L2_TTS_InferEvaluate_Magpietts_ZeroShot.sh

@@ -0,0 +1,35 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUBLAS_WORKSPACE_CONFIG=:4096:8 HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts_inference.py \
+    --deterministic \
+    --codecmodel_path /home/TestData/tts/21fps_causal_codecmodel.nemo \
+    --datasets_json_path examples/tts/evalset_config.json \
+    --datasets an4_val_ci \
+    --out_dir ./mp_zs_0 \
+    --batch_size 4 \
+    --use_cfg \
+    --cfg_scale 2.5 \
+    --num_repeats 1 \
+    --temperature 0.6 \
+    --nemo_files /home/TestData/tts/2602_MagpieTTS/feb26_Magpie-TTS-ML-V1--val_cer_gt=0.3258-step=1000.nemo \
+    --apply_attention_prior \
+    --apply_prior_to_layers "3,4,5,6,7,8,9" \
+    --estimate_alignment_from_layers "3,4,5,6" \
+    --use_local_transformer \
+    --run_evaluation \
+    --clean_up_disk \
+    --cer_target 0.09 \
+    --ssim_target 0.74 \
+    --asr_model_name /home/TestData/tts/pretrained_models/parakeet-tdt-1.1b/parakeet-tdt-1.1b.nemo \
+    --eou_model_name /home/TestData/tts/pretrained_models/wav2vec2-base-960h

tests/functional_tests/L2_TTS_InferEvaluatelongform_Magpietts_MoE_ZeroShot.sh

@@ -0,0 +1,30 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+CUBLAS_WORKSPACE_CONFIG=:4096:8 HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts_inference.py \
+    --deterministic \
+    --nemo_files "/home/TestData/tts/2602_MoE/moe16_sinkhorn_top1_valLoss5.0469_step2625132_epoch524.nemo" \
+    --codecmodel_path "/home/TestData/tts/21fps_causal_codecmodel.nemo" \
+    --datasets_json_path "examples/tts/evalset_config.json" \
+    --datasets "an4_val_ci_longform_tiny" \
+    --out_dir "./mplf_moe_zs_0" \
+    --batch_size 6 \
+    --use_cfg \
+    --cfg_scale 2.5 \
+    --apply_attention_prior \
+    --run_evaluation \
+    --clean_up_disk \
+    --cer_target 0.02 \
+    --ssim_target 0.48 \
+    --asr_model_name /home/TestData/tts/pretrained_models/parakeet-tdt-1.1b/parakeet-tdt-1.1b.nemo \
+    --eou_model_name /home/TestData/tts/pretrained_models/wav2vec2-base-960h

tests/functional_tests/L2_TTS_InferEvaluatelongform_Magpietts_ZeroShot.sh

@@ -0,0 +1,35 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+CUBLAS_WORKSPACE_CONFIG=:4096:8 HF_HUB_OFFLINE=1 TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/tts/magpietts_inference.py \
+    --deterministic \
+    --codecmodel_path /home/TestData/tts/21fps_causal_codecmodel.nemo \
+    --datasets_json_path examples/tts/evalset_config.json \
+    --datasets an4_val_ci_longform_tiny \
+    --out_dir ./mplf_zs_0 \
+    --batch_size 6 \
+    --use_cfg \
+    --cfg_scale 2.5 \
+    --num_repeats 1 \
+    --temperature 0.6 \
+    --nemo_files /home/TestData/tts/2602_MagpieTTS/feb26_Magpie-TTS-ML-V1--val_cer_gt=0.3258-step=1000.nemo \
+    --apply_attention_prior \
+    --apply_prior_to_layers "3,4,5,6,7,8,9" \
+    --estimate_alignment_from_layers "3,4,5,6" \
+    --use_local_transformer \
+    --run_evaluation \
+    --clean_up_disk \
+    --cer_target 0.02 \
+    --ssim_target 0.44 \
+    --asr_model_name /home/TestData/tts/pretrained_models/parakeet-tdt-1.1b/parakeet-tdt-1.1b.nemo \
+    --eou_model_name /home/TestData/tts/pretrained_models/wav2vec2-base-960h

tests/functional_tests/Optional_ASR_dev_run_Speech_To_Text_Finetuning.sh

@@ -0,0 +1,23 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/speech_to_text_finetune.py \
+    --config-path="conf/asr_finetune" --config-name="speech_to_text_finetune" \
+    model.train_ds.manifest_filepath=/home/TestData/an4_dataset/an4_train.json \
+    model.validation_ds.manifest_filepath=/home/TestData/an4_dataset/an4_val.json \
+    init_from_nemo_model=/home/TestData/asr/stt_en_fastconformer_transducer_large.nemo \
+    model.tokenizer.update_tokenizer=False \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_finetuning_results

tests/functional_tests/Optional_ASR_dev_run_Speech_To_Text_HF_Finetuning.sh

@@ -0,0 +1,43 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo examples/asr/speech_to_text_finetune.py \
+    --config-path="conf/asr_finetune" --config-name="speech_to_text_hf_finetune" \
+    ~model.train_ds.hf_data_cfg \
+    model.train_ds.num_workers=1 \
+    model.train_ds.batch_size=2 model.validation_ds.batch_size=2 \
+    model.train_ds.streaming=true \
+    +model.train_ds.hf_data_cfg.path="librispeech_asr" \
+    +model.train_ds.hf_data_cfg.name=null \
+    +model.train_ds.hf_data_cfg.split="test.clean" \
+    +model.train_ds.hf_data_cfg.streaming=true \
+    +model.train_ds.hf_data_cfg.trust_remote_code=True \
+    ++model.train_ds.hf_data_cfg.cache_dir=/home/TestData/HF_HOME \
+    ~model.validation_ds.hf_data_cfg \
+    model.validation_ds.streaming=true \
+    +model.validation_ds.hf_data_cfg.path="librispeech_asr" \
+    +model.validation_ds.hf_data_cfg.name=null \
+    +model.validation_ds.hf_data_cfg.split="test.clean" \
+    +model.validation_ds.hf_data_cfg.streaming=true \
+    +model.validation_ds.hf_data_cfg.trust_remote_code=True \
+    ++model.validation_ds.hf_data_cfg.cache_dir=/home/TestData/HF_HOME \
+    ~model.test_ds \
+    init_from_nemo_model=/home/TestData/asr/stt_en_fastconformer_transducer_large.nemo \
+    model.tokenizer.update_tokenizer=False \
+    model.optim.sched.warmup_steps=0 \
+    +model.optim.sched.max_steps=3 \
+    trainer.max_epochs=null \
+    trainer.devices=1 \
+    trainer.accelerator="gpu" \
+    +trainer.fast_dev_run=True \
+    exp_manager.exp_dir=/tmp/speech_finetuning_results

tests/functional_tests/Optional_L2_Speech_Batch_Size_OOMptimizer_Canary.sh

@@ -0,0 +1,17 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/speech_recognition/oomptimizer.py \
+  -c /home/TestData/oomptimizer/fast-conformer_aed.yaml \
+  -m nemo.collections.asr.models.EncDecMultiTaskModel \
+  -b "[[5.0,30],[5.0,45],[10.0,57],[10.0,71]]"

tests/functional_tests/SPEECHLM_Automodel_Compiled_Deps_Smoke.sh

@@ -0,0 +1,27 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.
+#
+# 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.
+
+set -e
+
+SMOKE_TEST=tests/functional_tests/speechlm_automodel_compiled_deps_smoke.py
+export SMOKE_TEST
+
+timeout 300s torchrun --nproc-per-node 2 --no-python python -c \
+  '
+import os
+import pytest
+
+args = [os.environ["SMOKE_TEST"], "-m", "integration and not pleasefixme", "--with_downloads"]
+os._exit(pytest.main(args))
+'

tests/functional_tests/SPEECHLM_Automodel_Training_SALM.sh

@@ -0,0 +1,50 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+# Run training with SALMAutomodel
+torchrun --nproc-per-node 1 --no-python \
+  coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/speechlm2/salm_train.py --config-name=salm_automodel \
+      model.pretrained_llm=/home/TestData/speechlm/pretrained_models/TinyLlama--TinyLlama_v1.1 \
+      model.pretrained_asr=/home/TestData/speechlm/pretrained_models/canary-1b-flash.nemo \
+      data.train_ds.input_cfg.0.cuts_path=/home/TestData/speechlm/lhotse/libri/librispeech_cuts_lower_train-clean-5.jsonl.gz \
+      data.validation_ds.datasets.val_set_0.input_cfg.0.cuts_path=/home/TestData/speechlm/lhotse/libri/librispeech_cuts_lower_dev-clean-2.jsonl.gz \
+      trainer.devices=1 \
+      trainer.max_steps=10
+
+# Convert to HF format
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+  examples/speechlm2/to_hf.py \
+  class_path=nemo.collections.speechlm2.models.SALMAutomodel \
+  ckpt_path=salm_results/checkpoints/step\=10-last.ckpt \
+  ckpt_config=salm_results/exp_config.yaml \
+  output_dir=test_salm_automodel_hf_model
+
+# Run generation (auto-detects SALMAutomodel from config.json)
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+  examples/speechlm2/salm_generate.py \
+  pretrained_name=test_salm_automodel_hf_model \
+  inputs=/home/TestData/speechlm/lhotse/libri/librispeech_cuts_lower_dev-clean-2-first10.jsonl.gz \
+  batch_size=4 \
+  output_manifest=generations_automodel.jsonl
+head generations_automodel.jsonl
+
+# Run generation + WER eval (auto-detects SALMAutomodel from config.json)
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+  examples/speechlm2/salm_eval.py \
+  pretrained_name=test_salm_automodel_hf_model \
+  inputs=/home/TestData/speechlm/lhotse/libri/librispeech_cuts_lower_dev-clean-2-first10.jsonl.gz \
+  batch_size=4 \
+  output_manifest=generations_automodel_wer.jsonl
+head generations_automodel_wer.jsonl

tests/functional_tests/SPEECHLM_HF_Training_DuplexS2S.sh

@@ -0,0 +1,34 @@
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
+#
+# 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.
+
+# Run training
+torchrun --nproc-per-node 1 --no-python \
+  coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+    examples/speechlm2/s2s_duplex_train.py \
+      model.pretrained_llm=/home/TestData/speechlm/pretrained_models/TinyLlama--TinyLlama_v1.1 \
+      model.pretrained_audio_codec=/home/TestData/speechlm/pretrained_models/Low_Frame-rate_Speech_Codec++_bf16_nodiscrim.nemo \
+      model.pretrained_asr=/home/TestData/speechlm/pretrained_models/stt_en_fastconformer_hybrid_large_streaming_80ms.nemo \
+      model.scoring_asr=/home/TestData/speechlm/pretrained_models/stt_en_fastconformer_transducer_large.nemo \
+      data.train_ds.input_cfg.0.shar_path=/home/TestData/speechlm/lhotse/speechlm2/train_micro \
+      data.validation_ds.datasets.val_set_0.shar_path=/home/TestData/speechlm/lhotse/speechlm2/train_micro \
+      trainer.devices=1 \
+      trainer.max_steps=10
+
+# Convert to HF format
+coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo \
+  examples/speechlm2/to_hf.py \
+    class_path=nemo.collections.speechlm2.models.DuplexS2SModel \
+    ckpt_path=s2s_results/checkpoints/step\\=10-last.ckpt \
+    ckpt_config=s2s_results/exp_config.yaml \
+    output_dir=test_speechlm2_hf_model