wehub-resource-sync
e06fe8e8c6
Secret Leaks / trufflehog (push) Failing after 1s
Build documentation / build (push) Failing after 1s
Build documentation / build_other_lang (push) Failing after 0s
CodeQL Security Analysis / CodeQL Analysis (push) Failing after 0s
PR CI / pr-ci (push) Failing after 1s
Slow tests on important models (on Push - A10) / Get all modified files (push) Failing after 1s
Slow tests on important models (on Push - A10) / Model CI (push) Has been skipped
Self-hosted runner (benchmark) / Benchmark (aws-g5-4xlarge-cache) (push) Has been cancelled
New model PR merged notification / Notify new model (push) Has been cancelled
Update Transformers metadata / build_and_package (push) Has been cancelled
110 lines
4.4 KiB
Python
110 lines
4.4 KiB
Python
# Copyright 2026 The HuggingFace Team. All rights reserved.
|
||
#
|
||
# 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.
|
||
|
||
import unittest
|
||
|
||
import torch
|
||
|
||
from transformers import PI0Processor
|
||
from transformers.testing_utils import get_tests_dir, require_torch, require_vision
|
||
from transformers.utils import is_vision_available
|
||
|
||
from ...test_processing_common import ProcessorTesterMixin
|
||
|
||
|
||
SAMPLE_VOCAB = get_tests_dir("fixtures/test_sentencepiece.model")
|
||
|
||
|
||
if is_vision_available():
|
||
from transformers import GemmaTokenizer, SiglipImageProcessor
|
||
|
||
|
||
@require_vision
|
||
class PI0ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
|
||
processor_class = PI0Processor
|
||
|
||
@classmethod
|
||
def _setup_image_processor(cls):
|
||
# Use 64×64 instead of the default 224×224 to avoid large tensors.
|
||
# image_seq_length=0 matches the processor attribute so token-count tests pass.
|
||
image_processor = SiglipImageProcessor(size={"height": 64, "width": 64})
|
||
image_processor.image_seq_length = 0
|
||
return image_processor
|
||
|
||
@classmethod
|
||
def _setup_tokenizer(cls):
|
||
tokenizer = GemmaTokenizer.from_pretrained(SAMPLE_VOCAB, keep_accents=True)
|
||
tokenizer.add_special_tokens({"additional_special_tokens": ["<image>"]})
|
||
return tokenizer
|
||
|
||
@classmethod
|
||
def _setup_test_attributes(cls, processor):
|
||
cls.image_token = processor.image_token
|
||
|
||
@require_vision
|
||
def prepare_image_inputs(self, batch_size: int | None = None, nested: bool = True):
|
||
return super().prepare_image_inputs(batch_size, nested=nested)
|
||
|
||
def test_image_processor_defaults(self):
|
||
image_processor = self.get_component("image_processor")
|
||
processor = self.get_processor()
|
||
image_input = self.prepare_image_inputs()
|
||
|
||
input_image_proc = image_processor(image_input, return_tensors="pt")
|
||
input_processor = processor(images=image_input, text="", return_tensors="pt")
|
||
|
||
for key in input_image_proc:
|
||
torch.testing.assert_close(input_image_proc[key], input_processor[key][:, 0])
|
||
self.assertTrue(torch.equal(input_processor["pixel_attention_mask"], torch.tensor([[True]])))
|
||
|
||
@require_torch
|
||
def test_single_camera_output_is_5d(self):
|
||
processor = self.get_processor()
|
||
image = self.prepare_image_inputs()
|
||
outputs = processor(images=image, text="task", return_tensors="pt")
|
||
self.assertEqual(outputs["pixel_values"].ndim, 5)
|
||
self.assertEqual(outputs["pixel_values"].shape[0], 1)
|
||
self.assertEqual(outputs["pixel_values"].shape[1], 1)
|
||
self.assertTrue(torch.equal(outputs["pixel_attention_mask"], torch.tensor([[True]])))
|
||
|
||
@require_torch
|
||
def test_multi_camera_padding_and_masks(self):
|
||
processor = self.get_processor()
|
||
image_a = self.prepare_image_inputs()
|
||
image_b = self.prepare_image_inputs()
|
||
image_c = self.prepare_image_inputs()
|
||
|
||
outputs = processor(
|
||
images=[[image_a, image_b], [image_c]],
|
||
text=["task a", "task b"],
|
||
return_tensors="pt",
|
||
)
|
||
|
||
self.assertEqual(outputs["pixel_values"].ndim, 5)
|
||
self.assertEqual(outputs["pixel_values"].shape[:2], torch.Size([2, 2]))
|
||
self.assertTrue(torch.equal(outputs["pixel_attention_mask"], torch.tensor([[True, True], [True, False]])))
|
||
|
||
@require_torch
|
||
def test_newline_normalization(self):
|
||
processor = self.get_processor()
|
||
image = self.prepare_image_inputs()
|
||
out_no_newline = processor(images=image, text="pick object", return_tensors="pt")
|
||
out_with_newline = processor(images=image, text="pick object\n", return_tensors="pt")
|
||
self.assertTrue(torch.equal(out_no_newline["input_ids"], out_with_newline["input_ids"]))
|
||
self.assertTrue(torch.equal(out_no_newline["attention_mask"], out_with_newline["attention_mask"]))
|
||
|
||
@unittest.skip("PI0 doesn't need vLLM integration")
|
||
def test_get_num_multimodal_tokens_matches_processor_call(self):
|
||
pass
|