Haocheng lu2

[LuHC409]

What does this PR do?

This PR fixes an issue in the _preprocess function of the Qwen2VLImageProcessor class, located in:
transformers/src/transformers/models/qwen2_vl/image_processing_qwen2_vl.py
Previously, when the number of patches was not divisible by temporal_patch_size, the code mistakenly repeated the last patch temporal_patch_size - 1 times, which could cause overshooting. This PR corrects the padding logic by computing the exact number of repeats needed:
pad_len = temporal_patch_size - (patches.shape[0] % temporal_patch_size)
repeats = np.repeat(patches[-1][np.newaxis], pad_len, axis=0)
Motivation and context
This change ensures that the total number of temporal patches is always divisible by temporal_patch_size, without introducing unnecessary extra patches. It avoids shape mismatch or over-padding problems in the later reshape steps.

Local Testing
✅ I have tested this change locally and confirmed that all tests pass.

Fixes # (issue)
#37064

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• [yes ] Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[github-actions]

Hi 👋, thank you for opening this pull request! The pull request is converted to draft by default. The CI will be paused while the PR is in draft mode. When it is ready for review, please click the Ready for review button (at the bottom of the PR page). This will assign reviewers and trigger CI.

[qubvel]

Thanks, looks good to me!

[zucchini-nlp]

Thanks! Can you also update the test here, so it checks that video processing works for any input shape?

Existing test:

transformers/tests/models/qwen2_vl/test_image_processing_qwen2_vl.py

Lines 277 to 288 in e2b0224

	def test_video_inputs(self):
	for image_processing_class in self.image_processor_list:
	image_processing = image_processing_class(**self.image_processor_dict)
	expected_dims_by_frames = {1: 34300, 2: 34300, 3: 68600, 4: 68600, 5: 102900, 6: 102900}
	for num_frames, expected_dims in expected_dims_by_frames.items():
	image_processor_tester = Qwen2VLImageProcessingTester(self, num_frames=num_frames)
	video_inputs = image_processor_tester.prepare_video_inputs(equal_resolution=True)
	prcocess_out = image_processing(None, videos=video_inputs, return_tensors="pt")
	encoded_video = prcocess_out.pixel_values_videos
	expected_output_video_shape = (expected_dims, 1176)
	self.assertEqual(tuple(encoded_video.shape), expected_output_video_shape)

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.