Fix Qwen2.5-VL Video Processor

[yeliudev]

What does this PR do?

This PR fixes the bug in Qwen2_5_VLProcessor, which would consistently logging "Unused or unrecognized kwargs: fps, return_tensors." during training. This issue seems to be introduced in #35206.

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?
• 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?

@zucchini-nlp

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.

[zucchini-nlp]

Thanks for the fix! Yes, I've been seeing the issue with return_tensors for a while and had a bigger PR to refactor further.

The problem isn't specific for Qwen models, and needs to be fixed in video_processing_utils.py. Can you try to update the PR to fix all video processors from base class?

[yeliudev]

Thanks for your comment! I'm afraid that I have no idea how to fix it in video_processing_utils.py as I'm not familiar with this new validate_kwargs feature. I've reverted the fix for return_tensors and am leaving that for fps only.

[zucchini-nlp]

@yeliudev since it is the only common kwarg and is not in VideosKwargs, we can manually pass it to valid_kwargs as follows. LMK if that works and doesn't flood output with warnings

# `return_tensors` is a common kwargs thus not in processing_utils.VideosKwargs
        validate_kwargs(
            captured_kwargs=kwargs.keys(),
            valid_processor_keys=list(self.valid_kwargs.__annotations__.keys()) + ["return_tensors"],
        )

[yeliudev]

@zucchini-nlp Yes it works. Many thanks!

[zucchini-nlp]

Perfect thanks!

[ArthurZucker]

Merging as test is unrelated! not super fan of adding the return_tensors like that but can be done afterwards