Fix AutoVideoProcessor class lookup when torchvision is unavailable

[Abhijeetsingh610]

What does this PR do?

Fixes a crash in AutoVideoProcessor when torchvision is unavailable.

VIDEO_PROCESSOR_MAPPING_NAMES can contain None, and video_processor_class_from_name was doing if class_name in extractors, which raises TypeError for None.

This PR:

• skips None entries in video_processor_class_from_name
• adds a regression test in tests/models/auto/test_video_processing_auto.py

Fixes #43986 #43986

[christopher5106]

Hi,

Does this PR also solve this issue: huggingface/diffusers#13140 (comment)

?

Thanks

[Abhijeetsingh610]

Thanks for checking. Yes, this PR addresses the same root cause as huggingface/diffusers#13140.
In that issue, the crash happens in transformers.models.auto.video_processing_auto.video_processor_class_from_name when a mapping entry is None (typically when torchvision is unavailable) and code checks class_name in extractors.
This PR adds a None guard for that lookup and includes a regression test.
So once this is merged and released in transformers, that diffusers issue should be resolved when using the updated transformers version.

[Frojdholm]

@christopher5106 note that the root cause for the crash is you not having torchvision installed.

If your model requires a video processor it will not work since this will not create one. If the video processor is optional this might solve your issue.

[christopher5106]

@Frojdholm the bug appears on an image model, Qwen-Image and Qwen-Image-Edit. In that case, there is no video in inputs and outputs.

[zucchini-nlp]

I don't think this is the fix we want. Video processors were originally copied from image processor that have slow/fast versions. But since videos entirely rely on torchvision, we need to raise an error if it's not installed and not skip until the last error message

Same might apply for image processors, only in case when the model has a fast-only processor implemented (we have a couple models only). We need to check if torchvision is installed in L363 elif video_processor_class is None and not is_torchvision_available(): raise ValueError(informative and actioanable message)

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: auto