fix(testing): Fix Kyutai Speech-To-Text and LongCatFlash test failures on main CI

[harshaljanjani]

What does this PR do?

The following failing tests were identified and fixed in this PR:

→ Kyutai Speech-To-Text: The PR [processors] Unbloating simple processors, refactored ProcessorMixin.call to use explicit keyword-only params instead of accepting positional arguments; but the KyutaiSTT integration tests were still calling processor(samples) positionally; the audio samples in the current state mapped to the images param.
→ LLaVA-OneVision (Removed from the scope of this PR; requires more information from maintainers 🟡): The PR Load a tiny video to make CI faster introduced local video file path mappings. LlavaOnevision's setUpClass was still building paths to Big_Buck_Bunny_720_10s_10MB.mp4 and sample_demo_1.mp4 in the repo root.
→ LongCatFlash: The PR [V5] Return a BatchEncoding dict from apply_chat_template by default again changed apply_chat_template to return BatchEncoding dict instead of a tensor. The test was passing this dict directly to model.generate and tried to access .shape on the dict; this fixes that :)

Note: The test still fails with an AssertionError, I'm not too sure and it could be flaky, but the crash should be resolved :)

cc: @Rocketknight1 @zucchini-nlp

CI Failures

Before the fix (feel free to cross-check; these errors are reproducible):

After the fix (feel free to cross-check):

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 fix any necessary existing tests?

[harshaljanjani]

cc: @Rocketknight1 @zucchini-nlp Just a gentle ping :)

[zucchini-nlp]

left one comment, otherwise lgmt

[zucchini-nlp]

we also load local images above from the same root path. IIRC we made sure these artifacts are cached when loading from hub so I dont know why they are being created here

cc @ydshieh for this

[harshaljanjani]

++, I didn't notice this previously but now that I've read into it a bit more, I guess even this image creation block isn't needed either for the same reason (ref: 05c0e1d); just the local_tiny_video logic staying intact should suffice but I'd love to know if I'm missing something.

[zucchini-nlp]

Would love to hear from YihDar, since I am not aware if we are supposed to do it like this instead of allowing hub to get the correct cache from the hub

[harshaljanjani]

Good day @ydshieh,
I'm checking in to see if I might ask for some clarification regarding this thread at your convenience; thank you!

[github-actions]

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

run-slow: kyutai_speech_to_text, longcat_flash

[harshaljanjani]

Good day @zucchini-nlp!
Was wondering if we could remove the LLaVA-OneVision fix from the scope of this PR to unblock it. Once we receive updates on that PR, I'll raise another one for LLaVA-OneVision with the new information in mind. Please let me know if that works :)

[zucchini-nlp]

Oke, lets merge wihtout llava

[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.

[harshaljanjani]

@zucchini-nlp Just bumping this up since it was dequeued a couple of times :)