add uniform processors for altclip + chinese_clip

[molbap]

What does this PR do?

Adds two models for #30511 , see parent PR for more details

[molbap]

Tests will succeed when #31197 is merged. See #30511 tests to confirm.

[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.

[amyeroberts]

Thanks for working on this!

Overall looks good to me. Main thought is that we probably want to move a lot of the kwarg preparation logic outside of the main __call__ method, and it can probably be abstracted into something more general for most processors.

Regarding tests, it would be good to check defaults and updates are correctly passed for all of the possible kwargs, particularly defaulting to the tokenizer.init_kwargs and ensuring these can be overridden.

Would be good to get a second opinion from @qubvel here!

[molbap]

Thanks! I'm currently moving the kwargs merging in a common method actually, and explaining order of operations wrt to kwargs priorities. Will be in #31197 in a minute

[qubvel]

Great work 🤗 Very structured and clean code!
I have the same concern as @amyeroberts that the logic of merging is worth split/move to separate methods. It is a bit hard to follow merging steps of different kwargs.

What do you think about grouping merging strategy per modality instead of per step?
not sure if its possible, because some kwargs, probably, have to be popped in advance.

Something like this:

images_kwargs = {} if images_kwargs is None else images_kwargs
default_images_kwargs = ChineseClipProcessorKwargs._defaults.get("images_kwargs", {}).copy()

# merging vision kwargs
images_kwargs = {**default_images_kwargs, **images_kwargs}   # with default
images_kwargs = {**images_kwargs, **kwargs.pop("images_kwargs", {})}  # with passed dict
images_kwargs = merge_with_kwargs_by_key(images_kwargs, ...)
images_kwargs = merge_with_common(images_kwargs, common_kwargs)

[amyeroberts]

Nice! Thanks for all the work on this 🤗

[molbap]

Re-pinging @amyeroberts on that one and @qubvel and @zucchini-nlp optionally if you want to take another look - not much changed, just was between sprints and didn't merge that one 😬 should be good to go now (I know you approved, Amy, but that was a while ago!)

[molbap]

Tests that were working are broken since passing from crop_size to size in common tests I believe - will look into this soon

[amyeroberts]

Thanks for adding!