Support auto_doctring in Processors

[yonigozlan]

What does this PR do?

Add support for Processors in @auto_docstring, and many other improvements to auto_docstring.py and check_docstrings.py, including more robust auto-fix with check_docstrings for missing, redundant, or unnecessary docstrings.

For processors, auto_docstring will pull custom args docstrings from custom "Kwargs" TypeDicts and add them to the .__doc__. For example, for processing_aria, we have:

class AriaImagesKwargs(ImagesKwargs, total=False):
    """
    split_image (`bool`, *optional*, defaults to `False`):
        Whether to split large images into multiple crops. When enabled, images exceeding the maximum size are
        divided into overlapping crops that are processed separately and then combined. This allows processing
        of very high-resolution images that exceed the model's input size limits.
    max_image_size (`int`, *optional*, defaults to `980`):
        Maximum image size (in pixels) for a single image crop. Images larger than this will be split into
        multiple crops when `split_image=True`, or resized if splitting is disabled. This parameter controls
        the maximum resolution of individual image patches processed by the model.
    min_image_size (`int`, *optional*):
        Minimum image size (in pixels) for a single image crop. Images smaller than this will be upscaled to
        meet the minimum requirement. If not specified, images are processed at their original size (subject
        to the maximum size constraint).
    """

    split_image: bool
    max_image_size: int
    min_image_size: int


class AriaProcessorKwargs(ProcessingKwargs, total=False):
    images_kwargs: AriaImagesKwargs

    _defaults = {
        "text_kwargs": {
            "padding": False,
            "return_mm_token_type_ids": False,
        },
        "images_kwargs": {
            "max_image_size": 980,
            "split_image": False,
        },
        "return_tensors": TensorType.PYTORCH,
    }


@auto_docstring
class AriaProcessor(ProcessorMixin):
    ...

    @auto_docstring
    def __call__(
        self,
        text: Union[TextInput, PreTokenizedInput, list[TextInput], list[PreTokenizedInput]],
        images: Optional[ImageInput] = None,
        **kwargs: Unpack[AriaProcessorKwargs],
    ) -> BatchFeature:
        r"""
        Returns:
            [`BatchFeature`]: A [`BatchFeature`] with the following fields:
            - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`.
            - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
            `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
            `None`).
            - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
            - **pixel_mask** -- Pixel mask to be fed to a model. Returned when `images` is not `None`.
        """
        ...

which results in the following docstring:

print(AriaProcessor.__call__.__doc__)

        Args:
            text (`Union[str, list, list]`):
                The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
                (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set
                `is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
            images (`Union[PIL.Image.Image, numpy.ndarray, torch.Tensor, list, list, list]`, *optional*):
                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
                passing in images with pixel values between 0 and 1, set `do_rescale=False`.
            split_image (`bool`, *optional*, defaults to `False`):
                Whether to split large images into multiple crops. When enabled, images exceeding the maximum size are
                divided into overlapping crops that are processed separately and then combined. This allows processing
                of very high-resolution images that exceed the model's input size limits.
            max_image_size (`int`, *optional*, defaults to `980`):
                Maximum image size (in pixels) for a single image crop. Images larger than this will be split into
                multiple crops when `split_image=True`, or resized if splitting is disabled. This parameter controls
                the maximum resolution of individual image patches processed by the model.
            min_image_size (`int`, *optional*):
                Minimum image size (in pixels) for a single image crop. Images smaller than this will be upscaled to
                meet the minimum requirement. If not specified, images are processed at their original size (subject
                to the maximum size constraint).
            return_tensors (`str` or [`~utils.TensorType`], *optional*):
                If set, will return tensors of a particular framework. Acceptable values are:

                - `'pt'`: Return PyTorch `torch.Tensor` objects.
                - `'np'`: Return NumPy `np.ndarray` objects.
        Returns:
            [`BatchFeature`]: A [`BatchFeature`] with the following fields:
            - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`.
            - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
            `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
            `None`).
            - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
            - **pixel_mask** -- Pixel mask to be fed to a model. Returned when `images` is not `None`.

[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.

[github-actions]

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

run-slow: align, altclip, aria, aya_vision, bamba, bark, blip, blip_2, bridgetower, bros, chameleon, chinese_clip, clap, clip, clipseg, clvp

[yonigozlan]

Also Cc @stevhliu :)

[Cyrilvallez]

Trusting you on that, but I think it would be time to add some proper tests no? I see a very old test_auto_docstrings.py but that does not run any tests -> probably a very nice idea to start rewriting it!

[Cyrilvallez]

nit: can we use textwrap.dedent here, so that the string respects the function indentation?

[yonigozlan]

Yep it's done right after

[Cyrilvallez]

Humm, I don't see it 😅 I meant doing something like

    intro = textwrap.dedent(
        """
        bla
        bla

        more bla
        """
    ).strip()

so that the indentation stays inside the function

[yonigozlan]

Trusting you on that, but I think it would be time to add some proper tests no? I see a very old test_auto_docstrings.py but that does not run any tests -> probably a very nice idea to start rewriting it!

Yes clearly! I'll add tests in the next autodocstring PR ;)

[stevhliu]

nice, thanks! added a few nits to the parameter definitions :)

[stevhliu]

Suggested change

	A Jinja template which will be used to convert lists of messages in a chat into a tokenizable string.
	A Jinja template to convert lists of messages in a chat into a tokenizable string.

[stevhliu]

Suggested change

	The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
	(pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set
	`is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
	The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
	(pretokenized string). If you pass a pretokenized input, set `is_split_into_words=True` to avoid ambiguity with batched inputs.

[stevhliu]

just curious, whats the difference between audio and audios below it?

[yonigozlan]

I think audios is deprecated but still present in some places

[stevhliu]

Suggested change

	This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability
	This is especially useful to enable using Tensor Cores on NVIDIA hardware with compute capability

[stevhliu]

Suggested change

	`PretrainedTokenizerBase.build_inputs_with_special_tokens` function, which defines which tokens are
	[`PretrainedTokenizerBase.build_inputs_with_special_tokens`] function, which defines which tokens are

[stevhliu]

Suggested change

	list of strings (pretokenized string). If the sequences are provided as list of strings (pretokenized),
	you must set `is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
	list of strings (pretokenized string). If you pass pretokenized input, set is_split_into_words=True to avoid ambiguity with batched inputs.

[stevhliu]

Suggested change

	list of strings (pretokenized string). If the sequences are provided as list of strings (pretokenized),
	you must set `is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
	list of strings (pretokenized string). If you pass pretokenized input, set is_split_into_words=True to avoid ambiguity with batched inputs.