processing_travisionlm.py

"""
Processor class for TraVisionLM.
"""

import logging
from typing import List, Optional, Union

from transformers.feature_extraction_utils import BatchFeature
from transformers.image_utils import ImageInput, is_valid_image
from transformers.processing_utils import ProcessorMixin
from transformers.tokenization_utils import (
    AddedToken,
    PaddingStrategy,
    PreTokenizedInput,
    TextInput,
    TruncationStrategy,  
)
from transformers.utils import TensorType

logger = logging.getLogger(__name__)

IMAGE_TOKEN = "<image>"
EXTRA_TOKENS = [f"<loc{i:0>4}>" for i in range(1024)] # for object detection task

# Copied from transformers.models.idefics2.processing_idefics2.is_url
def is_url(val) -> bool:
    return isinstance(val, str) and val.startswith("http")


# Copied from transformers.models.idefics2.processing_idefics2.is_image_or_image_url
def is_image_or_image_url(elem):
    return is_url(elem) or is_valid_image(elem)

# Copied from transformers.models.paligemma.processing_paligemma._is_str_or_image
def _is_str_or_image(elem):
    return isinstance(elem, (str)) or is_image_or_image_url(elem)


def build_string_from_input(image_seq_len, image_token):
    """
    Builds a string from the input prompt and image tokens.
    For example, for the call:
    build_string_from_input(
        image_seq_len=3,
        image_token="<im>",
    )
    The output will be:
    "<im><im><im>"
    Args:
        image_seq_len (`int`): The length of the image sequence.
        image_token (`str`): The image token.
    """
    return f"{image_token * image_seq_len}"


class TraVisionProcessor(ProcessorMixin):
    r"""
    Constructs a TraVision processor which wraps a SigLIP image processor and a GPT2 tokenizer into a single processor.

    [`TraVisionProcessor`] offers all the functionalities of [`SiglipImageProcessor`] and [`GPT2TokenizerFast`]. See the
    [`~TraVisionProcessor.__call__`] and [`~TraVisionProcessor.decode`] for more information.

    Args:
        image_processor ([`SiglipImageProcessor`], *optional*):
            The image processor is a required input.
        tokenizer ([`GPT2TokenizerFast`], *optional*):
            The tokenizer is a required input.
        chat_template (`str`, *optional*): A Jinja template which will be used to convert lists of messages
            in a chat into a tokenizable string.
    """

    attributes = ["image_processor", "tokenizer"]
    valid_kwargs = ["chat_template"]
    image_processor_class = "SiglipImageProcessor"
    tokenizer_class = ("GPT2Tokenizer", "GPT2TokenizerFast")

    def __init__(
        self,
        image_processor=None,
        tokenizer=None,
        chat_template=None,
        **kwargs,
    ):
        if image_processor is None:
            raise ValueError("You need to specify an `image_processor`.")
        if tokenizer is None:
            raise ValueError("You need to specify a `tokenizer`.")
        if not hasattr(image_processor, "image_seq_length"):
            raise ValueError("Image processor is missing an `image_seq_length` attribute.")

        self.image_seq_length = image_processor.image_seq_length

        image_token = AddedToken(IMAGE_TOKEN, normalized=False, special=True)
        tokens_to_add = {"additional_special_tokens": [image_token]}
        tokenizer.add_special_tokens(tokens_to_add)
        tokenizer.add_tokens(EXTRA_TOKENS)
        self.image_token_id = tokenizer.convert_tokens_to_ids(IMAGE_TOKEN)
        tokenizer.add_bos_token = False
        tokenizer.add_eos_token = False

        super().__init__(image_processor, tokenizer, chat_template=chat_template)

    def __call__(
        self,
        text: Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]] = None,
        images: ImageInput = None,
        tokenize_newline_separately: bool = True,
        padding: Union[bool, str, PaddingStrategy] = False,
        truncation: Union[bool, str, TruncationStrategy] = None,
        max_length=None,
        return_tensors: Optional[Union[str, TensorType]] = TensorType.PYTORCH,
        do_resize: bool = None,
        do_normalize: bool = None,
        image_mean: Optional[Union[float, List[float]]] = None,
        image_std: Optional[Union[float, List[float]]] = None,
        data_format: Optional["ChannelDimension"] = "channels_first",  # noqa: F821
        input_data_format: Optional[
            Union[str, "ChannelDimension"]  # noqa: F821
        ] = None,
        resample: "PILImageResampling" = None,  # noqa: F821
        do_convert_rgb: bool = None,
        do_thumbnail: bool = None,
        do_align_long_axis: bool = None,
        do_rescale: bool = None,
        labels: Optional[Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]]] = None,
    ) -> BatchFeature:
        """
        Main method to prepare for the model one or several sequences(s) and image(s). This method forwards the `text`
        and `kwargs` arguments to GPT2TokenizerFast's [`~GPT2TokenizerFast.__call__`] if `text` is not `None` to encode
        the text. To prepare the image(s), this method forwards the `images` and `kwrags` arguments to
        SiglipImageProcessor's [`~SiglipImageProcessor.__call__`] if `images` is not `None`. Please refer to the doctsring
        of the above two methods for more information.

        The usage for TraVisionLM fine-tuning preparation follows a standard 4D causal mask where only the prompt and label tokens
        are attended in an auto-regressive manner. The label in `text` are to be passed separately to the __call__ function and 
        will be placed after the prompt, which is the instruction to steer the model generation. 

        Args:
            text (`str`, `List[str]`, `List[List[str]]`):
                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 (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `List[PIL.Image.Image]`, `List[np.ndarray]`, `List[torch.Tensor]`):
                The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
                tensor. In case of a NumPy array/PyTorch tensor, each image should be of shape (C, H, W), where C is a
                number of channels, H and W are image height and width.
            tokenize_newline_separately (`bool`, defaults to `True`):
                Adds a separately tokenized '\n' at the end of the prompt.
            padding (`bool`, `str` or [`~utils.PaddingStrategy`], *optional*, defaults to `False`):
                Select a strategy to pad the returned sequences (according to the model's padding side and padding
                index) among:
                - `True` or `'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
                  sequence if provided).
                - `'max_length'`: Pad to a maximum length specified with the argument `max_length` or to the maximum
                  acceptable input length for the model if that argument is not provided.
                - `False` or `'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of different
                  lengths).
            max_length (`int`, *optional*):
                Maximum length of the returned list and optionally padding length (see above).
            truncation (`bool`, *optional*):
                Activates truncation to cut input sequences longer than `max_length` to `max_length`.
            return_tensors (`str` or [`~utils.TensorType`], *optional*):
                If set, will return tensors of a particular framework. Acceptable values are:

                - `'tf'`: Return TensorFlow `tf.constant` objects.
                - `'pt'`: Return PyTorch `torch.Tensor` objects.
                - `'np'`: Return NumPy `np.ndarray` objects.
                - `'jax'`: Return JAX `jnp.ndarray` objects.
            label (`str`, `List[str]`, `List[List[str]]`):
                The label or batch of labels to be encoded. Only necessary for training. 
                for more information. If your prompt is "<image> Resimde ne var", the label corresponds to the expected prediction "çimlerde uzanan bir köpek".

        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`. If `label`
              is provided, the `input_ids` will also contain the label input ids.
            - **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`.
            - **labels** -- Labels compatible with training if `label` is not None
        """

        # return_token_type_ids = True if labels is not None else False
        return_token_type_ids = True

        if images is None:
            raise ValueError("`images` are expected as arguments to a `TraVisionProcessor` instance.")
        if text is None:
            logger.warning_once(
                "You are using TraVisionLM without a text prefix. It will perform as a picture-captioning model."
            )
            text = "Açıkla"  # default prompt if it is not provided as an argument

        if isinstance(text, List) and isinstance(images, List):
            if len(images) < len(text):
                raise ValueError(
                    f"Received {len(images)} images for {len(text)} prompts. Each prompt should be associated with an image."
                )
        if _is_str_or_image(text):
            text = [text]
        elif isinstance(text, list) and _is_str_or_image(text[0]):
            pass
        text = [f"{prompt}\n" for prompt in text]

        if labels is not None and _is_str_or_image(labels):
            labels = [labels]
        if labels is not None:
            labels = [label + self.tokenizer.eos_token for label in labels]
        
            text = [f"{prompt}{label}" for prompt, label in zip(text, labels)]

        input_strings = [
            build_string_from_input(
                image_seq_len=self.image_seq_length,
                image_token=IMAGE_TOKEN,
            )            
            for _ in text
        ]

        pixel_values = self.image_processor(
            images,
            do_resize=do_resize,
            do_normalize=do_normalize,
            return_tensors=return_tensors,
            image_mean=image_mean,
            image_std=image_std,
            input_data_format=input_data_format,
            data_format=data_format,
            resample=resample,
            do_convert_rgb=do_convert_rgb,
        )["pixel_values"]

        if max_length is not None:
            max_length += self.image_seq_length  # max_length has to account for the image tokens

        inputs = self.tokenizer(
            input_strings,
            text_pair=text,
            return_tensors=return_tensors,
            padding=padding,
            max_length=max_length,
            truncation=truncation,
            return_token_type_ids=return_token_type_ids,
        )

        return_data = {**inputs, "pixel_values": pixel_values}

        if labels is not None:
            labels = inputs["input_ids"].masked_fill(inputs["token_type_ids"] == 0, -100)
            return_data.update({"labels": labels})
        return BatchFeature(data=return_data)

    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.batch_decode with CLIP->GPT2
    def batch_decode(self, *args, **kwargs):
        """
        This method forwards all its arguments to GPT2TokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please
        refer to the docstring of this method for more information.
        """
        return self.tokenizer.batch_decode(*args, **kwargs)

    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.decode with CLIP->GPT2
    def decode(self, *args, **kwargs):
        """
        This method forwards all its arguments to GPT2TokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to
        the docstring of this method for more information.
        """
        return self.tokenizer.decode(*args, **kwargs)

    @property
    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.model_input_names with CLIP->TraVision
    def model_input_names(self):
        tokenizer_input_names = self.tokenizer.model_input_names
        image_processor_input_names = self.image_processor.model_input_names
        return list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))