#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/colpali/modular_colpali.py.
#               Do NOT edit this file manually as any edits will be overwritten by the generation of
#             the file from the modular. If any change should be done, please apply the change to the
#                          modular_colpali.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# Copyright 2024 The HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from typing import Optional, Union

from ...feature_extraction_utils import BatchFeature
from ...image_utils import ImageInput, make_flat_list_of_images
from ...processing_utils import MultiModalData, ProcessingKwargs, ProcessorMixin, Unpack
from ...tokenization_utils_base import AddedToken, PreTokenizedInput, TextInput
from ...utils import auto_docstring, is_torch_available


if is_torch_available():
    import torch


class ColPaliProcessorKwargs(ProcessingKwargs, total=False):
    _defaults = {
        "text_kwargs": {
            "padding": "longest",
            "return_mm_token_type_ids": False,
            "return_text_replacement_offsets": False,
        },
        "images_kwargs": {
            "data_format": "channels_first",
            "do_convert_rgb": True,
        },
        "common_kwargs": {"return_tensors": "pt"},
    }


IMAGE_TOKEN = "<image>"
EXTRA_TOKENS = [f"<loc{i:0>4}>" for i in range(1024)] + [f"<seg{i:0>3}>" for i in range(128)]


@auto_docstring
class ColPaliProcessor(ProcessorMixin):
    valid_processor_kwargs = ColPaliProcessorKwargs

    def __init__(
        self,
        image_processor=None,
        tokenizer=None,
        chat_template=None,
        visual_prompt_prefix: str = "Describe the image.",
        query_prefix: str = "Question: ",
    ):
        r"""
        visual_prompt_prefix (`str`, *optional*, defaults to `"Describe the image."`):
            A string that gets tokenized and prepended to the image tokens.
        query_prefix (`str`, *optional*, defaults to `"Question: "`):
            A prefix to be used for the query.
        """
        self.visual_prompt_prefix = visual_prompt_prefix
        self.query_prefix = query_prefix
        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

        if not hasattr(tokenizer, "image_token"):
            image_token = AddedToken(IMAGE_TOKEN, normalized=False, special=True)
            tokens_to_add = {"additional_special_tokens": [image_token]}
            tokenizer.add_special_tokens(tokens_to_add)
            self.image_token_id = tokenizer.convert_tokens_to_ids(IMAGE_TOKEN)
            self.image_token = IMAGE_TOKEN
        else:
            self.image_token_id = tokenizer.image_token_id
            self.image_token = tokenizer.image_token

        tokenizer.add_tokens(EXTRA_TOKENS)
        tokenizer.add_bos_token = False
        tokenizer.add_eos_token = False

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

    @auto_docstring
    def __call__(
        self,
        images: ImageInput | None = None,
        text: TextInput | PreTokenizedInput | list[TextInput] | list[PreTokenizedInput] = None,
        **kwargs: Unpack[ColPaliProcessorKwargs],
    ) -> 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`. If `suffix`
              is provided, the `input_ids` will also contain the suffix 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 `suffix` is not None
        """
        if text is not None and images is not None:
            raise ValueError("Only one of text or images can be processed at a time")

        kwargs["return_token_type_ids"] = True
        output_kwargs = self._merge_kwargs(
            self.valid_processor_kwargs,
            tokenizer_init_kwargs=self.tokenizer.init_kwargs,
            **kwargs,
        )
        suffix = output_kwargs["text_kwargs"].pop("suffix", self.query_augmentation_token * 10)

        if text is not None:
            text = [f"{self.tokenizer.bos_token}{self.query_prefix}{sample}{suffix}\n" for sample in text]
            if output_kwargs["text_kwargs"].get("max_length", None) is not None:
                output_kwargs["text_kwargs"]["max_length"] += self.image_seq_length

        model_inputs = super().__call__(images=images, text=text, **output_kwargs)
        if images is not None:
            model_inputs["labels"] = model_inputs["input_ids"].masked_fill(model_inputs["token_type_ids"] == 0, -100)
        return model_inputs

    def prepare_inputs_layout(self, images=None, text=None, **kwargs):
        images, text, *_ = super().prepare_inputs_layout(images=images, text=text, **kwargs)
        if images is not None:
            images = make_flat_list_of_images(images)
            text = [
                f"{self.image_token}{self.tokenizer.bos_token}{self.visual_prompt_prefix}\n"
                for _ in range(len(images))
            ]
        return images, text, None, None

    def validate_inputs(
        self,
        images: ImageInput | None = None,
        text: TextInput | PreTokenizedInput | list[TextInput] | list[PreTokenizedInput] | None = None,
        **kwargs: Unpack[ProcessingKwargs],
    ):
        super().validate_inputs(images=images, text=text)
        if text is None and images is None:
            raise ValueError("Either text or images must be provided")

    def replace_image_token(self, image_inputs: dict, image_idx: int, **kwargs) -> str:
        return self.image_token * self.image_seq_length

    def _get_num_multimodal_tokens(self, image_sizes=None, **kwargs):
        """
        Computes the number of placeholder tokens needed for multimodal inputs with the given sizes.

        Args:
            image_sizes (list[list[str]], *optional*):
                The input sizes formatted as (height, width) per each image.
        Returns:
            `MultiModalData`: A `MultiModalData` object holding number of tokens per each of the provided
            input modalities, along with other useful data.
        """
        vision_data = {}
        if image_sizes is not None:
            num_image_tokens = [self.image_seq_length] * len(image_sizes)
            num_image_patches = [1] * len(image_sizes)
            vision_data.update({"num_image_tokens": num_image_tokens, "num_image_patches": num_image_patches})
        return MultiModalData(**vision_data)

    @property
    def model_input_names(self):
        return super().model_input_names + ["token_type_ids", "labels"]

    @property
    def query_augmentation_token(self) -> str:
        """
        Return the query augmentation token.

        Query augmentation buffers are used as reasoning buffers during inference.
        """
        return self.tokenizer.pad_token

    @auto_docstring(
        custom_intro="This method forwards the `images` and `kwargs` arguments to ColPaliProcessor's [`~ColPaliProcessor.__call__`]."
    )
    def process_images(
        self,
        images: ImageInput | None = None,
        **kwargs,
    ) -> BatchFeature:
        return self.__call__(images=images, **kwargs)

    @auto_docstring(
        custom_intro="This method forwards the `text` and `kwargs` arguments to ColPaliProcessor's [`~ColPaliProcessor.__call__`]."
    )
    def process_queries(
        self,
        text: TextInput | list[TextInput],
        **kwargs,
    ) -> BatchFeature:
        return self.__call__(text=text, **kwargs)

    def score_retrieval(
        self,
        query_embeddings: Union["torch.Tensor", list["torch.Tensor"]],
        passage_embeddings: Union["torch.Tensor", list["torch.Tensor"]],
        batch_size: int = 128,
        output_dtype: Optional["torch.dtype"] = None,
        output_device: Union["torch.device", str] = "cpu",
    ) -> "torch.Tensor":
        """
        Compute the late-interaction/MaxSim score (ColBERT-like) for the given multi-vector
        query embeddings (`qs`) and passage embeddings (`ps`). For ColPali, a passage is the
        image of a document page.

        Because the embedding tensors are multi-vector and can thus have different shapes, they
        should be fed as:
        (1) a list of tensors, where the i-th tensor is of shape (sequence_length_i, embedding_dim)
        (2) a single tensor of shape (n_passages, max_sequence_length, embedding_dim) -> usually
            obtained by padding the list of tensors.

        Args:
            query_embeddings (`Union[torch.Tensor, list[torch.Tensor]`): Query embeddings.
            passage_embeddings (`Union[torch.Tensor, list[torch.Tensor]`): Passage embeddings.
            batch_size (`int`, *optional*, defaults to 128): Batch size for computing scores.
            output_dtype (`torch.dtype`, *optional*, defaults to `torch.float32`): The dtype of the output tensor.
                If `None`, the dtype of the input embeddings is used.
            output_device (`torch.device` or `str`, *optional*, defaults to "cpu"): The device of the output tensor.

        Returns:
            `torch.Tensor`: A tensor of shape `(n_queries, n_passages)` containing the scores. The score
            tensor is saved on the "cpu" device.
        """

        if len(query_embeddings) == 0:
            raise ValueError("No queries provided")
        if len(passage_embeddings) == 0:
            raise ValueError("No passages provided")

        if query_embeddings[0].device != passage_embeddings[0].device:
            raise ValueError("Queries and passages must be on the same device")

        if query_embeddings[0].dtype != passage_embeddings[0].dtype:
            raise ValueError("Queries and passages must have the same dtype")

        if output_dtype is None:
            output_dtype = query_embeddings[0].dtype

        scores: list[torch.Tensor] = []

        for i in range(0, len(query_embeddings), batch_size):
            batch_scores: list[torch.Tensor] = []
            batch_queries = torch.nn.utils.rnn.pad_sequence(
                query_embeddings[i : i + batch_size], batch_first=True, padding_value=0
            )
            for j in range(0, len(passage_embeddings), batch_size):
                batch_passages = torch.nn.utils.rnn.pad_sequence(
                    passage_embeddings[j : j + batch_size], batch_first=True, padding_value=0
                )
                batch_scores.append(
                    torch.einsum("bnd,csd->bcns", batch_queries, batch_passages).max(dim=3)[0].sum(dim=2)
                )
            scores.append(torch.cat(batch_scores, dim=1).to(output_dtype).to(output_device))

        return torch.cat(scores, dim=0)


__all__ = ["ColPaliProcessor"]