modeling_tf_vision_encoder_decoder.py

# coding=utf-8
# Copyright 2022 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.
""" Classes to support TF Vision-Encoder-Text-Decoder architectures"""


from __future__ import annotations

import re
import warnings
from typing import Optional, Tuple, Union

import numpy as np
import tensorflow as tf

from ...configuration_utils import PretrainedConfig
from ...modeling_tf_outputs import TFBaseModelOutput, TFSeq2SeqLMOutput
from ...modeling_tf_utils import TFCausalLanguageModelingLoss, TFPreTrainedModel, get_initializer, keras, unpack_inputs
from ...tf_utils import shape_list
from ...utils import (
    ModelOutput,
    add_start_docstrings,
    add_start_docstrings_to_model_forward,
    logging,
    replace_return_docstrings,
)
from ..auto.configuration_auto import AutoConfig
from ..auto.modeling_tf_auto import TFAutoModel, TFAutoModelForCausalLM
from .configuration_vision_encoder_decoder import VisionEncoderDecoderConfig


logger = logging.get_logger(__name__)

_CONFIG_FOR_DOC = "VisionEncoderDecoderConfig"

DEPRECATION_WARNING = (
    "Version v4.17.0 introduces a better way to train encoder-decoder models by computing the loss inside the"
    " encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if"
    " fine-tuning a model trained with versions anterior to 4.17.0. The decoder_input_ids are now created based on the"
    " labels, no need to pass them yourself anymore."
)

VISION_ENCODER_DECODER_START_DOCSTRING = r"""
    This class can be used to initialize an image-to-text-sequence model with any pretrained vision autoencoding model
    as the encoder and any pretrained text autoregressive model as the decoder. The encoder is loaded via
    [`~TFAutoModel.from_pretrained`] function and the decoder is loaded via [`~TFAutoModelForCausalLM.from_pretrained`]
    function. Cross-attention layers are automatically added to the decoder and should be fine-tuned on a downstream
    generative task, like image captioning.

    The effectiveness of initializing sequence-to-sequence models with pretrained checkpoints for sequence generation
    tasks was shown in [Leveraging Pre-trained Checkpoints for Sequence Generation
    Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn. Michael Matena, Yanqi
    Zhou, Wei Li, Peter J. Liu.

    Additionally, in [TrOCR: Transformer-based Optical Character Recognition with Pre-trained
    Models](https://arxiv.org/abs/2109.10282) it is shown how leveraging large pretrained vision models for optical
    character recognition (OCR) yields a significant performance improvement.

    After such a Vision-Encoder-Text-Decoder model has been trained/fine-tuned, it can be saved/loaded just like any
    other models (see the examples for more information).

    This model inherits from [`TFPreTrainedModel`]. Check the superclass documentation for the generic methods the
    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
    etc.)

    This model is also a [keras.Model](https://www.tensorflow.org/api_docs/python/tf/keras/Model) subclass. Use it
    as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and
    behavior.

    Parameters:
        config ([`VisionEncoderDecoderConfig`]): Model configuration class with all the parameters of the model.
            Initializing with a config file does not load the weights associated with the model, only the
            configuration. Check out the [`~TFPreTrainedModel.from_pretrained`] method to load the model weights.
"""

VISION_ENCODER_DECODER_INPUTS_DOCSTRING = r"""
    Args:
        pixel_values (`np.ndarray`, `tf.Tensor`, `List[tf.Tensor]` ``Dict[str, tf.Tensor]` or `Dict[str, np.ndarray]` and each example must have the shape `(batch_size, num_channels, height, width)`):
            Pixel values. Pixel values can be obtained using the vision's model's image processor. For example, using
            [`AutoImageProcessor`]. See [`ViTImageProcessor.__call__`] for details.
        decoder_input_ids (`np.ndarray` or `tf.Tensor` of shape `(batch_size, target_sequence_length)`, *optional*):
            Indices of decoder input sequence tokens in the vocabulary.

            Indices can be obtained using [`PreTrainedTokenizer`]. See [`PreTrainedTokenizer.encode`] and
            [`PreTrainedTokenizer.__call__`] for details.

            [What are input IDs?](../glossary#input-ids)

            If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
            `past_key_values`).

            Provide for sequence to sequence training to the decoder. Indices can be obtained using
            [`PreTrainedTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for
            details.
        decoder_attention_mask (`np.ndarray` or `tf.Tensor` of shape `(batch_size, target_sequence_length)`, *optional*):
            Default behavior: generate a tensor that ignores pad tokens in `decoder_input_ids`. Causal mask will also
            be used by default.
        encoder_outputs (`tuple(tuple(tf.Tensor)`, *optional*):
            This tuple must consist of (`last_hidden_state`, *optional*: `hidden_states`, *optional*: `attentions`)
            `last_hidden_state` (`tf.Tensor` of shape `({0}, hidden_size)`) is a tensor of hidden-states at the output
            of the last layer of the encoder. Used in the cross-attention of the decoder.
        past_key_values (`tuple(tuple(tf.Tensor))` of length `config.n_layers` with each tuple having 4 tensors of shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.

            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
            don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
            `decoder_input_ids` of shape `({0})`.
        decoder_inputs_embeds (`np.ndarray` or `tf.Tensor` of shape `(batch_size, target_sequence_length, hidden_size)`, *optional*):
            Optionally, instead of passing `decoder_input_ids` you can choose to directly pass an embedded
            representation. This is useful if you want more control over how to convert `decoder_input_ids` indices
            into associated vectors than the model's internal embedding lookup matrix.
        labels (`np.ndarray` or `tf.Tensor` of shape `({0})`, *optional*):
            Labels for computing the masked language modeling loss for the decoder. Indices should be in `[-100, 0,
            ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored
            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`
        use_cache (`bool`, *optional*):
            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
            `past_key_values`).
        output_attentions (`bool`, *optional*):
            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
            tensors for more detail.
        output_hidden_states (`bool`, *optional*):
            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
            more detail.
        return_dict (`bool`, *optional*):
            If set to `True`, the model will return a [`~utils.Seq2SeqLMOutput`] instead of a plain tuple.
        training (`bool`, *optional*, defaults to `False`):
            Whether or not to use the model in training mode (some modules like dropout modules have different
            behaviors between training and evaluation).
        kwargs (*optional*): Remaining dictionary of keyword arguments. Keyword arguments come in two flavors:

            - Without a prefix which will be input as `**encoder_kwargs` for the encoder forward function.
            - With a *decoder_* prefix which will be input as `**decoder_kwargs` for the decoder forward function.
"""


# Copied from transformers.models.encoder_decoder.modeling_tf_encoder_decoder.shift_tokens_right
def shift_tokens_right(input_ids: tf.Tensor, pad_token_id: int, decoder_start_token_id: int):
    if pad_token_id is None:
        raise ValueError("Make sure to set the pad_token_id attribute of the model's configuration.")
    pad_token_id = tf.cast(pad_token_id, input_ids.dtype)

    if decoder_start_token_id is None:
        raise ValueError("Make sure to set the decoder_start_token_id attribute of the model's configuration.")
    decoder_start_token_id = tf.cast(decoder_start_token_id, input_ids.dtype)

    start_tokens = tf.fill((shape_list(input_ids)[0], 1), decoder_start_token_id)
    shifted_input_ids = tf.concat([start_tokens, input_ids[:, :-1]], -1)
    # replace possible -100 values in labels by `pad_token_id`
    shifted_input_ids = tf.where(
        shifted_input_ids == -100, tf.fill(shape_list(shifted_input_ids), pad_token_id), shifted_input_ids
    )

    # "Verify that `labels` has only positive values and -100"
    assert_gte0 = tf.debugging.assert_greater_equal(shifted_input_ids, tf.constant(0, dtype=input_ids.dtype))

    # Make sure the assertion op is called by wrapping the result in an identity no-op
    with tf.control_dependencies([assert_gte0]):
        shifted_input_ids = tf.identity(shifted_input_ids)

    return shifted_input_ids


@add_start_docstrings(VISION_ENCODER_DECODER_START_DOCSTRING)
class TFVisionEncoderDecoderModel(TFPreTrainedModel, TFCausalLanguageModelingLoss):
    r"""
    [`TFVisionEncoderDecoderModel`] is a generic model class that will be instantiated as a transformer architecture
    with one of the base vision model classes of the library as encoder and another one of the base model classes as
    decoder when created with the [`~TFAutoModel.from_pretrained`] class method for the encoder and
    [`~TFAutoModelForCausalLM.from_pretrained`] class method for the decoder.
    """

    config_class = VisionEncoderDecoderConfig
    base_model_prefix = "vision_encoder_decoder"
    load_weight_prefix = "tf_vision_encoder_decoder_model"
    main_input_name = "pixel_values"

    def __init__(
        self,
        config: Optional[PretrainedConfig] = None,
        encoder: Optional[TFPreTrainedModel] = None,
        decoder: Optional[TFPreTrainedModel] = None,
    ):
        if config is None and (encoder is None or decoder is None):
            raise ValueError("Either a configuration or an encoder and a decoder has to be provided.")
        if config is None:
            config = VisionEncoderDecoderConfig.from_encoder_decoder_configs(encoder.config, decoder.config)
        else:
            if not isinstance(config, self.config_class):
                raise ValueError(f"config: {config} has to be of type {self.config_class}")

        if config.decoder.cross_attention_hidden_size is not None:
            if config.decoder.cross_attention_hidden_size != config.encoder.hidden_size:
                raise ValueError(
                    "If `cross_attention_hidden_size` is specified in the decoder's configuration, it has to be equal"
                    f" to the encoder's `hidden_size`. Got {config.decoder.cross_attention_hidden_size} for"
                    f" `config.decoder.cross_attention_hidden_size` and {config.encoder.hidden_size} for"
                    " `config.encoder.hidden_size`."
                )

        # initialize with config
        super().__init__(config)

        if encoder is None:
            encoder = TFAutoModel.from_config(config.encoder, name="encoder")

        if decoder is None:
            decoder = TFAutoModelForCausalLM.from_config(config.decoder, name="decoder")

        self.encoder = encoder
        self.decoder = decoder

        if self.encoder.config.to_dict() != self.config.encoder.to_dict():
            logger.warning(
                f"Config of the encoder: {self.encoder.__class__} is overwritten by shared encoder config:"
                f" {self.config.encoder}"
            )
        if self.decoder.config.to_dict() != self.config.decoder.to_dict():
            logger.warning(
                f"Config of the decoder: {self.decoder.__class__} is overwritten by shared decoder config:"
                f" {self.config.decoder}"
            )

        # make sure that the individual model's config refers to the shared config
        # so that the updates to the config will be synced
        self.encoder.config = self.config.encoder
        self.decoder.config = self.config.decoder

        # encoder outputs might need to be projected to different dimension for decoder
        if (
            self.encoder.config.hidden_size != self.decoder.config.hidden_size
            and self.decoder.config.cross_attention_hidden_size is None
        ):
            self.enc_to_dec_proj = keras.layers.Dense(
                units=self.decoder.config.hidden_size,
                kernel_initializer=get_initializer(config.encoder.initializer_range),
                name="enc_to_dec_proj",
            )

        if self.encoder.get_output_embeddings() is not None:
            raise ValueError(
                f"The encoder {self.encoder} should not have a LM Head. Please use a model without LM Head"
            )

    @property
    def input_signature(self):
        vision_config = self.config.encoder
        if hasattr(vision_config, "vision_config"):
            vision_config = vision_config.vision_config
        if hasattr(vision_config, "image_size"):
            image_size = vision_config.image_size
        else:
            image_size = vision_config.input_size
        return {
            "pixel_values": tf.TensorSpec(
                shape=(
                    None,
                    vision_config.num_channels,
                    image_size,
                    image_size,
                ),
                dtype=tf.float32,
            ),
            "decoder_input_ids": tf.TensorSpec(shape=(None, None), dtype=tf.int32, name="decoder_input_ids"),
        }

    def get_encoder(self):
        return self.encoder

    def get_decoder(self):
        return self.decoder

    def get_input_embeddings(self):
        return self.encoder.get_input_embeddings()

    def get_output_embeddings(self):
        return self.decoder.get_output_embeddings()

    def set_output_embeddings(self, new_embeddings):
        return self.decoder.set_output_embeddings(new_embeddings)

    def tf_to_pt_weight_rename(self, tf_weight):
        # Matt: The TF and PT weights don't align because our TF base classes have an extra layer compared to PT models
        # (the main model stem is in the MainLayer class). If we remove that layer, then weight names sync up as normal.
        # However, the name of that extra layer is the name of the MainLayer in the base model. We make the assumption
        # here that the config model_type is the same as the name of the MainLayer. I don't know of anywhere that's
        # not the case, and I wasn't sure how else to go from the config to the correct MainLayer name!

        # This override is only needed in the case where we're crossloading weights from PT. However, since weights are
        # often safetensors now, we don't know if we're going to be crossloading until we sniff the weights file.
        # Therefore, we specify tf_to_pt_weight_rename anyway, and let the super method figure out if it needs it
        # or not.
        encoder_model_type = self.config.encoder.model_type
        if "encoder" in tf_weight and "decoder" not in tf_weight:
            return (re.sub(rf"encoder\.{encoder_model_type}\.", "encoder.", tf_weight),)
        else:
            return (tf_weight,)

    @classmethod
    def from_encoder_decoder_pretrained(
        cls,
        encoder_pretrained_model_name_or_path: str = None,
        decoder_pretrained_model_name_or_path: str = None,
        *model_args,
        **kwargs,
    ) -> TFPreTrainedModel:
        r"""
        Instantiate an encoder and a decoder from one or two base classes of the library from pretrained model
        checkpoints.


        Params:
            encoder_pretrained_model_name_or_path (`str`, *optional*):
                Information necessary to initiate the encoder. Can be either:

                    - A string, the *model id* of a pretrained model hosted inside a model repo on huggingface.co. An
                      example is `google/vit-base-patch16-224-in21k`.
                    - A path to a *directory* containing model weights saved using
                      [`~TFPreTrainedModel.save_pretrained`], e.g., `./my_model_directory/`.
                    - A path or url to a *pytorch index checkpoint file* (e.g, `./pt_model/`). In this case,
                      `encoder_from_pt` should be set to `True`.

            decoder_pretrained_model_name_or_path (`str`, *optional*, defaults to *None*):
                Information necessary to initiate the decoder. Can be either:

                    - A string, the *model id* of a pretrained model hosted inside a model repo on huggingface.co.
                    - A path to a *directory* containing model weights saved using
                      [`~TFPreTrainedModel.save_pretrained`], e.g., `./my_model_directory/`.
                    - A path or url to a *pytorch checkpoint file* (e.g, `./pt_model/`). In this case,
                      `decoder_from_pt` should be set to `True`.

            model_args (remaining positional arguments, *optional*):
                All remaning positional arguments will be passed to the underlying model's `__init__` method.

            kwargs (remaining dictionary of keyword arguments, *optional*):
                Can be used to update the configuration object (after it being loaded) and initiate the model (e.g.,
                `output_attentions=True`).

                - To update the encoder configuration, use the prefix *encoder_* for each configuration parameter.
                - To update the decoder configuration, use the prefix *decoder_* for each configuration parameter.
                - To update the parent model configuration, do not use a prefix for each configuration parameter.

                Behaves differently depending on whether a `config` is provided or automatically loaded.

        Example:

        ```python
        >>> from transformers import TFVisionEncoderDecoderModel

        >>> # initialize a vit-bert from a pretrained ViT and a pretrained BERT model. Note that the cross-attention layers will be randomly initialized
        >>> model = TFVisionEncoderDecoderModel.from_encoder_decoder_pretrained(
        ...     "google/vit-base-patch16-224-in21k", "google-bert/bert-base-uncased"
        ... )
        >>> # saving model after fine-tuning
        >>> model.save_pretrained("./vit-bert")
        >>> # load fine-tuned model
        >>> model = TFVisionEncoderDecoderModel.from_pretrained("./vit-bert")
        ```"""

        kwargs_encoder = {
            argument[len("encoder_") :]: value for argument, value in kwargs.items() if argument.startswith("encoder_")
        }

        kwargs_decoder = {
            argument[len("decoder_") :]: value for argument, value in kwargs.items() if argument.startswith("decoder_")
        }

        # remove encoder, decoder kwargs from kwargs
        for key in kwargs_encoder.keys():
            del kwargs["encoder_" + key]
        for key in kwargs_decoder.keys():
            del kwargs["decoder_" + key]

        # Load and initialize the encoder and decoder
        # The distinction between encoder and decoder at the model level is made
        # by the value of the flag `is_decoder` that we need to set correctly.
        encoder = kwargs_encoder.pop("model", None)
        if encoder is None:
            if encoder_pretrained_model_name_or_path is None:
                raise ValueError(
                    "If `encoder_model` is not defined as an argument, a `encoder_pretrained_model_name_or_path` has "
                    "to be defined."
                )

            if "config" not in kwargs_encoder:
                encoder_config = AutoConfig.from_pretrained(encoder_pretrained_model_name_or_path)
                if encoder_config.is_decoder is True or encoder_config.add_cross_attention is True:
                    logger.info(
                        f"Initializing {encoder_pretrained_model_name_or_path} as a encoder model "
                        "from a decoder model. Cross-attention and casual mask are disabled."
                    )
                    encoder_config.is_decoder = False
                    encoder_config.add_cross_attention = False

                kwargs_encoder["config"] = encoder_config

            kwargs_encoder["name"] = "encoder"
            kwargs_encoder["load_weight_prefix"] = cls.load_weight_prefix
            encoder = TFAutoModel.from_pretrained(encoder_pretrained_model_name_or_path, *model_args, **kwargs_encoder)

        decoder = kwargs_decoder.pop("model", None)
        if decoder is None:
            if decoder_pretrained_model_name_or_path is None:
                raise ValueError(
                    "If `decoder_model` is not defined as an argument, a `decoder_pretrained_model_name_or_path` has "
                    "to be defined."
                )

            if "config" not in kwargs_decoder:
                decoder_config = AutoConfig.from_pretrained(decoder_pretrained_model_name_or_path)
                if decoder_config.is_decoder is False or decoder_config.add_cross_attention is False:
                    logger.info(
                        f"Initializing {decoder_pretrained_model_name_or_path} as a decoder model. Cross attention"
                        f" layers are added to {decoder_pretrained_model_name_or_path} and randomly initialized if"
                        f" {decoder_pretrained_model_name_or_path}'s architecture allows for cross attention layers."
                    )
                    decoder_config.is_decoder = True
                    decoder_config.add_cross_attention = True

                kwargs_decoder["config"] = decoder_config

            if kwargs_decoder["config"].is_decoder is False or kwargs_decoder["config"].add_cross_attention is False:
                logger.warning(
                    f"Decoder model {decoder_pretrained_model_name_or_path} is not initialized as a decoder. "
                    f"In order to initialize {decoder_pretrained_model_name_or_path} as a decoder, "
                    "make sure that the attributes `is_decoder` and `add_cross_attention` of `decoder_config` "
                    "passed to `.from_encoder_decoder_pretrained(...)` are set to `True` or do not pass a "
                    "`decoder_config` to `.from_encoder_decoder_pretrained(...)`"
                )

            kwargs_decoder["name"] = "decoder"
            kwargs_decoder["load_weight_prefix"] = cls.load_weight_prefix
            decoder = TFAutoModelForCausalLM.from_pretrained(decoder_pretrained_model_name_or_path, **kwargs_decoder)

        # Make sure these 2 `keras.Model` have fixed names so `from_pretrained` could load model weights correctly.
        if encoder.name != "encoder":
            raise ValueError("encoder model must be created with the name `encoder`.")
        if decoder.name != "decoder":
            raise ValueError("decoder model must be created with the name `decoder`.")

        # instantiate config with corresponding kwargs
        config = VisionEncoderDecoderConfig.from_encoder_decoder_configs(encoder.config, decoder.config, **kwargs)
        return cls(encoder=encoder, decoder=decoder, config=config)

    @unpack_inputs
    @add_start_docstrings_to_model_forward(
        VISION_ENCODER_DECODER_INPUTS_DOCSTRING.format("batch_size, sequence_length")
    )
    @replace_return_docstrings(output_type=TFSeq2SeqLMOutput, config_class=_CONFIG_FOR_DOC)
    def call(
        self,
        pixel_values: np.ndarray | tf.Tensor | None = None,
        decoder_input_ids: np.ndarray | tf.Tensor | None = None,
        decoder_attention_mask: np.ndarray | tf.Tensor | None = None,
        encoder_outputs: Optional[Union[Tuple, TFBaseModelOutput]] = None,
        past_key_values: Optional[Tuple[Tuple[Union[np.ndarray, tf.Tensor]]]] = None,
        decoder_inputs_embeds: np.ndarray | tf.Tensor | None = None,
        labels: np.ndarray | tf.Tensor | None = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        training: bool = False,
        **kwargs,
    ) -> Union[TFSeq2SeqLMOutput, Tuple[tf.Tensor]]:
        r"""
        Returns:

        Examples:

        ```python
        >>> from transformers import AutoImageProcessor, AutoTokenizer, TFVisionEncoderDecoderModel
        >>> from PIL import Image
        >>> import requests

        >>> image_processor = AutoImageProcessor.from_pretrained("google/vit-base-patch16-224-in21k")
        >>> decoder_tokenizer = AutoTokenizer.from_pretrained("openai-community/gpt2")

        >>> # initialize a bert2gpt2 from a pretrained BERT and GPT2 models. Note that the cross-attention layers will be randomly initialized
        >>> model = TFVisionEncoderDecoderModel.from_encoder_decoder_pretrained(
        ...     "google/vit-base-patch16-224-in21k", "openai-community/gpt2"
        ... )

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> img = Image.open(requests.get(url, stream=True).raw)

        >>> # forward
        >>> pixel_values = image_processor(images=img, return_tensors="tf").pixel_values  # Batch size 1
        >>> decoder_input_ids = decoder_tokenizer("Linda Davis", return_tensors="tf").input_ids  # Batch size 1
        >>> outputs = model(pixel_values=pixel_values, decoder_input_ids=decoder_input_ids)

        >>> # training
        >>> outputs = model(pixel_values=pixel_values, decoder_input_ids=decoder_input_ids, labels=decoder_input_ids)
        >>> loss, logits = outputs.loss, outputs.logits

        >>> # save and load from pretrained
        >>> model.save_pretrained("vit-gpt2")
        >>> model = TFVisionEncoderDecoderModel.from_pretrained("vit-gpt2")

        >>> # generation
        >>> generated = model.generate(pixel_values, decoder_start_token_id=model.config.decoder.bos_token_id)
        ```"""
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        kwargs_encoder = {argument: value for argument, value in kwargs.items() if not argument.startswith("decoder_")}

        kwargs_decoder = {
            argument[len("decoder_") :]: value for argument, value in kwargs.items() if argument.startswith("decoder_")
        }

        # Let the user be responsible for the expected format.
        if encoder_outputs is not None:
            if return_dict and not isinstance(encoder_outputs, ModelOutput):
                raise ValueError(
                    "If `return_dict=True` and `encoder_outputs` is provided, it should be an instance of "
                    f"`ModelOutput`. Got an instance {type(encoder_outputs)} for `encoder_outputs`."
                )

        if encoder_outputs is None:
            encoder_inputs = {
                "input_ids": pixel_values,
                "output_attentions": output_attentions,
                "output_hidden_states": output_hidden_states,
                "return_dict": return_dict,
                "training": training,
            }

            # Add arguments to encoder from `kwargs_encoder`
            encoder_inputs.update(kwargs_encoder)

            if "input_ids" in encoder_inputs:
                encoder_inputs["pixel_values"] = encoder_inputs.pop("input_ids")

            if encoder_inputs["pixel_values"] is None:
                raise ValueError("You have to specify pixel_values")

            # Handle the case where the inputs are passed as a single dict which contains `labels`.
            # The `labels` shouldn't be passed to `self.encoder` below, because it is a based model without this
            # parameter (otherwise, an error occurs when `input_processing` is called inside `self.encoder.call()`).
            if "labels" in encoder_inputs:
                labels = encoder_inputs.pop("labels")

            # handle the init case where `dummy_inputs` returns a dict containing `decoder_input_ids`.
            if "decoder_input_ids" in encoder_inputs:
                decoder_input_ids = encoder_inputs.pop("decoder_input_ids")
            # handle the init case where `dummy_inputs` returns a dict containing `decoder_input_ids`.
            if "decoder_attention_mask" in encoder_inputs:
                decoder_attention_mask = encoder_inputs.pop("decoder_attention_mask")

            encoder_outputs = self.encoder(**encoder_inputs)

        encoder_hidden_states = encoder_outputs[0]

        # optionally project encoder_hidden_states
        if (
            self.encoder.config.hidden_size != self.decoder.config.hidden_size
            and self.decoder.config.cross_attention_hidden_size is None
        ):
            encoder_hidden_states = self.enc_to_dec_proj(encoder_hidden_states)

        if (labels is not None) and (decoder_input_ids is None and decoder_inputs_embeds is None):
            decoder_input_ids = shift_tokens_right(
                labels, self.config.pad_token_id, self.config.decoder_start_token_id
            )

        batch_size, sequence_length = shape_list(encoder_hidden_states)[:2]
        encoder_attention_mask = tf.ones(shape=(batch_size, sequence_length), dtype=tf.int32)

        decoder_inputs = {
            "input_ids": decoder_input_ids,
            "attention_mask": decoder_attention_mask,
            "encoder_hidden_states": encoder_hidden_states,
            "encoder_attention_mask": encoder_attention_mask,
            "inputs_embeds": decoder_inputs_embeds,
            "output_attentions": output_attentions,
            "output_hidden_states": output_hidden_states,
            "use_cache": use_cache,
            "past_key_values": past_key_values,
            "return_dict": return_dict,
            "training": training,
        }

        # Add arguments to decoder from `kwargs_decoder`
        decoder_inputs.update(kwargs_decoder)

        decoder_outputs = self.decoder(**decoder_inputs)

        logits = decoder_outputs[0]

        # Compute loss independent from decoder (as some shift the logits inside them)
        loss = None
        if labels is not None:
            warnings.warn(DEPRECATION_WARNING, FutureWarning)
            loss = self.hf_compute_loss(labels, logits)

        if not return_dict:
            past_key_values = None
            if use_cache:
                past_key_values = decoder_outputs[1]
            # The starting index of the remaining elements in `decoder_outputs`
            start_index = sum([1 if x is not None else 0 for x in (loss, logits, past_key_values)])

            if not isinstance(encoder_outputs, tuple):
                encoder_outputs = encoder_outputs.to_tuple()
            output = (loss, logits, past_key_values) + decoder_outputs[start_index:] + encoder_outputs
            output = tuple([x for x in output if x is not None])
            return output

        return TFSeq2SeqLMOutput(
            loss=loss,
            logits=decoder_outputs.logits,
            past_key_values=decoder_outputs.past_key_values,
            decoder_hidden_states=decoder_outputs.hidden_states,
            decoder_attentions=decoder_outputs.attentions,
            cross_attentions=decoder_outputs.cross_attentions,
            encoder_last_hidden_state=encoder_outputs.last_hidden_state,
            encoder_hidden_states=encoder_outputs.hidden_states,
            encoder_attentions=encoder_outputs.attentions,
        )

    def serving_output(self, output):
        pkv = tf.tuple(output.past_key_values)[1] if self.config.decoder.use_cache else None
        dec_hs = (
            tf.convert_to_tensor(output.decoder_hidden_states) if self.config.decoder.output_hidden_states else None
        )
        dec_attns = tf.convert_to_tensor(output.decoder_attentions) if self.config.decoder.output_attentions else None
        enc_hs = (
            tf.convert_to_tensor(output.encoder_hidden_states) if self.config.encoder.output_hidden_states else None
        )
        enc_attns = tf.convert_to_tensor(output.encoder_attentions) if self.config.encoder.output_attentions else None
        cross_attns = (
            tf.convert_to_tensor(output.cross_attentions)
            if self.config.decoder.output_attentions and output.cross_attentions is not None
            else None
        )

        return TFSeq2SeqLMOutput(
            logits=output.logits,
            past_key_values=pkv,
            decoder_hidden_states=dec_hs,
            decoder_attentions=dec_attns,
            encoder_last_hidden_state=output.encoder_last_hidden_state,
            encoder_hidden_states=enc_hs,
            encoder_attentions=enc_attns,
            cross_attentions=cross_attns,
        )

    def prepare_inputs_for_generation(
        self, input_ids, past_key_values=None, attention_mask=None, use_cache=None, encoder_outputs=None, **kwargs
    ):
        decoder_inputs = self.decoder.prepare_inputs_for_generation(input_ids, past_key_values=past_key_values)
        decoder_attention_mask = decoder_inputs["attention_mask"] if "attention_mask" in decoder_inputs else None
        past_key_values = decoder_inputs.get("past_key_values")
        input_dict = {
            "pixel_values": None,  # needs to be passed to make Keras.layer.__call__ happy
            "attention_mask": attention_mask,
            "decoder_attention_mask": decoder_attention_mask,
            "decoder_input_ids": decoder_inputs["input_ids"],
            # TODO (joao): the `TFBaseModelOutput` wrapper should not be needed after the generate refactor is complete
            "encoder_outputs": TFBaseModelOutput(last_hidden_state=encoder_outputs[0]),
            "past_key_values": past_key_values,
            "use_cache": use_cache,
        }
        return input_dict

    def prepare_decoder_input_ids_from_labels(self, labels: tf.Tensor):
        return shift_tokens_right(labels, self.config.pad_token_id, self.config.decoder_start_token_id)

    def resize_token_embeddings(self, *args, **kwargs):
        raise NotImplementedError(
            "Resizing the embedding layers via the TFVisionEncoderDecoderModel directly is not supported. "
            "Please use the respective methods of the wrapped objects (model.decoder.resize_token_embeddings(...))"
        )

    def build(self, input_shape=None):
        if self.built:
            return
        self.built = True
        if getattr(self, "enc_to_dec_proj", None) is not None:
            with tf.name_scope(self.enc_to_dec_proj.name):
                self.enc_to_dec_proj.build([None, None, self.encoder.config.hidden_size])
        if getattr(self, "encoder", None) is not None:
            with tf.name_scope(self.encoder.name):
                self.encoder.build(None)
        if getattr(self, "decoder", None) is not None:
            with tf.name_scope(self.decoder.name):
                self.decoder.build(None)