# Copyright 2020 The HuggingFace Team. All rights reserved.
#
# 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 collections import OrderedDict
from dataclasses import dataclass
from dataclasses import fields
from typing import Optional
from typing import Tuple

import paddle


class ModelOutput(OrderedDict):
    """
    Base class for all model outputs as dataclass. Has a `__getitem__` that allows indexing by integer or slice (like a
    tuple) or strings (like a dictionary) that will ignore the `None` attributes. Otherwise behaves like a regular
    python dictionary.

    <Tip warning={true}>

    You can't unpack a `ModelOutput` directly. Use the [`~utils.ModelOutput.to_tuple`] method to convert it to a tuple
    before.

    </Tip>
    """

    def __post_init__(self):
        class_fields = fields(self)

        # Safety and consistency checks
        if not len(class_fields):
            raise ValueError(f"{self.__class__.__name__} has no fields.")
        if not all(field.default is None for field in class_fields[1:]):
            raise ValueError(
                f"{self.__class__.__name__} should not have more than one required field."
            )

        first_field = getattr(self, class_fields[0].name)
        other_fields_are_none = all(
            getattr(self, field.name) is None for field in class_fields[1:])

        if other_fields_are_none and not paddle.is_tensor(first_field):
            if isinstance(first_field, dict):
                iterator = first_field.items()
                first_field_iterator = True
            else:
                try:
                    iterator = iter(first_field)
                    first_field_iterator = True
                except TypeError:
                    first_field_iterator = False

            # if we provided an iterator as first field and the iterator is a (key, value) iterator
            # set the associated fields
            if first_field_iterator:
                for element in iterator:
                    if (not isinstance(element, (list, tuple)) or
                            not len(element) == 2 or
                            not isinstance(element[0], str)):
                        break
                    setattr(self, element[0], element[1])
                    if element[1] is not None:
                        self[element[0]] = element[1]
            elif first_field is not None:
                self[class_fields[0].name] = first_field
        else:
            for field in class_fields:
                v = getattr(self, field.name)
                if v is not None:
                    self[field.name] = v

    def __delitem__(self, *args, **kwargs):
        raise Exception(
            f"You cannot use ``__delitem__`` on a {self.__class__.__name__} instance."
        )

    def setdefault(self, *args, **kwargs):
        raise Exception(
            f"You cannot use ``setdefault`` on a {self.__class__.__name__} instance."
        )

    def pop(self, *args, **kwargs):
        raise Exception(
            f"You cannot use ``pop`` on a {self.__class__.__name__} instance.")

    def update(self, *args, **kwargs):
        raise Exception(
            f"You cannot use ``update`` on a {self.__class__.__name__} instance."
        )

    def __getitem__(self, k):
        if isinstance(k, str):
            inner_dict = {k: v for (k, v) in self.items()}
            return inner_dict[k]
        else:
            return self.to_tuple()[k]

    def __setattr__(self, name, value):
        if name in self.keys() and value is not None:
            # Don't call self.__setitem__ to avoid recursion errors
            super().__setitem__(name, value)
        super().__setattr__(name, value)

    def __setitem__(self, key, value):
        # Will raise a KeyException if needed
        super().__setitem__(key, value)
        # Don't call self.__setattr__ to avoid recursion errors
        super().__setattr__(key, value)

    def to_tuple(self) -> Tuple:
        """
        Convert self to a tuple containing all the attributes/keys that are not `None`.
        """
        return tuple(self[k] for k in self.keys())


@dataclass
class BaseModelOutput(ModelOutput):
    """
    Base class for model's outputs, with potential hidden states and attentions.

    Args:
        last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    last_hidden_state: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class BaseModelOutputWithNoAttention(ModelOutput):
    """
    Base class for model's outputs, with potential hidden states.

    Args:
        last_hidden_state (`paddle.Tensor` of shape `(batch_size, num_channels, height, width)`):
            Sequence of hidden-states at the output of the last layer of the model.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, num_channels, height, width)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
    """

    last_hidden_state: paddle = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class BaseModelOutputWithPooling(ModelOutput):
    """
    Base class for model's outputs that also contains a pooling of the last hidden states.

    Args:
        last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.
        pooler_output (`paddle.Tensor` of shape `(batch_size, hidden_size)`):
            Last layer hidden-state of the first token of the sequence (classification token) after further processing
            through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns
            the classification token after processing through a linear layer and a tanh activation function. The linear
            layer weights are trained from the next sentence prediction (classification) objective during pretraining.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    last_hidden_state: paddle.Tensor = None
    pooler_output: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class BaseModelOutputWithPoolingAndNoAttention(ModelOutput):
    """
    Base class for model's outputs that also contains a pooling of the last hidden states.

    Args:
        last_hidden_state (`paddle.Tensor` of shape `(batch_size, num_channels, height, width)`):
            Sequence of hidden-states at the output of the last layer of the model.
        pooler_output (`paddle.Tensor` of shape `(batch_size, hidden_size)`):
            Last layer hidden-state after a pooling operation on the spatial dimensions.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, num_channels, height, width)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
    """

    last_hidden_state: paddle.Tensor = None
    pooler_output: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class BaseModelOutputWithPast(ModelOutput):
    """
    Base class for model's outputs that may also contain a past key/values (to speed up sequential decoding).

    Args:
        last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.

            If `past_key_values` is used only the last hidden-state of the sequences of shape `(batch_size, 1,
            hidden_size)` is output.
        past_key_values (`tuple(tuple(paddle.Tensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `tuple(paddle.Tensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and optionally if
            `config.is_encoder_decoder=True` 2 additional tensors of shape `(batch_size, num_heads,
            encoder_sequence_length, embed_size_per_head)`.

            Contains pre-computed hidden-states (key and values in the self-attention blocks and optionally if
            `config.is_encoder_decoder=True` in the cross-attention blocks) that can be used (see `past_key_values`
            input) to speed up sequential decoding.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    last_hidden_state: paddle.Tensor = None
    past_key_values: Optional[Tuple[Tuple[paddle.Tensor]]] = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class BaseModelOutputWithCrossAttentions(ModelOutput):
    """
    Base class for model's outputs, with potential hidden states and attentions.

    Args:
        last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
        cross_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` and `config.add_cross_attention=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
            weighted average in the cross-attention heads.
    """

    last_hidden_state: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None
    cross_attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class BaseModelOutputWithPoolingAndCrossAttentions(ModelOutput):
    """
    Base class for model's outputs that also contains a pooling of the last hidden states.

    Args:
        last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.
        pooler_output (`paddle.Tensor` of shape `(batch_size, hidden_size)`):
            Last layer hidden-state of the first token of the sequence (classification token) after further processing
            through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns
            the classification token after processing through a linear layer and a tanh activation function. The linear
            layer weights are trained from the next sentence prediction (classification) objective during pretraining.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
        cross_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` and `config.add_cross_attention=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
            weighted average in the cross-attention heads.
        past_key_values (`tuple(tuple(paddle.Tensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `tuple(paddle.Tensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and optionally if
            `config.is_encoder_decoder=True` 2 additional tensors of shape `(batch_size, num_heads,
            encoder_sequence_length, embed_size_per_head)`.

            Contains pre-computed hidden-states (key and values in the self-attention blocks and optionally if
            `config.is_encoder_decoder=True` in the cross-attention blocks) that can be used (see `past_key_values`
            input) to speed up sequential decoding.
    """

    last_hidden_state: paddle.Tensor = None
    pooler_output: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    past_key_values: Optional[Tuple[Tuple[paddle.Tensor]]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None
    cross_attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class BaseModelOutputWithPastAndCrossAttentions(ModelOutput):
    """
    Base class for model's outputs that may also contain a past key/values (to speed up sequential decoding).

    Args:
        last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.

            If `past_key_values` is used only the last hidden-state of the sequences of shape `(batch_size, 1,
            hidden_size)` is output.
        past_key_values (`tuple(tuple(paddle.Tensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `tuple(paddle.Tensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and optionally if
            `config.is_encoder_decoder=True` 2 additional tensors of shape `(batch_size, num_heads,
            encoder_sequence_length, embed_size_per_head)`.

            Contains pre-computed hidden-states (key and values in the self-attention blocks and optionally if
            `config.is_encoder_decoder=True` in the cross-attention blocks) that can be used (see `past_key_values`
            input) to speed up sequential decoding.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
        cross_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` and `config.add_cross_attention=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
            weighted average in the cross-attention heads.
    """

    last_hidden_state: paddle.Tensor = None
    past_key_values: Optional[Tuple[Tuple[paddle.Tensor]]] = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None
    cross_attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class Seq2SeqModelOutput(ModelOutput):
    """
    Base class for model encoder's outputs that also contains : pre-computed hidden states that can speed up sequential
    decoding.

    Args:
        last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the decoder of the model.

            If `past_key_values` is used only the last hidden-state of the sequences of shape `(batch_size, 1,
            hidden_size)` is output.
        past_key_values (`tuple(tuple(paddle.Tensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `tuple(paddle.Tensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.

            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
        decoder_hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the decoder at the output of each layer plus the optional initial embedding outputs.
        decoder_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the
            self-attention heads.
        cross_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
            weighted average in the cross-attention heads.
        encoder_last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
            Sequence of hidden-states at the output of the last layer of the encoder of the model.
        encoder_hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the encoder at the output of each layer plus the optional initial embedding outputs.
        encoder_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the
            self-attention heads.
    """

    last_hidden_state: paddle.Tensor = None
    past_key_values: Optional[Tuple[Tuple[paddle.Tensor]]] = None
    decoder_hidden_states: Optional[Tuple[paddle.Tensor]] = None
    decoder_attentions: Optional[Tuple[paddle.Tensor]] = None
    cross_attentions: Optional[Tuple[paddle.Tensor]] = None
    encoder_last_hidden_state: Optional[paddle.Tensor] = None
    encoder_hidden_states: Optional[Tuple[paddle.Tensor]] = None
    encoder_attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class CausalLMOutput(ModelOutput):
    """
    Base class for causal language model (or autoregressive) outputs.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Language modeling loss (for next-token prediction).
        logits (`paddle.Tensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class CausalLMOutputWithPast(ModelOutput):
    """
    Base class for causal language model (or autoregressive) outputs.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Language modeling loss (for next-token prediction).
        logits (`paddle.Tensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        past_key_values (`tuple(tuple(paddle.Tensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `tuple(paddle.Tensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
            `(batch_size, num_heads, sequence_length, embed_size_per_head)`)

            Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
            `past_key_values` input) to speed up sequential decoding.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    past_key_values: Optional[Tuple[Tuple[paddle.Tensor]]] = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class CausalLMOutputWithCrossAttentions(ModelOutput):
    """
    Base class for causal language model (or autoregressive) outputs.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Language modeling loss (for next-token prediction).
        logits (`paddle.Tensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
        cross_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Cross attentions weights after the attention softmax, used to compute the weighted average in the
            cross-attention heads.
        past_key_values (`tuple(tuple(paddle.Tensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `paddle.Tensor` tuples of length `config.n_layers`, with each tuple containing the cached key,
            value states of the self-attention and the cross-attention layers if model is used in encoder-decoder
            setting. Only relevant if `config.is_decoder = True`.

            Contains pre-computed hidden-states (key and values in the attention blocks) that can be used (see
            `past_key_values` input) to speed up sequential decoding.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    past_key_values: Optional[Tuple[Tuple[paddle.Tensor]]] = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None
    cross_attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class SequenceClassifierOutputWithPast(ModelOutput):
    """
    Base class for outputs of sentence classification models.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Classification (or regression if config.num_labels==1) loss.
        logits (`paddle.Tensor` of shape `(batch_size, config.num_labels)`):
            Classification (or regression if config.num_labels==1) scores (before SoftMax).
        past_key_values (`tuple(tuple(paddle.Tensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `tuple(paddle.Tensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
            `(batch_size, num_heads, sequence_length, embed_size_per_head)`)

            Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
            `past_key_values` input) to speed up sequential decoding.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    past_key_values: Optional[Tuple[Tuple[paddle.Tensor]]] = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class MaskedLMOutput(ModelOutput):
    """
    Base class for masked language models outputs.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Masked language modeling (MLM) loss.
        logits (`paddle.Tensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class Seq2SeqLMOutput(ModelOutput):
    """
    Base class for sequence-to-sequence language models outputs.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Language modeling loss.
        logits (`paddle.Tensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        past_key_values (`tuple(tuple(paddle.Tensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `tuple(paddle.Tensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.

            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
        decoder_hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
        decoder_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the
            self-attention heads.
        cross_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
            weighted average in the cross-attention heads.
        encoder_last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
            Sequence of hidden-states at the output of the last layer of the encoder of the model.
        encoder_hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
        encoder_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the
            self-attention heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    past_key_values: Optional[Tuple[Tuple[paddle.Tensor]]] = None
    decoder_hidden_states: Optional[Tuple[paddle.Tensor]] = None
    decoder_attentions: Optional[Tuple[paddle.Tensor]] = None
    cross_attentions: Optional[Tuple[paddle.Tensor]] = None
    encoder_last_hidden_state: Optional[paddle.Tensor] = None
    encoder_hidden_states: Optional[Tuple[paddle.Tensor]] = None
    encoder_attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class NextSentencePredictorOutput(ModelOutput):
    """
    Base class for outputs of models predicting if two sentences are consecutive or not.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `next_sentence_label` is provided):
            Next sequence prediction (classification) loss.
        logits (`paddle.Tensor` of shape `(batch_size, 2)`):
            Prediction scores of the next sequence prediction (classification) head (scores of True/False continuation
            before SoftMax).
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class SequenceClassifierOutput(ModelOutput):
    """
    Base class for outputs of sentence classification models.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Classification (or regression if config.num_labels==1) loss.
        logits (`paddle.Tensor` of shape `(batch_size, config.num_labels)`):
            Classification (or regression if config.num_labels==1) scores (before SoftMax).
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class Seq2SeqSequenceClassifierOutput(ModelOutput):
    """
    Base class for outputs of sequence-to-sequence sentence classification models.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `label` is provided):
            Classification (or regression if config.num_labels==1) loss.
        logits (`paddle.Tensor` of shape `(batch_size, config.num_labels)`):
            Classification (or regression if config.num_labels==1) scores (before SoftMax).
        past_key_values (`tuple(tuple(paddle.Tensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `tuple(paddle.Tensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.

            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
        decoder_hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
        decoder_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the
            self-attention heads.
        cross_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
            weighted average in the cross-attention heads.
        encoder_last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
            Sequence of hidden-states at the output of the last layer of the encoder of the model.
        encoder_hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
        encoder_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the
            self-attention heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    past_key_values: Optional[Tuple[Tuple[paddle.Tensor]]] = None
    decoder_hidden_states: Optional[Tuple[paddle.Tensor]] = None
    decoder_attentions: Optional[Tuple[paddle.Tensor]] = None
    cross_attentions: Optional[Tuple[paddle.Tensor]] = None
    encoder_last_hidden_state: Optional[paddle.Tensor] = None
    encoder_hidden_states: Optional[Tuple[paddle.Tensor]] = None
    encoder_attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class MultipleChoiceModelOutput(ModelOutput):
    """
    Base class for outputs of multiple choice models.

    Args:
        loss (`paddle.Tensor` of shape *(1,)*, *optional*, returned when `labels` is provided):
            Classification loss.
        logits (`paddle.Tensor` of shape `(batch_size, num_choices)`):
            *num_choices* is the second dimension of the input tensors. (see *input_ids* above).

            Classification scores (before SoftMax).
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class TokenClassifierOutput(ModelOutput):
    """
    Base class for outputs of token classification models.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided) :
            Classification loss.
        logits (`paddle.Tensor` of shape `(batch_size, sequence_length, config.num_labels)`):
            Classification scores (before SoftMax).
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class QuestionAnsweringModelOutput(ModelOutput):
    """
    Base class for outputs of question answering models.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.
        start_logits (`paddle.Tensor` of shape `(batch_size, sequence_length)`):
            Span-start scores (before SoftMax).
        end_logits (`paddle.Tensor` of shape `(batch_size, sequence_length)`):
            Span-end scores (before SoftMax).
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    start_logits: paddle.Tensor = None
    end_logits: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class Seq2SeqQuestionAnsweringModelOutput(ModelOutput):
    """
    Base class for outputs of sequence-to-sequence question answering models.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.
        start_logits (`paddle.Tensor` of shape `(batch_size, sequence_length)`):
            Span-start scores (before SoftMax).
        end_logits (`paddle.Tensor` of shape `(batch_size, sequence_length)`):
            Span-end scores (before SoftMax).
        past_key_values (`tuple(tuple(paddle.Tensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `tuple(paddle.Tensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.

            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
        decoder_hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
        decoder_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the
            self-attention heads.
        cross_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
            weighted average in the cross-attention heads.
        encoder_last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
            Sequence of hidden-states at the output of the last layer of the encoder of the model.
        encoder_hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
        encoder_attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the
            self-attention heads.
    """

    loss: Optional[paddle.Tensor] = None
    start_logits: paddle.Tensor = None
    end_logits: paddle.Tensor = None
    past_key_values: Optional[Tuple[Tuple[paddle.Tensor]]] = None
    decoder_hidden_states: Optional[Tuple[paddle.Tensor]] = None
    decoder_attentions: Optional[Tuple[paddle.Tensor]] = None
    cross_attentions: Optional[Tuple[paddle.Tensor]] = None
    encoder_last_hidden_state: Optional[paddle.Tensor] = None
    encoder_hidden_states: Optional[Tuple[paddle.Tensor]] = None
    encoder_attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class SemanticSegmenterOutput(ModelOutput):
    """
    Base class for outputs of semantic segmentation models.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Classification (or regression if config.num_labels==1) loss.
        logits (`paddle.Tensor` of shape `(batch_size, config.num_labels, logits_height, logits_width)`):
            Classification scores for each pixel.

            <Tip warning={true}>

            The logits returned do not necessarily have the same size as the `pixel_values` passed as inputs. This is
            to avoid doing two interpolations and lose some quality when a user needs to resize the logits to the
            original image size as post-processing. You should always check your logits shape and resize as needed.

            </Tip>

        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, patch_size, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, patch_size,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class ImageClassifierOutput(ModelOutput):
    """
    Base class for outputs of image classification models.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Classification (or regression if config.num_labels==1) loss.
        logits (`paddle.Tensor` of shape `(batch_size, config.num_labels)`):
            Classification (or regression if config.num_labels==1) scores (before SoftMax).
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states
            (also called feature maps) of the model at the output of each stage.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, patch_size,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class ImageClassifierOutputWithNoAttention(ModelOutput):
    """
    Base class for outputs of image classification models.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Classification (or regression if config.num_labels==1) loss.
        logits (`paddle.Tensor` of shape `(batch_size, config.num_labels)`):
            Classification (or regression if config.num_labels==1) scores (before SoftMax).
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each stage) of shape `(batch_size, num_channels, height, width)`. Hidden-states (also
            called feature maps) of the model at the output of each stage.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class DepthEstimatorOutput(ModelOutput):
    """
    Base class for outputs of depth estimation models.

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Classification (or regression if config.num_labels==1) loss.
        predicted_depth (`paddle.Tensor` of shape `(batch_size, height, width)`):
            Predicted depth for each pixel.

        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, num_channels, height, width)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, patch_size,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    predicted_depth: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class Wav2Vec2BaseModelOutput(ModelOutput):
    """
    Base class for models that have been trained with the Wav2Vec2 loss objective.

    Args:
        last_hidden_state (`paddle.Tensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.
        extract_features (`paddle.Tensor` of shape `(batch_size, sequence_length, conv_dim[-1])`):
            Sequence of extracted feature vectors of the last convolutional layer of the model.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings + one for the output of each layer) of
            shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    last_hidden_state: paddle.Tensor = None
    extract_features: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None


@dataclass
class XVectorOutput(ModelOutput):
    """
    Output type of [`Wav2Vec2ForXVector`].

    Args:
        loss (`paddle.Tensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Classification loss.
        logits (`paddle.Tensor` of shape `(batch_size, config.xvector_output_dim)`):
            Classification hidden states before AMSoftmax.
        embeddings (`paddle.Tensor` of shape `(batch_size, config.xvector_output_dim)`):
            Utterance embeddings used for vector similarity-based retrieval.
        hidden_states (`tuple(paddle.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `paddle.Tensor` (one for the output of the embeddings + one for the output of each layer) of
            shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        attentions (`tuple(paddle.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `paddle.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[paddle.Tensor] = None
    logits: paddle.Tensor = None
    embeddings: paddle.Tensor = None
    hidden_states: Optional[Tuple[paddle.Tensor]] = None
    attentions: Optional[Tuple[paddle.Tensor]] = None