modeling_pegasus_x.py

# coding=utf-8
# Copyright 2022, Google and The HuggingFace Inc. 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.
""" PyTorch PEGASUS-X model."""

import dataclasses
import math
import random
from typing import Optional, Tuple, Union

import numpy as np
import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import CrossEntropyLoss

from ...activations import ACT2FN
from ...modeling_outputs import (
    BaseModelOutput,
    BaseModelOutputWithPastAndCrossAttentions,
    Seq2SeqLMOutput,
    Seq2SeqModelOutput,
)
from ...modeling_utils import PreTrainedModel
from ...utils import (
    add_end_docstrings,
    add_start_docstrings,
    add_start_docstrings_to_model_forward,
    logging,
    replace_return_docstrings,
)
from .configuration_pegasus_x import PegasusXConfig


logger = logging.get_logger(__name__)

_CHECKPOINT_FOR_DOC = "google/pegasus-x-base"
_CONFIG_FOR_DOC = "PegasusXConfig"


PEGASUS_X_PRETRAINED_MODEL_ARCHIVE_LIST = [
    "google/pegasus-x-base",
    "google/pegasus-x-large",
    # See all PEGASUS models at https://huggingface.co/models?filter=pegasus-x
]


@dataclasses.dataclass
class DimensionInfo:
    """Wrapper for dimension info."""

    batch_size: int  # batch size
    seq_len: int  # token length
    block_size: int  # block size
    num_heads: int  # num heads
    hidden_dim: int  # hidden dim
    dim_per_head: int  # dim per head
    num_blocks: int  # num blocks
    global_len: int  # global length
    padded_seq_len: int  # padded token seq length

    # Note: Compared to the original Flax implementation, we will pad the token representations to
    #       a multiple of block size at the start of the encoder layers, so T=P always.


# Copied from transformers.models.bart.modeling_bart.shift_tokens_right
def shift_tokens_right(input_ids: torch.Tensor, pad_token_id: int, decoder_start_token_id: int):
    """
    Shift input ids one token to the right.
    """
    shifted_input_ids = input_ids.new_zeros(input_ids.shape)
    shifted_input_ids[:, 1:] = input_ids[:, :-1].clone()
    shifted_input_ids[:, 0] = decoder_start_token_id

    if pad_token_id is None:
        raise ValueError("self.model.config.pad_token_id has to be defined.")
    # replace possible -100 values in labels by `pad_token_id`
    shifted_input_ids.masked_fill_(shifted_input_ids == -100, pad_token_id)

    return shifted_input_ids


# Copied from transformers.models.bart.modeling_bart._make_causal_mask
def _make_causal_mask(
    input_ids_shape: torch.Size, dtype: torch.dtype, device: torch.device, past_key_values_length: int = 0
):
    """
    Make causal mask used for bi-directional self-attention.
    """
    bsz, tgt_len = input_ids_shape
    mask = torch.full((tgt_len, tgt_len), torch.tensor(torch.finfo(dtype).min, device=device), device=device)
    mask_cond = torch.arange(mask.size(-1), device=device)
    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
    mask = mask.to(dtype)

    if past_key_values_length > 0:
        mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
    return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)


# Copied from transformers.models.bart.modeling_bart._expand_mask
def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
    """
    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
    """
    bsz, src_len = mask.size()
    tgt_len = tgt_len if tgt_len is not None else src_len

    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)

    inverted_mask = 1.0 - expanded_mask

    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)


class PegasusXSinusoidalPositionalEmbedding(nn.Module):
    """This module produces sinusoidal positional embeddings of any length."""

    def __init__(self, embed_dim, max_scale: int = 10000.0):
        super().__init__()
        self.embed_dim = embed_dim
        self.max_scale = max_scale

    @torch.no_grad()
    def forward(self, input_embeds: torch.Tensor, past_key_values_length: int = 0) -> torch.Tensor:
        """`input_ids_shape` is expected to be [bsz x seqlen]."""
        batch_size, seq_len = input_embeds.shape[:2]
        positions = torch.arange(
            past_key_values_length, past_key_values_length + seq_len, dtype=torch.long, device=input_embeds.device
        )[:, None]
        pe = torch.zeros((seq_len, self.embed_dim), device=input_embeds.device, dtype=input_embeds.dtype)
        half_d_feature = self.embed_dim // 2
        div_term = torch.exp(
            torch.arange(half_d_feature, device=input_embeds.device, dtype=input_embeds.dtype)
            * -(np.log(float(self.max_scale)) / (half_d_feature - 1))
        )
        pe[:, :half_d_feature] = torch.sin(positions * div_term)
        pe[:, half_d_feature:] = torch.cos(positions * div_term)
        return pe[None].expand(batch_size, -1, -1)


# Copied from transformers.models.bart.modeling_bart.BartAttention with Bart->PegasusX
class PegasusXAttention(nn.Module):
    """Multi-headed attention from 'Attention Is All You Need' paper"""

    def __init__(
        self,
        embed_dim: int,
        num_heads: int,
        dropout: float = 0.0,
        is_decoder: bool = False,
        bias: bool = True,
    ):
        super().__init__()
        self.embed_dim = embed_dim
        self.num_heads = num_heads
        self.dropout = dropout
        self.head_dim = embed_dim // num_heads

        if (self.head_dim * num_heads) != self.embed_dim:
            raise ValueError(
                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim}"
                f" and `num_heads`: {num_heads})."
            )
        self.scaling = self.head_dim**-0.5
        self.is_decoder = is_decoder

        self.k_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
        self.v_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
        self.q_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
        self.out_proj = nn.Linear(embed_dim, embed_dim, bias=bias)

    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()

    def forward(
        self,
        hidden_states: torch.Tensor,
        key_value_states: Optional[torch.Tensor] = None,
        past_key_value: Optional[Tuple[torch.Tensor]] = None,
        attention_mask: Optional[torch.Tensor] = None,
        layer_head_mask: Optional[torch.Tensor] = None,
        output_attentions: bool = False,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
        """Input shape: Batch x Time x Channel"""

        # if key_value_states are provided this layer is used as a cross-attention layer
        # for the decoder
        is_cross_attention = key_value_states is not None

        bsz, tgt_len, _ = hidden_states.size()

        # get query proj
        query_states = self.q_proj(hidden_states) * self.scaling
        # get key, value proj
        # `past_key_value[0].shape[2] == key_value_states.shape[1]`
        # is checking that the `sequence_length` of the `past_key_value` is the same as
        # the provided `key_value_states` to support prefix tuning
        if (
            is_cross_attention
            and past_key_value is not None
            and past_key_value[0].shape[2] == key_value_states.shape[1]
        ):
            # reuse k,v, cross_attentions
            key_states = past_key_value[0]
            value_states = past_key_value[1]
        elif is_cross_attention:
            # cross_attentions
            key_states = self._shape(self.k_proj(key_value_states), -1, bsz)
            value_states = self._shape(self.v_proj(key_value_states), -1, bsz)
        elif past_key_value is not None:
            # reuse k, v, self_attention
            key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
            value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
            key_states = torch.cat([past_key_value[0], key_states], dim=2)
            value_states = torch.cat([past_key_value[1], value_states], dim=2)
        else:
            # self_attention
            key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
            value_states = self._shape(self.v_proj(hidden_states), -1, bsz)

        if self.is_decoder:
            # if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states.
            # Further calls to cross_attention layer can then reuse all cross-attention
            # key/value_states (first "if" case)
            # if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of
            # all previous decoder key/value_states. Further calls to uni-directional self-attention
            # can concat previous decoder key/value_states to current projected key/value_states (third "elif" case)
            # if encoder bi-directional self-attention `past_key_value` is always `None`
            past_key_value = (key_states, value_states)

        proj_shape = (bsz * self.num_heads, -1, self.head_dim)
        query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
        key_states = key_states.reshape(*proj_shape)
        value_states = value_states.reshape(*proj_shape)

        src_len = key_states.size(1)
        attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

        if attn_weights.size() != (bsz * self.num_heads, tgt_len, src_len):
            raise ValueError(
                f"Attention weights should be of size {(bsz * self.num_heads, tgt_len, src_len)}, but is"
                f" {attn_weights.size()}"
            )

        if attention_mask is not None:
            if attention_mask.size() != (bsz, 1, tgt_len, src_len):
                raise ValueError(
                    f"Attention mask should be of size {(bsz, 1, tgt_len, src_len)}, but is {attention_mask.size()}"
                )
            attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len, src_len) + attention_mask
            attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)

        attn_weights = nn.functional.softmax(attn_weights, dim=-1)

        if layer_head_mask is not None:
            if layer_head_mask.size() != (self.num_heads,):
                raise ValueError(
                    f"Head mask for a single layer should be of size {(self.num_heads,)}, but is"
                    f" {layer_head_mask.size()}"
                )
            attn_weights = layer_head_mask.view(1, -1, 1, 1) * attn_weights.view(bsz, self.num_heads, tgt_len, src_len)
            attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)

        if output_attentions:
            # this operation is a bit awkward, but it's required to
            # make sure that attn_weights keeps its gradient.
            # In order to do so, attn_weights have to be reshaped
            # twice and have to be reused in the following
            attn_weights_reshaped = attn_weights.view(bsz, self.num_heads, tgt_len, src_len)
            attn_weights = attn_weights_reshaped.view(bsz * self.num_heads, tgt_len, src_len)
        else:
            attn_weights_reshaped = None

        attn_probs = nn.functional.dropout(attn_weights, p=self.dropout, training=self.training)

        attn_output = torch.bmm(attn_probs, value_states)

        if attn_output.size() != (bsz * self.num_heads, tgt_len, self.head_dim):
            raise ValueError(
                f"`attn_output` should be of size {(bsz * self.num_heads, tgt_len, self.head_dim)}, but is"
                f" {attn_output.size()}"
            )

        attn_output = attn_output.view(bsz, self.num_heads, tgt_len, self.head_dim)
        attn_output = attn_output.transpose(1, 2)

        # Use the `embed_dim` from the config (stored in the class) rather than `hidden_state` because `attn_output` can be
        # partitioned across GPUs when using tensor-parallelism.
        attn_output = attn_output.reshape(bsz, tgt_len, self.embed_dim)

        attn_output = self.out_proj(attn_output)

        return attn_output, attn_weights_reshaped, past_key_value


class PegasusXGlobalLocalAttention(nn.Module):
    """Global + Local attention. For use with Encoder only."""

    def __init__(
        self,
        embed_dim: int,
        num_heads: int,
        block_size: int,
        dropout: float = 0.0,
        is_decoder: bool = False,
    ):
        super().__init__()
        self.embed_dim = embed_dim
        self.num_heads = num_heads
        self.block_size = block_size
        self.dropout = dropout
        self.head_dim = embed_dim // num_heads

        if (self.head_dim * num_heads) != self.embed_dim:
            raise ValueError(
                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim}"
                f" and `num_heads`: {num_heads})."
            )
        self.scaling = self.head_dim**-0.5
        self.is_decoder = is_decoder

        self.k_proj = nn.Linear(embed_dim, embed_dim, bias=False)
        self.v_proj = nn.Linear(embed_dim, embed_dim, bias=False)
        self.q_proj = nn.Linear(embed_dim, embed_dim, bias=False)
        self.out_proj = nn.Linear(embed_dim, embed_dim, bias=False)

    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()

    def forward(
        self,
        token_hidden_states: torch.Tensor,
        global_hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        output_attentions: bool = False,
    ) -> Tuple[torch.Tensor, torch.Tensor, Optional[torch.Tensor]]:
        """Input shape: Batch x Time x Channel"""
        dim = DimensionInfo(
            batch_size=token_hidden_states.shape[0],
            seq_len=token_hidden_states.shape[1],
            block_size=self.block_size,
            num_heads=self.num_heads,
            hidden_dim=token_hidden_states.shape[2],
            dim_per_head=self.head_dim,
            num_blocks=token_hidden_states.shape[1] // self.block_size,
            global_len=global_hidden_states.shape[1],
            padded_seq_len=token_hidden_states.shape[1],
        )

        # [batch_size, num_heads, padded_seq_len, dim_per_head]
        local_q = self._shape(
            self.q_proj(token_hidden_states) * self.scaling,
            seq_len=dim.padded_seq_len,
            bsz=dim.batch_size,
        )
        local_k = self._shape(
            self.k_proj(token_hidden_states),
            seq_len=dim.padded_seq_len,
            bsz=dim.batch_size,
        )
        local_v = self._shape(
            self.v_proj(token_hidden_states),
            seq_len=dim.padded_seq_len,
            bsz=dim.batch_size,
        )

        # [batch_size, num_heads, global_len, dim_per_head]
        global_q = self._shape(
            self.q_proj(global_hidden_states) * self.scaling,
            seq_len=dim.global_len,
            bsz=dim.batch_size,
        )
        global_k = self._shape(
            self.k_proj(global_hidden_states),
            seq_len=dim.global_len,
            bsz=dim.batch_size,
        )
        global_v = self._shape(
            self.v_proj(global_hidden_states),
            seq_len=dim.global_len,
            bsz=dim.batch_size,
        )

        global_attn_output, global_attn_probs = self.compute_global_attention_representations(
            global_q=global_q,
            global_k=global_k,
            global_v=global_v,
            local_k=local_k,
            local_v=local_v,
            mask=attention_mask,
            dim=dim,
        )
        local_attn_output, local_attn_probs = self.compute_local_attention_representations(
            global_k=global_k,
            global_v=global_v,
            local_q=local_q,
            local_k=local_k,
            local_v=local_v,
            mask=attention_mask,
            dim=dim,
        )

        # [batch_size, global_len, hidden_dim]
        global_attn_output = (
            global_attn_output.transpose(1, 2).contiguous().view(dim.batch_size, dim.global_len, dim.hidden_dim)
        )
        # [batch_size, global_len, hidden_dim]
        global_attn_output = self.out_proj(global_attn_output)
        # [batch_size, num_heads, block_size, num_heads, dim_per_head]
        local_attn_output = local_attn_output.permute(0, 2, 3, 1, 4).contiguous()
        # [batch_size, padded_seq_len, hidden_dim]
        local_attn_output = local_attn_output.view(dim.batch_size, dim.padded_seq_len, dim.hidden_dim)
        # [batch_size, padded_seq_len, hidden_dim]
        local_attn_output = self.out_proj(local_attn_output)

        if output_attentions:
            attn_probs = {"global": global_attn_probs, "local": local_attn_probs}
        else:
            attn_probs = None

        return local_attn_output, global_attn_output, attn_probs

    def compute_global_attention_representations(
        self, global_q, global_k, global_v, local_k, local_v, mask, dim: DimensionInfo
    ):
        """Compute attention representations for global tokens.

        Global tokens will attend to both global tokens as well as all input sequence tokens. Because the input
        sequence tokens are arranged in blocks for local attention, we unblock them and compute attention.

        Args:
            global_q (`torch.FloatTensor`) of shape [batch_size, num_heads, global_len, dim_per_head]:
                query vectors from global tokens
            global_k (`torch.FloatTensor`) of shape [batch_size, num_heads, global_len, dim_per_head]:
                key vectors from global tokens
            global_v (`torch.FloatTensor`) of shape [batch_size, num_heads, global_len, dim_per_head]:
                value vectors from global tokens
            local_k (`torch.FloatTensor`) of shape [batch_size, num_heads, padded_seq_len, dim_per_head]:
                key vectors from local tokens
            local_v (`torch.FloatTensor`) of shape [batch_size, num_heads, padded_seq_len, dim_per_head]:
                value vectors from local tokens
            mask (`torch.FloatTensor`) of shape [batch_size, padded_seq_len]: attention mask
            dim (DimensionInfo): DimensionInfo wrapper for dimensions

        Returns:
            output of shape `[batch_sizes, length, features]`. where length will be padded to a multiple of block_size
        """
        # [batch_size, num_heads, global_len+padded_seq_len, dim_per_head]
        global_and_local_k = torch.cat([global_k, local_k], dim=2)
        # [batch_size, num_heads, global_len+padded_seq_len, dim_per_head]
        global_and_local_v = torch.cat([global_v, local_v], dim=2)

        # [batch_size, global_len+padded_seq_len]
        extended_mask = nn.functional.pad(mask, pad=(dim.global_len, 0), value=0)

        # [batch_size, num_heads, global_len, global_len+padded_seq_len]
        attn_weights = torch.einsum("BHGF,BHXF->BHGX", global_q, global_and_local_k)
        attn_weights = attn_weights + extended_mask[:, None, None, :]
        attn_probs = nn.functional.softmax(attn_weights, dim=-1)
        attn_probs = nn.functional.dropout(attn_probs, p=self.dropout, training=self.training)

        # [batch_size, num_heads, global_len, F]
        attn_output = torch.einsum("BHGX,BHXF->BHGF", attn_probs, global_and_local_v)
        return attn_output, attn_probs

    def compute_local_attention_representations(
        self, global_k, global_v, local_q, local_k, local_v, mask, dim: DimensionInfo
    ):
        """Compute attention representations for local tokens.

        Local tokens will attend to both global tokens as well as all other tokens within the same local block. Hence,
        we need to tile and concatenate the global tokens to every local block

        Args:
            global_k (`torch.FloatTensor`) of shape [batch_size, num_heads, global_len, dim_per_head]:
                key vectors from global tokens
            global_v (`torch.FloatTensor`) of shape [batch_size, num_heads, global_len, dim_per_head]:
                value vectors from global tokens
            local_q (`torch.FloatTensor`) of shape [batch_size, num_heads, padded_seq_len, dim_per_head]:
                query vectors from local tokens
            local_k (`torch.FloatTensor`) of shape [batch_size, num_heads, padded_seq_len, dim_per_head]:
                key vectors from local tokens
            local_v (`torch.FloatTensor`) of shape [batch_size, num_heads, padded_seq_len, dim_per_head]:
                value vectors from local tokens
            mask (`torch.FloatTensor`) of shape [batch_size, padded_seq_len]: attention mask
            dim (DimensionInfo): DimensionInfo wrapper for dimensions

        Returns:
            output of shape `[batch_sizes, length, features]`. where length will be padded to a multiple of block_size
        """
        # [batch_size, num_heads, num_blocks, block_size, dim_per_head]
        blocked_local_q = local_q.view(dim.batch_size, dim.num_heads, dim.num_blocks, dim.block_size, dim.dim_per_head)
        # [batch_size, num_heads, num_blocks, block_size, dim_per_head]
        blocked_local_k = local_k.view(dim.batch_size, dim.num_heads, dim.num_blocks, dim.block_size, dim.dim_per_head)
        # [batch_size, num_heads, num_blocks, block_size, dim_per_head]
        blocked_local_v = local_v.view(dim.batch_size, dim.num_heads, dim.num_blocks, dim.block_size, dim.dim_per_head)

        # [batch_size, num_blocks, global_len+block_size]
        extended_mask = nn.functional.pad(
            mask.view(dim.batch_size, dim.num_blocks, dim.block_size),
            pad=(dim.global_len, 0),
            value=0,
        )

        # [batch_size, num_heads, num_blocks, block_size, global_len]
        blocked_local2global = torch.einsum("BHNKF,BHGF->BHNKG", blocked_local_q, global_k)
        # [batch_size, num_heads, num_blocks, block_size, block_size]
        blocked_local2local = torch.einsum("BHNKF,BHNXF->BHNKX", blocked_local_q, blocked_local_k)

        # [batch_size, num_heads, num_blocks, block_size, global_len+block_size]
        attn_weights = torch.cat([blocked_local2global, blocked_local2local], dim=-1)
        attn_weights = attn_weights + extended_mask[:, None, :, None, :]
        attn_probs = nn.functional.softmax(attn_weights, dim=-1)
        attn_probs = nn.functional.dropout(attn_probs, p=self.dropout, training=self.training)

        # [batch_size, num_heads, num_blocks, block_size, global_len]
        local2global_attn_probs = attn_probs[:, :, :, :, : dim.global_len]
        # [batch_size, num_heads, num_blocks, block_size, block_size]
        local2local_attn_probs = attn_probs[:, :, :, :, dim.global_len :]

        # [batch_size, num_heads, num_blocks, block_size, dim_per_head]
        local2global_attn_output = torch.einsum("BHNKG,BHGF->BHNKF", local2global_attn_probs, global_v)
        # [batch_size, num_heads, num_blocks, block_size, dim_per_head]
        local2local_attn_output = torch.einsum("BHNKX,BHNXF->BHNKF", local2local_attn_probs, blocked_local_v)
        # [batch_size, num_heads, num_blocks, block_size, dim_per_head]
        attn_output = local2global_attn_output + local2local_attn_output
        return attn_output, attn_probs


class PegasusXEncoderLayer(nn.Module):
    def __init__(self, stagger_blocks_this_layer: bool, config: PegasusXConfig):
        super().__init__()
        self.embed_dim = config.d_model
        self.self_attn = PegasusXGlobalLocalAttention(
            embed_dim=self.embed_dim,
            num_heads=config.encoder_attention_heads,
            block_size=config.block_size,
            dropout=config.attention_dropout,
        )
        self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
        self.global_self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
        self.dropout = config.dropout
        self.activation_fn = ACT2FN[config.activation_function]
        self.activation_dropout = config.activation_dropout
        self.fc1 = nn.Linear(self.embed_dim, config.encoder_ffn_dim)
        self.fc2 = nn.Linear(config.encoder_ffn_dim, self.embed_dim)
        self.final_layer_norm = nn.LayerNorm(self.embed_dim)
        self.stagger_blocks_this_layer = stagger_blocks_this_layer
        self.block_size = config.block_size

    def forward(
        self,
        hidden_states: torch.Tensor,
        global_hidden_states: torch.Tensor,
        attention_mask: torch.Tensor,
        output_attentions: bool = False,
    ) -> torch.Tensor:
        """
        Args:
            hidden_states (`torch.FloatTensor`): input to the layer of shape *(seq_len, batch, embed_dim)*
            global_hidden_states (`torch.FloatTensor`): global token hidden states
                *(seq_len, num_global_tokens, embed_dim)*
            attention_mask (`torch.FloatTensor`): attention mask of size
                *(batch, 1, tgt_len, src_len)* where padding elements are indicated by very large negative 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.
        """
        residual = hidden_states
        global_residual = global_hidden_states

        hidden_states = self.self_attn_layer_norm(hidden_states)
        global_hidden_states = self.global_self_attn_layer_norm(global_hidden_states)

        if self.stagger_blocks_this_layer:
            # Pad the blocks to simulate staggering
            hidden_states, attention_mask = self.pad_local_tokens(
                hidden_states=hidden_states, attention_mask=attention_mask, block_size=self.block_size
            )

        hidden_states, global_hidden_states, attn_weights = self.self_attn(
            token_hidden_states=hidden_states,
            global_hidden_states=global_hidden_states,
            attention_mask=attention_mask,
            output_attentions=output_attentions,
        )

        if self.stagger_blocks_this_layer:
            # Undo the padding
            hidden_states = self.unpad_local_tokens(padded_hidden_states=hidden_states, block_size=self.block_size)

        hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
        hidden_states = residual + hidden_states

        global_hidden_states = nn.functional.dropout(global_hidden_states, p=self.dropout, training=self.training)
        global_hidden_states = global_residual + global_hidden_states

        residual = hidden_states
        hidden_states = self.final_layer_norm(hidden_states)
        hidden_states = self.activation_fn(self.fc1(hidden_states))
        hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
        hidden_states = self.fc2(hidden_states)
        hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
        hidden_states = residual + hidden_states

        global_residual = global_hidden_states
        global_hidden_states = self.final_layer_norm(global_hidden_states)
        global_hidden_states = self.activation_fn(self.fc1(global_hidden_states))
        global_hidden_states = nn.functional.dropout(
            global_hidden_states, p=self.activation_dropout, training=self.training
        )
        global_hidden_states = self.fc2(global_hidden_states)
        global_hidden_states = nn.functional.dropout(global_hidden_states, p=self.dropout, training=self.training)
        global_hidden_states = global_residual + global_hidden_states
        outputs = (hidden_states, global_hidden_states)

        if output_attentions:
            outputs += (attn_weights,)

        return outputs

    @classmethod
    def pad_local_tokens(cls, hidden_states, attention_mask, block_size):
        # hidden_states: [batch_size, seq_len, hidden_dim]
        pad_size = block_size // 2
        mask_min_value = torch.finfo(hidden_states.dtype).min
        padded_hidden_states = torch.nn.functional.pad(
            hidden_states,
            pad=(0, 0, pad_size, pad_size),
        )
        padded_mask = torch.nn.functional.pad(
            attention_mask,
            pad=(pad_size, pad_size),
            value=mask_min_value,
        )
        return padded_hidden_states, padded_mask

    @classmethod
    def unpad_local_tokens(cls, padded_hidden_states, block_size):
        # padded_hidden_states: [batch_size, padded seq_len, hidden_dim]
        pad_size = block_size // 2
        return padded_hidden_states[:, pad_size:-pad_size, :]


class PegasusXDecoderLayer(nn.Module):
    def __init__(self, config: PegasusXConfig):
        super().__init__()
        self.embed_dim = config.d_model

        self.self_attn = PegasusXAttention(
            embed_dim=self.embed_dim,
            num_heads=config.decoder_attention_heads,
            dropout=config.attention_dropout,
            is_decoder=True,
            bias=False,
        )
        self.dropout = config.dropout
        self.activation_fn = ACT2FN[config.activation_function]
        self.activation_dropout = config.activation_dropout

        self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
        self.encoder_attn = PegasusXAttention(
            self.embed_dim,
            config.decoder_attention_heads,
            dropout=config.attention_dropout,
            is_decoder=True,
            bias=False,
        )
        self.encoder_attn_layer_norm = nn.LayerNorm(self.embed_dim)
        self.fc1 = nn.Linear(self.embed_dim, config.decoder_ffn_dim)
        self.fc2 = nn.Linear(config.decoder_ffn_dim, self.embed_dim)
        self.final_layer_norm = nn.LayerNorm(self.embed_dim)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        encoder_hidden_states: Optional[torch.Tensor] = None,
        encoder_attention_mask: Optional[torch.Tensor] = None,
        past_key_value: Optional[Tuple[torch.Tensor]] = None,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = True,
    ) -> torch.Tensor:
        """
        Args:
            hidden_states (`torch.FloatTensor`): input to the layer of shape *(seq_len, batch, embed_dim)*
            attention_mask (`torch.FloatTensor`): attention mask of size
                *(batch, 1, tgt_len, src_len)* where padding elements are indicated by very large negative values.
            encoder_hidden_states (`torch.FloatTensor`):
                cross attention input to the layer of shape *(seq_len, batch, embed_dim)*
            encoder_attention_mask (`torch.FloatTensor`): encoder attention mask of size
                *(batch, 1, tgt_len, src_len)* where padding elements are indicated by very large negative values.
            past_key_value (`Tuple(torch.FloatTensor)`): cached past key and value projection states
            output_attentions (`bool`, *optional*):
                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                returned tensors for more detail.
            use_cache: Whether to us KV cache for decoding
        """
        residual = hidden_states
        hidden_states = self.self_attn_layer_norm(hidden_states)

        # Self Attention
        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
        self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
        # add present self-attn cache to positions 1,2 of present_key_value tuple
        hidden_states, self_attn_weights, present_key_value = self.self_attn(
            hidden_states=hidden_states,
            past_key_value=self_attn_past_key_value,
            attention_mask=attention_mask,
            output_attentions=output_attentions,
        )
        hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
        hidden_states = residual + hidden_states

        # Cross-Attention Block
        cross_attn_present_key_value = None
        cross_attn_weights = None
        if encoder_hidden_states is not None:
            residual = hidden_states
            hidden_states = self.encoder_attn_layer_norm(hidden_states)

            # cross_attn cached key/values tuple is at positions 3,4 of present_key_value tuple
            cross_attn_past_key_value = past_key_value[-2:] if past_key_value is not None else None
            hidden_states, cross_attn_weights, cross_attn_present_key_value = self.encoder_attn(
                hidden_states=hidden_states,
                key_value_states=encoder_hidden_states,
                attention_mask=encoder_attention_mask,
                past_key_value=cross_attn_past_key_value,
                output_attentions=output_attentions,
            )
            hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
            hidden_states = residual + hidden_states

            # add cross-attn to positions 3,4 of present_key_value tuple
            present_key_value = present_key_value + cross_attn_present_key_value

        # Fully Connected
        residual = hidden_states
        hidden_states = self.final_layer_norm(hidden_states)
        hidden_states = self.activation_fn(self.fc1(hidden_states))
        hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
        hidden_states = self.fc2(hidden_states)
        hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (self_attn_weights, cross_attn_weights)

        if use_cache:
            outputs += (present_key_value,)

        return outputs


class PegasusXPreTrainedModel(PreTrainedModel):
    config_class = PegasusXConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True

    def _init_weights(self, module):
        std = self.config.init_std
        if isinstance(module, nn.Linear):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=std)

    def _set_gradient_checkpointing(self, module, value=False):
        if isinstance(module, (PegasusXDecoder, PegasusXEncoder)):
            module.gradient_checkpointing = value


PEGASUS_X_START_DOCSTRING = r"""
    This model inherits from [`PreTrainedModel`]. 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 PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
    and behavior.

    Parameters:
        config ([`PegasusXConfig`]):
            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
            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
"""

PEGASUS_X_GENERATION_EXAMPLE = r"""
    Summarization example:

    ```python
    >>> from transformers import AutoTokenizer, PegasusXForConditionalGeneration

    >>> model = PegasusXForConditionalGeneration.from_pretrained("google/pegasus-x-base")
    >>> tokenizer = AutoTokenizer.from_pretrained("google/pegasus-x-large")

    >>> ARTICLE_TO_SUMMARIZE = (
    ...     "PG&E stated it scheduled the blackouts in response to forecasts for high winds "
    ...     "amid dry conditions. The aim is to reduce the risk of wildfires. Nearly 800 thousand customers were "
    ...     "scheduled to be affected by the shutoffs which were expected to last through at least midday tomorrow."
    ... )
    >>> inputs = tokenizer(ARTICLE_TO_SUMMARIZE, max_length=1024, return_tensors="pt")

    >>> # Generate Summary
    >>> summary_ids = model.generate(inputs["input_ids"])
    >>> tokenizer.batch_decode(summary_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
    "California's largest electricity provider has turned off power to hundreds of thousands of customers."
    ```
"""

PEGASUS_X_INPUTS_DOCSTRING = r"""
    Args:
        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
            it.

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

            [What are input IDs?](../glossary#input-ids)
        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
        attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:

            - 1 for tokens that are **not masked**,
            - 0 for tokens that are **masked**.

            [What are attention masks?](../glossary#attention-mask)
        decoder_input_ids (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
            Indices of decoder input sequence tokens in the vocabulary.

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

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

            PEGASUS-X uses the `pad_token_id` as the starting token for `decoder_input_ids` generation. If
            `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
            `past_key_values`).
        decoder_attention_mask (`torch.LongTensor` 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(torch.FloatTensor)`, *optional*):
            Tuple consists of (`last_hidden_state`, *optional*: `hidden_states`, *optional*: `attentions`)
            `last_hidden_state` of shape `(batch_size, sequence_length, hidden_size)`, *optional*) is a sequence 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(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
            Tuple of `tuple(torch.FloatTensor)` 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.

            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 `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of shape
            `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing `input_ids` you
            can choose to directly pass an embedded representation. This is useful if you want more control over how to
            convert `input_ids` indices into associated vectors than the model's internal embedding lookup matrix.
        decoder_inputs_embeds (`torch.FloatTensor` 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. If `past_key_values` is used, optionally only the last `decoder_inputs_embeds` have to be
            input (see `past_key_values`). 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.

            If `decoder_input_ids` and `decoder_inputs_embeds` are both unset, `decoder_inputs_embeds` takes the value
            of `inputs_embeds`.
        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*):
            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
"""


class PegasusXEncoder(PegasusXPreTrainedModel):
    """
    Transformer encoder consisting of *config.encoder_layers* self attention layers. Each layer is a
    [`PegasusXEncoderLayer`].

    Args:
        config: PegasusXConfig
        embed_tokens (nn.Embedding): output embedding
    """

    def __init__(self, config: PegasusXConfig, embed_tokens: Optional[nn.Embedding] = None):
        super().__init__(config)

        self.dropout = config.dropout
        self.layerdrop = config.encoder_layerdrop

        embed_dim = config.d_model
        self.max_source_positions = config.max_position_embeddings
        self.embed_scale = math.sqrt(embed_dim) if config.scale_embedding else 1.0

        if embed_tokens is not None:
            self.embed_tokens = embed_tokens
        else:
            self.embed_tokens = nn.Embedding(config.vocab_size, embed_dim)

        self.embed_global = nn.Embedding(config.num_global_tokens, embed_dim)
        self.embed_positions = PegasusXSinusoidalPositionalEmbedding(embed_dim)
        self.layers = nn.ModuleList(
            [
                PegasusXEncoderLayer(
                    stagger_blocks_this_layer=i % 2 == 1 and config.stagger_local_blocks, config=config
                )
                for i in range(config.encoder_layers)
            ]
        )
        self.layer_norm = nn.LayerNorm(config.d_model)

        self.gradient_checkpointing = False
        # Initialize weights and apply final processing
        self.post_init()

    def resize_position_embeddings(self, new_num_position_embeddings: int):
        """
        Resizes position embeddings matrix of the model if `new_num_position_embeddings !=
        config.max_position_embeddings`.

        Arguments:
            new_num_position_embeddings (`int`):
                The number of new position embeddings. If position embeddings are learned, increasing the size will add
                newly initialized vectors at the end, whereas reducing the size will remove vectors from the end. If
                position embeddings are not learned (*e.g.* sinusoidal position embeddings), increasing the size will
                add correct vectors at the end following the position encoding algorithm, whereas reducing the size
                will remove vectors from the end.
        """
        logger.info(f"Setting `config.max_position_embeddings={new_num_position_embeddings}`...")
        self.config.max_position_embeddings = new_num_position_embeddings

        self.embed_positions = PegasusXSinusoidalPositionalEmbedding(self.config.d_model)
        self.embed_positions.to(self.device)

    def get_position_embeddings(self) -> nn.Embedding:
        """
        Returns the position embeddings matrix
        """
        return self.embed_positions

    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        inputs_embeds=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):
        r"""
        Args:
            input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
                Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you
                provide it.

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

                [What are input IDs?](../glossary#input-ids)
            attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
                Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:

                - 1 for tokens that are **not masked**,
                - 0 for tokens that are **masked**.

                [What are attention masks?](../glossary#attention-mask)

            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
                Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
                This is useful if you want more control over how to convert `input_ids` indices into associated vectors
                than the model's internal embedding lookup matrix.
            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*):
                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
        """
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # retrieve input_ids and inputs_embeds
        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
        elif input_ids is not None:
            input_shape = input_ids.size()
            input_ids = input_ids.view(-1, input_shape[-1])
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

        if inputs_embeds is None:
            inputs_embeds = self.embed_tokens(input_ids) * self.embed_scale

        embed_pos = self.embed_positions(inputs_embeds)

        hidden_states = inputs_embeds + embed_pos
        hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)

        batch_size, seq_len, _ = hidden_states.shape

        # Setup mask
        if attention_mask is None:
            attention_mask = torch.ones(*input_shape, dtype=inputs_embeds.dtype, device=inputs_embeds.device)
        attention_mask = attention_mask.to(dtype=hidden_states.dtype)
        mask_min_value = torch.finfo(hidden_states.dtype).min
        inverted_mask = 1.0 - attention_mask
        attention_mask = inverted_mask.masked_fill(
            inverted_mask.to(torch.bool),
            mask_min_value,
        )

        # padding to block_size
        if seq_len % self.config.block_size != 0:
            pad_len = self.config.block_size - seq_len % self.config.block_size
            hidden_states = nn.functional.pad(hidden_states, pad=(0, 0, 0, pad_len), value=0)
            attention_mask = nn.functional.pad(attention_mask, pad=(0, pad_len), value=mask_min_value)

        # Global tokens
        global_hidden_states = self.embed_global(
            torch.arange(self.config.num_global_tokens, device=hidden_states.device)[None].expand(batch_size, -1)
        )

        encoder_states = () if output_hidden_states else None
        all_attentions = () if output_attentions else None

        for idx, encoder_layer in enumerate(self.layers):
            if output_hidden_states:
                encoder_states = encoder_states + (hidden_states,)
            # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
            dropout_probability = random.uniform(0, 1)
            if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                layer_outputs = (None, None)
            else:
                if self.gradient_checkpointing and self.training:

                    def create_custom_forward(module):
                        def custom_forward(*inputs):
                            return module(*inputs, output_attentions)

                        return custom_forward

                    layer_outputs = torch.utils.checkpoint.checkpoint(
                        create_custom_forward(encoder_layer),
                        hidden_states,
                        global_hidden_states,
                        attention_mask,
                    )
                else:
                    layer_outputs = encoder_layer(
                        hidden_states,
                        global_hidden_states,
                        attention_mask,
                        output_attentions=output_attentions,
                    )

                hidden_states = layer_outputs[0]
                global_hidden_states = layer_outputs[1]

            if output_attentions:
                all_attentions = all_attentions + (layer_outputs[2],)

        # Undo padding-to-block-size
        hidden_states = hidden_states[:, :seq_len]

        hidden_states = self.layer_norm(hidden_states)

        if output_hidden_states:
            encoder_states = encoder_states + ((hidden_states, global_hidden_states),)

        if not return_dict:
            return tuple(v for v in [hidden_states, encoder_states, all_attentions] if v is not None)
        return BaseModelOutput(
            last_hidden_state=hidden_states, hidden_states=encoder_states, attentions=all_attentions
        )


class PegasusXDecoder(PegasusXPreTrainedModel):
    """
    Transformer decoder consisting of *config.decoder_layers* layers. Each layer is a [`PegasusDecoderLayer`]

    Args:
        config: PegasusXConfig
        embed_tokens (nn.Embedding): output embedding
    """

    def __init__(self, config: PegasusXConfig, embed_tokens: Optional[nn.Embedding] = None):
        super().__init__(config)
        self.dropout = config.dropout
        self.layerdrop = config.decoder_layerdrop
        self.max_target_positions = config.max_position_embeddings
        self.embed_scale = math.sqrt(config.d_model) if config.scale_embedding else 1.0

        if embed_tokens is not None:
            self.embed_tokens = embed_tokens
        else:
            self.embed_tokens = nn.Embedding(config.vocab_size, config.d_model)

        self.embed_positions = PegasusXSinusoidalPositionalEmbedding(config.d_model)
        self.layers = nn.ModuleList([PegasusXDecoderLayer(config) for _ in range(config.decoder_layers)])
        self.layer_norm = nn.LayerNorm(config.d_model)

        self.gradient_checkpointing = False
        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.embed_tokens

    def set_input_embeddings(self, value):
        self.embed_tokens = value

    # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
    def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
        # create causal mask
        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
        combined_attention_mask = None
        if input_shape[-1] > 1:
            combined_attention_mask = _make_causal_mask(
                input_shape,
                inputs_embeds.dtype,
                device=inputs_embeds.device,
                past_key_values_length=past_key_values_length,
            )

        if attention_mask is not None:
            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
            expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
                inputs_embeds.device
            )
            combined_attention_mask = (
                expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
            )

        return combined_attention_mask

    def resize_position_embeddings(self, new_num_position_embeddings: int):
        """
        Resizes position embeddings matrix of the model if `new_num_position_embeddings !=
        config.max_position_embeddings`.

        Arguments:
            new_num_position_embeddings (`int`):
                The number of new position embeddings. If position embeddings are learned, increasing the size will add
                newly initialized vectors at the end, whereas reducing the size will remove vectors from the end. If
                position embeddings are not learned (*e.g.* sinusoidal position embeddings), increasing the size will
                add correct vectors at the end following the position encoding algorithm, whereas reducing the size
                will remove vectors from the end.
        """
        logger.info(f"Setting `config.max_position_embeddings={new_num_position_embeddings}`...")
        self.config.max_position_embeddings = new_num_position_embeddings

        self.embed_positions = PegasusXSinusoidalPositionalEmbedding(self.config.d_model)
        self.embed_positions.to(self.device)

    def get_position_embeddings(self) -> nn.Embedding:
        """
        Returns the position embeddings matrix
        """
        return self.embed_positions

    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_values=None,
        inputs_embeds=None,
        use_cache=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):
        r"""
        Args:
            input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
                Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you
                provide it.

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

                [What are input IDs?](../glossary#input-ids)
            attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
                Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:

                - 1 for tokens that are **not masked**,
                - 0 for tokens that are **masked**.

                [What are attention masks?](../glossary#attention-mask)
            encoder_hidden_states (`torch.FloatTensor` of shape `(batch_size, encoder_sequence_length, hidden_size)`, *optional*):
                Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention
                of the decoder.
            encoder_attention_mask (`torch.LongTensor` of shape `(batch_size, encoder_sequence_length)`, *optional*):
                Mask to avoid performing cross-attention on padding tokens indices of encoder input_ids. Mask values
                selected in `[0, 1]`:

                - 1 for tokens that are **not masked**,
                - 0 for tokens that are **masked**.

                [What are attention masks?](../glossary#attention-mask)

            past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
                Tuple of `tuple(torch.FloatTensor)` 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.

                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 `(batch_size, sequence_length)`.
            inputs_embeds (`torch.FloatTensor` of
                shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing
                `input_ids` you can choose to directly pass an embedded representation. This is useful if you want more
                control over how to convert `input_ids` indices into associated vectors than the model's internal
                embedding lookup matrix.
            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*):
                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
        """
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # retrieve input_ids and inputs_embeds
        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
        elif input_ids is not None:
            input_shape = input_ids.size()
            input_ids = input_ids.view(-1, input_shape[-1])
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
        else:
            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")

        # past_key_values_length
        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0

        if inputs_embeds is None:
            inputs_embeds = self.embed_tokens(input_ids) * self.embed_scale

        attention_mask = self._prepare_decoder_attention_mask(
            attention_mask, input_shape, inputs_embeds, past_key_values_length
        )

        # expand encoder attention mask
        if encoder_hidden_states is not None and encoder_attention_mask is not None:
            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
            encoder_attention_mask = _expand_mask(encoder_attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1])

        # embed positions
        positions = self.embed_positions(inputs_embeds, past_key_values_length)

        hidden_states = inputs_embeds + positions

        hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)

        if self.gradient_checkpointing and self.training:
            if use_cache:
                logger.warning_once(
                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                )
                use_cache = False

        # decoder layers
        all_hidden_states = () if output_hidden_states else None
        all_self_attns = () if output_attentions else None
        all_cross_attentions = () if (output_attentions and encoder_hidden_states is not None) else None
        next_decoder_cache = () if use_cache else None

        for idx, decoder_layer in enumerate(self.layers):
            # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
            if output_hidden_states:
                all_hidden_states += (hidden_states,)
            dropout_probability = random.uniform(0, 1)
            if self.training and (dropout_probability < self.layerdrop):
                continue

            past_key_value = past_key_values[idx] if past_key_values is not None else None

            if self.gradient_checkpointing and self.training:

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        # None for past_key_value
                        return module(*inputs, output_attentions, use_cache)

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(decoder_layer),
                    hidden_states,
                    attention_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                    None,
                )
            else:
                layer_outputs = decoder_layer(
                    hidden_states,
                    attention_mask=attention_mask,
                    encoder_hidden_states=encoder_hidden_states,
                    encoder_attention_mask=encoder_attention_mask,
                    past_key_value=past_key_value,
                    output_attentions=output_attentions,
                    use_cache=use_cache,
                )
            hidden_states = layer_outputs[0]

            if use_cache:
                next_decoder_cache += (layer_outputs[3 if output_attentions else 1],)

            if output_attentions:
                all_self_attns += (layer_outputs[1],)

                if encoder_hidden_states is not None:
                    all_cross_attentions += (layer_outputs[2],)

        hidden_states = self.layer_norm(hidden_states)

        # add hidden states from the last decoder layer
        if output_hidden_states:
            all_hidden_states += (hidden_states,)

        next_cache = next_decoder_cache if use_cache else None
        if not return_dict:
            return tuple(
                v
                for v in [hidden_states, next_cache, all_hidden_states, all_self_attns, all_cross_attentions]
                if v is not None
            )
        return BaseModelOutputWithPastAndCrossAttentions(
            last_hidden_state=hidden_states,
            past_key_values=next_cache,
            hidden_states=all_hidden_states,
            attentions=all_self_attns,
            cross_attentions=all_cross_attentions,
        )


@add_start_docstrings(
    "The bare PEGASUS-X Model outputting raw hidden-states without any specific head on top.",
    PEGASUS_X_START_DOCSTRING,
)
class PegasusXModel(PegasusXPreTrainedModel):
    _keys_to_ignore_on_load_missing = ["decoder.embed_tokens.weight", "encoder.embed_tokens.weight"]

    def __init__(self, config: PegasusXConfig):
        super().__init__(config)

        vocab_size = config.vocab_size
        self.shared = nn.Embedding(vocab_size, config.d_model)

        self.encoder = PegasusXEncoder(config, self.shared)
        self.decoder = PegasusXDecoder(config, self.shared)

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.shared

    def set_input_embeddings(self, value):
        self.shared = value
        self.encoder.embed_tokens = self.shared
        self.decoder.embed_tokens = self.shared

    def get_encoder(self):
        return self.encoder

    def get_decoder(self):
        return self.decoder

    def resize_position_embeddings(self, new_num_position_embeddings: int):
        """
        Resizes position embeddings matrix of the model if `new_num_position_embeddings !=
        config.max_position_embeddings`.

        Arguments:
            new_num_position_embeddings (`int`):
                The number of new position embeddings. If position embeddings are learned, increasing the size will add
                newly initialized vectors at the end, whereas reducing the size will remove vectors from the end. If
                position embeddings are not learned (*e.g.* sinusoidal position embeddings), increasing the size will
                add correct vectors at the end following the position encoding algorithm, whereas reducing the size
                will remove vectors from the end.
        """
        self.config.max_position_embeddings = new_num_position_embeddings
        self.encoder.resize_position_embeddings(new_num_position_embeddings)
        self.decoder.resize_position_embeddings(new_num_position_embeddings)

    def get_position_embeddings(self) -> Tuple[nn.Embedding]:
        """
        Returns the position embeddings matrix
        """
        return (self.encoder.get_position_embeddings(), self.decoder.get_position_embeddings())

    @add_start_docstrings_to_model_forward(PEGASUS_X_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=Seq2SeqModelOutput, config_class=_CONFIG_FOR_DOC)
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        decoder_input_ids: Optional[torch.Tensor] = None,
        decoder_attention_mask: Optional[torch.Tensor] = None,
        encoder_outputs: Optional[Tuple[torch.FloatTensor]] = None,
        past_key_values: Optional[Tuple[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        decoder_inputs_embeds: Optional[torch.Tensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, Seq2SeqModelOutput]:
        r"""
        Returns:

        Example:

        ```python
        >>> from transformers import AutoTokenizer, PegasusModel

        >>> tokenizer = AutoTokenizer.from_pretrained("google/pegasus-x-large")
        >>> model = PegasusModel.from_pretrained("google/pegasus-x-large")

        >>> inputs = tokenizer("Studies have been shown that owning a dog is good for you", return_tensors="pt")
        >>> decoder_inputs = tokenizer("Studies show that", return_tensors="pt")
        >>> outputs = model(input_ids=inputs.input_ids, decoder_input_ids=decoder_inputs.input_ids)

        >>> last_hidden_states = outputs.last_hidden_state
        >>> list(last_hidden_states.shape)
        [1, 4, 1024]
        ```"""

        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if encoder_outputs is None:
            encoder_outputs = self.encoder(
                input_ids=input_ids,
                attention_mask=attention_mask,
                inputs_embeds=inputs_embeds,
                output_attentions=output_attentions,
                output_hidden_states=output_hidden_states,
                return_dict=return_dict,
            )
        # If the user passed a tuple for encoder_outputs, we wrap it in a BaseModelOutput when return_dict=True
        elif return_dict and not isinstance(encoder_outputs, BaseModelOutput):
            encoder_outputs = BaseModelOutput(
                last_hidden_state=encoder_outputs[0],
                hidden_states=encoder_outputs[1] if len(encoder_outputs) > 1 else None,
                attentions=encoder_outputs[2] if len(encoder_outputs) > 2 else None,
            )

        # decoder outputs consists of (dec_features, past_key_value, dec_hidden, dec_attn)
        decoder_outputs = self.decoder(
            input_ids=decoder_input_ids,
            attention_mask=decoder_attention_mask,
            encoder_hidden_states=encoder_outputs[0],
            encoder_attention_mask=attention_mask,
            past_key_values=past_key_values,
            inputs_embeds=decoder_inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        if not return_dict:
            return decoder_outputs + encoder_outputs

        return Seq2SeqModelOutput(
            last_hidden_state=decoder_outputs.last_hidden_state,
            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,
        )


@add_start_docstrings("The PEGASUS-X for conditional generation (e.g. summarization).", PEGASUS_X_START_DOCSTRING)
class PegasusXForConditionalGeneration(PegasusXPreTrainedModel):
    base_model_prefix = "model"
    _keys_to_ignore_on_load_missing = [
        r"encoder.version",
        r"decoder.version",
        r"lm_head.weight",
        r"embed_positions.weight",
        "decoder.embed_tokens.weight",
        "encoder.embed_tokens.weight",
    ]

    def __init__(self, config: PegasusXConfig):
        super().__init__(config)
        self.model = PegasusXModel(config)
        self.lm_head = nn.Linear(config.d_model, self.model.shared.num_embeddings, bias=False)

        # Initialize weights and apply final processing
        self.post_init()

    def get_encoder(self):
        return self.model.get_encoder()

    def get_decoder(self):
        return self.model.get_decoder()

    def resize_token_embeddings(self, new_num_tokens: int) -> nn.Embedding:
        new_embeddings = super().resize_token_embeddings(new_num_tokens)
        return new_embeddings

    def get_output_embeddings(self):
        return self.lm_head

    def set_output_embeddings(self, new_embeddings):
        self.lm_head = new_embeddings

    def resize_position_embeddings(self, new_num_position_embeddings: int):
        """
        Resizes position embeddings matrix of the model if `new_num_position_embeddings !=
        config.max_position_embeddings`.

        Arguments:
            new_num_position_embeddings (`int`):
                The number of new position embeddings. If position embeddings are learned, increasing the size will add
                newly initialized vectors at the end, whereas reducing the size will remove vectors from the end. If
                position embeddings are not learned (*e.g.* sinusoidal position embeddings), increasing the size will
                add correct vectors at the end following the position encoding algorithm, whereas reducing the size
                will remove vectors from the end.
        """
        self.config.max_position_embeddings = new_num_position_embeddings
        self.model.encoder.resize_position_embeddings(new_num_position_embeddings)
        self.model.decoder.resize_position_embeddings(new_num_position_embeddings)

    def get_position_embeddings(self) -> Tuple[nn.Embedding]:
        """
        Returns the position embeddings matrix
        """
        return (self.model.encoder.get_position_embeddings(), self.model.decoder.get_position_embeddings())

    @add_start_docstrings_to_model_forward(PEGASUS_X_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=Seq2SeqLMOutput, config_class=_CONFIG_FOR_DOC)
    @add_end_docstrings(PEGASUS_X_GENERATION_EXAMPLE)
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        decoder_input_ids: Optional[torch.Tensor] = None,
        decoder_attention_mask: Optional[torch.Tensor] = None,
        encoder_outputs: Optional[Tuple[torch.FloatTensor]] = None,
        past_key_values: Optional[Tuple[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        decoder_inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, Seq2SeqLMOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
            config.vocab_size]` or -100 (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]`.

        Returns:

        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if labels is not None:
            if use_cache:
                logger.warning("The `use_cache` argument is changed to `False` since `labels` is provided.")
            use_cache = False
            if 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
                )

        outputs = self.model(
            input_ids,
            attention_mask=attention_mask,
            decoder_input_ids=decoder_input_ids,
            encoder_outputs=encoder_outputs,
            decoder_attention_mask=decoder_attention_mask,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            decoder_inputs_embeds=decoder_inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )
        lm_logits = self.lm_head(outputs[0])

        masked_lm_loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            masked_lm_loss = loss_fct(lm_logits.view(-1, self.config.vocab_size), labels.view(-1))

        if not return_dict:
            output = (lm_logits,) + outputs[1:]
            return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output

        return Seq2SeqLMOutput(
            loss=masked_lm_loss,
            logits=lm_logits,
            past_key_values=outputs.past_key_values,
            decoder_hidden_states=outputs.decoder_hidden_states,
            decoder_attentions=outputs.decoder_attentions,
            cross_attentions=outputs.cross_attentions,
            encoder_last_hidden_state=outputs.encoder_last_hidden_state,
            encoder_hidden_states=outputs.encoder_hidden_states,
            encoder_attentions=outputs.encoder_attentions,
        )

    def prepare_inputs_for_generation(
        self,
        decoder_input_ids,
        past_key_values=None,
        attention_mask=None,
        use_cache=None,
        encoder_outputs=None,
        **kwargs,
    ):
        # cut decoder_input_ids if past is used
        if past_key_values is not None:
            decoder_input_ids = decoder_input_ids[:, -1:]

        return {
            "input_ids": None,  # encoder_outputs is defined. input_ids not needed
            "encoder_outputs": encoder_outputs,
            "past_key_values": past_key_values,
            "decoder_input_ids": decoder_input_ids,
            "attention_mask": attention_mask,
            "use_cache": use_cache,  # change this to avoid caching (presumably for debugging)
        }

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

    @staticmethod
    def _reorder_cache(past_key_values, beam_idx):
        reordered_past = ()
        for layer_past in past_key_values:
            # cached cross_attention states don't have to be reordered -> they are always the same
            reordered_past += (
                tuple(past_state.index_select(0, beam_idx) for past_state in layer_past[:2]) + layer_past[2:],
            )
        return reordered_past


# Copied from transformers.models.bart.modeling_bart.BartDecoderWrapper with Bart->PegasusX
class PegasusXDecoderWrapper(PegasusXPreTrainedModel):
    """
    This wrapper class is a helper class to correctly load pretrained checkpoints when the causal language model is
    used in combination with the [`EncoderDecoderModel`] framework.
    """

    def __init__(self, config):
        super().__init__(config)
        self.decoder = PegasusXDecoder(config)

    def forward(self, *args, **kwargs):
        return self.decoder(*args, **kwargs)