# coding=utf-8
# Copyright 2021 The Fairseq Authors 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 XGLM model."""

import math
from typing import Optional, Union

import torch
import torch.utils.checkpoint
from torch import nn

from ...activations import ACT2FN
from ...cache_utils import Cache, EncoderDecoderCache
from ...generation import GenerationMixin
from ...modeling_attn_mask_utils import _prepare_4d_attention_mask, _prepare_4d_causal_attention_mask
from ...modeling_layers import GradientCheckpointingLayer
from ...modeling_outputs import BaseModelOutputWithPastAndCrossAttentions, CausalLMOutputWithCrossAttentions
from ...modeling_utils import PreTrainedModel
from ...utils import auto_docstring, logging
from .configuration_xglm import XGLMConfig


logger = logging.get_logger(__name__)


# Copied from transformers.models.bart.modeling_bart.BartScaledWordEmbedding with Bart->XGLM
class XGLMScaledWordEmbedding(nn.Embedding):
    """
    This module overrides nn.Embeddings' forward by multiplying with embeddings scale.
    """

    def __init__(self, num_embeddings: int, embedding_dim: int, padding_idx: int, embed_scale: Optional[float] = 1.0):
        super().__init__(num_embeddings, embedding_dim, padding_idx)
        self.embed_scale = embed_scale

    def forward(self, input_ids: torch.Tensor):
        return super().forward(input_ids) * self.embed_scale


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

    def __init__(self, num_positions: int, embedding_dim: int, padding_idx: Optional[int] = None):
        super().__init__()
        self.offset = 2
        self.embedding_dim = embedding_dim
        self.padding_idx = padding_idx
        self.make_weights(num_positions + self.offset, embedding_dim, padding_idx)

    def make_weights(self, num_embeddings: int, embedding_dim: int, padding_idx: Optional[int] = None):
        emb_weights = self.get_embedding(num_embeddings, embedding_dim, padding_idx)
        if hasattr(self, "weights"):
            # in forward put the weights on the correct dtype and device of the param
            emb_weights = emb_weights.to(dtype=self.weights.dtype, device=self.weights.device)

        self.register_buffer("weights", emb_weights, persistent=False)

    @staticmethod
    def get_embedding(num_embeddings: int, embedding_dim: int, padding_idx: Optional[int] = None):
        """
        Build sinusoidal embeddings.

        This matches the implementation in tensor2tensor, but differs slightly from the description in Section 3.5 of
        "Attention Is All You Need".
        """
        half_dim = embedding_dim // 2
        emb = math.log(10000) / (half_dim - 1)
        emb = torch.exp(torch.arange(half_dim, dtype=torch.int64).float() * -emb)
        emb = torch.arange(num_embeddings, dtype=torch.int64).float().unsqueeze(1) * emb.unsqueeze(0)
        emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1).view(num_embeddings, -1)
        if embedding_dim % 2 == 1:
            # zero pad
            emb = torch.cat([emb, torch.zeros(num_embeddings, 1)], dim=1)
        if padding_idx is not None:
            emb[padding_idx, :] = 0

        return emb.to(torch.get_default_dtype())

    @torch.no_grad()
    def forward(self, position_ids: Optional[torch.Tensor] = None, past_key_values_length: int = 0):
        bsz, seq_len = position_ids.size()
        position_ids += self.offset

        # Expand embeddings if needed. `position_ids.max()` is NOT used to keep torch.fx compatibility.
        max_pos = 2 + seq_len + past_key_values_length
        if max_pos > self.weights.size(0):
            self.make_weights(max_pos, self.embedding_dim, self.padding_idx)

        return self.weights.index_select(0, position_ids.view(-1)).view(bsz, seq_len, self.weights.shape[-1]).detach()


class XGLMAttention(nn.Module):
    """Multi-headed attention from 'Attention Is All You Need' paper"""

    def __init__(
        self,
        embed_dim: int,
        num_heads: int,
        dropout: Optional[float] = 0.0,
        is_decoder: Optional[bool] = False,
        bias: Optional[bool] = True,
        layer_idx: Optional[bool] = None,
    ):
        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.layer_idx = layer_idx

        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 forward(
        self,
        hidden_states: torch.Tensor,
        key_value_states: Optional[torch.Tensor] = None,
        past_key_value: Optional[Cache] = None,
        attention_mask: Optional[torch.Tensor] = None,
        layer_head_mask: Optional[torch.Tensor] = None,
        output_attentions: bool = False,
        cache_position: Optional[torch.Tensor] = None,
    ) -> 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()
        src_len = key_value_states.shape[1] if is_cross_attention else tgt_len

        # get query proj
        query_states = self.q_proj(hidden_states) * self.scaling

        if past_key_value is not None:
            if isinstance(past_key_value, EncoderDecoderCache):
                is_updated = past_key_value.is_updated.get(self.layer_idx)
                if is_cross_attention:
                    # after the first generated id, we can subsequently re-use all key/value_states from cache
                    curr_past_key_value = past_key_value.cross_attention_cache
                else:
                    curr_past_key_value = past_key_value.self_attention_cache
            else:
                curr_past_key_value = past_key_value

        current_states = key_value_states if is_cross_attention else hidden_states
        if is_cross_attention and past_key_value is not None and is_updated:
            # reuse k,v, cross_attentions
            key_states = curr_past_key_value.key_cache[self.layer_idx]
            value_states = curr_past_key_value.value_cache[self.layer_idx]
        else:
            key_states = self.k_proj(current_states)
            value_states = self.v_proj(current_states)
            key_states = key_states.view(bsz, src_len, -1, self.head_dim).transpose(1, 2)
            value_states = value_states.view(bsz, src_len, -1, self.head_dim).transpose(1, 2)

            if past_key_value is not None:
                # save all key/value_states to cache to be re-used for fast auto-regressive generation
                cache_position = cache_position if not is_cross_attention else None
                key_states, value_states = curr_past_key_value.update(
                    key_states, value_states, self.layer_idx, {"cache_position": cache_position}
                )
                # set flag that curr layer for cross-attn is already updated so we can re-use in subsequent calls
                if is_cross_attention:
                    past_key_value.is_updated[self.layer_idx] = True

        proj_shape = (bsz * self.num_heads, -1, self.head_dim)
        query_states = query_states.view(bsz, tgt_len, self.num_heads, self.head_dim).transpose(1, 2)
        query_states = query_states.reshape(*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 = torch.max(
                attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min, device=attn_weights.device)
            )
            attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)

        # upcast to fp32 if the weights are in fp16. Please see https://github.com/huggingface/transformers/pull/17437
        if attn_weights.dtype == torch.float16:
            attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(torch.float16)
        else:
            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 aross 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


class XGLMDecoderLayer(GradientCheckpointingLayer):
    def __init__(self, config: XGLMConfig, layer_idx=None):
        super().__init__()
        self.embed_dim = config.d_model

        self.self_attn = XGLMAttention(
            embed_dim=self.embed_dim,
            num_heads=config.attention_heads,
            dropout=config.attention_dropout,
            is_decoder=True,
            layer_idx=layer_idx,
        )
        self.dropout = config.dropout
        self.activation_fn = ACT2FN[config.activation_function]
        self.activation_dropout = config.activation_dropout

        if config.add_cross_attention:
            self.encoder_attn = XGLMAttention(
                embed_dim=self.embed_dim,
                num_heads=config.attention_heads,
                dropout=config.attention_dropout,
                is_decoder=True,
                layer_idx=layer_idx,
            )
            self.encoder_attn_layer_norm = nn.LayerNorm(self.embed_dim)

        self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
        self.fc1 = nn.Linear(self.embed_dim, config.ffn_dim)
        self.fc2 = nn.Linear(config.ffn_dim, self.embed_dim)
        self.final_layer_norm = nn.LayerNorm(self.embed_dim)

    # Copied from transformers.models.musicgen.modeling_musicgen.MusicgenDecoderLayer.forward
    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,
        layer_head_mask: Optional[torch.Tensor] = None,
        cross_attn_layer_head_mask: Optional[torch.Tensor] = None,
        past_key_value: Optional[Cache] = None,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = True,
        cache_position: Optional[torch.Tensor] = None,
    ) -> torch.Tensor:
        """
        Args:
            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, 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 `(batch, seq_len, 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.
            layer_head_mask (`torch.FloatTensor`): mask for attention heads in a given layer of size
                `(encoder_attention_heads,)`.
            cross_attn_layer_head_mask (`torch.FloatTensor`): mask for cross-attention heads in a given layer of
                size `(decoder_attention_heads,)`.
            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.
        """
        residual = hidden_states
        hidden_states = self.self_attn_layer_norm(hidden_states)

        # Self Attention
        hidden_states, self_attn_weights = self.self_attn(
            hidden_states=hidden_states,
            past_key_value=past_key_value,
            attention_mask=attention_mask,
            layer_head_mask=layer_head_mask,
            output_attentions=output_attentions,
            cache_position=cache_position,
        )
        hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
        hidden_states = residual + hidden_states

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

            hidden_states, cross_attn_weights = self.encoder_attn(
                hidden_states=hidden_states,
                key_value_states=encoder_hidden_states,
                attention_mask=encoder_attention_mask,
                layer_head_mask=cross_attn_layer_head_mask,
                past_key_value=past_key_value,
                output_attentions=output_attentions,
                cache_position=cache_position,
            )
            hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
            hidden_states = residual + hidden_states

        # 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)

        return outputs


@auto_docstring
class XGLMPreTrainedModel(PreTrainedModel):
    config: XGLMConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["XGLMDecoderLayer"]

    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)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()


@auto_docstring
class XGLMModel(XGLMPreTrainedModel):
    def __init__(self, config: XGLMConfig, embed_tokens: Optional[nn.Embedding] = None):
        r"""
        embed_tokens (`nn.Embedding`, *optional*):
            output embeddings
        """
        super().__init__(config)
        self.dropout = config.dropout
        self.layerdrop = config.layerdrop
        self.padding_idx = config.pad_token_id
        self.max_target_positions = config.max_position_embeddings
        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 = XGLMScaledWordEmbedding(
                config.vocab_size, config.d_model, self.padding_idx, embed_scale=embed_scale
            )

        self.embed_positions = XGLMSinusoidalPositionalEmbedding(
            config.max_position_embeddings,
            config.d_model,
            config.pad_token_id,
        )
        self.layers = nn.ModuleList([XGLMDecoderLayer(config, layer_idx=i) for i in range(config.num_layers)])
        self.layer_norm = nn.LayerNorm(config.d_model)

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

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        encoder_hidden_states: Optional[torch.Tensor] = None,
        encoder_attention_mask: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        cross_attn_head_mask: Optional[torch.Tensor] = None,
        past_key_values: Optional[list[torch.FloatTensor]] = None,
        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,
        cache_position: Optional[torch.Tensor] = None,
    ) -> Union[tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]:
        r"""
        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)
        cross_attn_head_mask (`torch.Tensor` of shape `(num_layers, attention_heads)`, *optional*):
            Mask to nullify selected heads of the cross-attention modules. Mask values selected in `[0, 1]`:

            - 1 indicates the head is **not masked**,
            - 0 indicates the head is **masked**.
        """
        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 input_ids and inputs_embeds at the same time")
        elif input_ids is not None:
            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
            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)

        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

        # initialize `past_key_values`
        return_legacy_cache = False
        if use_cache and not isinstance(past_key_values, Cache):
            return_legacy_cache = True
            logger.warning_once(
                "Passing a tuple of `past_key_values` is deprecated and will be removed in Transformers v4.58.0. "
                "You should pass an instance of `EncoderDecoderCache` instead, e.g. "
                "`past_key_values=EncoderDecoderCache.from_legacy_cache(past_key_values)`."
            )
            past_key_values = EncoderDecoderCache.from_legacy_cache(past_key_values)

        past_key_values_length = past_key_values.get_seq_length() if past_key_values is not None else 0
        attention_mask = _prepare_4d_causal_attention_mask(
            attention_mask, input_shape, inputs_embeds, past_key_values_length
        )

        if position_ids is None:
            position_ids = torch.arange(
                past_key_values_length,
                input_shape[-1] + past_key_values_length,
                dtype=torch.long,
                device=input_ids.device if input_ids is not None else inputs_embeds.device,
            )
            position_ids = position_ids.unsqueeze(0)

        # 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 = _prepare_4d_attention_mask(
                encoder_attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
            )

        hidden_states = inputs_embeds + self.embed_positions(position_ids, past_key_values_length).to(
            inputs_embeds.device
        )
        hidden_states = nn.functional.dropout(hidden_states, p=float(self.dropout), training=self.training)

        # 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

        # check if head_mask/cross_attn_head_mask has a correct number of layers specified if desired
        for attn_mask, mask_name in zip([head_mask, cross_attn_head_mask], ["head_mask", "cross_attn_head_mask"]):
            if attn_mask is not None:
                if attn_mask.size()[0] != len(self.layers):
                    raise ValueError(
                        f"The `{mask_name}` should be specified for {len(self.layers)} layers, but it is for"
                        f" {head_mask.size()[0]}."
                    )
        for idx, decoder_layer in enumerate(self.layers):
            # add LayerDrop (see https://huggingface.co/papers/1909.11556 for description)
            if output_hidden_states:
                all_hidden_states += (hidden_states,)
            if self.training:
                dropout_probability = torch.rand([])
                if dropout_probability < self.layerdrop:
                    continue

            layer_outputs = decoder_layer(
                hidden_states,
                attention_mask,
                encoder_hidden_states,  # as a positional argument for gradient checkpointing
                encoder_attention_mask=encoder_attention_mask,
                layer_head_mask=(head_mask[idx] if head_mask is not None else None),
                cross_attn_layer_head_mask=(cross_attn_head_mask[idx] if cross_attn_head_mask is not None else None),
                past_key_value=past_key_values,
                output_attentions=output_attentions,
                use_cache=use_cache,
                cache_position=cache_position,
            )
            hidden_states = layer_outputs[0]

            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,)

        if return_legacy_cache:
            past_key_values = past_key_values.to_legacy_cache()

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


@auto_docstring(
    custom_intro="""
    The XGLM Model transformer with a language modeling head on top (linear layer with weights tied to the input
    embeddings).
    """
)
class XGLMForCausalLM(XGLMPreTrainedModel, GenerationMixin):
    base_model_prefix = "model"
    _tied_weights_keys = ["lm_head.weight"]

    def __init__(self, config):
        super().__init__(config)
        self.model = XGLMModel(config)
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

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

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        encoder_hidden_states: Optional[torch.Tensor] = None,
        encoder_attention_mask: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        cross_attn_head_mask: Optional[torch.Tensor] = None,
        past_key_values: Optional[list[torch.FloatTensor]] = None,
        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,
        cache_position: Optional[torch.Tensor] = None,
        **kwargs,
    ) -> Union[tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]:
        r"""
        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)
        cross_attn_head_mask (`torch.Tensor` of shape `(num_layers, attention_heads)`, *optional*):
            Mask to nullify selected heads of the cross-attention modules. Mask values selected in `[0, 1]`:

            - 1 indicates the head is **not masked**,
            - 0 indicates the head is **masked**.
        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]`.
        """

        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

        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs = self.model(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_attention_mask,
            head_mask=head_mask,
            cross_attn_head_mask=cross_attn_head_mask,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            cache_position=cache_position,
        )

        logits = self.lm_head(outputs[0])

        loss = None
        if labels is not None:
            loss = self.loss_function(
                logits,
                labels,
                vocab_size=self.config.vocab_size,
                pad_token_id=self.config.pad_token_id,
                **kwargs,
            )

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

        return CausalLMOutputWithCrossAttentions(
            loss=loss,
            logits=logits,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
            cross_attentions=outputs.cross_attentions,
        )


__all__ = ["XGLMForCausalLM", "XGLMModel", "XGLMPreTrainedModel"]