src/opt.py

# coding=utf-8
# Copyright 2018 The OpenAI Team Authors and HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION.  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 OpenAI GPT-2 model."""

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

import torch
import torch.utils.checkpoint
from packaging import version
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss

from transformers.models.gpt2.modeling_gpt2 import load_tf_weights_in_gpt2, GPT2LMHeadModel, GPT2MLP, GPT2Attention, GPT2Block, GPT2Model 

from transformers.models.opt.modeling_opt import OPTForCausalLM, OPTAttention, OPTDecoderLayer, OPTModel, OPTDecoder 

from transformers.activations import ACT2FN
from transformers.modeling_outputs import (
    BaseModelOutputWithPastAndCrossAttentions,
    CausalLMOutputWithCrossAttentions,
    SequenceClassifierOutputWithPast,
    TokenClassifierOutput,
)

from transformers.modeling_outputs import (
    BaseModelOutputWithPast,
    CausalLMOutputWithPast,
    QuestionAnsweringModelOutput,
    SequenceClassifierOutputWithPast,
)

from transformers.modeling_utils import PreTrainedModel, SequenceSummary
from transformers.pytorch_utils import Conv1D, find_pruneable_heads_and_indices, prune_conv1d_layer
from transformers.utils import (
    ModelOutput,
    logging,
)
from transformers.utils.model_parallel_utils import assert_device_map, get_device_map
from transformers.models.opt.configuration_opt import OPTConfig


if version.parse(torch.__version__) >= version.parse("1.6"):
    is_amp_available = True
    from torch.cuda.amp import autocast
else:
    is_amp_available = False

    
class ThisOPTConfig(OPTConfig):
    model_type = "this_opt"

    def __init__(
        self,
        cross_attention_reduce_factor = 1,
        **kwargs,
    ):
        super().__init__(**kwargs)
        self.cross_attention_reduce_factor = cross_attention_reduce_factor
        

class ThisOPTAttention(OPTAttention):
    """Multi-headed attention from 'Attention Is All You Need' paper"""

    def __init__(
        self,
        embed_dim,
        num_heads,
        dropout = 0.0,
        is_decoder = False,
        bias = True,
        config=None,
        is_cross_attention=False,
    ):
        super().__init__(embed_dim,num_heads, dropout,is_decoder,bias)
        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.cross_attention_reduce_factor = config.cross_attention_reduce_factor
        self.head_dim = int(self.head_dim / self.cross_attention_reduce_factor)


        if is_cross_attention:
            #print("self", int(embed_dim / self.cross_attention_reduce_factor))
            self.k_proj = nn.Linear(768, int(embed_dim / self.cross_attention_reduce_factor), bias=bias)
            #print("self.k_proj",self.k_proj)
            self.v_proj = nn.Linear(768, int(embed_dim / self.cross_attention_reduce_factor), bias=bias)
            self.q_proj = nn.Linear(embed_dim, int(embed_dim / self.cross_attention_reduce_factor), bias=bias)
            self.out_proj = nn.Linear(int(embed_dim / self.cross_attention_reduce_factor),embed_dim, bias=bias)

            self.embed_dim=int(embed_dim / self.cross_attention_reduce_factor)
        else:
            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,
        key_value_states = None,
        past_key_value = None,
        attention_mask = None,
        layer_head_mask = None,
        output_attentions= False,
    ):
        """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
        if is_cross_attention and past_key_value is not None:
            # 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.view(*proj_shape)
        value_states = value_states.view(*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))
            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, past_key_value


class ThisOPTDecoderLayer(OPTDecoderLayer):
    def __init__(self, config):
        super().__init__(config)

        if config.add_cross_attention:
            self.encoder_attn = ThisOPTAttention(
                embed_dim=self.embed_dim,
                num_heads=config.num_attention_heads,
                dropout=config.attention_dropout,
                is_decoder=True,
                #bias=config.enable_bias,
                config=config,
                is_cross_attention=True
            )
            self.encoder_attn_layer_norm = nn.LayerNorm(self.embed_dim,elementwise_affine=config.layer_norm_elementwise_affine)


    def forward(
        self,
        hidden_states,
        attention_mask= None,
        encoder_hidden_states = None,
        encoder_attention_mask = None,
        layer_head_mask = None,
        cross_attn_head_mask = None,
        output_attentions = False,
        use_cache = False,
        past_key_value = None,
    ):
        """
        Args:
            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
            attention_mask (`torch.FloatTensor`, *optional*): 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`, *optional*): mask for attention heads in a given layer of size
                `(encoder_attention_heads,)`.
            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 (`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`).
            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
        """

        residual = hidden_states

        # 125m, 1.7B, ..., 175B applies layer norm BEFORE attention
        if self.do_layer_norm_before:
            hidden_states = self.self_attn_layer_norm(hidden_states)

        # Self Attention
        hidden_states, self_attn_weights, present_key_value = 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,
        )
        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,
                layer_head_mask=cross_attn_head_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



        # 350m applies layer norm AFTER attention
        if not self.do_layer_norm_before:
            hidden_states = self.self_attn_layer_norm(hidden_states)

        # Fully Connected
        hidden_states_shape = hidden_states.shape
        hidden_states = hidden_states.reshape(-1, hidden_states.size(-1))
        residual = hidden_states

        # 125m, 1.7B, ..., 175B applies layer norm BEFORE attention
        if self.do_layer_norm_before:
            hidden_states = self.final_layer_norm(hidden_states)

        hidden_states = self.fc1(hidden_states)
        hidden_states = self.activation_fn(hidden_states)

        hidden_states = self.fc2(hidden_states)
        hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)

        hidden_states = (residual + hidden_states).view(hidden_states_shape)

        # 350m applies layer norm AFTER attention
        if not self.do_layer_norm_before:
            hidden_states = self.final_layer_norm(hidden_states)

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (self_attn_weights,)

        if use_cache:
            outputs += (present_key_value,)

        return outputs

class ThisOPTDecoder(OPTDecoder):
    def __init__(self, config):
        super().__init__(config)
        self.layers = nn.ModuleList([ThisOPTDecoderLayer(config) for _ in range(config.num_hidden_layers)])


    def forward(
        self,
        input_ids = None,
        attention_mask = None,
        encoder_hidden_states=None,
        encoder_attention_mask = None,
        head_mask = None,
        cross_attn_head_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)
            head_mask (`torch.Tensor` of shape `(num_hidden_layers, num_attention_heads)`, *optional*):
                Mask to nullify selected heads of the attention modules. Mask values selected in `[0, 1]`:
                - 1 indicates the head is **not masked**,
                - 0 indicates the head is **masked**.
            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
                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[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)

        # embed positions
        if attention_mask is None:
            attention_mask = torch.ones(inputs_embeds.shape[:2], dtype=torch.bool, device=inputs_embeds.device)
        pos_embeds = self.embed_positions(attention_mask, past_key_values_length)

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

        if self.project_in is not None:
            inputs_embeds = self.project_in(inputs_embeds)

        hidden_states = inputs_embeds + pos_embeds

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


        # check if head_mask has a correct number of layers specified if desired
        for attn_mask, mask_name in zip([head_mask], ["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://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:

                if use_cache:
                    logger.warning(
                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                    )
                    use_cache = False

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

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(decoder_layer),
                    hidden_states,
                    attention_mask,
                    head_mask[idx] if head_mask is not None else None,
                    None,
                )
            else:

                layer_outputs = decoder_layer(
                    hidden_states,
                    encoder_attention_mask=encoder_attention_mask,
                    encoder_hidden_states=encoder_hidden_states,
                    cross_attn_head_mask=cross_attn_head_mask[idx] if cross_attn_head_mask is not None else None,
                    attention_mask=attention_mask,
                    layer_head_mask=(head_mask[idx] if head_mask is not None else None),
                    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[2 if output_attentions else 1],)

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

        if self.final_layer_norm is not None:
            hidden_states = self.final_layer_norm(hidden_states)

        if self.project_out is not None:
            hidden_states = self.project_out(hidden_states)

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

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

        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] 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,
        )
        



class ThisOPTModel(OPTModel):

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

    def forward(
        self,
        input_ids = None,
        attention_mask = None,
        encoder_hidden_states=None,
        encoder_attention_mask = None,
        head_mask = None,
        cross_attn_head_mask = None,
        past_key_values = None,
        inputs_embeds = None,
        use_cache = None,
        output_attentions = None,
        output_hidden_states = None,
        return_dict = None,
    ):

        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_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])


        # decoder outputs consists of (dec_features, past_key_value, dec_hidden, dec_attn)
        decoder_outputs = self.decoder(
            input_ids=input_ids,
            attention_mask=attention_mask,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_attention_mask,
            cross_attn_head_mask=(
                        cross_attn_head_mask[idx] if cross_attn_head_mask is not None else None
            ),
            head_mask=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,
        )


        if not return_dict:
            return decoder_outputs

        return BaseModelOutputWithPast(
            last_hidden_state=decoder_outputs.last_hidden_state,
            past_key_values=decoder_outputs.past_key_values,
            hidden_states=decoder_outputs.hidden_states,
            attentions=decoder_outputs.attentions,
        )

class ThisOPTForCausalLM(OPTForCausalLM):
    config_class = ThisOPTConfig
    
    def __init__(self, config):
        super().__init__(config)
        self.model = ThisOPTModel(config)


    def forward(
        self,
        input_ids = None,
        attention_mask = None,
        encoder_hidden_states=None,
        encoder_attention_mask = None,
        head_mask = None,
        cross_attn_head_mask = None,
        past_key_values = None,
        inputs_embeds = None,
        labels = None,
        use_cache = None,
        output_attentions = None,
        output_hidden_states = None,
        return_dict = None,
    ) :


        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.decoder(
            input_ids=input_ids,
            attention_mask=attention_mask,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask = encoder_attention_mask,
            cross_attn_head_mask = cross_attn_head_mask,
            head_mask=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,
        )

        logits = self.lm_head(outputs[0]).contiguous()

        loss = None
        if labels is not None:
            # Shift so that tokens < n predict n
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            # Flatten the tokens
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(shift_logits.view(-1, self.config.vocab_size), shift_labels.view(-1))

        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,
        )