Upload 2 files

configuration_deltalm.py

@@ -0,0 +1,99 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+
+""" deltalm model configuration"""
+
+import warnings
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+logger = logging.get_logger(__name__)
+
+class DeltalmConfig(PretrainedConfig):
+
+    model_type = "Deltalm"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {"num_attention_heads": "encoder_attention_heads", "hidden_size": "d_model"}
+
+    def __init__(
+        self,
+        vocab_size=250001,
+        max_position_embeddings=1024,
+        encoder_layers=12,
+        encoder_ffn_dim=3072,
+        encoder_attention_heads=12,
+        decoder_layers=6,
+        decoder_ffn_dim=3072,
+        decoder_attention_heads=12,
+        encoder_layerdrop=0.0,
+        decoder_layerdrop=0.0,
+        activation_function="gelu",
+        d_model=1024,
+        dropout=0.1,
+        attention_dropout=0.0,
+        activation_dropout=0.0,
+        init_std=0.02,
+        classifier_dropout=0.0,
+        scale_embedding=False,
+        use_cache=True,
+        num_labels=3,
+        pad_token_id=1,
+        bos_token_id=0,
+        eos_token_id=2,
+        is_encoder_decoder=True,
+        decoder_start_token_id=0,
+        forced_eos_token_id=2,
+        label_smoothing=0.1,
+        length_penalty=1.0,
+        encoder_normalize_before=False,
+        **kwargs
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.d_model = d_model
+        self.encoder_ffn_dim = encoder_ffn_dim
+        self.encoder_layers = encoder_layers
+        self.encoder_attention_heads = encoder_attention_heads
+        self.decoder_ffn_dim = decoder_ffn_dim
+        self.decoder_layers = decoder_layers
+        self.decoder_attention_heads = decoder_attention_heads
+        self.dropout = dropout
+        self.attention_dropout = attention_dropout
+        self.activation_dropout = activation_dropout
+        self.activation_function = activation_function
+        self.init_std = init_std
+        self.encoder_layerdrop = encoder_layerdrop
+        self.decoder_layerdrop = decoder_layerdrop
+        self.classifier_dropout = classifier_dropout
+        self.use_cache = use_cache
+        self.num_hidden_layers = encoder_layers
+        self.scale_embedding = scale_embedding  # scale factor will be sqrt(d_model) if True
+        self.label_smoothing = label_smoothing
+        self.encoder_normalize_before = encoder_normalize_before
+
+        super().__init__(
+            num_labels=num_labels,
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            is_encoder_decoder=is_encoder_decoder,
+            decoder_start_token_id=decoder_start_token_id,
+            forced_eos_token_id=forced_eos_token_id,
+            length_penalty=length_penalty,
+            **kwargs,
+        )
+
+        # ensure backward compatibility for BART CNN models
+        if self.forced_bos_token_id is None and kwargs.get("force_bos_token_to_be_generated", False):
+            self.forced_bos_token_id = self.bos_token_id
+            warnings.warn(
+                f"Please make sure the config includes `forced_bos_token_id={self.bos_token_id}` in future versions. "
+                "The config can simply be saved and uploaded again to be fixed."
+            )
+
+    @property
+    def num_attention_heads(self) -> int:
+        return self.encoder_attention_heads
+
+    @property
+    def hidden_size(self) -> int:
+        return self.d_model

modeling_deltalm.py

@@ -0,0 +1,1551 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+
+
+import copy
+import math
+import random
+import torch
+import torch.nn as nn
+import torch.utils.checkpoint
+from torch.nn import CrossEntropyLoss
+from typing import List, Optional, Tuple, Union
+
+from transformers.modeling_utils import PreTrainedModel
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import (
+    BaseModelOutput,
+    BaseModelOutputWithPastAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions,
+    Seq2SeqModelOutput,
+    Seq2SeqLMOutput,
+)
+from transformers.file_utils import (
+    add_end_docstrings,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    replace_return_docstrings,
+)
+
+import logging
+from .configuration_deltalm import DeltalmConfig
+logger = logging.getLogger(__name__)
+
+_CHECKPOINT_FOR_DOC = "IDEA-CCNL/Randeng-Deltalm-362M-En-Zn"
+_CONFIG_FOR_DOC = "DeltalmConfig"
+_TOKENIZER_FOR_DOC = "DeltalmTokenizer"
+
+# Base model docstring
+_EXPECTED_OUTPUT_SHAPE = [1, 8, 768]
+
+
+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
+
+
+def _make_causal_mask(input_ids_shape: torch.Size, dtype: torch.dtype, 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(float("-inf")))
+    mask_cond = torch.arange(mask.size(-1))
+    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), mask], dim=-1)
+    return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)
+
+
+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 DeltalmLearnedPositionalEmbedding(nn.Embedding):
+    """
+    This module learns positional embeddings up to a fixed maximum size.
+    """
+
+    def __init__(self, num_embeddings: int, embedding_dim: int):
+        # Deltalm is set up so that if padding_idx is specified then offset the embedding ids by 2
+        # and adjust num_embeddings appropriately. Other models don't have this hack
+        self.offset = 2
+        super().__init__(num_embeddings + self.offset, embedding_dim)
+
+    def forward(self, input_ids_shape: torch.Size, past_key_values_length: int = 0):
+        """`input_ids_shape` is expected to be [bsz x seqlen]."""
+        bsz, seq_len = input_ids_shape[:2]
+        positions = torch.arange(
+            past_key_values_length, past_key_values_length + seq_len, dtype=torch.long, device=self.weight.device
+        )
+        return super().forward(positions + self.offset)
+
+
+class DeltalmAttention(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
+        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 = 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 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 DeltalmEncoderLayer(nn.Module):
+    def __init__(self, config: DeltalmConfig):
+        super().__init__()
+        self.embed_dim = config.d_model
+        self.self_attn = DeltalmAttention(
+            embed_dim=self.embed_dim,
+            num_heads=config.encoder_attention_heads,
+            dropout=config.attention_dropout,
+        )
+        self.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)
+
+    def forward(
+        self,
+        hidden_states: torch.FloatTensor,
+        attention_mask: torch.FloatTensor,
+        layer_head_mask: torch.FloatTensor,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.FloatTensor, Optional[torch.FloatTensor]]:
+        """
+        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.
+            layer_head_mask (`torch.FloatTensor`): 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.
+        """
+        residual = hidden_states
+        hidden_states, attn_weights, _ = self.self_attn(
+            hidden_states=hidden_states,
+            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
+        hidden_states = self.self_attn_layer_norm(hidden_states)
+
+        residual = 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
+        hidden_states = self.final_layer_norm(hidden_states)
+
+        if hidden_states.dtype == torch.float16 and (
+            torch.isinf(hidden_states).any() or torch.isnan(hidden_states).any()
+        ):
+            clamp_value = torch.finfo(hidden_states.dtype).max - 1000
+            hidden_states = torch.clamp(hidden_states, min=-clamp_value, max=clamp_value)
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (attn_weights,)
+
+        return outputs
+
+
+class DeltalmDecoderLayer(nn.Module):
+    def __init__(self, config: DeltalmConfig):
+        super().__init__()
+        self.embed_dim = config.d_model
+
+        self.self_attn = DeltalmAttention(
+            embed_dim=self.embed_dim,
+            num_heads=config.decoder_attention_heads,
+            dropout=config.attention_dropout,
+            is_decoder=True,
+        )
+        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 = DeltalmAttention(
+            self.embed_dim,
+            config.decoder_attention_heads,
+            dropout=config.attention_dropout,
+            is_decoder=True,
+        )
+        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.fc3 = nn.Linear(self.embed_dim, config.decoder_ffn_dim)
+        self.fc4 = nn.Linear(config.decoder_ffn_dim, self.embed_dim)
+
+        self.ffn_layer_norm = nn.LayerNorm(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,
+        layer_head_mask: Optional[torch.Tensor] = None,
+        cross_attn_layer_head_mask: Optional[torch.Tensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = True,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        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
+
+        # 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,
+            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
+        hidden_states = self.self_attn_layer_norm(hidden_states)
+
+        # Add another ffn after self-attention to keep the structure same to encoder-layer
+        residual = hidden_states
+        hidden_states = self.activation_fn(self.fc3(hidden_states))
+        hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
+        hidden_states = self.fc4(hidden_states)
+        hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
+        hidden_states = residual + hidden_states
+        hidden_states = self.ffn_layer_norm(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
+
+            # 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_layer_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
+            hidden_states = self.encoder_attn_layer_norm(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.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
+        hidden_states = self.final_layer_norm(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 DeltalmPretrainedModel(PreTrainedModel):
+    config_class = DeltalmConfig
+    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)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, (DeltalmDecoder, DeltalmEncoder)):
+            module.gradient_checkpointing = value
+
+
+class DeltalmDecoder(DeltalmPretrainedModel):
+    """
+    Transformer decoder consisting of *config.decoder_layers* layers. Each layer is a [`DeltalmDecoderLayer`]
+    Args:
+        config: DeltalmConfig
+        embed_tokens (nn.Embedding): output embedding
+    """
+
+    def __init__(self, config: DeltalmConfig, embed_tokens: Optional[nn.Embedding] = None):
+        super().__init__(config)
+        self.dropout = config.dropout
+        self.layerdrop = config.decoder_layerdrop
+        self.padding_idx = config.pad_token_id
+        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.padding_idx)
+
+        self.embed_positions = DeltalmLearnedPositionalEmbedding(
+            config.max_position_embeddings,
+            config.d_model,
+        )
+        self.layers = nn.ModuleList([DeltalmDecoderLayer(config) for _ in range(config.decoder_layers)])
+        self.layernorm_embedding = nn.LayerNorm(config.d_model)
+
+        self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        self.post_init()
+
+        # fairseq实现了一个 nn.init.normal_(self.output_projection.weight, mean=0, std=self.output_embed_dim ** -0.5) 对最后的output权重做正态分布转换？
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    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, past_key_values_length=past_key_values_length
+            ).to(inputs_embeds.device)
+
+        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])
+            combined_attention_mask = (
+                expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
+            )
+
+        return combined_attention_mask
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.FloatTensor] = None,
+        encoder_attention_mask: Optional[torch.LongTensor] = 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.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPastAndCrossAttentions]:
+        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 [`DeltalmTokenizer`]. 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)
+            head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_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**.
+            cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*):
+                Mask to nullify selected heads of the cross-attention modules in the decoder to avoid performing
+                cross-attention on hidden heads. 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
+                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(input_shape, past_key_values_length)
+
+        hidden_states = inputs_embeds + positions
+        hidden_states = self.layernorm_embedding(hidden_states)
+
+        hidden_states = nn.functional.dropout(hidden_states, p=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
+        next_decoder_cache = () if use_cache 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://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, 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,
+                    head_mask[idx] if head_mask is not None else None,
+                    cross_attn_head_mask[idx] if cross_attn_head_mask is not None else None,
+                    None,
+                )
+            else:
+
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    encoder_hidden_states=encoder_hidden_states,
+                    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_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],)
+
+        # 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,
+        )
+
+
+DELTALM_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 ([`DeltalmConfig`]):
+            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.
+"""
+
+DELTALM_GENERATION_EXAMPLE = r"""
+    Summarization example:
+    ```python
+    >>> from transformers import DeltalmTokenizer, DeltalmForConditionalGeneration
+    >>> model = DeltalmForConditionalGeneration.from_pretrained("facebook/deltalm-large-cnn")
+    >>> tokenizer = DeltalmTokenizer.from_pretrained("facebook/deltalm-large-cnn")
+    >>> 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"], num_beams=2, min_length=0, max_length=20)
+    >>> tokenizer.batch_decode(summary_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    'PG&E scheduled the blackouts in response to forecasts for high winds amid dry conditions'
+    ```
+    Mask filling example:
+    ```python
+    >>> from transformers import DeltalmTokenizer, DeltalmForConditionalGeneration
+    >>> tokenizer = DeltalmTokenizer.from_pretrained("facebook/deltalm-base")
+    >>> model = DeltalmForConditionalGeneration.from_pretrained("facebook/deltalm-base")
+    >>> TXT = "My friends are <mask> but they eat too many carbs."
+    >>> input_ids = tokenizer([TXT], return_tensors="pt")["input_ids"]
+    >>> logits = model(input_ids).logits
+    >>> masked_index = (input_ids[0] == tokenizer.mask_token_id).nonzero().item()
+    >>> probs = logits[0, masked_index].softmax(dim=0)
+    >>> values, predictions = probs.topk(5)
+    >>> tokenizer.decode(predictions).split()
+    ['not', 'good', 'healthy', 'great', 'very']
+    ```
+"""
+
+DELTALM_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 [`DeltalmTokenizer`]. 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)
+        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 [`DeltalmTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+            [What are decoder input IDs?](../glossary#decoder-input-ids)
+            Deltalm uses the `eos_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`).
+            For translation and summarization training, `decoder_input_ids` should be provided. If no
+            `decoder_input_ids` is provided, the model will create this tensor by shifting the `input_ids` to the right
+            for denoising pre-training following the paper.
+        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.
+            If you want to change padding behavior, you should read [`modeling_deltalm._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+        head_mask (`torch.Tensor` of shape `(encoder_layers, encoder_attention_heads)`, *optional*):
+            Mask to nullify selected heads of the attention modules in the encoder. Mask values selected in `[0, 1]`:
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        decoder_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*):
+            Mask to nullify selected heads of the attention modules in the decoder. Mask values selected in `[0, 1]`:
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*):
+            Mask to nullify selected heads of the cross-attention modules in the decoder. Mask values selected in `[0,
+            1]`:
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        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 DeltalmEncoder(DeltalmPretrainedModel):
+    """
+    Transformer encoder consisting of *config.encoder_layers* self attention layers. Each layer is a
+    [`DeltalmEncoderLayer`].
+    Args:
+        config: DeltalmConfig
+        embed_tokens (nn.Embedding): output embedding
+    """
+
+    def __init__(self, config: DeltalmConfig, embed_tokens: Optional[nn.Embedding] = None):
+        super().__init__(config)
+
+        self.dropout = config.dropout
+        self.layerdrop = config.encoder_layerdrop
+
+        embed_dim = config.d_model
+        self.padding_idx = config.pad_token_id
+        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.padding_idx)
+
+        self.embed_positions = DeltalmLearnedPositionalEmbedding(
+            config.max_position_embeddings,
+            embed_dim,
+        )
+        self.layers = nn.ModuleList([DeltalmEncoderLayer(config) for _ in range(config.encoder_layers)])
+        self.layernorm_embedding = nn.LayerNorm(embed_dim)
+
+        self.gradient_checkpointing = False
+        if config.encoder_normalize_before:
+            self.layer_norm = nn.LayerNorm(embed_dim)
+        else:
+            self.layer_norm = None
+        # 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
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutput]:
+        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 [`DeltalmTokenizer`]. 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 `(encoder_layers, encoder_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**.
+            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(input_shape)
+
+        hidden_states = inputs_embeds + embed_pos
+        hidden_states = self.layernorm_embedding(hidden_states)
+        hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
+
+        # expand attention_mask
+        if attention_mask is not None:
+            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+            attention_mask = _expand_mask(attention_mask, inputs_embeds.dtype)
+
+        encoder_states = () if output_hidden_states else None
+        all_attentions = () if output_attentions else None
+
+        # check if head_mask has a correct number of layers specified if desired
+        if head_mask is not None:
+            if head_mask.size()[0] != (len(self.layers)):
+                raise ValueError(
+                    f"The head_mask should be specified for {len(self.layers)} layers, but it is for"
+                    f" {head_mask.size()[0]}."
+                )
+
+        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,
+                        attention_mask,
+                        (head_mask[idx] if head_mask is not None else None),
+                    )
+                else:
+                    layer_outputs = encoder_layer(
+                        hidden_states,
+                        attention_mask,
+                        layer_head_mask=(head_mask[idx] if head_mask is not None else None),
+                        output_attentions=output_attentions,
+                    )
+
+                hidden_states = layer_outputs[0]
+
+            if output_attentions:
+                all_attentions = all_attentions + (layer_outputs[1],)
+
+        if self.layer_norm is not None:
+            hidden_states = self.layer_norm(hidden_states)
+        # hidden_states = self.layernorm_embedding(hidden_states)
+
+        if output_hidden_states:
+            encoder_states = encoder_states + (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 DeltalmModel(DeltalmPretrainedModel):
+    def __init__(self, config: DeltalmConfig):
+        super().__init__(config)
+
+        padding_idx, vocab_size = config.pad_token_id, config.vocab_size
+        self.shared = nn.Embedding(vocab_size, config.d_model, padding_idx)
+
+        self.encoder = DeltalmEncoder(config, self.shared)
+        self.decoder = DeltalmDecoder(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
+
+    @add_start_docstrings_to_model_forward(DELTALM_INPUTS_DOCSTRING)
+    # @add_code_sample_docstrings(
+    #     processor_class=_TOKENIZER_FOR_DOC,
+    #     checkpoint=_CHECKPOINT_FOR_DOC,
+    #     output_type=Seq2SeqModelOutput,
+    #     config_class=_CONFIG_FOR_DOC,
+    #     expected_output=_EXPECTED_OUTPUT_SHAPE,
+    # )
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        decoder_input_ids: Optional[torch.LongTensor] = None,
+        decoder_attention_mask: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        decoder_head_mask: Optional[torch.Tensor] = None,
+        cross_attn_head_mask: Optional[torch.Tensor] = None,
+        encoder_outputs: Optional[List[torch.FloatTensor]] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        decoder_inputs_embeds: Optional[torch.FloatTensor] = 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]:
+
+        # different to other models, Deltalm automatically creates decoder_input_ids from
+        # input_ids if no decoder_input_ids are provided
+        if decoder_input_ids is None and decoder_inputs_embeds is None:
+            if input_ids is None:
+                raise ValueError(
+                    "If no `decoder_input_ids` or `decoder_inputs_embeds` are "
+                    "passed, `input_ids` cannot be `None`. Please pass either "
+                    "`input_ids` or `decoder_input_ids` or `decoder_inputs_embeds`."
+                )
+
+            decoder_input_ids = shift_tokens_right(
+                input_ids, self.config.pad_token_id, self.config.decoder_start_token_id
+            )
+
+        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,
+                head_mask=head_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,
+            head_mask=decoder_head_mask,
+            cross_attn_head_mask=cross_attn_head_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
+
+        logger.debug("last_hidden_state.size:  %s", decoder_outputs.last_hidden_state)
+        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 DELTALM Model with a language modeling head. Can be used for translation.", DELTALM_START_DOCSTRING
+)
+class DeltalmForConditionalGeneration(DeltalmPretrainedModel):
+    base_model_prefix = "model"
+    _keys_to_ignore_on_load_missing = [r"final_logits_bias", r"lm_head.weight"]
+
+    def __init__(self, config: DeltalmConfig):
+        super().__init__(config)
+        self.model = DeltalmModel(config)
+        self.register_buffer("final_logits_bias", torch.zeros((1, self.model.shared.num_embeddings)))
+        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)
+        self._resize_final_logits_bias(new_num_tokens)
+        return new_embeddings
+
+    def _resize_final_logits_bias(self, new_num_tokens: int) -> None:
+        logger.debug("Debug: coming to _resize_final_logits_bias")
+        old_num_tokens = self.final_logits_bias.shape[-1]
+        if new_num_tokens <= old_num_tokens:
+            new_bias = self.final_logits_bias[:, :new_num_tokens]
+        else:
+            extra_bias = torch.zeros((1, new_num_tokens - old_num_tokens), device=self.final_logits_bias.device)
+            new_bias = torch.cat([self.final_logits_bias, extra_bias], dim=1)
+        self.register_buffer("final_logits_bias", new_bias)
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    @add_start_docstrings_to_model_forward(DELTALM_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=Seq2SeqLMOutput, config_class=_CONFIG_FOR_DOC)
+    @add_end_docstrings(DELTALM_GENERATION_EXAMPLE)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        decoder_input_ids: Optional[torch.LongTensor] = None,
+        decoder_attention_mask: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        decoder_head_mask: Optional[torch.Tensor] = None,
+        cross_attn_head_mask: Optional[torch.Tensor] = None,
+        encoder_outputs: Optional[List[torch.FloatTensor]] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        decoder_inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = 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
+
+        logger.debug("Comming to Generation!")
+
+        if labels is not None:
+            logger.debug("Debug: *************** Before label ***************** ")
+            logger.debug("Debug: %s", labels.size())
+            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
+                )
+
+            logger.debug("Debug: ************ After labels ************")
+            logger.debug("Debug: %s", labels.size())
+
+        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,
+            head_mask=head_mask,
+            decoder_head_mask=decoder_head_mask,
+            cross_attn_head_mask=cross_attn_head_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]) + self.final_logits_bias
+        # print(self.lm_head)
+        logger.debug("Debug: logit_size: %s", lm_logits.size())
+
+        # logger.debug("Debug: change logit size: ", lm_logits.view(-1, self.config.vocab_size).size())
+        # logger.debug("Debug: change label size: ", labels.view(-1).size())
+        masked_lm_loss = None
+
+        if labels is not None:
+            # logger.debug("Debug: model label_size: %s", labels.size())
+            # loss_fct = CrossEntropyLoss()
+            # masked_lm_loss = loss_fct(lm_logits.view(-1, self.config.vocab_size), labels.view(-1))
+            loss_fct = CrossEntropyLoss()
+            masked_lm_loss = loss_fct(lm_logits.view(-1, self.config.vocab_size), labels.view(-1))
+            # label_smoothing = self.config.label_smoothing
+            # # logger.debug("Debug: label.size: ", )
+            # if label_smoothing == 0:
+            #     # compute label smoothed loss
+            #     loss_fct = CrossEntropyLoss()
+            #     masked_lm_loss = loss_fct(lm_logits.view(-1, self.config.vocab_size), labels.view(-1))
+            # else:
+            #     m = torch.nn.LogSoftmax(dim=-1)
+            #     lprobs = m(lm_logits.float())
+            #     # lprobs = m(lm_logits)
+            #     # # torch.set_printoptions(linewidth=200)
+            #     loss_fn = label_smoothed_nll_loss
+            #     masked_lm_loss, _ = loss_fn(lprobs.view(-1, lprobs.size(-1)), labels.view(-1), label_smoothing, self.config.pad_token_id)
+
+        if not return_dict:
+            logger.debug("Debug: 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=None,
+        attention_mask=None,
+        head_mask=None,
+        decoder_head_mask=None,
+        cross_attn_head_mask=None,
+        use_cache=None,
+        encoder_outputs=None,
+        **kwargs
+    ):
+        # cut decoder_input_ids if past is used
+        if past 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,
+            "decoder_input_ids": decoder_input_ids,
+            "attention_mask": attention_mask,
+            "head_mask": head_mask,
+            "decoder_head_mask": decoder_head_mask,
+            "cross_attn_head_mask": cross_attn_head_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, beam_idx):
+        reordered_past = ()
+        for layer_past in past:
+            # 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
+
+
+class DeltalmDecoderWrapper(DeltalmPretrainedModel):
+    """
+    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 = DeltalmDecoder(config)
+
+    def forward(self, *args, **kwargs):
+        return self.decoder(*args, **kwargs)
+
+
+class DeltalmForCausalLM(DeltalmPretrainedModel):
+    def __init__(self, config):
+        config = copy.deepcopy(config)
+        config.is_decoder = True
+        config.is_encoder_decoder = False
+        super().__init__(config)
+        self.model = DeltalmDecoderWrapper(config)
+
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.decoder.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.decoder.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model.decoder = decoder
+
+    def get_decoder(self):
+        return self.model.decoder
+
+    @replace_return_docstrings(output_type=CausalLMOutputWithCrossAttentions, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.FloatTensor] = None,
+        encoder_attention_mask: Optional[torch.FloatTensor] = 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.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, CausalLMOutputWithCrossAttentions]:
+        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 [`DeltalmTokenizer`]. 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, sequence_length, hidden_size)`, *optional*):
+                Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention
+                if the model is configured as a decoder.
+            encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used
+                in the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`:
+            head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_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**.
+            cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_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**.
+            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)`. The two additional
+                tensors are only required when the model is used as a decoder in a Sequence to Sequence model.
+                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)`.
+            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]`.
+            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`).
+                - 1 for tokens that are **not masked**,
+                - 0 for tokens that are **masked**.
+            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.
+        Returns:
+        Example:
+        ```python
+        >>> from transformers import DeltalmTokenizer, DeltalmForCausalLM
+        >>> tokenizer = DeltalmTokenizer.from_pretrained("facebook/deltalm-base")
+        >>> model = DeltalmForCausalLM.from_pretrained("facebook/deltalm-base", add_cross_attention=False)
+        >>> assert model.config.is_decoder, f"{model.__class__} has to be configured as a decoder."
+        >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
+        >>> outputs = model(**inputs)
+        >>> logits = outputs.logits
+        >>> expected_shape = [1, inputs.input_ids.shape[-1], model.config.vocab_size]
+        >>> list(logits.shape) == expected_shape
+        True
+        ```"""
+
+        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,
+            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,
+        )
+
+        logits = self.lm_head(outputs[0])
+
+        loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(logits.view(-1, self.config.vocab_size), 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,
+        )
+
+    def prepare_inputs_for_generation(self, input_ids, past=None, attention_mask=None, use_cache=None, **kwargs):
+        # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
+        if attention_mask is None:
+            attention_mask = input_ids.new_ones(input_ids.shape)
+
+        if past:
+            input_ids = input_ids[:, -1:]
+        # first step, decoder_cached_states are empty
+        return {
+            "input_ids": input_ids,  # encoder_outputs is defined. input_ids not needed
+            "attention_mask": attention_mask,
+            "past_key_values": past,
+            "use_cache": use_cache,
+        }
+
+    @staticmethod
+    def _reorder_cache(past, beam_idx):
+        reordered_past = ()
+        for layer_past in past:
+            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
+        return reordered_past