Upload 2 files

configuration_codet5p.py

@@ -0,0 +1,113 @@
+# coding=utf-8
+# Copyright 2023 Salesforce authors, The EleutherAI, and HuggingFace Teams. All rights reserved.
+
+""" CodeT5+ model configuration"""
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+import copy
+
+logger = logging.get_logger(__name__)
+
+
+# Adapted from transformers.models.codegen.configuration_codegen.CodeGenConfig
+class CodeT5pModuleConfig(PretrainedConfig):
+    model_type = "codet5p_module"
+    attribute_map = {
+        "max_position_embeddings": "n_positions",
+        "hidden_size": "n_embd",
+        "num_attention_heads": "n_head",
+        "num_hidden_layers": "n_layer",
+    }
+
+    def __init__(
+            self,
+            vocab_size=50400,
+            n_positions=2048,
+            n_ctx=2048,
+            n_embd=4096,
+            n_layer=28,
+            n_head=16,
+            rotary_dim=64,
+            n_inner=None,
+            activation_function="gelu_new",
+            resid_pdrop=0.0,
+            embd_pdrop=0.0,
+            attn_pdrop=0.0,
+            layer_norm_epsilon=1e-5,
+            initializer_range=0.02,
+            scale_attn_weights=True,
+            use_cache=True,
+            bos_token_id=50256,
+            eos_token_id=50256,
+            tie_word_embeddings=False,
+            **kwargs
+    ):
+        self.vocab_size = vocab_size
+        self.n_ctx = n_ctx
+        self.n_positions = n_positions
+        self.n_embd = n_embd
+        self.n_layer = n_layer
+        self.n_head = n_head
+        self.n_inner = n_inner
+        self.rotary_dim = rotary_dim
+        self.activation_function = activation_function
+        self.resid_pdrop = resid_pdrop
+        self.embd_pdrop = embd_pdrop
+        self.attn_pdrop = attn_pdrop
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+        self.scale_attn_weights = scale_attn_weights
+        self.use_cache = use_cache
+
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+
+        super().__init__(
+            bos_token_id=bos_token_id, eos_token_id=eos_token_id, tie_word_embeddings=tie_word_embeddings, **kwargs
+        )
+
+
+# Adapted from transformers.models.encoder_decoder.configuration_encoder_decoder.EncoderDecoderConfig
+class CodeT5pConfig(PretrainedConfig):
+    model_type = "codet5p"
+    is_composition = True
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        assert (
+                "encoder" in kwargs and "decoder" in kwargs
+        ), "Config has to be initialized with encoder and decoder config"
+        encoder_config = kwargs.pop("encoder")
+        decoder_config = kwargs.pop("decoder")
+        encoder_model_type = encoder_config.pop("model_type")
+        decoder_model_type = decoder_config.pop("model_type")
+
+        if encoder_model_type != decoder_model_type:
+            logger.warning("Encoder and decoder model types are different")
+
+        self.encoder = CodeT5pModuleConfig(**encoder_config)
+        self.decoder = CodeT5pModuleConfig(**decoder_config)
+        self.is_encoder_decoder = True
+
+    @classmethod
+    def from_encoder_decoder_configs(
+            cls, encoder_config: PretrainedConfig, decoder_config: PretrainedConfig, **kwargs
+    ) -> PretrainedConfig:
+        logger.info("Set `config.is_decoder=True` and `config.add_cross_attention=True` for decoder_config")
+        decoder_config.is_decoder = True
+        decoder_config.add_cross_attention = True
+
+        return cls(encoder=encoder_config.to_dict(), decoder=decoder_config.to_dict(), **kwargs)
+
+    def to_dict(self):
+        """
+        Serializes this instance to a Python dictionary. Override the default *to_dict()* from *PretrainedConfig*.
+
+        Returns:
+            `Dict[str, any]`: Dictionary of all the attributes that make up this configuration instance,
+        """
+        output = copy.deepcopy(self.__dict__)
+        output["encoder"] = self.encoder.to_dict()
+        output["decoder"] = self.decoder.to_dict()
+        output["model_type"] = self.__class__.model_type
+        return output

modeling_codet5p.py

@@ -0,0 +1,982 @@
+# coding=utf-8
+# Copyright 2023 Salesforce authors, The EleutherAI, and HuggingFace Teams. All rights reserved.
+""" PyTorch CodeT5+ 2B 6B 16B models.
+The implementation is mainly based on transformers.models.codegen.modeling_codegen by adding cross-attention
+and transformers.models.encoder_decoder.modeling_encoder_decoder.EncoderDecoderModel.
+"""
+from typing import Optional, Tuple, Union
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import BaseModelOutput, Seq2SeqLMOutput, \
+    BaseModelOutputWithPast, CausalLMOutputWithPast, \
+    BaseModelOutputWithPastAndCrossAttentions, CausalLMOutputWithCrossAttentions
+from transformers.modeling_utils import PreTrainedModel
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import add_code_sample_docstrings, add_start_docstrings, logging
+from configuration_codet5p import CodeT5pConfig, CodeT5pModuleConfig
+
+logger = logging.get_logger(__name__)
+
+CODET5P_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "Salesforce/codet5p-220m",
+    "Salesforce/codet5p-770m",
+    "Salesforce/codet5p-220m-py",
+    "Salesforce/codet5p-770m-py",
+    "Salesforce/codet5p-2b",
+    "Salesforce/codet5p-6b",
+    "Salesforce/codet5p-16b",
+    "Salesforce/instructcodet5p-16b",
+    # See all CodeT5+ models at https://huggingface.co/models?filter=codet5p
+]
+
+
+# Copied from transformers.models.gptj.modeling_gptj.fixed_pos_embedding
+def fixed_pos_embedding(x, seq_dim=1, seq_len=None):
+    dim = x.shape[-1]
+    if seq_len is None:
+        seq_len = x.shape[seq_dim]
+    inv_freq = 1.0 / (10000 ** (torch.arange(0, dim, 2) / dim))
+    sinusoid_inp = (
+        torch.einsum("i , j -> i j", torch.arange(seq_len, dtype=torch.float), inv_freq).to(x.device).float()
+    )
+    return torch.sin(sinusoid_inp), torch.cos(sinusoid_inp)
+
+
+# Copied from transformers.models.gptj.modeling_gptj.rotate_every_two
+def rotate_every_two(x):
+    x1 = x[:, :, :, ::2]
+    x2 = x[:, :, :, 1::2]
+    x = torch.stack((-x2, x1), dim=-1)
+    return x.flatten(-2)  # in einsum notation: rearrange(x, '... d j -> ... (d j)')
+
+
+# Copied from transformers.models.gptj.modeling_gptj.duplicate_interleave
+def duplicate_interleave(m):
+    """
+    A simple version of `torch.repeat_interleave` for duplicating a matrix while interleaving the copy.
+    """
+    dim0 = m.shape[0]
+    m = m.view(-1, 1)  # flatten the matrix
+    m = m.repeat(1, 2)  # repeat all elements into the 2nd dimension
+    m = m.view(dim0, -1)  # reshape into a matrix, interleaving the copy
+    return m
+
+
+# Copied from transformers.models.gptj.modeling_gptj.apply_rotary_pos_emb
+def apply_rotary_pos_emb(x, sincos, offset=0):
+    sin, cos = (duplicate_interleave(t)[None, offset: x.shape[1] + offset, None, :] for t in sincos)
+    # einsum notation for lambda t: repeat(t[offset:x.shape[1]+offset,:], "n d -> () n () (d j)", j=2)
+    return (x * cos) + (rotate_every_two(x) * sin)
+
+
+# Adapted from transformers.models.codegen.modeling_codegen.CodeGenAttention
+class CodeT5pAttention(nn.Module):
+    def __init__(self, config, is_cross_attention=False, is_decoder=True):
+        super().__init__()
+
+        max_positions = config.max_position_embeddings
+        self.register_buffer(
+            "causal_mask",
+            torch.tril(torch.ones((max_positions, max_positions), dtype=torch.uint8)).view(
+                1, 1, max_positions, max_positions
+            ),
+        )
+
+        self.attn_dropout = nn.Dropout(config.attn_pdrop)
+        self.resid_dropout = nn.Dropout(config.resid_pdrop)
+
+        self.embed_dim = config.hidden_size
+        self.num_attention_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_attention_heads
+        if self.head_dim * self.num_attention_heads != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_attention_heads (got `embed_dim`: {self.embed_dim} and"
+                f" `num_attention_heads`: {self.num_attention_heads})."
+            )
+
+        self.scale_attn = torch.sqrt(torch.tensor(self.head_dim, dtype=torch.float32)).to(torch.get_default_dtype())
+        self.is_decoder = is_decoder
+        self.is_cross_attention = is_cross_attention
+        if self.is_cross_attention:
+            self.qkv_proj = nn.Linear(self.embed_dim, self.embed_dim * 2, bias=False)
+            self.q_attn = nn.Linear(self.embed_dim, self.embed_dim, bias=False)
+        else:
+            self.qkv_proj = nn.Linear(self.embed_dim, self.embed_dim * 3, bias=False)
+
+        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False)
+        self.rotary_dim = None
+        if config.rotary_dim is not None:
+            self.rotary_dim = config.rotary_dim
+
+    def _split_heads(self, x, n_head, dim_head, mp_num):
+        reshaped = x.reshape(x.shape[:-1] + (n_head // mp_num, dim_head))
+        reshaped = reshaped.reshape(x.shape[:-2] + (-1,) + reshaped.shape[-1:])
+        return reshaped
+
+    def _merge_heads(self, tensor, num_attention_heads, attn_head_size):
+        """
+        Merges attn_head_size dim and num_attn_heads dim into n_ctx
+        """
+        if len(tensor.shape) == 5:
+            tensor = tensor.permute(0, 1, 3, 2, 4).contiguous()
+        elif len(tensor.shape) == 4:
+            tensor = tensor.permute(0, 2, 1, 3).contiguous()
+        else:
+            raise ValueError(f"Input tensor rank should be one of [4, 5], but is: {len(tensor.shape)}")
+        new_shape = tensor.size()[:-2] + (num_attention_heads * attn_head_size,)
+        return tensor.view(new_shape)
+
+    def _attn(
+            self,
+            query,
+            key,
+            value,
+            attention_mask=None,
+            head_mask=None,
+    ):
+        # Keep the attention weights computation in fp32 to avoid overflow issues
+        query = query.to(torch.float32)
+        key = key.to(torch.float32)
+
+        attn_weights = torch.matmul(query, key.transpose(-1, -2))
+        attn_weights = attn_weights / self.scale_attn
+
+        if not self.is_cross_attention and self.is_decoder:
+            # compute causal mask from causal mask buffer
+            query_length, key_length = query.size(-2), key.size(-2)
+            causal_mask = self.causal_mask[:, :, key_length - query_length: key_length, :key_length]
+            mask_value = torch.finfo(attn_weights.dtype).min
+            # Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`.
+            # Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device`
+            mask_value = torch.tensor(mask_value, dtype=attn_weights.dtype).to(attn_weights.device)
+            attn_weights = torch.where(causal_mask.bool(), attn_weights, mask_value)
+
+        if attention_mask is not None:
+            # Apply the attention mask
+            attn_weights = attn_weights + attention_mask
+
+        attn_weights = nn.Softmax(dim=-1)(attn_weights)
+        attn_weights = attn_weights.to(value.dtype)
+        attn_weights = self.attn_dropout(attn_weights)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attn_weights = attn_weights * head_mask
+
+        attn_output = torch.matmul(attn_weights, value)
+
+        return attn_output, attn_weights
+
+    def forward(
+            self,
+            hidden_states: Optional[torch.FloatTensor],
+            attention_mask: Optional[torch.FloatTensor] = None,
+            layer_past: Optional[Tuple[torch.Tensor]] = None,
+            head_mask: Optional[torch.FloatTensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
+            encoder_attention_mask: Optional[torch.FloatTensor] = None,
+            use_cache: Optional[bool] = False,
+            output_attentions: Optional[bool] = False,
+    ) -> Union[
+        Tuple[torch.Tensor, Tuple[torch.Tensor]],
+        Optional[Tuple[torch.Tensor, Tuple[torch.Tensor], Tuple[torch.Tensor, ...]]],
+    ]:
+
+        if encoder_hidden_states is not None:
+            if not hasattr(self, "q_attn"):
+                raise ValueError(
+                    "If class is used as cross attention, the weights `q_attn` have to be defined. "
+                    "Please make sure to instantiate class with `GPT2Attention(..., is_cross_attention=True)`."
+                )
+
+            mp_num = 4
+            local_dim = self.head_dim * self.num_attention_heads // mp_num
+            q = self.q_attn(hidden_states)
+            q_split = q.reshape(q.shape[:-1] + (mp_num, -1))
+            query = torch.split(q_split, local_dim, dim=-1)[0]
+
+            qkv = self.qkv_proj(encoder_hidden_states)
+            qkv_split = qkv.reshape(qkv.shape[:-1] + (mp_num, -1))
+            value, key = torch.split(qkv_split, local_dim, dim=-1)
+
+            attention_mask = encoder_attention_mask
+        else:
+            qkv = self.qkv_proj(hidden_states)
+            mp_num = 4
+            qkv_split = qkv.reshape(qkv.shape[:-1] + (mp_num, -1))
+
+            local_dim = self.head_dim * self.num_attention_heads // mp_num
+            query, value, key = torch.split(qkv_split, local_dim, dim=-1)
+
+        query = self._split_heads(query, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+        key = self._split_heads(key, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+
+        value = self._split_heads(value, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+        value = value.permute(0, 2, 1, 3)
+
+        seq_len = key.shape[1]
+        offset = 0
+
+        if layer_past is not None:
+            offset = layer_past[0].shape[-2]
+            seq_len += offset
+
+        if self.rotary_dim is not None:
+            k_rot = key[:, :, :, : self.rotary_dim]
+            k_pass = key[:, :, :, self.rotary_dim:]
+
+            q_rot = query[:, :, :, : self.rotary_dim]
+            q_pass = query[:, :, :, self.rotary_dim:]
+
+            sincos = fixed_pos_embedding(k_rot, 1, seq_len=seq_len)
+            k_rot = apply_rotary_pos_emb(k_rot, sincos, offset=offset)
+            seq_len_q = query.shape[1]
+            sincos_q = fixed_pos_embedding(q_rot, 1, seq_len=seq_len_q)
+            q_rot = apply_rotary_pos_emb(q_rot, sincos_q, offset=offset)
+
+            key = torch.cat([k_rot, k_pass], dim=-1)
+            query = torch.cat([q_rot, q_pass], dim=-1)
+        else:
+            sincos = fixed_pos_embedding(key, 1, seq_len=seq_len)
+            key = apply_rotary_pos_emb(key, sincos, offset=offset)
+            query = apply_rotary_pos_emb(query, sincos, offset=offset)
+
+        key = key.permute(0, 2, 1, 3)
+        query = query.permute(0, 2, 1, 3)
+
+        if layer_past is not None:
+            past_key = layer_past[0]
+            past_value = layer_past[1]
+            key = torch.cat((past_key, key), dim=-2)
+            value = torch.cat((past_value, value), dim=-2)
+
+        if use_cache is True:
+            present = (key, value)
+        else:
+            present = None
+
+        # compute self-attention: V x Softmax(QK^T)
+        attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask)
+
+        attn_output = self._merge_heads(attn_output, self.num_attention_heads, self.head_dim)
+        attn_output = self.out_proj(attn_output)
+        attn_output = self.resid_dropout(attn_output)
+
+        outputs = (attn_output, present)
+        if output_attentions:
+            outputs += (attn_weights,)
+
+        return outputs  # a, present, (attentions)
+
+
+# Adapted from transformers.models.codegen.modeling_codegen.CodeGenMLP
+class CodeT5pMLP(nn.Module):
+    def __init__(self, intermediate_size, config):  # in MLP: intermediate_size= 4 * embed_dim
+        super().__init__()
+        embed_dim = config.n_embd
+
+        self.fc_in = nn.Linear(embed_dim, intermediate_size)
+        self.fc_out = nn.Linear(intermediate_size, embed_dim)
+
+        self.act = ACT2FN[config.activation_function]
+        self.dropout = nn.Dropout(config.resid_pdrop)
+
+    def forward(self, hidden_states: Optional[torch.FloatTensor]) -> torch.FloatTensor:
+        hidden_states = self.fc_in(hidden_states)
+        hidden_states = self.act(hidden_states)
+        hidden_states = self.fc_out(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        return hidden_states
+
+
+# Adapted from transformers.models.codegen.modeling_codegen.CodeGenBlock
+class CodeT5pBlock(nn.Module):
+    def __init__(self, config, layer_idx=None):
+        super().__init__()
+        inner_dim = config.n_inner if config.n_inner is not None else 4 * config.n_embd
+        self.ln_1 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+
+        if config.is_decoder is False:
+            self.attn = CodeT5pAttention(config, is_cross_attention=False, is_decoder=False)
+        else:
+            self.attn = CodeT5pAttention(config)
+        self.mlp = CodeT5pMLP(inner_dim, config)
+
+        # Adding 1 cross-attention layer at the final decoder layer
+        self.add_cross_attention_by_layer = True \
+            if config.add_cross_attention and layer_idx == config.n_layer - 1 else False
+
+        if config.add_cross_attention and self.add_cross_attention_by_layer:
+            self.crossattention = CodeT5pAttention(config, is_cross_attention=True)
+
+    def forward(
+            self,
+            hidden_states: Optional[torch.FloatTensor],
+            layer_past: Optional[Tuple[torch.Tensor]] = None,
+            attention_mask: Optional[torch.FloatTensor] = None,
+            head_mask: Optional[torch.FloatTensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
+            encoder_attention_mask: Optional[torch.FloatTensor] = None,
+            use_cache: Optional[bool] = False,
+            output_attentions: Optional[bool] = False,
+    ) -> Union[Tuple[torch.Tensor], Optional[Tuple[torch.Tensor, Tuple[torch.FloatTensor, ...]]]]:
+        residual = hidden_states
+        hidden_states = self.ln_1(hidden_states)
+        attn_outputs = self.attn(
+            hidden_states,
+            layer_past=layer_past,
+            attention_mask=attention_mask,
+            head_mask=head_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+        )
+        attn_output = attn_outputs[0]  # output_attn: a, present, (attentions)
+        outputs = attn_outputs[1:]
+        feed_forward_hidden_states = self.mlp(hidden_states)
+
+        if encoder_hidden_states is not None and self.add_cross_attention_by_layer:
+            # add one self-attention block for cross-attention
+            if not hasattr(self, "crossattention"):
+                raise ValueError(
+                    f"If `encoder_hidden_states` are passed, {self} has to be instantiated with "
+                    "cross-attention layers by setting `config.add_cross_attention=True`"
+                )
+            # residual = hidden_states
+            # hidden_states = self.ln_cross_attn(residual)
+            cross_attn_outputs = self.crossattention(
+                hidden_states,
+                attention_mask=attention_mask,
+                head_mask=head_mask,
+                encoder_hidden_states=encoder_hidden_states,
+                encoder_attention_mask=encoder_attention_mask,
+                output_attentions=output_attentions,
+            )
+            xattn_output = cross_attn_outputs[0]
+            attn_output = attn_output + xattn_output
+            outputs = outputs + cross_attn_outputs[2:]  # add cross attentions if we output attention weights
+
+        hidden_states = attn_output + feed_forward_hidden_states + residual
+
+        if use_cache:
+            outputs = (hidden_states,) + outputs
+        else:
+            outputs = (hidden_states,) + outputs[1:]
+
+        return outputs  # hidden_states, present, (attentions)
+
+
+# Adapted from transformers.models.codegen.modeling_codegen.CodeGenPreTrainedModel
+class CodeT5pPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+    config_class = CodeT5pModuleConfig
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["CodeT5pBlock"]
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module):
+        """Initialize the weights."""
+        if isinstance(module, (nn.Linear,)):
+            # Slightly different from Mesh Transformer JAX which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, CodeT5pModel):
+            module.gradient_checkpointing = value
+
+
+# Adapted from transformers.models.codegen.modeling_codegen.CodeGenModel
+class CodeT5pModel(CodeT5pPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.embed_dim = config.n_embd
+        self.vocab_size = config.vocab_size
+        self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
+        self.drop = nn.Dropout(config.embd_pdrop)
+        self.h = nn.ModuleList([CodeT5pBlock(config, idx) for idx in range(config.n_layer)])
+        self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
+        self.rotary_dim = min(config.rotary_dim, config.n_ctx // config.num_attention_heads)
+
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.wte
+
+    def set_input_embeddings(self, new_embeddings):
+        self.wte = new_embeddings
+
+    def forward(
+            self,
+            input_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
+            attention_mask: Optional[torch.FloatTensor] = None,
+            token_type_ids: Optional[torch.LongTensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            head_mask: Optional[torch.FloatTensor] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
+            encoder_attention_mask: 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]:
+        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 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])
+            batch_size = input_ids.shape[0]
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size = inputs_embeds.shape[0]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
+        if token_type_ids is not None:
+            token_type_ids = token_type_ids.view(-1, input_shape[-1])
+
+        if position_ids is not None:
+            position_ids = position_ids.view(-1, input_shape[-1])
+
+        if past_key_values is None:
+            past_length = 0
+            past_key_values = tuple([None] * len(self.h))
+        else:
+            past_length = past_key_values[0][0].size(-2)
+
+        if position_ids is None:
+            position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
+            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+
+        # Attention mask.
+        if attention_mask is not None:
+            if batch_size <= 0:
+                raise ValueError("batch_size has to be defined and > 0")
+            attention_mask = attention_mask.view(batch_size, -1)
+            # We create a 3D attention mask from a 2D tensor mask.
+            # Sizes are [batch_size, 1, 1, to_seq_length]
+            # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
+            # this attention mask is more simple than the triangular masking of causal attention
+            # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
+            attention_mask = attention_mask[:, None, None, :]
+
+            # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
+            # masked positions, this operation will create a tensor which is 0.0 for
+            # positions we want to attend and the dtype's smallest value for masked positions.
+            # Since we are adding it to the raw scores before the softmax, this is
+            # effectively the same as removing these entirely.
+            attention_mask = attention_mask.to(dtype=self.dtype)  # fp16 compatibility
+            attention_mask = (1.0 - attention_mask) * torch.finfo(self.dtype).min
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if self.config.add_cross_attention and encoder_hidden_states is not None:
+            encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            if encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+            encoder_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x num_attention_heads x N x N
+        # head_mask has shape n_layer x batch x num_attention_heads x N x N
+        head_mask = self.get_head_mask(head_mask, self.config.n_layer)
+
+        if inputs_embeds is None:
+            inputs_embeds = self.wte(input_ids)
+
+        hidden_states = inputs_embeds
+
+        if token_type_ids is not None:
+            token_type_embeds = self.wte(token_type_ids)
+            hidden_states = hidden_states + token_type_embeds
+
+        hidden_states = self.drop(hidden_states)
+
+        output_shape = input_shape + (hidden_states.size(-1),)
+
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+        all_hidden_states = () if output_hidden_states else None
+        for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                if use_cache:
+                    logger.warning(
+                        "`use_cache=True` is incompatible with `config.gradient_checkpointing=True`. Setting "
+                        "`use_cache=False`..."
+                    )
+                    use_cache = False
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, use_cache, output_attentions)
+
+                    return custom_forward
+
+                outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(block),
+                    hidden_states,
+                    None,
+                    attention_mask,
+                    head_mask[i],
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                )
+            else:
+                outputs = block(
+                    hidden_states,
+                    layer_past=layer_past,
+                    attention_mask=attention_mask,
+                    head_mask=head_mask[i],
+                    encoder_hidden_states=encoder_hidden_states,
+                    encoder_attention_mask=encoder_attention_mask,
+                    use_cache=use_cache,
+                    output_attentions=output_attentions,
+                )
+
+            hidden_states = outputs[0]
+            if use_cache is True:
+                presents = presents + (outputs[1],)
+
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (outputs[2 if use_cache else 1],)
+                if self.config.add_cross_attention and self.add_cross_attention_by_layer:
+                    all_cross_attentions = all_cross_attentions + (outputs[3 if use_cache else 2],)
+
+        hidden_states = self.ln_f(hidden_states)
+
+        hidden_states = hidden_states.view(output_shape)
+        # Add last hidden state
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v for v in [hidden_states, presents, all_hidden_states, all_self_attentions, all_cross_attentions] if
+                v is not None)
+
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+# Adapted from transformers.models.codegen.modeling_codegen.CodeGenForCausalLM
+class CodeT5pForCausalLM(CodeT5pPreTrainedModel):
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.causal_mask"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.transformer = CodeT5pModel(config)
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, **kwargs):
+        token_type_ids = kwargs.get("token_type_ids", None)
+        # only last token for inputs_ids if past is defined in kwargs
+        if past_key_values:
+            input_ids = input_ids[:, -1].unsqueeze(-1)
+            if token_type_ids is not None:
+                token_type_ids = token_type_ids[:, -1].unsqueeze(-1)
+
+        attention_mask = kwargs.get("attention_mask", None)
+        position_ids = kwargs.get("position_ids", None)
+
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+        else:
+            position_ids = None
+        return {
+            "input_ids": input_ids,
+            "past_key_values": past_key_values,
+            "use_cache": kwargs.get("use_cache"),
+            "position_ids": position_ids,
+            "attention_mask": attention_mask,
+            "token_type_ids": token_type_ids,
+        }
+
+    def forward(
+            self,
+            input_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
+            attention_mask: Optional[torch.FloatTensor] = None,
+            token_type_ids: Optional[torch.LongTensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            head_mask: Optional[torch.FloatTensor] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
+            encoder_attention_mask: 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"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
+            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
+            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.transformer(
+            input_ids,
+            past_key_values=past_key_values,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = transformer_outputs[0]
+
+        # make sure sampling in fp16 works correctly and
+        # compute loss in fp32 to match with mesh-tf version
+        # https://github.com/EleutherAI/gpt-neo/blob/89ce74164da2fb16179106f54e2269b5da8db333/models/gpt2/gpt2.py#L179
+        lm_logits = self.lm_head(hidden_states).to(torch.float32)
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = lm_logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+            loss = loss.to(hidden_states.dtype)
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithCrossAttentions(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+            cross_attentions=transformer_outputs.cross_attentions,
+        )
+
+    @staticmethod
+    def _reorder_cache(
+            past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor
+    ) -> Tuple[Tuple[torch.Tensor]]:
+        """
+        This function is used to re-order the `past_key_values` cache if [`~PretrainedModel.beam_search`] or
+        [`~PretrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
+        beam_idx at every generation step.
+        """
+        return tuple(
+            tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past)
+            for layer_past in past_key_values
+        )
+
+
+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()
+    if decoder_start_token_id is None:
+        raise ValueError("Make sure to set the decoder_start_token_id attribute of the model's configuration.")
+    shifted_input_ids[:, 0] = decoder_start_token_id
+
+    if pad_token_id is None:
+        raise ValueError("Make sure to set the pad_token_id attribute of the model's configuration.")
+    # 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
+
+
+# Adapted from transformers.models.encoder_decoder.modeling_encoder_decoder.EncoderDecoderModel
+class CodeT5pEncoderDecoderModel(PreTrainedModel):
+    config_class = CodeT5pConfig
+
+    def __init__(
+            self,
+            config: Optional[PretrainedConfig] = None,
+            encoder: Optional[PreTrainedModel] = None,
+            decoder: Optional[PreTrainedModel] = None,
+    ):
+        if config is None and (encoder is None or decoder is None):
+            raise ValueError("Either a configuration or an encoder and a decoder has to be provided.")
+        if config is None:
+            config = CodeT5pConfig.from_encoder_decoder_configs(encoder.config, decoder.config)
+        else:
+            if not isinstance(config, self.config_class):
+                raise ValueError(f"Config: {config} has to be of type {self.config_class}")
+
+        if config.decoder.cross_attention_hidden_size is not None:
+            if config.decoder.cross_attention_hidden_size != config.encoder.hidden_size:
+                raise ValueError(
+                    "If `cross_attention_hidden_size` is specified in the decoder's configuration, it has to be equal"
+                    f" to the encoder's `hidden_size`. Got {config.decoder.cross_attention_hidden_size} for"
+                    f" `config.decoder.cross_attention_hidden_size` and {config.encoder.hidden_size} for"
+                    " `config.encoder.hidden_size`."
+                )
+
+        # initialize with config
+        super().__init__(config)
+
+        if encoder is None:
+            encoder = CodeT5pModel(config.encoder)
+
+        if decoder is None:
+            decoder = CodeT5pForCausalLM(config.decoder)
+
+        self.encoder = encoder
+        self.decoder = decoder
+
+        if self.encoder.config.to_dict() != self.config.encoder.to_dict():
+            logger.warning(
+                f"Config of the encoder: {self.encoder.__class__} is overwritten by shared encoder config:"
+                f" {self.config.encoder}"
+            )
+        if self.decoder.config.to_dict() != self.config.decoder.to_dict():
+            logger.warning(
+                f"Config of the decoder: {self.decoder.__class__} is overwritten by shared decoder config:"
+                f" {self.config.decoder}"
+            )
+
+        # make sure that the individual model's config refers to the shared config
+        # so that the updates to the config will be synced
+        self.encoder.config = self.config.encoder
+        self.decoder.config = self.config.decoder
+
+        # encoder outputs might need to be projected to different dimension for decoder
+        if (
+                self.encoder.config.hidden_size != self.decoder.config.hidden_size
+                and self.decoder.config.cross_attention_hidden_size is None
+        ):
+            self.enc_to_dec_proj = nn.Linear(self.encoder.config.hidden_size, self.decoder.config.hidden_size)
+
+        if self.encoder.get_output_embeddings() is not None:
+            raise ValueError(
+                f"The encoder {self.encoder} should not have a LM Head. Please use a model without LM Head"
+            )
+        # tie encoder, decoder weights if config set accordingly
+        self.tie_weights()
+
+    def tie_weights(self):
+        # tie encoder & decoder if needed
+        if self.config.tie_encoder_decoder:
+            # tie encoder and decoder base model
+            decoder_base_model_prefix = self.decoder.base_model_prefix
+            self._tie_encoder_decoder_weights(
+                self.encoder, self.decoder._modules[decoder_base_model_prefix], self.decoder.base_model_prefix
+            )
+
+    def get_encoder(self):
+        return self.encoder
+
+    def get_decoder(self):
+        return self.decoder
+
+    def get_input_embeddings(self):
+        return self.encoder.get_input_embeddings()
+
+    def get_output_embeddings(self):
+        return self.decoder.get_output_embeddings()
+
+    def set_output_embeddings(self, new_embeddings):
+        return self.decoder.set_output_embeddings(new_embeddings)
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        # At the moment fast initialization is not supported for composite models
+        if kwargs.get("_fast_init", False):
+            logger.warning(
+                "Fast initialization is currently not supported for EncoderDecoderModel. "
+                "Falling back to slow initialization..."
+            )
+        kwargs["_fast_init"] = False
+        return super().from_pretrained(*args, **kwargs)
+
+    def forward(
+            self,
+            input_ids: Optional[torch.LongTensor] = None,
+            attention_mask: Optional[torch.FloatTensor] = None,
+            decoder_input_ids: Optional[torch.LongTensor] = None,
+            decoder_attention_mask: Optional[torch.BoolTensor] = None,
+            encoder_outputs: Optional[Tuple[torch.FloatTensor]] = None,
+            past_key_values: Tuple[Tuple[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,
+            **kwargs,
+    ) -> Union[Tuple, Seq2SeqLMOutput]:
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        kwargs_encoder = {argument: value for argument, value in kwargs.items() if not argument.startswith("decoder_")}
+
+        kwargs_decoder = {
+            argument[len("decoder_"):]: value for argument, value in kwargs.items() if argument.startswith("decoder_")
+        }
+
+        if encoder_outputs is None:
+            encoder_outputs = self.encoder(
+                input_ids=input_ids,
+                attention_mask=attention_mask,
+                inputs_embeds=inputs_embeds,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+                **kwargs_encoder,
+            )
+        elif isinstance(encoder_outputs, tuple):
+            encoder_outputs = BaseModelOutput(*encoder_outputs)
+
+        encoder_hidden_states = encoder_outputs[0]
+
+        # optionally project encoder_hidden_states
+        if (
+                self.encoder.config.hidden_size != self.decoder.config.hidden_size
+                and self.decoder.config.cross_attention_hidden_size is None
+        ):
+            encoder_hidden_states = self.enc_to_dec_proj(encoder_hidden_states)
+
+        if (labels is not None) and (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
+            )
+
+        # Decode
+        decoder_outputs = self.decoder(
+            input_ids=decoder_input_ids,
+            attention_mask=decoder_attention_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=attention_mask,
+            inputs_embeds=decoder_inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            use_cache=use_cache,
+            past_key_values=past_key_values,
+            return_dict=return_dict,
+            **kwargs_decoder,
+        )
+
+        # Compute loss independent from decoder (as some shift the logits inside them)
+        loss = None
+        if labels is not None:
+            # warnings.warn(DEPRECATION_WARNING, FutureWarning)
+            logits = decoder_outputs.logits if return_dict else decoder_outputs[0]
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(logits.reshape(-1, self.decoder.config.vocab_size), labels.view(-1))
+
+        if not return_dict:
+            if loss is not None:
+                return (loss,) + decoder_outputs + encoder_outputs
+            else:
+                return decoder_outputs + encoder_outputs
+
+        return Seq2SeqLMOutput(
+            loss=loss,
+            logits=decoder_outputs.logits,
+            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,
+        )
+
+    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)
+
+    def prepare_inputs_for_generation(
+            self, input_ids, past=None, attention_mask=None, use_cache=None, encoder_outputs=None, **kwargs
+    ):
+        decoder_inputs = self.decoder.prepare_inputs_for_generation(input_ids, past=past)
+        decoder_attention_mask = decoder_inputs["attention_mask"] if "attention_mask" in decoder_inputs else None
+        input_dict = {
+            "attention_mask": attention_mask,
+            "decoder_attention_mask": decoder_attention_mask,
+            "decoder_input_ids": decoder_inputs["input_ids"],
+            "encoder_outputs": encoder_outputs,
+            "past_key_values": decoder_inputs["past_key_values"],
+            "use_cache": use_cache,
+        }
+        return input_dict
+
+    def resize_token_embeddings(self, *args, **kwargs):
+        raise NotImplementedError(
+            "Resizing the embedding layers via the EncoderDecoderModel directly is not supported. Please use the"
+            " respective methods of the wrapped objects (model.encoder.resize_token_embeddings(...) or"
+            " model.decoder.resize_token_embeddings(...))"
+        )
+
+    def _reorder_cache(self, past, beam_idx):
+        # apply decoder cache reordering here
+        return self.decoder._reorder_cache(past, beam_idx)