Upload 3 files

initial commit: library code

jargon_configuration.py

@@ -0,0 +1,88 @@
+
+from transformers.models.roberta.modeling_roberta import RobertaConfig
+
+
+class JargonConfig(RobertaConfig):
+    model_type = "jargon"
+
+    def __init__(
+        self,
+        compress_layer= 1,
+        shared_layer_kv_compressed=1,
+        shared_kv_compressed=0,
+        max_positions=512,
+        max_position_embeddings=512,
+        compressed=4,
+        vocab_size=30522,
+        freeze_compress=0,
+        embed_dim=768,
+        num_heads=16,
+        dim_feedforward=4096,
+        dropout=0.1,
+        activation="relu",
+        layer_norm_eps=1e-05,
+        self_attention=True,
+        encoder_decoder_attention=False,
+        bias=True,
+        q_noise=0,
+        qn_block_size=8,
+        add_bias_kv=False,
+        add_zero_attn=False,
+        num_layers=12,
+        untie_weights_roberta=False,
+        layernorm_embedding=False,
+        encoder_normalize_before=False,
+        encoder_embed_dim=768,
+        encoder_attention_heads=12,
+        quant_noise_pq=0.0,
+        quant_noise_pq_block_size=8,
+        quant_noise_scalar=0,
+        encoder_ffn_embed_dim=4096,
+        add_pooling_layer=False,
+        intermediate_size=4096,
+        intermediate_act_fn="relu",
+        hidden_act="relu",
+        output_hidden_states=False,
+        position_embedding_type="learned",
+        **kwargs
+    ):
+        super().__init__(**kwargs)
+
+        self.add_pooling_layer = add_pooling_layer
+        self.compress_layer = compress_layer
+        self.shared_layer_kv_compressed = shared_layer_kv_compressed
+        self.shared_kv_compressed = shared_kv_compressed
+        self.max_positions = max_positions
+        self.max_position_embeddings = max_position_embeddings
+        self.compressed = compressed
+        self.freeze_compress = freeze_compress
+        self.embed_dim = embed_dim
+        self.num_heads = num_heads
+        self.dim_feedforward=dim_feedforward
+        self.dropout = dropout
+        self.activation= activation 
+        self.layer_norm_eps = layer_norm_eps
+        self.self_attention = self_attention
+        self.encoder_decoder_attention = encoder_decoder_attention
+        self.bias = bias
+        self.q_noise = q_noise
+        self.qn_block_size = qn_block_size
+        self.add_bias_kv = add_bias_kv
+        self.add_zero_attn = add_zero_attn
+        self.num_layers = num_layers
+        self.untie_weights_roberta = untie_weights_roberta
+        self.layernorm_embedding=layernorm_embedding
+        self.encoder_embed_dim = encoder_embed_dim
+        self.encoder_attention_heads=encoder_attention_heads
+        self.quant_noise_pq = quant_noise_pq
+        self.quant_noise_pq_block_size=quant_noise_pq_block_size
+        self.quant_noise_scalar=quant_noise_scalar
+        self.encoder_normalize_before=encoder_normalize_before
+        self.encoder_ffn_embed_dim = encoder_ffn_embed_dim
+        self.vocab_size = vocab_size
+        self.intermediate_size = intermediate_size
+        self.intermediate_act_fn = intermediate_act_fn
+        self.output_hidden_states = output_hidden_states
+        self.hidden_act = hidden_act
+        self.position_embedding_type = position_embedding_type
+        self.encoder_normalize_before = encoder_normalize_before

jargon_model.py

@@ -0,0 +1,522 @@
+import math
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.nn import LayerNorm
+from fairseq.models.roberta import (
+    RobertaModel as RobertModel, 
+    RobertaEncoder as RobertaEncoderFS
+)
+from transformers.models.roberta.modeling_roberta import (
+    RobertaEncoder,
+    RobertaConfig,
+    RobertaModel,
+    RobertaLMHead,
+    RobertaForMaskedLM,
+    RobertaEmbeddings,
+    RobertaForTokenClassification,
+    RobertaForSequenceClassification
+)
+from transformers.modeling_outputs import (
+    MaskedLMOutput,   
+    BaseModelOutputWithPastAndCrossAttentions,
+    BaseModelOutputWithPoolingAndCrossAttentions,
+)
+
+from .linformer import LinformerTransformerEncoderLayer
+from .jargon_configuration import JargonConfig
+
+
+class JargonForSequenceClassification(RobertaForSequenceClassification):
+
+    config_class = JargonConfig
+    
+    def __init__(self, config,  **kwargs):
+        base_model_prefix = "jargon"
+        
+        super().__init__(config, **kwargs)
+
+        self.roberta = JargonModel(config, add_pooling_layer=False)
+        self.sbo_head = self.build_sbo_head(config)
+
+    def build_sbo_head(self, config):
+        return SBOHead(
+            config,
+            embedding_weights=(
+                self.roberta.embeddings.word_embeddings.weight
+                if not config.untie_weights_roberta
+                else None
+            )
+        )
+
+
+class JargonForTokenClassification(RobertaForTokenClassification):
+
+    config_class = JargonConfig
+    
+    def __init__(self, config,  **kwargs):
+        base_model_prefix = "jargon"
+        
+        super().__init__(config, **kwargs)
+
+        self.roberta = JargonModel(config, add_pooling_layer=False)
+        self.sbo_head = self.build_sbo_head(config)
+
+    def build_sbo_head(self, config):
+        return SBOHead(
+            config,
+            embedding_weights=(
+                self.roberta.embeddings.word_embeddings.weight
+                if not config.untie_weights_roberta
+                else None
+            )
+        )
+        
+
+class JargonForMaskedLM(RobertaForMaskedLM):
+
+    config_class = JargonConfig
+    
+    def __init__(self, config,  **kwargs):
+        base_model_prefix = "jargon"
+        
+        super().__init__(config, **kwargs)
+
+        self.roberta = JargonModel(config, add_pooling_layer=False)
+        self.sbo_head = self.build_sbo_head(config)
+
+    def build_sbo_head(self, config):
+        return SBOHead(
+            config,
+            embedding_weights=(
+                self.roberta.embeddings.word_embeddings.weight
+                if not config.untie_weights_roberta
+                else None
+            )
+        )
+
+
+class JargonForMaskedLMFS(RobertaForMaskedLM):
+
+    def __init__(self, config, dictionary, **kwargs):
+        config_class = JargonConfig
+        base_model_prefix = "jargon"
+        
+        super().__init__(config, **kwargs)
+
+        self.roberta = FlaubertEncoder(config, dictionary)
+
+    def build_sbo_head(self, config):
+        return SBOHead(
+            config,
+            embedding_weights=(
+                self.roberta.embeddings.word_embeddings.weight
+                if not config.untie_weights_roberta
+                else None
+            )
+        )
+
+
+class JargonEmbeddings(RobertaEmbeddings):
+
+    def __init__(self, config, **kwargs):
+        config_class = JargonConfig
+        base_model_prefix = "jargon"
+        super().__init__(config, **kwargs)
+    
+    def forward(
+        self, input_ids=None, token_type_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
+    ):
+        if position_ids is None:
+            if input_ids is not None:
+                # Create the position ids from the input token ids. Any padded tokens remain padded.
+                position_ids = create_position_ids_from_input_ids(input_ids, self.padding_idx, past_key_values_length)
+            else:
+                position_ids = self.create_position_ids_from_inputs_embeds(inputs_embeds)
+
+        if input_ids is not None:
+            input_shape = input_ids.size()
+        else:
+            input_shape = inputs_embeds.size()[:-1]
+
+        seq_length = input_shape[1]
+
+        # Setting the token_type_ids to the registered buffer in constructor where it is all zeros, which usually occurs
+        # when its auto-generated, registered buffer helps users when tracing the model without passing token_type_ids, solves
+        # issue #5664
+        if token_type_ids is None:
+            if hasattr(self, "token_type_ids"):
+                buffered_token_type_ids = self.token_type_ids[:, :seq_length]
+                buffered_token_type_ids_expanded = buffered_token_type_ids.expand(input_shape[0], seq_length)
+                token_type_ids = buffered_token_type_ids_expanded
+            else:
+                token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=self.position_ids.device)
+
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+        token_type_embeddings = self.token_type_embeddings(token_type_ids)
+        
+        embeddings = inputs_embeds + token_type_embeddings
+        position_embeddings = self.position_embeddings(position_ids)
+
+        embeddings += position_embeddings
+        embeddings = self.dropout(embeddings)
+        return embeddings
+
+
+class JargonEncoder(RobertaEncoder):
+
+    def __init__(self, args):
+        compress_layer = None
+        if args.shared_layer_kv_compressed == 1 and compress_layer is None:
+            compress_layer = nn.Linear(
+                args.max_positions,
+                args.max_positions // args.compressed
+            )
+            # intialize parameters for compressed layer
+            nn.init.xavier_uniform_(compress_layer.weight, gain=1 / math.sqrt(2))
+            if args.freeze_compress == 1:
+                compress_layer.weight.requires_grad = False
+            compress_layer = compress_layer
+
+        super().__init__(args)
+ 
+        self.layer = nn.ModuleList([LinformerTransformerEncoderLayer(args, compress_layer) for _ in range(args.num_layers)])
+        self.compress_layer = compress_layer
+
+        if args.encoder_normalize_before:
+            self.layer_norm = LayerNorm(args.embed_dim)
+        else:
+            self.layer_norm = None
+        
+        self.lm_head = None
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        encoder_hidden_states: Optional[torch.FloatTensor] = None,
+        encoder_attention_mask: Optional[torch.FloatTensor] = None,
+        past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = False,
+        output_hidden_states: Optional[bool] = False,
+        return_dict: Optional[bool] = True,
+    ) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]:
+
+        x = super().forward(hidden_states=hidden_states,
+                            attention_mask=attention_mask,
+                            head_mask=head_mask,
+                            encoder_hidden_states=encoder_hidden_states,
+                            encoder_attention_mask=encoder_attention_mask,
+                            past_key_values=past_key_values,
+                            use_cache=use_cache,
+                            output_attentions=output_attentions,
+                            output_hidden_states=output_hidden_states,
+                            return_dict=return_dict)
+        
+ 
+        if self.layer_norm is not None:
+            x.last_hidden_state = self.layer_norm(x.last_hidden_state)
+        
+        return x
+
+    def build_encoder(self, args, dictionary, embed_tokens):
+        encoder = LinformerTransformerEncoder(args)
+        return encoder
+        if args.use_linformer:
+            encoder = LinformerTransformerEncoder(args, dictionary, embed_tokens)
+        elif args.use_fft:
+            encoder = FourierTransformerEncoder(args, dictionary, embed_tokens)
+        else:
+            encoder = TransformerEncoder(args, dictionary, embed_tokens)
+
+        encoder.apply(init_bert_params)
+
+        return encoder
+
+    def output_layer(self, features, masked_tokens=None, pairs=None, **unused):
+        lm_out = self.lm_head(features, masked_tokens)
+        if pairs is not None:
+            sbo_out = self.sbo_head(features, pairs)
+            return lm_out, sbo_out
+        else:
+            return lm_out
+
+    
+class JargonModel(RobertaModel):
+    config_class = JargonConfig
+    def __init__(self, config, **kwargs):
+        config_class = JargonConfig
+        base_model_prefix = "jargon"
+        
+        super().__init__(config, **kwargs)
+        self.embeddings = JargonEmbeddings(config)
+        self.encoder = JargonEncoder(config)
+    # Copied from modeling_roberta.py
+    # Add transpose of embeddings as implemented in fairseq
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[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[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]:
+        r"""
+        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]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (`tuple(tuple(torch.FloatTensor))` of length `config.n_layers` with each tuple having 4 tensors of shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up 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)`.
+        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 = 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
+
+        if self.config.is_decoder:
+            use_cache = use_cache if use_cache is not None else self.config.use_cache
+        else:
+            use_cache = False
+
+        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()
+        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")
+
+        batch_size, seq_length = input_shape
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
+        # 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 attention_mask is None:
+            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
+
+        if token_type_ids is None:
+            if hasattr(self.embeddings, "token_type_ids"):
+                buffered_token_type_ids = self.embeddings.token_type_ids[:, :seq_length]
+                buffered_token_type_ids_expanded = buffered_token_type_ids.expand(batch_size, seq_length)
+                token_type_ids = buffered_token_type_ids_expanded
+            else:
+                token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
+
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
+
+        # 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.is_decoder 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_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_extended_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
+        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
+
+        embedding_output = self.embeddings(
+            input_ids=input_ids,
+            position_ids=position_ids,
+            token_type_ids=token_type_ids,
+            inputs_embeds=inputs_embeds,
+            past_key_values_length=past_key_values_length,
+        )
+
+
+        embedding_output = embedding_output.transpose(0,1)
+        encoder_outputs = self.encoder(
+            embedding_output,
+            attention_mask=extended_attention_mask,
+            head_mask=head_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_extended_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+        )
+
+        sequence_output = encoder_outputs[0].transpose(0,1)
+
+        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
+
+        # Fairseq Linformer implementation works with transposed hidden states -> we transpose them back for HF implementation.
+        if output_hidden_states:
+            encoder_outputs.hidden_states = [h.transpose(0,1) for h in encoder_outputs.hidden_states]
+
+        if not return_dict:
+            return (sequence_output, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            past_key_values=encoder_outputs.past_key_values,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+            cross_attentions=encoder_outputs.cross_attentions,
+        )
+
+
+class SBOLayer(nn.Module):
+
+    def __init__(self, input_size, hidden_size, activation, export):
+        super().__init__()
+        self.layer = nn.Linear(input_size, hidden_size)
+        self.activ = get_activation_fn(activation)
+        self.norm = LayerNorm(hidden_size)
+
+    def forward(self, x):
+        return self.norm(self.activ(self.layer(x)))
+
+
+class SBONetwork(nn.Module):
+
+    def __init__(self, input_size, hidden_size, activation, export):
+        super().__init__()
+        self.layers = nn.ModuleList([
+            self.build_sbo_layer(input_size, hidden_size, activation, export),
+            self.build_sbo_layer(hidden_size, hidden_size, activation, export)
+        ])
+        self.layers = nn.Sequential(*self.layers)
+
+    def build_sbo_layer(self, input_size, output_size, activation, export):
+        return SBOLayer(input_size, output_size, activation, export)
+
+    def forward(self, x):
+        return self.layers(x)
+ 
+
+class SBOHead(nn.Module):
+
+    def __init__(self, args, embedding_weights, max_targets=10, position_embedding_size=200):
+        super().__init__()
+
+        self.position_embeddings = nn.Embedding(max_targets, position_embedding_size)
+
+        export = getattr(args, "export", False)
+        hidden_size = args.embed_dim
+        input_size = hidden_size * 2 + position_embedding_size
+        activation = getattr(args, "activation_fn", "relu") or "relu"
+
+        self.mlp_layer_norm = self.build_sbo_network(input_size, hidden_size, activation, export)
+
+        # The output weights are the same as the input embeddings, but there is
+        # an output-only bias for each token.
+        self.decoder = nn.Linear(
+            embedding_weights.size(1),
+            embedding_weights.size(0),
+            bias=False
+        )
+        if embedding_weights is not None:
+            self.decoder.weight = embedding_weights
+
+        self.bias = nn.Parameter(torch.zeros(embedding_weights.size(0)))
+        self.max_targets = max_targets
+
+    def build_sbo_network(self, input_size, hidden_size, activation, export):
+        return SBONetwork(input_size, hidden_size, activation, export)
+
+    def forward(self, hidden_states, pairs):
+        bs, num_pairs, _ = pairs.size()
+        bs, seq_len, dim = hidden_states.size()
+        # pair indices: (bs, num_pairs)
+        left, right = pairs[:,:, 0], pairs[:, :, 1]
+        # (bs, num_pairs, dim)
+        left_hidden = torch.gather(hidden_states, 1, left.unsqueeze(2).repeat(1, 1, dim))
+        # pair states: bs * num_pairs, max_targets, dim
+        left_hidden = left_hidden.contiguous().view(bs * num_pairs, dim).unsqueeze(1).repeat(1, self.max_targets, 1)
+        
+        right_hidden = torch.gather(hidden_states, 1, right.unsqueeze(2).repeat(1, 1, dim))
+        # bs * num_pairs, max_targets, dim
+        right_hidden = right_hidden.contiguous().view(bs * num_pairs, dim).unsqueeze(1).repeat(1, self.max_targets, 1)
+
+        # (max_targets, dim)
+        position_embeddings = self.position_embeddings.weight
+       
+        z = torch.cat((left_hidden, right_hidden, position_embeddings.unsqueeze(0).repeat(bs * num_pairs, 1, 1)), -1)
+        
+        hidden_states = self.mlp_layer_norm(torch.cat((left_hidden, right_hidden, position_embeddings.unsqueeze(0).repeat(bs * num_pairs, 1, 1)), -1))
+        # target scores : bs * num_pairs, max_targets, vocab_size
+        target_scores = self.decoder(hidden_states) + self.bias
+        return target_scores
+
+
+def get_activation_fn(activation):
+    """Returns the activation function corresponding to `activation`"""
+
+    if activation == "relu":
+        return F.relu
+    elif activation == "relu_squared":
+        return F.relu_squared
+    elif activation == "gelu":
+        return F.gelu
+    elif activation == "gelu_fast":
+        deprecation_warning(
+            "--activation-fn=gelu_fast has been renamed to gelu_accurate"
+        )
+        return F.gelu_accurate
+    elif activation == "gelu_accurate":
+        return F.gelu_accurate
+    elif activation == "tanh":
+        return torch.tanh
+    elif activation == "linear":
+        return lambda x: x
+    elif activation == "swish":
+        return torch.nn.SiLU
+    else:
+        raise RuntimeError("--activation-fn {} not supported".format(activation))
+
+
+def create_position_ids_from_input_ids(input_ids, padding_idx, past_key_values_length=0):
+    """
+    Replace non-padding symbols with their position numbers. Position numbers begin at padding_idx+1. Padding symbols
+    are ignored. This is modified from fairseq's `utils.make_positions`.
+
+    Args:
+        x: torch.Tensor x:
+
+    Returns: torch.Tensor
+    """
+    # The series of casts and type-conversions here are carefully balanced to both work with ONNX export and XLA.
+    mask = input_ids.ne(padding_idx).int()
+    incremental_indices = (torch.cumsum(mask, dim=1).type_as(mask) + past_key_values_length) * mask
+    return incremental_indices.long() + padding_idx

linformer.py

@@ -0,0 +1,740 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+#
+# This source code is licensed under the MIT license found in the
+# LICENSE file in the root directory of this source tree.
+
+import math
+import inspect
+from typing import Callable, Dict, List, Optional, Set, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.nn import TransformerEncoder, TransformerEncoderLayer
+from fairseq import utils
+from fairseq.models.transformer import *
+from fairseq.incremental_decoding_utils import with_incremental_state
+from fairseq.modules.quant_noise import quant_noise
+from transformers.models.roberta.modeling_roberta import (
+    RobertaEncoder,
+    RobertaConfig,
+    RobertaModel,
+    RobertaLMHead,
+    RobertaForMaskedLM,
+    RobertaLayer
+)
+
+# from .multihead_linear_attention import MultiheadLinearAttention
+
+
+class LinformerTransformerEncoderLayer(RobertaLayer):
+    """
+    Implements a Linformer Encoder Layer used in BERT/XLM style pre-trained
+    models.
+    """
+
+    def __init__(self, config, shared_compress_layer):
+        # wrap in a list so it's not automatically registered by PyTorch
+        self.shared_compress_layer = [shared_compress_layer]
+        d_model=config.embed_dim
+        nhead=config.num_heads
+        dim_feedforward=config.dim_feedforward
+        dropout=config.dropout
+        activation=config.activation 
+        layer_norm_eps=config.layer_norm_eps
+        
+        super().__init__(config)
+        self.attention = self.build_self_attention(config.embed_dim, config)
+        self.attn_layer_norm = nn.LayerNorm(config.hidden_size, eps=1e-5)
+        self.final_layer_norm = nn.LayerNorm(config.hidden_size, eps=1e-5)
+        self.output = RobertaOutput(config)
+
+    def build_self_attention(self, embed_dim, args):
+
+        attn = MultiheadLinearAttention(
+            embed_dim,
+            args.encoder_attention_heads,
+            dropout=args.dropout,
+            self_attention=True,
+            q_noise=args.quant_noise_pq,
+            qn_block_size=args.quant_noise_pq_block_size,
+            compressed=args.compressed,
+            max_seq_len=args.max_positions,
+            shared_kv_compressed=args.shared_kv_compressed,
+            shared_compress_layer=self.shared_compress_layer[0],
+            freeze_compress=args.freeze_compress,
+        ) 
+        return attn
+
+    def feed_forward_chunk(self, attention_output):
+        residual = attention_output
+        x = self.intermediate(attention_output)
+        layer_output = self.output(x, residual)
+        return layer_output
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        encoder_hidden_states: Optional[torch.FloatTensor] = None,
+        encoder_attention_mask: Optional[torch.FloatTensor] = None,
+        past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.Tensor]:
+
+        residual = hidden_states
+ 
+        if self.attn_layer_norm is not None:
+            hidden_states = self.attn_layer_norm(hidden_states)
+
+        # 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
+        self_attention_outputs = self.attention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            output_attentions=output_attentions,
+            past_key_value=self_attn_past_key_value,
+        )
+        attention_output = self_attention_outputs[0]
+
+        # if decoder, the last output is tuple of self-attn cache
+        if self.is_decoder:
+            outputs = self_attention_outputs[1:-1]
+            present_key_value = self_attention_outputs[-1]
+        else:
+            outputs = self_attention_outputs[1:]  # add self attentions if we output attention weights
+
+        cross_attn_present_key_value = None
+        if self.is_decoder and encoder_hidden_states is not None:
+            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`"
+                )
+
+            # cross_attn cached key/values tuple is at positions 3,4 of past_key_value tuple
+            cross_attn_past_key_value = past_key_value[-2:] if past_key_value is not None else None
+            cross_attention_outputs = self.crossattention(
+                attention_output,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                cross_attn_past_key_value,
+                output_attentions,
+            )
+            attention_output = cross_attention_outputs[0]
+            outputs = outputs + cross_attention_outputs[1:-1]  # add cross attentions if we output attention weights
+
+            # add cross-attn cache to positions 3,4 of present_key_value tuple
+            cross_attn_present_key_value = cross_attention_outputs[-1]
+            present_key_value = present_key_value + cross_attn_present_key_value
+      
+        attention_output = attention_output + residual
+        residual = attention_output
+        attention_output = self.final_layer_norm(attention_output)
+        layer_output = apply_chunking_to_forward(
+            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
+        )
+        layer_output = layer_output + residual
+
+        outputs = (layer_output,) + outputs
+
+        # if decoder, return the attn key/values as the last output
+        if self.is_decoder:
+            outputs = outputs + (present_key_value,)
+        
+        return outputs
+
+
+class RobertaOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.dense(hidden_states)
+        return hidden_states
+
+
+class LinformerTransformerEncoder(RobertaEncoder):
+    """
+    Implementation for a Bi-directional Linformer based Sentence Encoder used
+    in BERT/XLM style pre-trained models.
+
+    This first computes the token embedding using the token embedding matrix,
+    position embeddings (if specified) and segment embeddings
+    (if specified). After applying the specified number of
+    LinformerEncoderLayers, it outputs all the internal states of the
+    encoder as well as the final representation associated with the first
+    token (usually CLS token).
+
+    Input:
+        - tokens: B x T matrix representing sentences
+        - segment_labels: B x T matrix representing segment label for tokens
+
+    Output:
+        - a tuple of the following:
+            - a list of internal model states used to compute the
+              predictions where each tensor has shape T x B x C
+            - sentence representation associated with first input token
+              in format B x C.
+    """
+
+    def __init__(self, config,**kwargs):
+        compress_layer = None
+        if config.shared_layer_kv_compressed == 1 and compress_layer is None:
+            compress_layer = nn.Linear(
+                config.max_positions,
+                config.max_positions // config.compressed
+            )
+            # intialize parameters for compressed layer
+            nn.init.xavier_uniform_(compress_layer.weight, gain=1 / math.sqrt(2))
+            if config.freeze_compress == 1:
+                compress_layer.weight.requires_grad = False
+            compress_layer = compress_layer
+        #encoder_layer = LinformerTransformerEncoderLayer(config, compress_layer)
+       
+        super().__init__(config)
+        
+        self.layer = nn.ModuleList([LinformerTransformerEncoderLayer(config, compress_layer) for _ in range(config.num_layers)])
+        self.compress_layer = compress_layer
+        self.layer_norm = nn.LayerNorm(config.embed_dim)
+
+
+@with_incremental_state
+class MultiheadLinearAttention(nn.Module):
+    def __init__(
+        self,
+        embed_dim,
+        num_heads,
+        kdim=None,
+        vdim=None,
+        dropout=0.0,
+        bias=True,
+        add_bias_kv=False,
+        add_zero_attn=False,
+        self_attention=False,
+        encoder_decoder_attention=False,
+        q_noise=0.0,
+        qn_block_size=8,
+        compressed=1,
+        max_seq_len=256,
+        shared_kv_compressed=0,
+        shared_compress_layer=None,
+        freeze_compress=0,
+    ):
+        super().__init__()
+        self.embed_dim = embed_dim
+        self.kdim = kdim if kdim is not None else embed_dim
+        self.vdim = vdim if vdim is not None else embed_dim
+        self.qkv_same_dim = self.kdim == embed_dim and self.vdim == embed_dim
+        self.num_heads = num_heads
+        self.dropout = dropout
+        self.head_dim = embed_dim // num_heads
+        assert (
+            self.head_dim * num_heads == self.embed_dim
+        ), "embed_dim must be divisible by num_heads"
+        self.scaling = self.head_dim ** -0.5
+
+        self.self_attention = self_attention
+        self.encoder_decoder_attention = encoder_decoder_attention
+        assert not self.self_attention or self.qkv_same_dim, (
+            "Self-attention requires query, key and " "value to be of the same size"
+        )
+
+        self.k_proj = quant_noise(
+            nn.Linear(self.kdim, embed_dim, bias=bias), q_noise, qn_block_size
+        )
+        self.v_proj = quant_noise(
+            nn.Linear(self.vdim, embed_dim, bias=bias), q_noise, qn_block_size
+        )
+        self.q_proj = quant_noise(
+            nn.Linear(embed_dim, embed_dim, bias=bias), q_noise, qn_block_size
+        )
+
+        # used for compress sequence to subsequence
+        if shared_compress_layer is None:
+            self.compress_seq_len = max_seq_len // compressed
+            self.compress_k = nn.Linear(max_seq_len, self.compress_seq_len, bias=False)
+            if shared_kv_compressed == 0:
+                self.compress_v = nn.Linear(
+                    max_seq_len, self.compress_seq_len, bias=False
+                )
+            self.layerwise_sharing = False
+        else:
+            self.compress_k = shared_compress_layer
+            if shared_kv_compressed == 0:
+                self.compress_v = shared_compress_layer
+            self.layerwise_sharing = True
+        self.shared_kv_compressed = shared_kv_compressed
+
+        self.out_proj = quant_noise(
+            nn.Linear(embed_dim, embed_dim, bias=bias), q_noise, qn_block_size)
+
+        if add_bias_kv:
+            self.bias_k = nn.Parameter(torch.Tensor(1, 1, embed_dim))
+            self.bias_v = nn.Parameter(torch.Tensor(1, 1, embed_dim))
+        else:
+            self.bias_k = self.bias_v = None
+
+        self.add_zero_attn = add_zero_attn
+
+        self.reset_parameters()
+
+        if freeze_compress == 1:
+            self.compress_k.weight.requires_grad = False
+            if shared_kv_compressed == 0:
+                self.compress_v.weight.requires_grad = False
+
+        self.onnx_trace = False
+    def reset_parameters(self):
+        if self.qkv_same_dim:
+            # Empirically observed the convergence to be much better with
+            # the scaled initialization
+            nn.init.xavier_uniform_(self.k_proj.weight, gain=1 / math.sqrt(2))
+            nn.init.xavier_uniform_(self.v_proj.weight, gain=1 / math.sqrt(2))
+            nn.init.xavier_uniform_(self.q_proj.weight, gain=1 / math.sqrt(2))
+            if (
+                not self.layerwise_sharing
+            ):  # otherwise, we already initialize the parameters
+                nn.init.xavier_uniform_(self.compress_k.weight, gain=1 / math.sqrt(2))
+                if self.shared_kv_compressed == 0:
+                    nn.init.xavier_uniform_(
+                        self.compress_v.weight, gain=1 / math.sqrt(2)
+                    )
+        else:
+            nn.init.xavier_uniform_(self.k_proj.weight)
+            nn.init.xavier_uniform_(self.v_proj.weight)
+            nn.init.xavier_uniform_(self.q_proj.weight)
+            if (
+                not self.layerwise_sharing
+            ):  # otherwise, we already initialize the parameters
+                nn.init.xavier_uniform_(self.compress_k.weight)
+                if self.shared_kv_compressed == 0:
+                    nn.init.xavier_uniform_(self.compress_v.weight)
+
+        nn.init.xavier_uniform_(self.out_proj.weight)
+        if self.out_proj.bias is not None:
+            nn.init.constant_(self.out_proj.bias, 0.0)
+        if self.bias_k is not None:
+            nn.init.xavier_normal_(self.bias_k)
+        if self.bias_v is not None:
+            nn.init.xavier_normal_(self.bias_v)
+
+    def prepare_for_onnx_export_(self):
+        self.onnx_trace = True
+
+    def forward(
+        self,
+        query,
+        key: Optional[torch.Tensor],
+        value: Optional[torch.Tensor],
+        key_padding_mask: Optional[torch.Tensor] = None,
+        incremental_state: Optional[Dict[str, Dict[str, Optional[torch.Tensor]]]] = None,
+        output_attentions: bool = True,
+        need_weights: bool = True,
+        static_kv: bool = False,
+        attn_mask: Optional[torch.Tensor] = None,
+        before_softmax: bool = False,
+        need_head_weights: bool = False,
+        past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
+        """Input shape: Time x Batch x Channel
+
+        Args:
+            key_padding_mask (ByteTensor, optional): mask to exclude
+                keys that are pads, of shape `(batch, src_len)`, where
+                padding elements are indicated by 1s.
+            need_weights (bool, optional): return the attention weights,
+                averaged over heads (default: False).
+            attn_mask (ByteTensor, optional): typically used to
+                implement causal attention, where the mask prevents the
+                attention from looking forward in time (default: None).
+            before_softmax (bool, optional): return the raw attention
+                weights and values before the attention softmax.
+            need_head_weights (bool, optional): return the attention
+                weights for each head. Implies *need_weights*. Default:
+                return the average attention weights over all heads.
+        """
+        
+        if need_head_weights:
+            need_weights = True
+
+        tgt_len, bsz, embed_dim = query.size()
+        assert embed_dim == self.embed_dim
+        assert list(query.size()) == [tgt_len, bsz, embed_dim]
+
+        if incremental_state is not None:
+            saved_state = self._get_input_buffer(incremental_state)
+            if saved_state is not None and "prev_key" in saved_state:
+                # previous time steps are cached - no need to recompute
+                # key and value if they are static
+                if static_kv:
+                    assert self.encoder_decoder_attention and not self.self_attention
+                    key = value = None
+        else:
+            saved_state = None
+
+        if self.self_attention:
+            q = self.q_proj(query)
+
+            k_input = query.permute(1, 2, 0).contiguous()  # B * C * T
+            k_input = (
+                F.linear(k_input, self.compress_k.weight[:, 0:tgt_len])
+                .permute(2, 0, 1)
+                .contiguous()
+            )
+            k = self.k_proj(k_input)
+
+            v_input = query.permute(1, 2, 0).contiguous()  # B * C * T
+            if self.shared_kv_compressed == 0:
+                v_input = (
+                    F.linear(v_input, self.compress_v.weight[:, 0:tgt_len])
+                    .permute(2, 0, 1)
+                    .contiguous()
+                )
+            if self.shared_kv_compressed == 1:  # use shared kv compressed linear layer
+                v_input = (
+                    F.linear(v_input, self.compress_k.weight[:, 0:tgt_len])
+                    .permute(2, 0, 1)
+                    .contiguous()
+                )
+            v = self.v_proj(v_input)
+        elif self.encoder_decoder_attention:
+            # encoder-decoder attention
+            q = self.q_proj(query)
+            if key is None:
+                assert value is None
+                k = v = None
+            else:
+                k = self.k_proj(key)
+                v = self.v_proj(key)
+
+        else:
+            assert key is not None and value is not None
+            q = self.q_proj(query)
+            k = self.k_proj(key)
+            v = self.v_proj(value)
+        q *= self.scaling
+
+        if self.bias_k is not None:
+            assert self.bias_v is not None
+            k = torch.cat([k, self.bias_k.repeat(1, bsz, 1)])
+            v = torch.cat([v, self.bias_v.repeat(1, bsz, 1)])
+            if attn_mask is not None:
+                attn_mask = torch.cat(
+                    [attn_mask, attn_mask.new_zeros(attn_mask.size(0), 1)], dim=1
+                )
+            if key_padding_mask is not None:
+                key_padding_mask = torch.cat(
+                    [
+                        key_padding_mask,
+                        key_padding_mask.new_zeros(key_padding_mask.size(0), 1),
+                    ],
+                    dim=1,
+                )
+
+        q = (
+            q.contiguous()
+            .view(tgt_len, bsz * self.num_heads, self.head_dim)
+            .transpose(0, 1)
+        )
+        if k is not None:
+            k = (
+                k.contiguous()
+                .view(-1, bsz * self.num_heads, self.head_dim)
+                .transpose(0, 1)
+            )
+        if v is not None:
+            v = (
+                v.contiguous()
+                .view(-1, bsz * self.num_heads, self.head_dim)
+                .transpose(0, 1)
+            )
+
+        if saved_state is not None:
+            # saved states are stored with shape (bsz, num_heads, seq_len, head_dim)
+            if "prev_key" in saved_state:
+                _prev_key = saved_state["prev_key"]
+                assert _prev_key is not None
+                prev_key = _prev_key.view(bsz * self.num_heads, -1, self.head_dim)
+                if static_kv:
+                    k = prev_key
+                else:
+                    assert k is not None
+                    k = torch.cat([prev_key, k], dim=1)
+            if "prev_value" in saved_state:
+                _prev_value = saved_state["prev_value"]
+                assert _prev_value is not None
+                prev_value = _prev_value.view(bsz * self.num_heads, -1, self.head_dim)
+                if static_kv:
+                    v = prev_value
+                else:
+                    assert v is not None
+                    v = torch.cat([prev_value, v], dim=1)
+            prev_key_padding_mask: Optional[torch.Tensor] = None
+            if "prev_key_padding_mask" in saved_state:
+                prev_key_padding_mask = saved_state["prev_key_padding_mask"]
+            assert k is not None and v is not None
+            key_padding_mask = MultiheadLinearAttention._append_prev_key_padding_mask(
+                key_padding_mask=key_padding_mask,
+                prev_key_padding_mask=prev_key_padding_mask,
+                batch_size=bsz,
+                src_len=k.size(1),
+                static_kv=static_kv,
+            )
+
+            saved_state["prev_key"] = k.view(bsz, self.num_heads, -1, self.head_dim)
+            saved_state["prev_value"] = v.view(bsz, self.num_heads, -1, self.head_dim)
+            saved_state["prev_key_padding_mask"] = key_padding_mask
+            # In this branch incremental_state is never None
+            assert incremental_state is not None
+            incremental_state = self._set_input_buffer(incremental_state, saved_state)
+        assert k is not None
+        src_len = k.size(1)
+
+        if self.add_zero_attn:
+            assert v is not None
+            src_len += 1
+            k = torch.cat([k, k.new_zeros((k.size(0), 1) + k.size()[2:])], dim=1)
+            v = torch.cat([v, v.new_zeros((v.size(0), 1) + v.size()[2:])], dim=1)
+            if attn_mask is not None:
+                attn_mask = torch.cat(
+                    [attn_mask, attn_mask.new_zeros(attn_mask.size(0), 1)], dim=1
+                )
+
+        attn_weights = torch.bmm(q, k.transpose(1, 2))
+        attn_weights = MultiheadLinearAttention.apply_sparse_mask(
+            attn_weights, tgt_len, src_len, bsz
+        )
+
+        assert list(attn_weights.size()) == [bsz * self.num_heads, tgt_len, src_len]
+
+        if attn_mask is not None:
+            attn_mask = attn_mask.unsqueeze(0)
+            if self.onnx_trace:
+                attn_mask = attn_mask.repeat(attn_weights.size(0), 1, 1)
+            attn_weights += attn_mask
+
+        if before_softmax:
+            return attn_weights, v
+
+        attn_weights_float = utils.softmax(
+            attn_weights, dim=-1, onnx_trace=self.onnx_trace
+        )
+        attn_weights = attn_weights_float.type_as(attn_weights)
+        attn_probs = F.dropout(
+            attn_weights,
+            p=self.dropout,
+            training=self.training,
+        )
+        assert v is not None
+        attn = torch.bmm(attn_probs, v)
+        assert list(attn.size()) == [bsz * self.num_heads, tgt_len, self.head_dim]
+        if self.onnx_trace and attn.size(1) == 1:
+            # when ONNX tracing a single decoder step (sequence length == 1)
+            # the transpose is a no-op copy before view, thus unnecessary
+            attn = attn.contiguous().view(tgt_len, bsz, embed_dim)
+        else:
+            attn = attn.transpose(0, 1).contiguous().view(tgt_len, bsz, embed_dim)
+        attn = self.out_proj(attn)
+        attn_weights: Optional[torch.Tensor] = None
+        if output_attentions:
+            attn_weights = attn_weights_float.view(
+                bsz, self.num_heads, tgt_len, src_len
+            ).transpose(1, 0)
+            if not need_head_weights:
+                # average attention weights over heads
+                attn_weights = attn_weights.mean(dim=0)
+
+
+        return attn, attn_weights
+
+    @staticmethod
+    def _append_prev_key_padding_mask(
+        key_padding_mask: Optional[torch.Tensor],
+        prev_key_padding_mask: Optional[torch.Tensor],
+        batch_size: int,
+        src_len: int,
+        static_kv: bool,
+    ) -> Optional[torch.Tensor]:
+        # saved key padding masks have shape (bsz, seq_len)
+        if prev_key_padding_mask is not None and static_kv:
+            new_key_padding_mask = prev_key_padding_mask
+        elif prev_key_padding_mask is not None and key_padding_mask is not None:
+            new_key_padding_mask = torch.cat(
+                [prev_key_padding_mask.float(), key_padding_mask.float()], dim=1
+            )
+        # During incremental decoding, as the padding token enters and
+        # leaves the frame, there will be a time when prev or current
+        # is None
+        elif prev_key_padding_mask is not None:
+            filler = torch.zeros(
+                (batch_size, src_len - prev_key_padding_mask.size(1)),
+                device=prev_key_padding_mask.device,
+            )
+            new_key_padding_mask = torch.cat(
+                [prev_key_padding_mask.float(), filler.float()], dim=1
+            )
+        elif key_padding_mask is not None:
+            filler = torch.zeros(
+                (batch_size, src_len - key_padding_mask.size(1)),
+                device=key_padding_mask.device,
+            )
+            new_key_padding_mask = torch.cat(
+                [filler.float(), key_padding_mask.float()], dim=1
+            )
+        else:
+            new_key_padding_mask = prev_key_padding_mask
+        return new_key_padding_mask
+
+    @torch.jit.export
+    def reorder_incremental_state(
+        self,
+        incremental_state: Dict[str, Dict[str, Optional[torch.Tensor]]],
+        new_order: torch.Tensor,
+    ):
+        """Reorder buffered internal state (for incremental generation)."""
+        input_buffer = self._get_input_buffer(incremental_state)
+        if input_buffer is not None:
+            for k in input_buffer.keys():
+                input_buffer_k = input_buffer[k]
+                if input_buffer_k is not None:
+                    if self.encoder_decoder_attention and input_buffer_k.size(
+                        0
+                    ) == new_order.size(0):
+                        break
+                    input_buffer[k] = input_buffer_k.index_select(0, new_order)
+            incremental_state = self._set_input_buffer(incremental_state, input_buffer)
+        return incremental_state
+
+    def _get_input_buffer(
+        self, incremental_state: Optional[Dict[str, Dict[str, Optional[torch.Tensor]]]]
+    ) -> Dict[str, Optional[torch.Tensor]]:
+        result = self.get_incremental_state(incremental_state, "attn_state")
+        if result is not None:
+            return result
+        else:
+            empty_result: Dict[str, Optional[torch.Tensor]] = {}
+            return empty_result
+
+    def _set_input_buffer(
+        self,
+        incremental_state: Dict[str, Dict[str, Optional[torch.Tensor]]],
+        buffer: Dict[str, Optional[torch.Tensor]],
+    ):
+        return self.set_incremental_state(incremental_state, "attn_state", buffer)
+
+    def apply_sparse_mask(attn_weights, tgt_len: int, src_len: int, bsz: int):
+        return attn_weights
+
+    def upgrade_state_dict_named(self, state_dict, name):
+        prefix = name + "." if name != "" else ""
+        items_to_add = {}
+        keys_to_remove = []
+        for k in state_dict.keys():
+            if k.endswith(prefix + "in_proj_weight"):
+                # in_proj_weight used to be q + k + v with same dimensions
+                dim = int(state_dict[k].shape[0] / 3)
+                items_to_add[prefix + "q_proj.weight"] = state_dict[k][:dim]
+                items_to_add[prefix + "k_proj.weight"] = state_dict[k][dim : 2 * dim]
+                items_to_add[prefix + "v_proj.weight"] = state_dict[k][2 * dim :]
+
+                keys_to_remove.append(k)
+
+                k_bias = prefix + "in_proj_bias"
+                if k_bias in state_dict.keys():
+                    dim = int(state_dict[k].shape[0] / 3)
+                    items_to_add[prefix + "q_proj.bias"] = state_dict[k_bias][:dim]
+                    items_to_add[prefix + "k_proj.bias"] = state_dict[k_bias][
+                        dim : 2 * dim
+                    ]
+                    items_to_add[prefix + "v_proj.bias"] = state_dict[k_bias][2 * dim :]
+
+                    keys_to_remove.append(prefix + "in_proj_bias")
+
+        for k in keys_to_remove:
+            del state_dict[k]
+
+        for key, value in items_to_add.items():
+            state_dict[key] = value
+
+
+
+def apply_chunking_to_forward(
+    forward_fn: Callable[..., torch.Tensor], chunk_size: int, chunk_dim: int, *input_tensors
+) -> torch.Tensor:
+    """
+    This function chunks the `input_tensors` into smaller input tensor parts of size `chunk_size` over the dimension
+    `chunk_dim`. It then applies a layer `forward_fn` to each chunk independently to save memory.
+
+    If the `forward_fn` is independent across the `chunk_dim` this function will yield the same result as directly
+    applying `forward_fn` to `input_tensors`.
+
+    Args:
+        forward_fn (`Callable[..., torch.Tensor]`):
+            The forward function of the model.
+        chunk_size (`int`):
+            The chunk size of a chunked tensor: `num_chunks = len(input_tensors[0]) / chunk_size`.
+        chunk_dim (`int`):
+            The dimension over which the `input_tensors` should be chunked.
+        input_tensors (`Tuple[torch.Tensor]`):
+            The input tensors of `forward_fn` which will be chunked
+
+    Returns:
+        `torch.Tensor`: A tensor with the same shape as the `forward_fn` would have given if applied`.
+
+
+    Examples:
+
+    ```python
+    # rename the usual forward() fn to forward_chunk()
+    def forward_chunk(self, hidden_states):
+        hidden_states = self.decoder(hidden_states)
+        return hidden_states
+
+
+    # implement a chunked forward function
+    def forward(self, hidden_states):
+        return apply_chunking_to_forward(self.forward_chunk, self.chunk_size_lm_head, self.seq_len_dim, hidden_states)
+    ```"""
+    assert len(input_tensors) > 0, f"{input_tensors} has to be a tuple/list of tensors"
+
+    # inspect.signature exist since python 3.5 and is a python method -> no problem with backward compatibility
+    num_args_in_forward_chunk_fn = len(inspect.signature(forward_fn).parameters)
+    if num_args_in_forward_chunk_fn != len(input_tensors):
+        raise ValueError(
+            f"forward_chunk_fn expects {num_args_in_forward_chunk_fn} arguments, but only {len(input_tensors)} input "
+            "tensors are given"
+        )
+
+    if chunk_size > 0:
+        tensor_shape = input_tensors[0].shape[chunk_dim]
+        for input_tensor in input_tensors:
+            if input_tensor.shape[chunk_dim] != tensor_shape:
+                raise ValueError(
+                    f"All input tenors have to be of the same shape: {tensor_shape}, "
+                    f"found shape {input_tensor.shape[chunk_dim]}"
+                )
+
+        if input_tensors[0].shape[chunk_dim] % chunk_size != 0:
+            raise ValueError(
+                f"The dimension to be chunked {input_tensors[0].shape[chunk_dim]} has to be a multiple of the chunk "
+                f"size {chunk_size}"
+            )
+
+        num_chunks = input_tensors[0].shape[chunk_dim] // chunk_size
+
+        # chunk input tensor into tuples
+        input_tensors_chunks = tuple(input_tensor.chunk(num_chunks, dim=chunk_dim) for input_tensor in input_tensors)
+        # apply forward fn to every tuple
+        output_chunks = tuple(forward_fn(*input_tensors_chunk) for input_tensors_chunk in zip(*input_tensors_chunks))
+        # concatenate output at same dimension
+        return torch.cat(output_chunks, dim=chunk_dim)
+
+    return forward_fn(*input_tensors)