modeling_bert.py

# coding=utf-8
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License
import math
import os
import numpy as np
import warnings
from dataclasses import dataclass
from typing import List, Optional, Tuple, Union
from transformers import AutoTokenizer
import torch
import torch.utils.checkpoint
from torch import nn

from transformers.activations import ACT2FN
from transformers.modeling_outputs import (
    BaseModelOutputWithPastAndCrossAttentions,
    BaseModelOutputWithPoolingAndCrossAttentions
)
from transformers.modeling_utils import PreTrainedModel
from transformers.pytorch_utils import apply_chunking_to_forward, find_pruneable_heads_and_indices, prune_linear_layer
from transformers.utils import (
    ModelOutput,
    logging)
from .configuration_bert import LongBertConfig

try:
    from torch.nn.functional import scaled_dot_product_attention
except ImportError:
    scaled_dot_product_attention = None
logger = logging.get_logger(__name__)

try:
    from tqdm.autonotebook import trange

    has_tqdm = True
except ImportError:
    has_tqdm = False


def load_tf_weights_in_bert(model, config, tf_checkpoint_path):
    """Load tf checkpoints in a pytorch model."""
    try:
        import re

        import numpy as np
        import tensorflow as tf
    except ImportError:
        logger.error(
            "Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see "
            "https://www.tensorflow.org/install/ for installation instructions."
        )
        raise
    tf_path = os.path.abspath(tf_checkpoint_path)
    logger.info(f"Converting TensorFlow checkpoint from {tf_path}")
    # Load weights from TF model
    init_vars = tf.train.list_variables(tf_path)
    names = []
    arrays = []
    for name, shape in init_vars:
        logger.info(f"Loading TF weight {name} with shape {shape}")
        array = tf.train.load_variable(tf_path, name)
        names.append(name)
        arrays.append(array)

    for name, array in zip(names, arrays):
        name = name.split("/")
        # adam_v and adam_m are variables used in AdamWeightDecayOptimizer to calculated m and v
        # which are not required for using pretrained model
        if any(
                n in ["adam_v", "adam_m", "AdamWeightDecayOptimizer", "AdamWeightDecayOptimizer_1", "global_step"]
                for n in name
        ):
            logger.info(f"Skipping {'/'.join(name)}")
            continue
        pointer = model
        for m_name in name:
            if re.fullmatch(r"[A-Za-z]+_\d+", m_name):
                scope_names = re.split(r"_(\d+)", m_name)
            else:
                scope_names = [m_name]
            if scope_names[0] == "kernel" or scope_names[0] == "gamma":
                pointer = getattr(pointer, "weight")
            elif scope_names[0] == "output_bias" or scope_names[0] == "beta":
                pointer = getattr(pointer, "bias")
            elif scope_names[0] == "output_weights":
                pointer = getattr(pointer, "weight")
            elif scope_names[0] == "squad":
                pointer = getattr(pointer, "classifier")
            else:
                try:
                    pointer = getattr(pointer, scope_names[0])
                except AttributeError:
                    logger.info(f"Skipping {'/'.join(name)}")
                    continue
            if len(scope_names) >= 2:
                num = int(scope_names[1])
                pointer = pointer[num]
        if m_name[-11:] == "_embeddings":
            pointer = getattr(pointer, "weight")
        elif m_name == "kernel":
            array = np.transpose(array)
        try:
            if pointer.shape != array.shape:
                raise ValueError(f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched")
        except AssertionError as e:
            e.args += (pointer.shape, array.shape)
            raise
        logger.info(f"Initialize PyTorch weight {name}")
        pointer.data = torch.from_numpy(array)
    return model


class LongBertEmbeddings(nn.Module):
    """Construct the embeddings from word, position and token_type embeddings."""

    def __init__(self, config):
        super().__init__()
        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)

        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
        # any TensorFlow checkpoint file
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        # position_ids (1, len position emb) is contiguous in memory and exported when serialized

    def forward(
            self,
            input_ids: Optional[torch.LongTensor] = None,
            token_type_ids: Optional[torch.LongTensor] = None,
            inputs_embeds: Optional[torch.FloatTensor] = None
    ) -> torch.Tensor:
        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
        embeddings = self.LayerNorm(embeddings)
        embeddings = self.dropout(embeddings)
        return embeddings


class LongBertSelfAttention(nn.Module):
    def __init__(self, config, position_embedding_type=None):
        super().__init__()
        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
            raise ValueError(
                f"The hidden size ({config.hidden_size}) is not a multiple of the number of attention "
                f"heads ({config.num_attention_heads})"
            )

        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.query = nn.Linear(config.hidden_size, self.all_head_size)
        self.key = nn.Linear(config.hidden_size, self.all_head_size)
        self.value = nn.Linear(config.hidden_size, self.all_head_size)

        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
        self.position_embedding_type = position_embedding_type or getattr(
            config, "position_embedding_type", "alibi"
        )
        self.is_decoder = config.is_decoder

    def transpose_for_scores(self, x: torch.Tensor) -> torch.Tensor:
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
        x = x.view(new_x_shape)
        return x.permute(0, 2, 1, 3)

    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,
            bias: Optional[torch.FloatTensor] = None,
    ) -> Tuple[torch.Tensor]:
        mixed_query_layer = self.query(hidden_states)

        # If this is instantiated as a cross-attention module, the keys
        # and values come from an encoder; the attention mask needs to be
        # such that the encoder's padding tokens are not attended to.
        is_cross_attention = encoder_hidden_states is not None

        if is_cross_attention and past_key_value is not None:
            # reuse k,v, cross_attentions
            key_layer = past_key_value[0]
            value_layer = past_key_value[1]
            attention_mask = encoder_attention_mask
        elif is_cross_attention:
            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
            attention_mask = encoder_attention_mask
        elif past_key_value is not None:
            key_layer = self.transpose_for_scores(self.key(hidden_states))
            value_layer = self.transpose_for_scores(self.value(hidden_states))
            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
        else:
            key_layer = self.transpose_for_scores(self.key(hidden_states))
            value_layer = self.transpose_for_scores(self.value(hidden_states))

        query_layer = self.transpose_for_scores(mixed_query_layer)

        use_cache = past_key_value is not None
        if self.is_decoder:
            # if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states.
            # Further calls to cross_attention layer can then reuse all cross-attention
            # key/value_states (first "if" case)
            # if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of
            # all previous decoder key/value_states. Further calls to uni-directional self-attention
            # can concat previous decoder key/value_states to current projected key/value_states (third "elif" case)
            # if encoder bi-directional self-attention `past_key_value` is always `None`
            past_key_value = (key_layer, value_layer)

        # Take the dot product between "query" and "key" to get the raw attention scores.
        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))

        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
            query_length, key_length = query_layer.shape[2], key_layer.shape[2]
            if use_cache:
                position_ids_l = torch.tensor(key_length - 1, dtype=torch.long, device=hidden_states.device).view(
                    -1, 1
                )
            else:
                position_ids_l = torch.arange(query_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
            position_ids_r = torch.arange(key_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
            distance = position_ids_l - position_ids_r

            positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
            positional_embedding = positional_embedding.to(dtype=query_layer.dtype)  # fp16 compatibility

            if self.position_embedding_type == "relative_key":
                relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
                attention_scores = attention_scores + relative_position_scores
            elif self.position_embedding_type == "relative_key_query":
                relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
                relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
                attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key

        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        if attention_mask is not None:
            # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
            attention_scores = attention_scores + attention_mask

        # Normalize the attention scores to probabilities.
        attention_probs = nn.functional.softmax(attention_scores + bias, dim=-1)

        # This is actually dropping out entire tokens to attend to, which might
        # seem a bit unusual, but is taken from the original Transformer paper.
        attention_probs = self.dropout(attention_probs)

        # Mask heads if we want to
        if head_mask is not None:
            attention_probs = attention_probs * head_mask

        context_layer = torch.matmul(attention_probs, value_layer)

        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
        context_layer = context_layer.view(new_context_layer_shape)

        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)

        if self.is_decoder:
            outputs = outputs + (past_key_value,)
        return outputs


class LongBertSelfOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_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)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


class LongBertAttention(nn.Module):
    def __init__(self, config, position_embedding_type=None):
        super().__init__()
        self.self = LongBertSelfAttention(config, position_embedding_type=position_embedding_type)
        self.output = LongBertSelfOutput(config)
        self.pruned_heads = set()

    def prune_heads(self, heads):
        if len(heads) == 0:
            return
        heads, index = find_pruneable_heads_and_indices(
            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
        )

        # Prune linear layers
        self.self.query = prune_linear_layer(self.self.query, index)
        self.self.key = prune_linear_layer(self.self.key, index)
        self.self.value = prune_linear_layer(self.self.value, index)
        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)

        # Update hyper params and store pruned heads
        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
        self.pruned_heads = self.pruned_heads.union(heads)

    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,
            bias: Optional[torch.FloatTensor] = None,
    ) -> Tuple[torch.Tensor]:
        self_outputs = self.self(
            hidden_states,
            attention_mask,
            head_mask,
            encoder_hidden_states,
            encoder_attention_mask,
            past_key_value,
            output_attentions,
            bias,
        )
        attention_output = self.output(self_outputs[0], hidden_states)
        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
        return outputs


class LongBertIntermediate(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
        if isinstance(config.hidden_act, str):
            self.intermediate_act_fn = ACT2FN[config.hidden_act]
        else:
            self.intermediate_act_fn = config.hidden_act

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states


class LongBertOutput(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)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


class LongBertLayer(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.chunk_size_feed_forward = config.chunk_size_feed_forward
        self.seq_len_dim = 1
        self.attention = LongBertAttention(config)
        self.is_decoder = config.is_decoder
        self.add_cross_attention = config.add_cross_attention
        if self.add_cross_attention:
            if not self.is_decoder:
                raise ValueError(f"{self} should be used as a decoder model if cross attention is added")
            self.crossattention = LongBertAttention(config, position_embedding_type="absolute")
        self.intermediate = LongBertIntermediate(config)
        self.output = LongBertOutput(config)

    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,
            bias: Optional[torch.FloatTensor] = None,
            past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
            output_attentions: Optional[bool] = False,
    ) -> Tuple[torch.Tensor]:
        # 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,
            bias=bias,
        )
        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

        layer_output = apply_chunking_to_forward(
            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
        )
        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

    def feed_forward_chunk(self, attention_output):
        intermediate_output = self.intermediate(attention_output)
        layer_output = self.output(intermediate_output, attention_output)
        return layer_output


class LongBertEncoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.layer = nn.ModuleList([LongBertLayer(config) for _ in range(config.num_hidden_layers)])
        self.gradient_checkpointing = False
        self.num_attention_heads = config.num_attention_heads
        self.register_buffer(
            "alibi",
            self.rebuild_alibi_tensor(size=config.max_position_embeddings),
            persistent=False,
        )

    def rebuild_alibi_tensor(
            self, size: int, device: Optional[Union[torch.device, str]] = None
    ):
        # Alibi
        # Following https://github.com/ofirpress/attention_with_linear_biases/issues/5 (Implementation 1)
        # In the causal case, you can exploit the fact that softmax is invariant to a uniform translation
        # of the logits, which makes the math work out *after* applying causal masking. If no causal masking
        # will be applied, it is necessary to construct the diagonal mask.
        n_heads = self.num_attention_heads

        def _get_alibi_head_slopes(n_heads: int) -> List[float]:
            def get_slopes_power_of_2(n):
                start = 2 ** (-(2 ** -(math.log2(n) - 3)))
                ratio = start
                return [start * ratio ** i for i in range(n)]

            if math.log2(n_heads).is_integer():
                return get_slopes_power_of_2(
                    n_heads
                )  # In the paper, we only train models that have 2^a heads for some a. This function has
            else:  # some good properties that only occur when the input is a power of 2. To maintain that even
                closest_power_of_2 = 2 ** math.floor(
                    math.log2(n_heads)
                )  # when the number of heads is not a power of 2, we use this workaround.
                return (
                        get_slopes_power_of_2(closest_power_of_2)
                        + _get_alibi_head_slopes(2 * closest_power_of_2)[0::2][
                          : n_heads - closest_power_of_2
                          ]
                )

        context_position = torch.arange(size, device=device)[:, None]
        memory_position = torch.arange(size, device=device)[None, :]
        relative_position = torch.abs(memory_position - context_position)
        # [n_heads, max_token_length, max_token_length]
        relative_position = relative_position.unsqueeze(0).expand(n_heads, -1, -1)
        slopes = torch.Tensor(_get_alibi_head_slopes(n_heads)).to(device) * -1
        alibi = slopes.unsqueeze(1).unsqueeze(1) * relative_position
        # [1, n_heads, max_token_length, max_token_length]
        alibi = alibi.unsqueeze(0)
        assert alibi.shape == torch.Size([1, n_heads, size, size])

        self._current_alibi_size = size
        return alibi

    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]:
        all_hidden_states = () if output_hidden_states else None
        all_self_attentions = () if output_attentions else None
        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
        # Add alibi matrix to extended_attention_mask
        _, seqlen, _ = hidden_states.size()
        if self._current_alibi_size < seqlen:
            # Rebuild the alibi tensor when needed
            warnings.warn(
                f'Increasing alibi size from {self._current_alibi_size} to {seqlen}.'
            )
            self.register_buffer(
                "alibi",
                self.rebuild_alibi_tensor(size=seqlen, device=hidden_states.device).to(
                    hidden_states.dtype
                ),
                persistent=False,
            )
        elif self.alibi.device != hidden_states.device:
            # Device catch-up
            self.alibi = self.alibi.to(hidden_states.device)

        alibi_bias = self.alibi[:, :, :seqlen, :seqlen]
        if self.gradient_checkpointing and self.training:
            if use_cache:
                logger.warning_once(
                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                )
                use_cache = False

        next_decoder_cache = () if use_cache else None
        for i, layer_module in enumerate(self.layer):
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            layer_head_mask = head_mask[i] if head_mask is not None else None
            past_key_value = past_key_values[i] if past_key_values is not None else None

            if self.gradient_checkpointing and self.training:

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        return module(*inputs, past_key_value, output_attentions)

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(layer_module),
                    hidden_states,
                    attention_mask,
                    layer_head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                )
            else:
                layer_outputs = layer_module(
                    hidden_states,
                    attention_mask,
                    layer_head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                    alibi_bias,
                    past_key_value,
                    output_attentions,
                )

            hidden_states = layer_outputs[0]
            if use_cache:
                next_decoder_cache += (layer_outputs[-1],)
            if output_attentions:
                all_self_attentions = all_self_attentions + (layer_outputs[1],)
                if self.config.add_cross_attention:
                    all_cross_attentions = all_cross_attentions + (layer_outputs[2],)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

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


class LongBertPooler(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.activation = nn.Tanh()

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        # We "pool" the model by simply taking the hidden state corresponding
        # to the first token.
        first_token_tensor = hidden_states[:, 0]
        pooled_output = self.dense(first_token_tensor)
        pooled_output = self.activation(pooled_output)
        return pooled_output


class LongBertPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = LongBertConfig
    load_tf_weights = load_tf_weights_in_bert
    base_model_prefix = "bert"
    supports_gradient_checkpointing = True

    def _init_weights(self, module):
        """Initialize the weights"""
        if isinstance(module, nn.Linear):
            # Slightly different from the TF version 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, LongBertEncoder):
            module.gradient_checkpointing = value


@dataclass
class LongBertForPreTrainingOutput(ModelOutput):
    """
    Output type of [`BertForPreTraining`].

    Args:
        loss (*optional*, returned when `labels` is provided, `torch.FloatTensor` of shape `(1,)`):
            Total loss as the sum of the masked language modeling loss and the next sequence prediction
            (classification) loss.
        prediction_logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        seq_relationship_logits (`torch.FloatTensor` of shape `(batch_size, 2)`):
            Prediction scores of the next sequence prediction (classification) head (scores of True/False continuation
            before SoftMax).
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of
            shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[torch.FloatTensor] = None
    prediction_logits: torch.FloatTensor = None
    seq_relationship_logits: torch.FloatTensor = None
    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    attentions: Optional[Tuple[torch.FloatTensor]] = None


class LongBertModel(LongBertPreTrainedModel):
    """

    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
    cross-attention is added between the self-attention layers, following the architecture described in [Attention is
    all you need](https://arxiv.org/abs/1706.03762) by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.

    To behave as an decoder the model needs to be initialized with the `is_decoder` argument of the configuration set
    to `True`. To be used in a Seq2Seq model, the model needs to initialized with both `is_decoder` argument and
    `add_cross_attention` set to `True`; an `encoder_hidden_states` is then expected as an input to the forward pass.
    """

    def __init__(self, config, add_pooling_layer=True):
        super().__init__(config)
        self.config = config
        self.embeddings = LongBertEmbeddings(config)
        self.encoder = LongBertEncoder(config)

        self.pooler = LongBertPooler(config) if add_pooling_layer else None
        self.tokenizer = AutoTokenizer.from_pretrained(config._name_or_path)
        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.embeddings.word_embeddings

    def set_input_embeddings(self, value):
        self.embeddings.word_embeddings = value

    def _prune_heads(self, heads_to_prune):
        """
        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
        class PreTrainedModel
        """
        for layer, heads in heads_to_prune.items():
            self.encoder.layer[layer].attention.prune_heads(heads)

    def forward(
            self,
            input_ids: Optional[torch.Tensor] = None,
            attention_mask: Optional[torch.Tensor] = None,
            token_type_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,
            token_type_ids=token_type_ids,
            inputs_embeds=inputs_embeds
        )
        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,
            return_dict=return_dict,
        )
        sequence_output = encoder_outputs[0]
        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None

        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,
        )

    def encode(self,
               sentences: Union[str, List[str]],
               batch_size: int = 32,
               show_progress_bar: Optional[bool] = None,
               output_value: str = 'sentence_embedding',
               convert_to_numpy: bool = True,
               convert_to_tensor: bool = False,
               device: Optional[torch.device] = "cpu",
               normalize_embeddings: bool = False,
               **tokenizer_kwargs,
               ) -> Union[List[torch.Tensor], np.ndarray, torch.Tensor]:
        """
        Computes sentence embeddings

        Args:
            sentences(`str` or `List[str]`):
                Sentence or sentences to be encoded
            batch_size(`int`, *optional*, defaults to 32):
                Batch size for the computation
            show_progress_bar(`bool`, *optional*, defaults to None):
                Show a progress bar when encoding sentences.
                If set to None, progress bar is only shown when `logger.level == logging.INFO` or `logger.level == logging.DEBUG`.
            output_value(`str`, *optional*, defaults to 'sentence_embedding'):
                Default sentence_embedding, to get sentence embeddings.
                Can be set to token_embeddings to get wordpiece token embeddings.
                Set to None, to get all output values
            convert_to_numpy(`bool`, *optional*, defaults to True):
                If true, the output is a list of numpy vectors.
                Else, it is a list of pytorch tensors.
            convert_to_tensor(`bool`, *optional*, defaults to False):
                If true, you get one large tensor as return.
                Overwrites any setting from convert_to_numpy
            device(`torch.device`, *optional*, defaults to None):
                Which torch.device to use for the computation
            normalize_embeddings(`bool`, *optional*, defaults to False):
                If set to true, returned vectors will have length 1. In that case, the faster dot-product (util.dot_score) instead of cosine similarity can be used.
            tokenizer_kwargs(`Dict[str, Any]`, *optional*, defaults to {}):
                Keyword arguments for the tokenizer

        Returns:
            By default, a list of tensors is returned.
            If convert_to_tensor, a stacked tensor is returned.
            If convert_to_numpy, a numpy matrix is returned.
        """

        if convert_to_tensor:
            convert_to_numpy = False

        if output_value != 'sentence_embedding':
            convert_to_tensor = False
            convert_to_numpy = False

        input_was_string = False
        if isinstance(sentences, str) or not hasattr(sentences, '__len__'):
            sentences = [sentences]
            input_was_string = True

        # TODO: Maybe use better length heuristic?
        permutation = np.argsort([-len(i) for i in sentences])
        inverse_permutation = np.argsort(permutation)
        sentences = [sentences[idx] for idx in permutation]

        tokenizer_kwargs['padding'] = tokenizer_kwargs.get('padding', True)
        tokenizer_kwargs['max_length'] = tokenizer_kwargs.get('max_length', 8192)
        tokenizer_kwargs['truncation'] = tokenizer_kwargs.get('truncation', True)

        all_embeddings = []

        if has_tqdm:
            range_iter = trange(
                0,
                len(sentences),
                batch_size,
                desc="Encoding",
                disable=not show_progress_bar,
            )
        else:
            range_iter = range(0, len(sentences), batch_size)

        for i in range_iter:
            encoded_input = self.tokenizer(
                sentences[i: i + batch_size],
                return_tensors='pt',
                **tokenizer_kwargs,
            )
            for key in encoded_input.keys():
                encoded_input[key] = encoded_input[key].to(self.device)
            token_embs = self.forward(**encoded_input)[0]
            # Accumulate in fp32 to avoid overflow
            token_embs = token_embs.float()

            if output_value == 'token_embeddings':
                raise NotImplementedError
            elif output_value is None:
                raise NotImplementedError
            else:
                embeddings = self.mean_pooling(
                    token_embs, encoded_input['attention_mask']
                )

                if normalize_embeddings:
                    embeddings = torch.nn.functional.normalize(embeddings, p=2, dim=1)

                if convert_to_numpy:
                    embeddings = embeddings.cpu()
            all_embeddings.extend(embeddings)

        all_embeddings = [all_embeddings[idx] for idx in inverse_permutation]

        if convert_to_tensor:
            all_embeddings = torch.stack(all_embeddings)
        elif convert_to_numpy:
            all_embeddings = np.asarray([emb.detach().numpy() for emb in all_embeddings])

        if input_was_string:
            all_embeddings = all_embeddings[0]
        return all_embeddings

    def mean_pooling(self, token_embeddings: torch.Tensor, attention_mask: torch.Tensor):
        input_mask_expanded = (
            attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
        )
        return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(
            input_mask_expanded.sum(1), min=1e-9
        )