Bert 源码（pytorch）超详细的解读_bert源码分析

model.py

对transformers的bert源码的解读 

# coding=utf-8
 
from __future__ import absolute_import, division, print_function, unicode_literals
 
import copy
import json
import logging
import math
import os
import shutil
import tarfile
import tempfile
import sys
from io import open
 
import torch
from torch import nn
from torch.nn import CrossEntropyLoss
 
from .file_utils import cached_path, WEIGHTS_NAME, CONFIG_NAME
 
logger = logging.getLogger(__name__)
 
 
# uncased不分大小写 multilingual多语种的
PRETRAINED_MODEL_ARCHIVE_MAP = {
    'bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased.tar.gz",
    'bert-large-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased.tar.gz",
    'bert-base-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased.tar.gz",
    'bert-large-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased.tar.gz",
    'bert-base-multilingual-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-uncased.tar.gz",
    'bert-base-multilingual-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-cased.tar.gz",
    'bert-base-chinese': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-chinese.tar.gz",
}
BERT_CONFIG_NAME = 'bert_config.json'
TF_WEIGHTS_NAME = 'model.ckpt'
 
def load_tf_weights_in_bert(model, tf_checkpoint_path):
    """ Load tf checkpoints in a pytorch model
    """
    try:
        import re
        import numpy as np
        import tensorflow as tf
    except ImportError:
        print("Loading a TensorFlow models 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)
    print("Converting TensorFlow checkpoint from {}".format(tf_path))
    # Load weights from TF model
    init_vars = tf.train.list_variables(tf_path)
    names = []
    arrays = []
    for name, shape in init_vars:
        print("Loading TF weight {} with shape {}".format(name, 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", "global_step"] for n in name):
            print("Skipping {}".format("/".join(name)))
            continue
        pointer = model
        for m_name in name:
            if re.fullmatch(r'[A-Za-z]+_\d+', m_name):
                l = re.split(r'_(\d+)', m_name)
            else:
                l = [m_name]
            if l[0] == 'kernel' or l[0] == 'gamma':
                pointer = getattr(pointer, 'weight')
            elif l[0] == 'output_bias' or l[0] == 'beta':
                pointer = getattr(pointer, 'bias')
            elif l[0] == 'output_weights':
                pointer = getattr(pointer, 'weight')
            elif l[0] == 'squad':
                pointer = getattr(pointer, 'classifier')
            else:
                try:
                    pointer = getattr(pointer, l[0])
                except AttributeError:
                    print("Skipping {}".format("/".join(name)))
                    continue
            if len(l) >= 2:
                num = int(l[1])
                pointer = pointer[num]
        if m_name[-11:] == '_embeddings':
            pointer = getattr(pointer, 'weight')
        elif m_name == 'kernel':
            array = np.transpose(array)
        try:
            assert pointer.shape == array.shape
        except AssertionError as e:
            e.args += (pointer.shape, array.shape)
            raise
        print("Initialize PyTorch weight {}".format(name))
        pointer.data = torch.from_numpy(array)
    return model
 
 
def gelu(x):  # bert的激活函数
    """Implementation of the gelu activation function.
        For information: OpenAI GPT's gelu is slightly different (and gives slightly different results):
        0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 * torch.pow(x, 3))))
        Also see https://arxiv.org/abs/1606.08415
    """
    return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
 
 
def swish(x):
    return x * torch.sigmoid(x)
 
 
ACT2FN = {"gelu": gelu, "relu": torch.nn.functional.relu, "swish": swish}
 
 
class BertConfig(object):
    """Configuration class to store the configuration of a `BertModel`.
    """
    def __init__(self,
                 vocab_size_or_config_json_file,
                 hidden_size=768,
                 num_hidden_layers=12,
                 num_attention_heads=12,
                 intermediate_size=3072,
                 hidden_act="gelu",
                 hidden_dropout_prob=0.1,
                 attention_probs_dropout_prob=0.1,
                 max_position_embeddings=512,
                 type_vocab_size=2,
                 initializer_range=0.02):
        """Constructs BertConfig.
        Args:
            vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `BertModel`.
            hidden_size: Size of the encoder layers and the pooler layer.
            num_hidden_layers: Number of hidden layers in the Transformer encoder.
            num_attention_heads: Number of attention heads for each attention layer in
                the Transformer encoder.
            intermediate_size: The size of the "intermediate" (i.e., feed-forward)
                layer in the Transformer encoder.
            hidden_act: The non-linear activation function (function or string) in the
                encoder and pooler. If string, "gelu", "relu" and "swish" are supported.
            hidden_dropout_prob: The dropout probabilitiy for all fully connected
                layers in the embeddings, encoder, and pooler.
            attention_probs_dropout_prob: The dropout ratio for the attention
                probabilities.
            max_position_embeddings: The maximum sequence length that this model might
                ever be used with. Typically set this to something large just in case
                (e.g., 512 or 1024 or 2048).
            type_vocab_size: The vocabulary size of the `token_type_ids` passed into
                `BertModel`.
            initializer_range: The sttdev of the truncated_normal_initializer for
                initializing all weight matrices.
        """
        if isinstance(vocab_size_or_config_json_file, str) or (sys.version_info[0] == 2
                        and isinstance(vocab_size_or_config_json_file, unicode)):
            with open(vocab_size_or_config_json_file, "r", encoding='utf-8') as reader:
                json_config = json.loads(reader.read())
            for key, value in json_config.items():
                self.__dict__[key] = value
        elif isinstance(vocab_size_or_config_json_file, int):    # isinstance() 函数来判断一个对象是否是一个已知的类型，类似 type()
            self.vocab_size = vocab_size_or_config_json_file
            self.hidden_size = hidden_size
            self.num_hidden_layers = num_hidden_layers
            self.num_attention_heads = num_attention_heads
            self.hidden_act = hidden_act
            self.intermediate_size = intermediate_size
            self.hidden_dropout_prob = hidden_dropout_prob
            self.attention_probs_dropout_prob = attention_probs_dropout_prob
            self.max_position_embeddings = max_position_embeddings
            self.type_vocab_size = type_vocab_size
            self.initializer_range = initializer_range
        else:
            raise ValueError("First argument must be either a vocabulary size (int)"
                             "or the path to a pretrained model config file (str)")
 
    @classmethod
    def from_dict(cls, json_object):
        """Constructs a `BertConfig` from a Python dictionary of parameters."""
        config = BertConfig(vocab_size_or_config_json_file=-1)
        for key, value in json_object.items():
            config.__dict__[key] = value
        return config
 
    @classmethod
    def from_json_file(cls, json_file):
        """Constructs a `BertConfig` from a json file of parameters."""
        with open(json_file, "r", encoding='utf-8') as reader:
            text = reader.read()
        return cls.from_dict(json.loads(text))
 
    def __repr__(self):
        return str(self.to_json_string())
 
    def to_dict(self):
        """Serializes this instance to a Python dictionary."""
        output = copy.deepcopy(self.__dict__)
        return output
 
    def to_json_string(self):
        """Serializes this instance to a JSON string."""
        return json.dumps(self.to_dict(), indent=2, sort_keys=True) + "\n"
 
    def to_json_file(self, json_file_path):
        """ Save this instance to a json file."""
        with open(json_file_path, "w", encoding='utf-8') as writer:
            writer.write(self.to_json_string())
 
try:
    from apex.normalization.fused_layer_norm import FusedLayerNorm as BertLayerNorm
except ImportError:
    logger.info("Better speed can be achieved with apex installed from https://www.github.com/nvidia/apex .")
    class BertLayerNorm(nn.Module):
        def __init__(self, hidden_size, eps=1e-12):
            """Construct a layernorm module in the TF style (epsilon inside the square root).
            """
            super(BertLayerNorm, self).__init__()
            self.weight = nn.Parameter(torch.ones(hidden_size))
            self.bias = nn.Parameter(torch.zeros(hidden_size))
            self.variance_epsilon = eps
 
        def forward(self, x):
            u = x.mean(-1, keepdim=True)  # 求均值
            s = (x - u).pow(2).mean(-1, keepdim=True)  # 求方差
            x = (x - u) / torch.sqrt(s + self.variance_epsilon)
            return self.weight * x + self.bias
 
class BertEmbeddings(nn.Module):
    """Construct the embeddings from word, position and token_type embeddings.
    """
    def __init__(self, config):
        super(BertEmbeddings, self).__init__()
        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=0)  # 21128,768
        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size) # 521,76
        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)  # 2,768
 
        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
        # any TensorFlow checkpoint file
        self.LayerNorm = BertLayerNorm(config.hidden_size, eps=1e-12)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
 
    '''
    BERT的模型结构里没有递归或循环，为了使模型能有效的利用模型特征，我们需要加入序列中各个token相对位置或token在序列中的绝对位置信息
    做法是对不同位置随机初始化一个position embedding，将其加到token embedding上输入模型，作为参数进行学习训练
    绝对位置编码的优点是简单，但位置之间没有约束关系，我们只能期待模型隐形的学习它们之间的关系
    在transformer中，提出了相对位置编码
    '''
    def forward(self, input_ids, token_type_ids=None):
        # print('input',input_ids.size())   torch.Size([128, 32])  输入为[batch_size, seq_len]
        # [[ 101,  860, 7741,  ...,    0,    0,    0],..
        #  [ 101, 8183, 2399,  ...,    0,    0,    0]]    101就是[CLS]对于的token_id
        seq_length = input_ids.size(1)
        position_ids = torch.arange(seq_length, dtype=torch.long, device=input_ids.device)
        # 初始化position [0,1,..,31]
        position_ids = position_ids.unsqueeze(0).expand_as(input_ids)
        # [128, 32]  [[ 0,  1,  2,  ..., 29, 30, 31],..
        #         [ 0,  1,  2,  ..., 29, 30, 31]]  (128个)
        if token_type_ids is None:   # 如果不需要区分token type，也就是说只有一个句子输入时
            token_type_ids = torch.zeros_like(input_ids)
 
        words_embeddings = self.word_embeddings(input_ids)
        position_embeddings = self.position_embeddings(position_ids)
        token_type_embeddings = self.token_type_embeddings(token_type_ids)
 
        embeddings = words_embeddings + position_embeddings + token_type_embeddings
        # 四个shape都是[128, 32, 768]，即(batch_size, sequence_length, hidden_size)
        embeddings = self.LayerNorm(embeddings)
        embeddings = self.dropout(embeddings)
        # 为什么这里用layernorm+dropout 而不是batchnorm？
        # https://www.zhihu.com/question/395811291/answer/1260290120
        return embeddings
 
 
class BertSelfAttention(nn.Module):
    def __init__(self, config):
        super(BertSelfAttention, self).__init__()
        if config.hidden_size % config.num_attention_heads != 0:
            raise ValueError(
                "The hidden size (%d) is not a multiple of the number of attention "
                "heads (%d)" % (config.hidden_size, config.num_attention_heads))
        # hidden_size必须是num_attention_heads的整数倍，以这里的bert-base为例，
        # 每个attention包含12个head，hidden_size是768，所以每个head大小即attention_head_size=768/12=64
        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
        # 12*64=768 之所以不用hidden_size,因为有剪枝prune_heads操作，这里代码没有写
 
        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)
 
    def transpose_for_scores(self, x):
        # 把hidden_size拆成多个头输出的形状，并且将中间两维转置以进行矩阵相乘；
        # x一般就是模型的输入，也就是我们刚才得到的embedding[128, 32, 768]
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size) # [128, 32, 12, 64]
        # 注意这里的写法，x.size()[:-1]的意思是得到x的前两个纬度，shape为[128, 32]
        # print(new_x_shape)就为torch.Size([128, 32, 12, 64])  这个变量就代表一个shape，它不是一个向量
        x = x.view(*new_x_shape)  # [128, 32, 12, 64]  这里是使x变为new_x_shape这个形状
        # *是为什么：多看官方文档  https://pytorch.org/docs/stable/tensors.html#torch.Tensor.view
        return x.permute(0, 2, 1, 3)
 
    def forward(self, hidden_states, attention_mask):
        # attention_mask：[128,1,1,32]就是bert的输入mask扩充了两个纬度
        # 比如y=[6,3],shape为[2],y.unsqueeze(0)，它的纬度就是[1,2],值为[[6,3]]
        mixed_query_layer = self.query(hidden_states)
        mixed_key_layer = self.key(hidden_states)
        mixed_value_layer = self.value(hidden_states)
 
        query_layer = self.transpose_for_scores(mixed_query_layer)  # 这里纬度怎么相乘的可以自己算一下
        key_layer = self.transpose_for_scores(mixed_key_layer)
        value_layer = self.transpose_for_scores(mixed_value_layer)
        # (batch_size, num_attention_heads, sequence_length, attention_head_size)
 
        # 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))
        # [128, 12, 32, 32]  (batch_size, num_attention_heads, sequence_length, sequence_length)
 
        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        # 除以根号dk
        # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
        attention_scores = attention_scores + attention_mask  # 这里的attention_mask是为了对句子进行padding
        # 但输入的mask不是[1, 1, 1, 1, 0, 0]这样的吗，形状[128, 32]，按理应该相乘的
        # 但其实attention_mask被动过手脚了  (BertModel函数的extended_attention_mask处)
        # print一下 [[    -0.,     -0.,     -0.,  ..., -10000., -10000., -10000.]]],..
        #         [[[    -0.,     -0.,     -0.,  ..., -10000., -10000., -10000.]]],..
        # 将原本为1的部分变为0，而原本为0的部分（即padding）变为一个较大的负数，这样相加就得到了一个较大的负值
        # 这样一来经过softmax操作以后这一项就会变成接近0的数，实现了padding的目的
 
        # Normalize the attention scores to probabilities.
        attention_probs = nn.Softmax(dim=-1)(attention_scores)
 
        # 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)
 
        context_layer = torch.matmul(attention_probs, value_layer)
        # (batch_size, num_attention_heads, sequence_length, attention_head_size)
        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)  # [128, 32, 768]
        context_layer = context_layer.view(*new_context_layer_shape)  # 多头注意力concat
        return context_layer
 
 
class BertSelfOutput(nn.Module):
    def __init__(self, config):
        super(BertSelfOutput, self).__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.LayerNorm = BertLayerNorm(config.hidden_size, eps=1e-12)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
 
    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        # 残差连接 对应公式LayerNorm(s+sublayer(x)) 也就是Add&Norm
        return hidden_states
 
 
class BertAttention(nn.Module):
    def __init__(self, config):
        super(BertAttention, self).__init__()
        self.self = BertSelfAttention(config)
        self.output = BertSelfOutput(config)
 
    def forward(self, input_tensor, attention_mask):  # bert的两个输入
        # input_tensor.size():[128, 32, 768] attention_mask.size():[128, 1, 1, 32]
        self_output = self.self(input_tensor, attention_mask)
        attention_output = self.output(self_output, input_tensor)
        return attention_output
 
 
class BertIntermediate(nn.Module):
    """
    全连接+激活 中间层的目的是为了对齐维度
    对应论文的FFN
    """
    def __init__(self, config):
        super(BertIntermediate, self).__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
        if isinstance(config.hidden_act, str) or (sys.version_info[0] == 2 and isinstance(config.hidden_act, unicode)):
            self.intermediate_act_fn = ACT2FN[config.hidden_act]
        else:
            self.intermediate_act_fn = config.hidden_act
 
    def forward(self, hidden_states):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states
 
 
class BertOutput(nn.Module):
    """又是一个全连接+dropout+LayerNorm，还有一个残差连接residual connect"""
    def __init__(self, config):
        super(BertOutput, self).__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
        self.LayerNorm = BertLayerNorm(config.hidden_size, eps=1e-12)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
 
    def forward(self, hidden_states, input_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 BertLayer(nn.Module):
    def __init__(self, config):
        super(BertLayer, self).__init__()
        self.attention = BertAttention(config)
        self.intermediate = BertIntermediate(config)
        self.output = BertOutput(config)
 
    def forward(self, hidden_states, attention_mask):
        attention_output = self.attention(hidden_states, attention_mask)
        intermediate_output = self.intermediate(attention_output)
        layer_output = self.output(intermediate_output, attention_output)
        return layer_output
 
 
class BertEncoder(nn.Module):
    def __init__(self, config):
        super(BertEncoder, self).__init__()
        layer = BertLayer(config)
        self.layer = nn.ModuleList([copy.deepcopy(layer) for _ in range(config.num_hidden_layers)])
        # 多层encoder的写法
 
    def forward(self, hidden_states, attention_mask, output_all_encoded_layers=True):
        all_encoder_layers = []
        for layer_module in self.layer:
            hidden_states = layer_module(hidden_states, attention_mask)
            if output_all_encoded_layers:
                all_encoder_layers.append(hidden_states)
        if not output_all_encoded_layers:
            all_encoder_layers.append(hidden_states)
        return all_encoder_layers
 
 
class BertPooler(nn.Module):
    """
    这一层只是简单地取出了句子的第一个token，即[CLS]对应的向量
    pooling还有其他方式，如avgpool，maxpool
    """
    def __init__(self, config):
        super(BertPooler, self).__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.activation = nn.Tanh()
 
    def forward(self, hidden_states):
        # 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 BertPredictionHeadTransform(nn.Module):
    def __init__(self, config):
        super(BertPredictionHeadTransform, self).__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        if isinstance(config.hidden_act, str) or (sys.version_info[0] == 2 and isinstance(config.hidden_act, unicode)):
            self.transform_act_fn = ACT2FN[config.hidden_act]
        else:
            self.transform_act_fn = config.hidden_act
        self.LayerNorm = BertLayerNorm(config.hidden_size, eps=1e-12)
 
    def forward(self, hidden_states):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.transform_act_fn(hidden_states)
        hidden_states = self.LayerNorm(hidden_states)
        return hidden_states
 
 
class BertLMPredictionHead(nn.Module):
    def __init__(self, config, bert_model_embedding_weights):
        super(BertLMPredictionHead, self).__init__()
        self.transform = BertPredictionHeadTransform(config)
 
        # The output weights are the same as the input embeddings, but there is
        # an output-only bias for each token.
        self.decoder = nn.Linear(bert_model_embedding_weights.size(1),
                                 bert_model_embedding_weights.size(0),
                                 bias=False)
        self.decoder.weight = bert_model_embedding_weights
        self.bias = nn.Parameter(torch.zeros(bert_model_embedding_weights.size(0)))
 
    def forward(self, hidden_states):
        hidden_states = self.transform(hidden_states)
        hidden_states = self.decoder(hidden_states) + self.bias
        return hidden_states  # [batch_size, seq_length, vocab_size]，即预测每个句子每个词是什么类别的概率值
 
 
class BertOnlyMLMHead(nn.Module):
    def __init__(self, config, bert_model_embedding_weights):
        super(BertOnlyMLMHead, self).__init__()
        self.predictions = BertLMPredictionHead(config, bert_model_embedding_weights)
 
    def forward(self, sequence_output):
        prediction_scores = self.predictions(sequence_output)
        return prediction_scores
 
 
class BertOnlyNSPHead(nn.Module):
    def __init__(self, config):
        super(BertOnlyNSPHead, self).__init__()
        self.seq_relationship = nn.Linear(config.hidden_size, 2)
 
    def forward(self, pooled_output):
        seq_relationship_score = self.seq_relationship(pooled_output)
        return seq_relationship_score
 
 
class BertPreTrainingHeads(nn.Module):
    def __init__(self, config, bert_model_embedding_weights):
        super(BertPreTrainingHeads, self).__init__()
        self.predictions = BertLMPredictionHead(config, bert_model_embedding_weights)
        self.seq_relationship = nn.Linear(config.hidden_size, 2)
 
    def forward(self, sequence_output, pooled_output):
        prediction_scores = self.predictions(sequence_output)
        seq_relationship_score = self.seq_relationship(pooled_output)
        return prediction_scores, seq_relationship_score
 
 
class BertPreTrainedModel(nn.Module):
    """ An abstract class to handle weights initialization and
        a simple interface for dowloading and loading pretrained models.
    """
    def __init__(self, config, *inputs, **kwargs):
        super(BertPreTrainedModel, self).__init__()
        if not isinstance(config, BertConfig):
            raise ValueError(
                "Parameter config in `{}(config)` should be an instance of class `BertConfig`. "
                "To create a model from a Google pretrained model use "
                "`model = {}.from_pretrained(PRETRAINED_MODEL_NAME)`".format(
                    self.__class__.__name__, self.__class__.__name__
                ))
        self.config = config
 
    def init_bert_weights(self, module):
        """ Initialize the weights.
        """
        if isinstance(module, (nn.Linear, nn.Embedding)):
            # 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)
        elif isinstance(module, BertLayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)
        if isinstance(module, nn.Linear) and module.bias is not None:
            module.bias.data.zero_()
 
    @classmethod
    def from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs):
        """
        Instantiate a BertPreTrainedModel from a pre-trained model file or a pytorch state dict.
        Download and cache the pre-trained model file if needed.
        Params:
            pretrained_model_name_or_path: either:
                - a str with the name of a pre-trained model to load selected in the list of:
                    . `bert-base-uncased`
                    . `bert-large-uncased`
                    . `bert-base-cased`
                    . `bert-large-cased`
                    . `bert-base-multilingual-uncased`
                    . `bert-base-multilingual-cased`
                    . `bert-base-chinese`
                - a path or url to a pretrained model archive containing:
                    . `bert_config.json` a configuration file for the model
                    . `pytorch_model.bin` a PyTorch dump of a BertForPreTraining instance
                - a path or url to a pretrained model archive containing:
                    . `bert_config.json` a configuration file for the model
                    . `model.chkpt` a TensorFlow checkpoint
            from_tf: should we load the weights from a locally saved TensorFlow checkpoint
            cache_dir: an optional path to a folder in which the pre-trained models will be cached.
            state_dict: an optional state dictionnary (collections.OrderedDict object) to use instead of Google pre-trained models
            *inputs, **kwargs: additional input for the specific Bert class
                (ex: num_labels for BertForSequenceClassification)
        """
        state_dict = kwargs.get('state_dict', None)
        kwargs.pop('state_dict', None)
        cache_dir = kwargs.get('cache_dir', None)
        kwargs.pop('cache_dir', None)
        from_tf = kwargs.get('from_tf', False)
        kwargs.pop('from_tf', None)
 
        if pretrained_model_name_or_path in PRETRAINED_MODEL_ARCHIVE_MAP:
            archive_file = PRETRAINED_MODEL_ARCHIVE_MAP[pretrained_model_name_or_path]
        else:
            archive_file = pretrained_model_name_or_path
        # redirect to the cache, if necessary
        try:
            resolved_archive_file = cached_path(archive_file, cache_dir=cache_dir)
        except EnvironmentError:
            logger.error(
                "Model name '{}' was not found in model name list ({}). "
                "We assumed '{}' was a path or url but couldn't find any file "
                "associated to this path or url.".format(
                    pretrained_model_name_or_path,
                    ', '.join(PRETRAINED_MODEL_ARCHIVE_MAP.keys()),
                    archive_file))
            return None
        if resolved_archive_file == archive_file:
            logger.info("loading archive file {}".format(archive_file))
        else:
            logger.info("loading archive file {} from cache at {}".format(
                archive_file, resolved_archive_file))
        tempdir = None
        if os.path.isdir(resolved_archive_file) or from_tf:
            serialization_dir = resolved_archive_file
        else:
            # Extract archive to temp dir
            tempdir = tempfile.mkdtemp()
            logger.info("extracting archive file {} to temp dir {}".format(
                resolved_archive_file, tempdir))
            with tarfile.open(resolved_archive_file, 'r:gz') as archive:
                archive.extractall(tempdir)
            serialization_dir = tempdir
        # Load config
        config_file = os.path.join(serialization_dir, CONFIG_NAME)
        if not os.path.exists(config_file):
            # Backward compatibility with old naming format
            config_file = os.path.join(serialization_dir, BERT_CONFIG_NAME)
        config = BertConfig.from_json_file(config_file)
        logger.info("Model config {}".format(config))
        # Instantiate model.
        model = cls(config, *inputs, **kwargs)
        if state_dict is None and not from_tf:
            weights_path = os.path.join(serialization_dir, WEIGHTS_NAME)
            state_dict = torch.load(weights_path, map_location='cpu')
        if tempdir:
            # Clean up temp dir
            shutil.rmtree(tempdir)
        if from_tf:
            # Directly load from a TensorFlow checkpoint
            weights_path = os.path.join(serialization_dir, TF_WEIGHTS_NAME)
            return load_tf_weights_in_bert(model, weights_path)
        # Load from a PyTorch state_dict
        old_keys = []
        new_keys = []
        for key in state_dict.keys():
            new_key = None
            if 'gamma' in key:
                new_key = key.replace('gamma', 'weight')
            if 'beta' in key:
                new_key = key.replace('beta', 'bias')
            if new_key:
                old_keys.append(key)
                new_keys.append(new_key)
        for old_key, new_key in zip(old_keys, new_keys):
            state_dict[new_key] = state_dict.pop(old_key)
 
        missing_keys = []
        unexpected_keys = []
        error_msgs = []
        # copy state_dict so _load_from_state_dict can modify it
        metadata = getattr(state_dict, '_metadata', None)
        state_dict = state_dict.copy()
        if metadata is not None:
            state_dict._metadata = metadata
 
        def load(module, prefix=''):
            local_metadata = {} if metadata is None else metadata.get(prefix[:-1], {})
            module._load_from_state_dict(
                state_dict, prefix, local_metadata, True, missing_keys, unexpected_keys, error_msgs)
            for name, child in module._modules.items():
                if child is not None:
                    load(child, prefix + name + '.')
        start_prefix = ''
        if not hasattr(model, 'bert') and any(s.startswith('bert.') for s in state_dict.keys()):
            start_prefix = 'bert.'
        load(model, prefix=start_prefix)
        if len(missing_keys) > 0:
            logger.info("Weights of {} not initialized from pretrained model: {}".format(
                model.__class__.__name__, missing_keys))
        if len(unexpected_keys) > 0:
            logger.info("Weights from pretrained model not used in {}: {}".format(
                model.__class__.__name__, unexpected_keys))
        if len(error_msgs) > 0:
            raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
                               model.__class__.__name__, "\n\t".join(error_msgs)))
        return model
 
 
class BertModel(BertPreTrainedModel):
    """BERT model ("Bidirectional Embedding Representations from a Transformer").
    Params:
        config: a BertConfig class instance with the configuration to build a new model
    Inputs:
        `input_ids`: a torch.LongTensor of shape [batch_size, sequence_length]
            with the word token indices in the vocabulary(see the tokens preprocessing logic in the scripts
            `extract_features.py`, `run_classifier.py` and `run_squad.py`)
        `token_type_ids`: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token
            types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to
            a `sentence B` token (see BERT paper for more details).
        `attention_mask`: an optional torch.LongTensor of shape [batch_size, sequence_length] with indices
            selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max
            input sequence length in the current batch. It's the mask that we typically use for attention when
            a batch has varying length sentences.
        `output_all_encoded_layers`: boolean which controls the content of the `encoded_layers` output as described below. Default: `True`.
    Outputs: Tuple of (encoded_layers, pooled_output)
        `encoded_layers`: controled by `output_all_encoded_layers` argument:
            - `output_all_encoded_layers=True`: outputs a list of the full sequences of encoded-hidden-states at the end
                of each attention block (i.e. 12 full sequences for BERT-base, 24 for BERT-large), each
                encoded-hidden-state is a torch.FloatTensor of size [batch_size, sequence_length, hidden_size],
            - `output_all_encoded_layers=False`: outputs only the full sequence of hidden-states corresponding
                to the last attention block of shape [batch_size, sequence_length, hidden_size],
        `pooled_output`: a torch.FloatTensor of size [batch_size, hidden_size] which is the output of a
            classifier pretrained on top of the hidden state associated to the first character of the
            input (`CLS`) to train on the Next-Sentence task (see BERT's paper).
    Example usage:
    ```python
    # Already been converted into WordPiece token ids
    input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]])
    input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]])
    token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]])
    config = modeling.BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768,
        num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072)
    model = modeling.BertModel(config=config)
    all_encoder_layers, pooled_output = model(input_ids, token_type_ids, input_mask)
    ```
    """
    def __init__(self, config):
        super(BertModel, self).__init__(config)
        self.embeddings = BertEmbeddings(config)
        self.encoder = BertEncoder(config)
        self.pooler = BertPooler(config)
        self.apply(self.init_bert_weights)
 
    def forward(self, input_ids, token_type_ids=None, attention_mask=None, output_all_encoded_layers=True):
        if attention_mask is None:
            attention_mask = torch.ones_like(input_ids)
        if token_type_ids is None:
            token_type_ids = torch.zeros_like(input_ids)
 
        # We create a 3D attention mask from a 2D tensor mask.
        # Sizes are [batch_size, 1, 1, to_seq_length]
        # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
        # this attention mask is more simple than the triangular masking of causal attention
        # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
        extended_attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)
 
        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
        # masked positions, this operation will create a tensor which is 0.0 for
        # positions we want to attend and -10000.0 for masked positions.
        # Since we are adding it to the raw scores before the softmax, this is
        # effectively the same as removing these entirely.
        extended_attention_mask = extended_attention_mask.to(dtype=next(self.parameters()).dtype) # fp16 compatibility
        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
 
        embedding_output = self.embeddings(input_ids, token_type_ids)
        encoded_layers = self.encoder(embedding_output,
                                      extended_attention_mask,
                                      output_all_encoded_layers=output_all_encoded_layers)
        sequence_output = encoded_layers[-1]  # 即句子对应的向量，也是CLS向量
        pooled_output = self.pooler(sequence_output)
        if not output_all_encoded_layers:
            encoded_layers = encoded_layers[-1]
        return encoded_layers, pooled_output
 
 
class BertForPreTraining(BertPreTrainedModel):
    """BERT model with pre-training heads.
    This module comprises the BERT model followed by the two pre-training heads:
        - the masked language modeling head, and
        - the next sentence classification head.
    Params:
        config: a BertConfig class instance with the configuration to build a new model.
    Inputs:
        `input_ids`: a torch.LongTensor of shape [batch_size, sequence_length]
            with the word token indices in the vocabulary(see the tokens preprocessing logic in the scripts
            `extract_features.py`, `run_classifier.py` and `run_squad.py`)
        `token_type_ids`: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token
            types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to
            a `sentence B` token (see BERT paper for more details).
        `attention_mask`: an optional torch.LongTensor of shape [batch_size, sequence_length] with indices
            selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max
            input sequence length in the current batch. It's the mask that we typically use for attention when
            a batch has varying length sentences.
        `masked_lm_labels`: optional masked language modeling labels: torch.LongTensor of shape [batch_size, sequence_length]
            with indices selected in [-1, 0, ..., vocab_size]. All labels set to -1 are ignored (masked), the loss
            is only computed for the labels set in [0, ..., vocab_size]
        `next_sentence_label`: optional next sentence classification loss: torch.LongTensor of shape [batch_size]
            with indices selected in [0, 1].
            0 => next sentence is the continuation, 1 => next sentence is a random sentence.
    Outputs:
        if `masked_lm_labels` and `next_sentence_label` are not `None`:
            Outputs the total_loss which is the sum of the masked language modeling loss and the next
            sentence classification loss.
        if `masked_lm_labels` or `next_sentence_label` is `None`:
            Outputs a tuple comprising
            - the masked language modeling logits of shape [batch_size, sequence_length, vocab_size], and
            - the next sentence classification logits of shape [batch_size, 2].
    Example usage:
    ```python
    # Already been converted into WordPiece token ids
    input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]])
    input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]])
    token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]])
    config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768,
        num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072)
    model = BertForPreTraining(config)
    masked_lm_logits_scores, seq_relationship_logits = model(input_ids, token_type_ids, input_mask)
    ```
    """
    def __init__(self, config):
        super(BertForPreTraining, self).__init__(config)
        self.bert = BertModel(config)
        self.cls = BertPreTrainingHeads(config, self.bert.embeddings.word_embeddings.weight)
        self.apply(self.init_bert_weights)
 
    def forward(self, input_ids, token_type_ids=None, attention_mask=None, masked_lm_labels=None, next_sentence_label=None):
        sequence_output, pooled_output = self.bert(input_ids, token_type_ids, attention_mask,
                                                   output_all_encoded_layers=False)
        prediction_scores, seq_relationship_score = self.cls(sequence_output, pooled_output)
 
        if masked_lm_labels is not None and next_sentence_label is not None:
            loss_fct = CrossEntropyLoss(ignore_index=-1)
            masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), masked_lm_labels.view(-1))
            next_sentence_loss = loss_fct(seq_relationship_score.view(-1, 2), next_sentence_label.view(-1))
            total_loss = masked_lm_loss + next_sentence_loss
            return total_loss
        else:
            return prediction_scores, seq_relationship_score
 
 
class BertForMaskedLM(BertPreTrainedModel):
    """BERT model with the masked language modeling head.
    This module comprises the BERT model followed by the masked language modeling head.
    Params:
        config: a BertConfig class instance with the configuration to build a new model.
    Inputs:
        `input_ids`: a torch.LongTensor of shape [batch_size, sequence_length]
            with the word token indices in the vocabulary(see the tokens preprocessing logic in the scripts
            `extract_features.py`, `run_classifier.py` and `run_squad.py`)
        `token_type_ids`: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token
            types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to
            a `sentence B` token (see BERT paper for more details).
        `attention_mask`: an optional torch.LongTensor of shape [batch_size, sequence_length] with indices
            selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max
            input sequence length in the current batch. It's the mask that we typically use for attention when
            a batch has varying length sentences.
        `masked_lm_labels`: masked language modeling labels: torch.LongTensor of shape [batch_size, sequence_length]
            with indices selected in [-1, 0, ..., vocab_size]. All labels set to -1 are ignored (masked), the loss
            is only computed for the labels set in [0, ..., vocab_size]
    Outputs:
        if `masked_lm_labels` is  not `None`:
            Outputs the masked language modeling loss.
        if `masked_lm_labels` is `None`:
            Outputs the masked language modeling logits of shape [batch_size, sequence_length, vocab_size].
    Example usage:
    ```python
    # Already been converted into WordPiece token ids
    input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]])
    input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]])
    token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]])
    config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768,
        num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072)
    model = BertForMaskedLM(config)
    masked_lm_logits_scores = model(input_ids, token_type_ids, input_mask)
    ```
    """
    def __init__(self, config):
        super(BertForMaskedLM, self).__init__(config)
        self.bert = BertModel(config)
        self.cls = BertOnlyMLMHead(config, self.bert.embeddings.word_embeddings.weight)
        self.apply(self.init_bert_weights)
 
    def forward(self, input_ids, token_type_ids=None, attention_mask=None, masked_lm_labels=None):
        sequence_output, _ = self.bert(input_ids, token_type_ids, attention_mask,
                                       output_all_encoded_layers=False)
        prediction_scores = self.cls(sequence_output)
 
        if masked_lm_labels is not None:
            loss_fct = CrossEntropyLoss(ignore_index=-1)
            masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), masked_lm_labels.view(-1))
            return masked_lm_loss
        else:
            return prediction_scores
 
 
class BertForNextSentencePrediction(BertPreTrainedModel):
    """BERT model with next sentence prediction head.
    This module comprises the BERT model followed by the next sentence classification head.
    Params:
        config: a BertConfig class instance with the configuration to build a new model.
    Inputs:
        `input_ids`: a torch.LongTensor of shape [batch_size, sequence_length]
            with the word token indices in the vocabulary(see the tokens preprocessing logic in the scripts
            `extract_features.py`, `run_classifier.py` and `run_squad.py`)
        `token_type_ids`: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token
            types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to
            a `sentence B` token (see BERT paper for more details).
        `attention_mask`: an optional torch.LongTensor of shape [batch_size, sequence_length] with indices
            selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max
            input sequence length in the current batch. It's the mask that we typically use for attention when
            a batch has varying length sentences.
        `next_sentence_label`: next sentence classification loss: torch.LongTensor of shape [batch_size]
            with indices selected in [0, 1].
            0 => next sentence is the continuation, 1 => next sentence is a random sentence.
    Outputs:
        if `next_sentence_label` is not `None`:
            Outputs the total_loss which is the sum of the masked language modeling loss and the next
            sentence classification loss.
        if `next_sentence_label` is `None`:
            Outputs the next sentence classification logits of shape [batch_size, 2].
    Example usage:
    ```python
    # Already been converted into WordPiece token ids
    input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]])
    input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]])
    token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]])
    config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768,
        num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072)
    model = BertForNextSentencePrediction(config)
    seq_relationship_logits = model(input_ids, token_type_ids, input_mask)
    ```
    """
    def __init__(self, config):
        super(BertForNextSentencePrediction, self).__init__(config)
        self.bert = BertModel(config)
        self.cls = BertOnlyNSPHead(config)
        self.apply(self.init_bert_weights)
 
    def forward(self, input_ids, token_type_ids=None, attention_mask=None, next_sentence_label=None):
        _, pooled_output = self.bert(input_ids, token_type_ids, attention_mask,
                                     output_all_encoded_layers=False)
        seq_relationship_score = self.cls( pooled_output)
 
        if next_sentence_label is not None:
            loss_fct = CrossEntropyLoss(ignore_index=-1)
            next_sentence_loss = loss_fct(seq_relationship_score.view(-1, 2), next_sentence_label.view(-1))
            return next_sentence_loss
        else:
            return seq_relationship_score
 
 
class BertForSequenceClassification(BertPreTrainedModel):
    """BERT model for classification.
    This module is composed of the BERT model with a linear layer on top of
    the pooled output.
    Params:
        `config`: a BertConfig class instance with the configuration to build a new model.
        `num_labels`: the number of classes for the classifier. Default = 2.
    Inputs:
        `input_ids`: a torch.LongTensor of shape [batch_size, sequence_length]
            with the word token indices in the vocabulary. Items in the batch should begin with the special "CLS" token. (see the tokens preprocessing logic in the scripts
            `extract_features.py`, `run_classifier.py` and `run_squad.py`)
        `token_type_ids`: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token
            types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to
            a `sentence B` token (see BERT paper for more details).
        `attention_mask`: an optional torch.LongTensor of shape [batch_size, sequence_length] with indices
            selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max
            input sequence length in the current batch. It's the mask that we typically use for attention when
            a batch has varying length sentences.
        `labels`: labels for the classification output: torch.LongTensor of shape [batch_size]
            with indices selected in [0, ..., num_labels].
    Outputs:
        if `labels` is not `None`:
            Outputs the CrossEntropy classification loss of the output with the labels.
        if `labels` is `None`:
            Outputs the classification logits of shape [batch_size, num_labels].
    Example usage:
    ```python
    # Already been converted into WordPiece token ids
    input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]])
    input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]])
    token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]])
    config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768,
        num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072)
    num_labels = 2
    model = BertForSequenceClassification(config, num_labels)
    logits = model(input_ids, token_type_ids, input_mask)
    ```
    """
    def __init__(self, config, num_labels):
        super(BertForSequenceClassification, self).__init__(config)
        self.num_labels = num_labels
        self.bert = BertModel(config)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        self.classifier = nn.Linear(config.hidden_size, num_labels)
        self.apply(self.init_bert_weights)
 
    def forward(self, input_ids, token_type_ids=None, attention_mask=None, labels=None):
        _, pooled_output = self.bert(input_ids, token_type_ids, attention_mask, output_all_encoded_layers=False)
        pooled_output = self.dropout(pooled_output)
        logits = self.classifier(pooled_output)
 
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
            return loss
        else:
            return logits
 
 
class BertForMultipleChoice(BertPreTrainedModel):
    """BERT model for multiple choice tasks.
    This module is composed of the BERT model with a linear layer on top of
    the pooled output.
    Params:
        `config`: a BertConfig class instance with the configuration to build a new model.
        `num_choices`: the number of classes for the classifier. Default = 2.
    Inputs:
        `input_ids`: a torch.LongTensor of shape [batch_size, num_choices, sequence_length]
            with the word token indices in the vocabulary(see the tokens preprocessing logic in the scripts
            `extract_features.py`, `run_classifier.py` and `run_squad.py`)
        `token_type_ids`: an optional torch.LongTensor of shape [batch_size, num_choices, sequence_length]
            with the token types indices selected in [0, 1]. Type 0 corresponds to a `sentence A`
            and type 1 corresponds to a `sentence B` token (see BERT paper for more details).
        `attention_mask`: an optional torch.LongTensor of shape [batch_size, num_choices, sequence_length] with indices
            selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max
            input sequence length in the current batch. It's the mask that we typically use for attention when
            a batch has varying length sentences.
        `labels`: labels for the classification output: torch.LongTensor of shape [batch_size]
            with indices selected in [0, ..., num_choices].
    Outputs:
        if `labels` is not `None`:
            Outputs the CrossEntropy classification loss of the output with the labels.
        if `labels` is `None`:
            Outputs the classification logits of shape [batch_size, num_labels].
    Example usage:
    ```python
    # Already been converted into WordPiece token ids
    input_ids = torch.LongTensor([[[31, 51, 99], [15, 5, 0]], [[12, 16, 42], [14, 28, 57]]])
    input_mask = torch.LongTensor([[[1, 1, 1], [1, 1, 0]],[[1,1,0], [1, 0, 0]]])
    token_type_ids = torch.LongTensor([[[0, 0, 1], [0, 1, 0]],[[0, 1, 1], [0, 0, 1]]])
    config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768,
        num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072)
    num_choices = 2
    model = BertForMultipleChoice(config, num_choices)
    logits = model(input_ids, token_type_ids, input_mask)
    ```
    """
    def __init__(self, config, num_choices):
        super(BertForMultipleChoice, self).__init__(config)
        self.num_choices = num_choices
        self.bert = BertModel(config)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        self.classifier = nn.Linear(config.hidden_size, 1)
        self.apply(self.init_bert_weights)
 
    def forward(self, input_ids, token_type_ids=None, attention_mask=None, labels=None):
        flat_input_ids = input_ids.view(-1, input_ids.size(-1))
        flat_token_type_ids = token_type_ids.view(-1, token_type_ids.size(-1)) if token_type_ids is not None else None
        flat_attention_mask = attention_mask.view(-1, attention_mask.size(-1)) if attention_mask is not None else None
        _, pooled_output = self.bert(flat_input_ids, flat_token_type_ids, flat_attention_mask, output_all_encoded_layers=False)
        pooled_output = self.dropout(pooled_output)
        logits = self.classifier(pooled_output)
        reshaped_logits = logits.view(-1, self.num_choices)
 
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(reshaped_logits, labels)
            return loss
        else:
            return reshaped_logits
 
 
class BertForTokenClassification(BertPreTrainedModel):
    """BERT model for token-level classification.
    This module is composed of the BERT model with a linear layer on top of
    the full hidden state of the last layer.
    Params:
        `config`: a BertConfig class instance with the configuration to build a new model.
        `num_labels`: the number of classes for the classifier. Default = 2.
    Inputs:
        `input_ids`: a torch.LongTensor of shape [batch_size, sequence_length]
            with the word token indices in the vocabulary(see the tokens preprocessing logic in the scripts
            `extract_features.py`, `run_classifier.py` and `run_squad.py`)
        `token_type_ids`: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token
            types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to
            a `sentence B` token (see BERT paper for more details).
        `attention_mask`: an optional torch.LongTensor of shape [batch_size, sequence_length] with indices
            selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max
            input sequence length in the current batch. It's the mask that we typically use for attention when
            a batch has varying length sentences.
        `labels`: labels for the classification output: torch.LongTensor of shape [batch_size, sequence_length]
            with indices selected in [0, ..., num_labels].
    Outputs:
        if `labels` is not `None`:
            Outputs the CrossEntropy classification loss of the output with the labels.
        if `labels` is `None`:
            Outputs the classification logits of shape [batch_size, sequence_length, num_labels].
    Example usage:
    ```python
    # Already been converted into WordPiece token ids
    input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]])
    input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]])
    token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]])
    config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768,
        num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072)
    num_labels = 2
    model = BertForTokenClassification(config, num_labels)
    logits = model(input_ids, token_type_ids, input_mask)
    ```
    """
    def __init__(self, config, num_labels):
        super(BertForTokenClassification, self).__init__(config)
        self.num_labels = num_labels
        self.bert = BertModel(config)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        self.classifier = nn.Linear(config.hidden_size, num_labels)
        self.apply(self.init_bert_weights)
 
    def forward(self, input_ids, token_type_ids=None, attention_mask=None, labels=None):
        sequence_output, _ = self.bert(input_ids, token_type_ids, attention_mask, output_all_encoded_layers=False)
        sequence_output = self.dropout(sequence_output)
        logits = self.classifier(sequence_output)
 
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            # Only keep active parts of the loss
            if attention_mask is not None:
                active_loss = attention_mask.view(-1) == 1
                active_logits = logits.view(-1, self.num_labels)[active_loss]
                active_labels = labels.view(-1)[active_loss]
                loss = loss_fct(active_logits, active_labels)
            else:
                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
            return loss
        else:
            return logits
 
 
class BertForQuestionAnswering(BertPreTrainedModel):
    """BERT model for Question Answering (span extraction).
    This module is composed of the BERT model with a linear layer on top of
    the sequence output that computes start_logits and end_logits
    Params:
        `config`: a BertConfig class instance with the configuration to build a new model.
    Inputs:
        `input_ids`: a torch.LongTensor of shape [batch_size, sequence_length]
            with the word token indices in the vocabulary(see the tokens preprocessing logic in the scripts
            `extract_features.py`, `run_classifier.py` and `run_squad.py`)
        `token_type_ids`: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token
            types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to
            a `sentence B` token (see BERT paper for more details).
        `attention_mask`: an optional torch.LongTensor of shape [batch_size, sequence_length] with indices
            selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max
            input sequence length in the current batch. It's the mask that we typically use for attention when
            a batch has varying length sentences.
        `start_positions`: position of the first token for the labeled span: torch.LongTensor of shape [batch_size].
            Positions are clamped to the length of the sequence and position outside of the sequence are not taken
            into account for computing the loss.
        `end_positions`: position of the last token for the labeled span: torch.LongTensor of shape [batch_size].
            Positions are clamped to the length of the sequence and position outside of the sequence are not taken
            into account for computing the loss.
    Outputs:
        if `start_positions` and `end_positions` are not `None`:
            Outputs the total_loss which is the sum of the CrossEntropy loss for the start and end token positions.
        if `start_positions` or `end_positions` is `None`:
            Outputs a tuple of start_logits, end_logits which are the logits respectively for the start and end
            position tokens of shape [batch_size, sequence_length].
    Example usage:
    ```python
    # Already been converted into WordPiece token ids
    input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]])
    input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]])
    token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]])
    config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768,
        num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072)
    model = BertForQuestionAnswering(config)
    start_logits, end_logits = model(input_ids, token_type_ids, input_mask)
    ```
    """
    def __init__(self, config):
        super(BertForQuestionAnswering, self).__init__(config)
        self.bert = BertModel(config)
        # TODO check with Google if it's normal there is no dropout on the token classifier of SQuAD in the TF version
        # self.dropout = nn.Dropout(config.hidden_dropout_prob)
        self.qa_outputs = nn.Linear(config.hidden_size, 2)
        self.apply(self.init_bert_weights)
    def forward(self, input_ids, token_type_ids=None, attention_mask=None, start_positions=None, end_positions=None):
        sequence_output, _ = self.bert(input_ids, token_type_ids, attention_mask, output_all_encoded_layers=False)
        logits = self.qa_outputs(sequence_output)
        start_logits, end_logits = logits.split(1, dim=-1)
        start_logits = start_logits.squeeze(-1)
        end_logits = end_logits.squeeze(-1)
        if start_positions is not None and end_positions is not None:
            # If we are on multi-GPU, split add a dimension
            if len(start_positions.size()) > 1:
                start_positions = start_positions.squeeze(-1)
            if len(end_positions.size()) > 1:
                end_positions = end_positions.squeeze(-1)
            # sometimes the start/end positions are outside our model inputs, we ignore these terms
            ignored_index = start_logits.size(1)
            start_positions.clamp_(0, ignored_index)
            end_positions.clamp_(0, ignored_index)
            loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
            start_loss = loss_fct(start_logits, start_positions)
            end_loss = loss_fct(end_logits, end_positions)
            total_loss = (start_loss + end_loss) / 2
            return total_loss
        else:
            return start_logits, end_logits