# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
#
# 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.
"""PyTorch BERT model. """
from __future__ import absolute_import, division, print_function, unicode_literals
import json
import logging
import math
import os
import sys
from io import open
import torch
from torch import nn
from torch.nn import CrossEntropyLoss, MSELoss
from .modeling_utils import (WEIGHTS_NAME, CONFIG_NAME, PretrainedConfig, PreTrainedModel,
prune_linear_layer, add_start_docstrings)
logger = logging.getLogger(__name__)
BERT_PRETRAINED_MODEL_ARCHIVE_MAP = {
'bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-pytorch_model.bin",
'bert-large-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-pytorch_model.bin",
'bert-base-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-pytorch_model.bin",
'bert-large-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-pytorch_model.bin",
'bert-base-multilingual-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-uncased-pytorch_model.bin",
'bert-base-multilingual-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-cased-pytorch_model.bin",
'bert-base-chinese': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-chinese-pytorch_model.bin",
'bert-base-german-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-german-cased-pytorch_model.bin",
'bert-large-uncased-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-pytorch_model.bin",
'bert-large-cased-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-pytorch_model.bin",
'bert-large-uncased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-finetuned-squad-pytorch_model.bin",
'bert-large-cased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-finetuned-squad-pytorch_model.bin",
'bert-base-cased-finetuned-mrpc': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-finetuned-mrpc-pytorch_model.bin",
}
BERT_PRETRAINED_CONFIG_ARCHIVE_MAP = {
'bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-config.json",
'bert-large-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-config.json",
'bert-base-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-config.json",
'bert-large-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-config.json",
'bert-base-multilingual-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-uncased-config.json",
'bert-base-multilingual-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-cased-config.json",
'bert-base-chinese': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-chinese-config.json",
'bert-base-german-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-german-cased-config.json",
'bert-large-uncased-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-config.json",
'bert-large-cased-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-config.json",
'bert-large-uncased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-finetuned-squad-config.json",
'bert-large-cased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-finetuned-squad-config.json",
'bert-base-cased-finetuned-mrpc': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-finetuned-mrpc-config.json",
}
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 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)
logger.info("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:
logger.info("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):
logger.info("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:
logger.info("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
logger.info("Initialize PyTorch weight {}".format(name))
pointer.data = torch.from_numpy(array)
return model
def gelu(x):
"""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}
[docs]class BertConfig(PretrainedConfig):
r"""
:class:`~pytorch_transformers.BertConfig` is the configuration class to store the configuration of a
`BertModel`.
Arguments:
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.
layer_norm_eps: The epsilon used by LayerNorm.
"""
pretrained_config_archive_map = BERT_PRETRAINED_CONFIG_ARCHIVE_MAP
def __init__(self,
vocab_size_or_config_json_file=30522,
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,
layer_norm_eps=1e-12,
**kwargs):
super(BertConfig, self).__init__(**kwargs)
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):
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
self.layer_norm_eps = layer_norm_eps
else:
raise ValueError("First argument must be either a vocabulary size (int)"
"or the path to a pretrained model config file (str)")
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)
s = (x - u).pow(2).mean(-1, keepdim=True)
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)
self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
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 = BertLayerNorm(config.hidden_size, eps=config.layer_norm_eps)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, input_ids, token_type_ids=None, position_ids=None):
seq_length = input_ids.size(1)
if position_ids is None:
position_ids = torch.arange(seq_length, dtype=torch.long, device=input_ids.device)
position_ids = position_ids.unsqueeze(0).expand_as(input_ids)
if token_type_ids is None:
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
embeddings = self.LayerNorm(embeddings)
embeddings = self.dropout(embeddings)
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))
self.output_attentions = config.output_attentions
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)
def transpose_for_scores(self, x):
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, attention_mask, head_mask=None):
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)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_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))
attention_scores = attention_scores / math.sqrt(self.attention_head_size)
# 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.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)
# 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 self.output_attentions else (context_layer,)
return outputs
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=config.layer_norm_eps)
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 BertAttention(nn.Module):
def __init__(self, config):
super(BertAttention, self).__init__()
self.self = BertSelfAttention(config)
self.output = BertSelfOutput(config)
def prune_heads(self, heads):
if len(heads) == 0:
return
mask = torch.ones(self.self.num_attention_heads, self.self.attention_head_size)
for head in heads:
mask[head] = 0
mask = mask.view(-1).contiguous().eq(1)
index = torch.arange(len(mask))[mask].long()
# 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
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
def forward(self, input_tensor, attention_mask, head_mask=None):
self_outputs = self.self(input_tensor, attention_mask, head_mask)
attention_output = self.output(self_outputs[0], input_tensor)
outputs = (attention_output,) + self_outputs[1:] # add attentions if we output them
return outputs
class BertIntermediate(nn.Module):
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):
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=config.layer_norm_eps)
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, head_mask=None):
attention_outputs = self.attention(hidden_states, attention_mask, head_mask)
attention_output = attention_outputs[0]
intermediate_output = self.intermediate(attention_output)
layer_output = self.output(intermediate_output, attention_output)
outputs = (layer_output,) + attention_outputs[1:] # add attentions if we output them
return outputs
class BertEncoder(nn.Module):
def __init__(self, config):
super(BertEncoder, self).__init__()
self.output_attentions = config.output_attentions
self.output_hidden_states = config.output_hidden_states
self.layer = nn.ModuleList([BertLayer(config) for _ in range(config.num_hidden_layers)])
def forward(self, hidden_states, attention_mask, head_mask=None):
all_hidden_states = ()
all_attentions = ()
for i, layer_module in enumerate(self.layer):
if self.output_hidden_states:
all_hidden_states = all_hidden_states + (hidden_states,)
layer_outputs = layer_module(hidden_states, attention_mask, head_mask[i])
hidden_states = layer_outputs[0]
if self.output_attentions:
all_attentions = all_attentions + (layer_outputs[1],)
# Add last layer
if self.output_hidden_states:
all_hidden_states = all_hidden_states + (hidden_states,)
outputs = (hidden_states,)
if self.output_hidden_states:
outputs = outputs + (all_hidden_states,)
if self.output_attentions:
outputs = outputs + (all_attentions,)
return outputs # outputs, (hidden states), (attentions)
class BertPooler(nn.Module):
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=config.layer_norm_eps)
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):
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(config.hidden_size,
config.vocab_size,
bias=False)
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
def forward(self, hidden_states):
hidden_states = self.transform(hidden_states)
hidden_states = self.decoder(hidden_states) + self.bias
return hidden_states
class BertOnlyMLMHead(nn.Module):
def __init__(self, config):
super(BertOnlyMLMHead, self).__init__()
self.predictions = BertLMPredictionHead(config)
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):
super(BertPreTrainingHeads, self).__init__()
self.predictions = BertLMPredictionHead(config)
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(PreTrainedModel):
""" An abstract class to handle weights initialization and
a simple interface for dowloading and loading pretrained models.
"""
config_class = BertConfig
pretrained_model_archive_map = BERT_PRETRAINED_MODEL_ARCHIVE_MAP
load_tf_weights = load_tf_weights_in_bert
base_model_prefix = "bert"
def __init__(self, *inputs, **kwargs):
super(BertPreTrainedModel, self).__init__(*inputs, **kwargs)
def init_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_()
BERT_START_DOCSTRING = r""" The BERT model was proposed in
`BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding`_
by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova. It's a bidirectional transformer
pre-trained using a combination of masked language modeling objective and next sentence prediction
on a large corpus comprising the Toronto Book Corpus and Wikipedia.
This model is a PyTorch `torch.nn.Module`_ sub-class. Use it as a regular PyTorch Module and
refer to the PyTorch documentation for all matter related to general usage and behavior.
.. _`BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding`:
https://arxiv.org/abs/1810.04805
.. _`torch.nn.Module`:
https://pytorch.org/docs/stable/nn.html#module
Parameters:
config (:class:`~pytorch_transformers.BertConfig`): Model configuration class with all the parameters of the model.
"""
BERT_INPUTS_DOCSTRING = r"""
Inputs:
**input_ids**: ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
Indices of input sequence tokens in the vocabulary.
To match pre-training, BERT input sequence should be formatted with [CLS] and [SEP] tokens as follows:
(a) For sequence pairs:
``tokens: [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]``
``token_type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1``
(b) For single sequences:
``tokens: [CLS] the dog is hairy . [SEP]``
``token_type_ids: 0 0 0 0 0 0 0``
Indices can be obtained using :class:`pytorch_transformers.BertTokenizer`.
See :func:`pytorch_transformers.PreTrainedTokenizer.encode` and
:func:`pytorch_transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.
**position_ids**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
Indices of positions of each input sequence tokens in the position embeddings.
Selected in the range ``[0, config.max_position_embeddings - 1[``.
**token_type_ids**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
Segment token indices to indicate first and second portions of the inputs.
Indices are selected in ``[0, 1]``: ``0`` corresponds to a `sentence A` token, ``1``
corresponds to a `sentence B` token
(see `BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding`_ for more details).
**attention_mask**: (`optional`) ``torch.Tensor`` of shape ``(batch_size, sequence_length)``:
Mask to avoid performing attention on padding token indices.
Mask values selected in ``[0, 1]``:
``1`` for tokens that are NOT MASKED, ``0`` for MASKED tokens.
**head_mask**: (`optional`) ``torch.Tensor`` of shape ``(num_heads,)`` or ``(num_layers, num_heads)``:
Mask to nullify selected heads of the self-attention modules.
Mask values selected in ``[0, 1]``:
``1`` indicates the head is **not masked**, ``0`` indicates the head is **masked**.
"""
[docs]@add_start_docstrings("The bare Bert Model transformer outputing raw hidden-states without any specific head on top.",
BERT_START_DOCSTRING, BERT_INPUTS_DOCSTRING)
class BertModel(BertPreTrainedModel):
r"""
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**last_hidden_state**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, hidden_size)``
Sequence of hidden-states at the last layer of the model.
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
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**: (`optional`, returned when ``config.output_attentions=True``)
list 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.
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>> model = BertModel(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids)
>>> last_hidden_states = outputs[0] # The last hidden-state is the first element of the output tuple
"""
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_weights)
def _resize_token_embeddings(self, new_num_tokens):
old_embeddings = self.embeddings.word_embeddings
new_embeddings = self._get_resized_embeddings(old_embeddings, new_num_tokens)
self.embeddings.word_embeddings = new_embeddings
return self.embeddings.word_embeddings
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)
[docs] def forward(self, input_ids, token_type_ids=None, attention_mask=None, position_ids=None, head_mask=None):
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
# 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]
if head_mask is not None:
if head_mask.dim() == 1:
head_mask = head_mask.unsqueeze(0).unsqueeze(0).unsqueeze(-1).unsqueeze(-1)
head_mask = head_mask.expand(self.config.num_hidden_layers, -1, -1, -1, -1)
elif head_mask.dim() == 2:
head_mask = head_mask.unsqueeze(1).unsqueeze(-1).unsqueeze(-1) # We can specify head_mask for each layer
head_mask = head_mask.to(dtype=next(self.parameters()).dtype) # switch to fload if need + fp16 compatibility
else:
head_mask = [None] * self.config.num_hidden_layers
embedding_output = self.embeddings(input_ids, position_ids=position_ids, token_type_ids=token_type_ids)
encoder_outputs = self.encoder(embedding_output,
extended_attention_mask,
head_mask=head_mask)
sequence_output = encoder_outputs[0]
pooled_output = self.pooler(sequence_output)
outputs = (sequence_output, pooled_output,) + encoder_outputs[1:] # add hidden_states and attentions if they are here
return outputs # sequence_output, pooled_output, (hidden_states), (attentions)
[docs]@add_start_docstrings("""Bert Model with two heads on top as done during the pre-training:
a `masked language modeling` head and a `next sentence prediction (classification)` head. """,
BERT_START_DOCSTRING, BERT_INPUTS_DOCSTRING)
class BertForPreTraining(BertPreTrainedModel):
r"""
**masked_lm_labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
Labels for computing the masked language modeling loss.
Indices should be in ``[-1, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring)
Tokens with indices set to ``-1`` are ignored (masked), the loss is only computed for the tokens with labels
in ``[0, ..., config.vocab_size]``
**next_sentence_label**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size,)``:
Labels for computing the next sequence prediction (classification) loss. Input should be a sequence pair (see ``input_ids`` docstring)
Indices should be in ``[0, 1]``.
``0`` indicates sequence B is a continuation of sequence A,
``1`` indicates sequence B is a random sequence.
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**loss**: (`optional`, returned when both ``masked_lm_labels`` and ``next_sentence_label`` are 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_scores**: ``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_scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, 2)``
Prediction scores of the next sequence prediction (classification) head (scores of True/False continuation before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
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**: (`optional`, returned when ``config.output_attentions=True``)
list 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.
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForPreTraining(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids)
>>> prediction_scores, seq_relationship_scores = outputs[:2]
"""
def __init__(self, config):
super(BertForPreTraining, self).__init__(config)
self.bert = BertModel(config)
self.cls = BertPreTrainingHeads(config)
self.apply(self.init_weights)
self.tie_weights()
[docs] def tie_weights(self):
""" Make sure we are sharing the input and output embeddings.
Export to TorchScript can't handle parameter sharing so we are cloning them instead.
"""
self._tie_or_clone_weights(self.cls.predictions.decoder,
self.bert.embeddings.word_embeddings)
[docs] def forward(self, input_ids, token_type_ids=None, attention_mask=None, masked_lm_labels=None,
next_sentence_label=None, position_ids=None, head_mask=None):
outputs = self.bert(input_ids, position_ids=position_ids, token_type_ids=token_type_ids,
attention_mask=attention_mask, head_mask=head_mask)
sequence_output, pooled_output = outputs[:2]
prediction_scores, seq_relationship_score = self.cls(sequence_output, pooled_output)
outputs = (prediction_scores, seq_relationship_score,) + outputs[2:] # add hidden states and attention if they are here
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
outputs = (total_loss,) + outputs
return outputs # (loss), prediction_scores, seq_relationship_score, (hidden_states), (attentions)
[docs]@add_start_docstrings("""Bert Model with a `next sentence prediction (classification)` head on top. """,
BERT_START_DOCSTRING, BERT_INPUTS_DOCSTRING)
class BertForNextSentencePrediction(BertPreTrainedModel):
r"""
**next_sentence_label**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size,)``:
Labels for computing the next sequence prediction (classification) loss. Input should be a sequence pair (see ``input_ids`` docstring)
Indices should be in ``[0, 1]``.
``0`` indicates sequence B is a continuation of sequence A,
``1`` indicates sequence B is a random sequence.
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**loss**: (`optional`, returned when ``next_sentence_label`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
Next sequence prediction (classification) loss.
**seq_relationship_scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, 2)``
Prediction scores of the next sequence prediction (classification) head (scores of True/False continuation before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
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**: (`optional`, returned when ``config.output_attentions=True``)
list 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.
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForNextSentencePrediction(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids)
>>> seq_relationship_scores = outputs[0]
"""
def __init__(self, config):
super(BertForNextSentencePrediction, self).__init__(config)
self.bert = BertModel(config)
self.cls = BertOnlyNSPHead(config)
self.apply(self.init_weights)
[docs] def forward(self, input_ids, token_type_ids=None, attention_mask=None, next_sentence_label=None,
position_ids=None, head_mask=None):
outputs = self.bert(input_ids, position_ids=position_ids, token_type_ids=token_type_ids,
attention_mask=attention_mask, head_mask=head_mask)
pooled_output = outputs[1]
seq_relationship_score = self.cls(pooled_output)
outputs = (seq_relationship_score,) + outputs[2:] # add hidden states and attention if they are here
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))
outputs = (next_sentence_loss,) + outputs
return outputs # (next_sentence_loss), seq_relationship_score, (hidden_states), (attentions)
[docs]@add_start_docstrings("""Bert Model transformer with a sequence classification/regression head on top (a linear layer on top of
the pooled output) e.g. for GLUE tasks. """,
BERT_START_DOCSTRING, BERT_INPUTS_DOCSTRING)
class BertForSequenceClassification(BertPreTrainedModel):
r"""
**labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size,)``:
Labels for computing the sequence classification/regression loss.
Indices should be in ``[0, ..., config.num_labels]``.
If ``config.num_labels == 1`` a regression loss is computed (Mean-Square loss),
If ``config.num_labels > 1`` a classification loss is computed (Cross-Entropy).
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**loss**: (`optional`, returned when ``labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
Classification (or regression if config.num_labels==1) loss.
**logits**: ``torch.FloatTensor`` of shape ``(batch_size, config.num_labels)``
Classification (or regression if config.num_labels==1) scores (before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
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**: (`optional`, returned when ``config.output_attentions=True``)
list 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.
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForSequenceClassification(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> labels = torch.tensor([1]).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids, labels=labels)
>>> loss, logits = outputs[:2]
"""
def __init__(self, config):
super(BertForSequenceClassification, self).__init__(config)
self.num_labels = config.num_labels
self.bert = BertModel(config)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
self.classifier = nn.Linear(config.hidden_size, self.config.num_labels)
self.apply(self.init_weights)
[docs] def forward(self, input_ids, token_type_ids=None, attention_mask=None, labels=None,
position_ids=None, head_mask=None):
outputs = self.bert(input_ids, position_ids=position_ids, token_type_ids=token_type_ids,
attention_mask=attention_mask, head_mask=head_mask)
pooled_output = outputs[1]
pooled_output = self.dropout(pooled_output)
logits = self.classifier(pooled_output)
outputs = (logits,) + outputs[2:] # add hidden states and attention if they are here
if labels is not None:
if self.num_labels == 1:
# We are doing regression
loss_fct = MSELoss()
loss = loss_fct(logits.view(-1), labels.view(-1))
else:
loss_fct = CrossEntropyLoss()
loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
outputs = (loss,) + outputs
return outputs # (loss), logits, (hidden_states), (attentions)
[docs]@add_start_docstrings("""Bert Model with a token classification head on top (a linear layer on top of
the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. """,
BERT_START_DOCSTRING, BERT_INPUTS_DOCSTRING)
class BertForTokenClassification(BertPreTrainedModel):
r"""
**labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
Labels for computing the token classification loss.
Indices should be in ``[0, ..., config.num_labels]``.
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**loss**: (`optional`, returned when ``labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
Classification loss.
**scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, config.num_labels)``
Classification scores (before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
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**: (`optional`, returned when ``config.output_attentions=True``)
list 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.
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForTokenClassification(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> labels = torch.tensor([1] * input_ids.size(1)).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids, labels=labels)
>>> loss, scores = outputs[:2]
"""
def __init__(self, config):
super(BertForTokenClassification, self).__init__(config)
self.num_labels = config.num_labels
self.bert = BertModel(config)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
self.classifier = nn.Linear(config.hidden_size, config.num_labels)
self.apply(self.init_weights)
[docs] def forward(self, input_ids, token_type_ids=None, attention_mask=None, labels=None,
position_ids=None, head_mask=None):
outputs = self.bert(input_ids, position_ids=position_ids, token_type_ids=token_type_ids,
attention_mask=attention_mask, head_mask=head_mask)
sequence_output = outputs[0]
sequence_output = self.dropout(sequence_output)
logits = self.classifier(sequence_output)
outputs = (logits,) + outputs[2:] # add hidden states and attention if they are here
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))
outputs = (loss,) + outputs
return outputs # (loss), scores, (hidden_states), (attentions)
[docs]@add_start_docstrings("""Bert Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of
the hidden-states output to compute `span start logits` and `span end logits`). """,
BERT_START_DOCSTRING, BERT_INPUTS_DOCSTRING)
class BertForQuestionAnswering(BertPreTrainedModel):
r"""
**start_positions**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size,)``:
Labels for position (index) of the start of the labelled span for computing the token classification loss.
Positions are clamped to the length of the sequence (`sequence_length`).
Position outside of the sequence are not taken into account for computing the loss.
**end_positions**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size,)``:
Labels for position (index) of the end of the labelled span for computing the token classification loss.
Positions are clamped to the length of the sequence (`sequence_length`).
Position outside of the sequence are not taken into account for computing the loss.
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**loss**: (`optional`, returned when ``labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.
**start_scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length,)``
Span-start scores (before SoftMax).
**end_scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length,)``
Span-end scores (before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
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**: (`optional`, returned when ``config.output_attentions=True``)
list 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.
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForQuestionAnswering(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> start_positions = torch.tensor([1])
>>> end_positions = torch.tensor([3])
>>> outputs = model(input_ids, start_positions=start_positions, end_positions=end_positions)
>>> loss, start_scores, end_scores = outputs[:2]
"""
def __init__(self, config):
super(BertForQuestionAnswering, self).__init__(config)
self.num_labels = config.num_labels
self.bert = BertModel(config)
self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)
self.apply(self.init_weights)
[docs] def forward(self, input_ids, token_type_ids=None, attention_mask=None, start_positions=None,
end_positions=None, position_ids=None, head_mask=None):
outputs = self.bert(input_ids, position_ids=position_ids, token_type_ids=token_type_ids,
attention_mask=attention_mask, head_mask=head_mask)
sequence_output = outputs[0]
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)
outputs = (start_logits, end_logits,) + outputs[2:]
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
outputs = (total_loss,) + outputs
return outputs # (loss), start_logits, end_logits, (hidden_states), (attentions)