From 43300ad90bc3cf4eb4ac0eb06216e951c0285806 Mon Sep 17 00:00:00 2001 From: yh_cc Date: Sun, 7 Jun 2020 20:35:09 +0800 Subject: [PATCH] =?UTF-8?q?=E4=BF=AE=E6=94=B9=E9=83=A8=E5=88=86=E6=B3=A8?= =?UTF-8?q?=E9=87=8A,=20=E5=87=86=E5=A4=87=E5=BC=80=E5=8F=910.5.6?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- fastNLP/__init__.py | 2 +- fastNLP/models/bert.py | 7 +++++++ fastNLP/modules/encoder/bert.py | 14 +++++++++++++- setup.py | 2 +- 4 files changed, 22 insertions(+), 3 deletions(-) diff --git a/fastNLP/__init__.py b/fastNLP/__init__.py index 4436e565..46d83a9d 100644 --- a/fastNLP/__init__.py +++ b/fastNLP/__init__.py @@ -81,7 +81,7 @@ __all__ = [ 'logger', "init_logger_dist", ] -__version__ = '0.5.5' +__version__ = '0.5.6' import sys diff --git a/fastNLP/models/bert.py b/fastNLP/models/bert.py index 976b4638..88dd021c 100644 --- a/fastNLP/models/bert.py +++ b/fastNLP/models/bert.py @@ -70,6 +70,7 @@ class BertForSequenceClassification(BaseModel): def forward(self, words): r""" + 输入为 [[w1, w2, w3, ...], [...]], BERTEmbedding会在开头和结尾额外加入[CLS]与[SEP] :param torch.LongTensor words: [batch_size, seq_len] :return: { :attr:`fastNLP.Const.OUTPUT` : logits}: torch.Tensor [batch_size, num_labels] """ @@ -115,6 +116,8 @@ class BertForSentenceMatching(BaseModel): def forward(self, words): r""" + 输入words的格式为 [sent1] + [SEP] + [sent2]（BertEmbedding会在开头加入[CLS]和在结尾加入[SEP]），输出为batch_size x num_labels + :param torch.LongTensor words: [batch_size, seq_len] :return: { :attr:`fastNLP.Const.OUTPUT` : logits}: torch.Tensor [batch_size, num_labels] """ @@ -247,6 +250,10 @@ class BertForQuestionAnswering(BaseModel): def forward(self, words): r""" + 输入words为question + [SEP] + [paragraph]，BERTEmbedding在之后会额外加入开头的[CLS]和结尾的[SEP]. note: + 如果BERTEmbedding中include_cls_sep=True，则输出的start和end index相对输入words会增加一位；如果为BERTEmbedding中 + include_cls_sep=False, 则输出start和end index的位置与输入words的顺序完全一致 + :param torch.LongTensor words: [batch_size, seq_len] :return: 一个包含num_labels个logit的dict，每一个logit的形状都是[batch_size, seq_len + 2] """ diff --git a/fastNLP/modules/encoder/bert.py b/fastNLP/modules/encoder/bert.py index 4e8d644f..62c18d48 100644 --- a/fastNLP/modules/encoder/bert.py +++ b/fastNLP/modules/encoder/bert.py @@ -473,6 +473,17 @@ class BertModel(nn.Module): module.bias.data.zero_() def forward(self, input_ids, token_type_ids=None, attention_mask=None, output_all_encoded_layers=True): + """ + + :param torch.LongTensor input_ids: bsz x max_len的输入id + :param torch.LongTensor token_type_ids: bsz x max_len，如果不输入认为全为0，一般第一个sep(含)及以前为0, 一个sep之后为1 + :param attention_mask: 需要attend的为1，不需要为0 + :param bool output_all_encoded_layers: 是否输出所有层，默认输出token embedding(包含bpe, position以及type embedding) + 及每一层的hidden states。如果为False，只输出最后一层的结果 + :return: encode_layers: 如果output_all_encoded_layers为True，返回list(共num_layers+1个元素)，每个元素为 + bsz x max_len x hidden_size否则返回bsz x max_len x hidden_size的tensor; + pooled_output: bsz x hidden_size为cls的表示，可以用于句子的分类 + """ if attention_mask is None: attention_mask = torch.ones_like(input_ids) if token_type_ids is None: @@ -504,7 +515,8 @@ class BertModel(nn.Module): pooled_output = sequence_output[:, 0] if not output_all_encoded_layers: encoded_layers = encoded_layers[-1] - encoded_layers.insert(0, embedding_output) + else: + encoded_layers.insert(0, embedding_output) return encoded_layers, pooled_output @classmethod diff --git a/setup.py b/setup.py index f10ec93d..150e8e22 100644 --- a/setup.py +++ b/setup.py @@ -16,7 +16,7 @@ print(pkgs) setup( name='FastNLP', - version='0.5.0', + version='0.5.6', url='https://github.com/fastnlp/fastNLP', description='fastNLP: Deep Learning Toolkit for NLP, developed by Fudan FastNLP Team', long_description=readme,