def _get_embbeder(self, weigths_name, special_tokens_fix):
embedders = {'bert': PretrainedBertEmbedder(
pretrained_model=weigths_name,
requires_grad=False,
top_layer_only=True,
special_tokens_fix=special_tokens_fix)
}
text_field_embedder = BasicTextFieldEmbedder(
token_embedders=embedders,
embedder_to_indexer_map={"bert": ["bert", "bert-offsets"]},
allow_unmatched_keys=True)
return text_field_embedder
def get_token_embedders(model_name, tune_bert=False, special_tokens_fix=0):
take_grads = True if tune_bert > 0 else False # bert参数是否也进行训练.
bert_token_emb = PretrainedBertEmbedder(
pretrained_model=model_name,
top_layer_only=True, requires_grad=take_grads, # 只训练最后一层.
special_tokens_fix=special_tokens_fix)# 这里面进行了token扩充.然后也补上了新的初始化权重,也复用了就的权重.
token_embedders = {'bert': bert_token_emb}
embedder_to_indexer_map = {"bert": ["bert", "bert-offsets"]}
text_filed_emd = BasicTextFieldEmbedder(token_embedders=token_embedders,
embedder_to_indexer_map=embedder_to_indexer_map,
allow_unmatched_keys=True)
return text_filed_emd
def get_token_embedders(model_name, tune_bert=False, special_tokens_fix=0):
take_grads = True if tune_bert > 0 else False
bert_token_emb = PretrainedBertEmbedder(
pretrained_model=model_name,
top_layer_only=True, requires_grad=take_grads,
special_tokens_fix=special_tokens_fix)
token_embedders = {'bert': bert_token_emb}
embedder_to_indexer_map = {"bert": ["bert", "bert-offsets"]}
text_filed_emd = BasicTextFieldEmbedder(token_embedders=token_embedders,
embedder_to_indexer_map=embedder_to_indexer_map,
allow_unmatched_keys=True)
return text_filed_emd