def get_bert(BERT_PT_PATH, bert_type, do_lower_case, no_pretraining):
bert_config_file = os.path.join(BERT_PT_PATH, f'bert_config_{bert_type}.json')
vocab_file = os.path.join(BERT_PT_PATH, f'vocab_{bert_type}.txt')
init_checkpoint = os.path.join(BERT_PT_PATH, f'pytorch_model_{bert_type}.bin')
bert_config = BertConfig.from_json_file(bert_config_file)
tokenizer = tokenization.FullTokenizer(
vocab_file=vocab_file, do_lower_case=do_lower_case)
bert_config.print_status()
model_bert = BertModel(bert_config)
# if no_pretraining:
# pass
# else:
# model_bert.load_state_dict(torch.load(init_checkpoint, map_location='cpu'))
# print("Load pre-trained parameters.")
# model_bert=torch.nn.DataParallel(model_bert, device_ids=[0, 4, 5])
model_bert.to(device)
# model_bert.cuda(2)
return model_bert, tokenizer, bert_config
def get_bert(path_bert):
bert_config_file = path_bert + 'bert_config_uncased_L-12_H-768_A-12.json'
vocab_file = path_bert + 'vocab_uncased_L-12_H-768_A-12.txt'
init_checkpoint = path_bert + 'pytorch_model_uncased_L-12_H-768_A-12.bin'
bert_config = BertConfig.from_json_file(bert_config_file)
tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file,
do_lower_case=True)
model_bert = BertModel(bert_config)
model_bert.load_state_dict(torch.load(init_checkpoint, map_location='cpu'))
print("Load pre-trained parameters.")
if gpu:
model_bert.to(device)
return model_bert, tokenizer, bert_config
def get_bert(BERT_PATH):
bert_config_file = BERT_PATH + "/bert_config_uncased_L-12_H-768_A-12.json"
vocab_file = BERT_PATH + "/vocab_uncased_L-12_H-768_A-12.txt"
init_checkpoint = BERT_PATH + "/pytorch_model_uncased_L-12_H-768_A-12.bin"
bert_config = BertConfig.from_json_file(bert_config_file)
tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file,
do_lower_case=True)
bert_config.print_status()
model_bert = BertModel(bert_config)
model_bert.load_state_dict(torch.load(init_checkpoint, map_location='cpu'))
print("Load pre-trained BERT parameters.")
model_bert.to(device)
return model_bert, tokenizer, bert_config
def get_bert(BERT_PT_PATH, bert_type, do_lower_case, no_pretraining):
bert_config_file = os.path.join(BERT_PT_PATH,
f'bert_config_{bert_type}.json')
vocab_file = os.path.join(BERT_PT_PATH, f'vocab_{bert_type}.txt')
init_checkpoint = os.path.join(BERT_PT_PATH,
f'pytorch_model_{bert_type}.bin')
bert_config = BertConfig.from_json_file(bert_config_file)
tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file,
do_lower_case=do_lower_case)
bert_config.print_status()
model_bert = BertModel(bert_config)
model_bert.load_state_dict(torch.load(init_checkpoint, map_location='cpu'))
model_bert.to(device)
return model_bert, tokenizer, bert_config
def get_bert(BERT_PT_PATH):
bert_config_file = os.path.join(BERT_PT_PATH, 'bert_config.json')
vocab_file = os.path.join(BERT_PT_PATH, 'vocab.txt')
init_checkpoint = os.path.join(BERT_PT_PATH, 'pytorch_model.bin')
bert_config = BertConfig.from_json_file(bert_config_file)
tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file)
bert_config.print_status()
model_bert = BertModel(bert_config)
model_bert.load_state_dict(torch.load(init_checkpoint, map_location='cpu'))
print("Load pre-trained parameters.")
model_bert.to(device)
return model_bert, tokenizer, bert_config
def get_bert(BERT_PT_PATH, bert_type, do_lower_case, no_pretraining):
bert_config_file = os.path.join(BERT_PT_PATH, f"bert_config_{bert_type}.json")
vocab_file = os.path.join(BERT_PT_PATH, f"vocab_{bert_type}.txt")
init_checkpoint = os.path.join(BERT_PT_PATH, f"pytorch_model_{bert_type}.bin")
bert_config = BertConfig.from_json_file(bert_config_file)
tokenizer = tokenization.FullTokenizer(
vocab_file=vocab_file, do_lower_case=do_lower_case
)
bert_config.print_status()
model_bert = BertModel(bert_config)
if no_pretraining:
pass
else:
model_bert.load_state_dict(torch.load(init_checkpoint, map_location="cpu"))
print("Load pre-trained parameters.")
model_bert.to(device)
return model_bert, tokenizer, bert_config
def get_bert(BERT_PT_PATH, bert_type, do_lower_case, no_pretraining):
bert_config_file = os.path.join(BERT_PT_PATH,
f'bert_config.json') #bert的配置文件
vocab_file = os.path.join(BERT_PT_PATH, f'vocab.txt') #bert的词汇文件
init_checkpoint = os.path.join(BERT_PT_PATH,
f'pytorch_model.bin') #bert的预训练模型(不一定有)
"""
==BertConfig==该类在bert文件里的modeling里，用bert的配置文件初始化（默认uS）
<from_json_file>方法用于读取bert配置文件的内容
"""
bert_config = BertConfig.from_json_file(bert_config_file)
"""
==tokenization==bert里的文件
==FullTokenizer==类，里面存放词汇信息
"""
tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file,
do_lower_case=do_lower_case)
#毫无作用的输出参数
bert_config.print_status()
"""
==BertModel==该类在bert文件里的modeling里，同样用bert的配置文件初始化，里面有一系列对bert模型的操作（例如添加层，加载参数等...）
"""
model_bert = BertModel(bert_config)
if no_pretraining: #如果不用bert预训练模型，只要它们团队的模型（不需要.bin）
pass
else:
model_bert.load_state_dict(
torch.load(init_checkpoint,
map_location='cpu')) #加载.bin文件，即加载预训练参数
print("Load pre-trained parameters.")
model_bert.to(device)
# bert模型 词汇 bert配置文件
return model_bert, tokenizer, bert_config