将model_name和train_from_scratch作为参数传递给Bi-Encoder对象时,需要以字典格式传递,并且需要使用update方法将该字典与其他参数合并。示例代码如下:
from transformers import BiEncoder, AutoTokenizer
model_name = 'sentence-transformers/bert-base-nli-mean-tokens' tokenizer = AutoTokenizer.from_pretrained(model_name)
bi_encoder = BiEncoder( **{'model_name_or_path': model_name, 'tokenizer': tokenizer}, update={'train_from_scratch': True} )