ZhuiyiTechnology / simbert

a bert for retrieval and generation

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

生成句子报错,维度不一致。

wjx-git opened this issue · comments

使用 simbert.py 训练得到模型,使用 gen_synonyms_test.py 测试时报错:
tensorflow.python.framework.errors_impl.InvalidArgumentError: Input to reshape is a tensor with 13584 values, but the requested shape has 13685

解决方法:
训练模型时将 keep_tokens 注释即可。
bert = build_transformer_model( config_path, checkpoint_path, with_pool='linear', application='unilm', # keep_tokens=keep_tokens, # 只保留keep_tokens中的字,精简原字表 return_keras_model=False, )