无法达到0.78+效果，使用 "hfl/chinese-bert-wwm" 五折交叉验证

1JasonZhang / Q-A-matching-of-real-estate-industry

MIT License

27 stars 7 forks source link

Open wentaozhu opened 3 years ago

wentaozhu commented 3 years ago

跑一下代码，发现无法达到0.78+ 的性能。请您指点一下。谢谢！

1JasonZhang commented 3 years ago

用这个预训练模型，hfl/chinese-bert-wwm-ext，然后epochs设为5应该能达到0.78左右的效果

1JasonZhang commented 3 years ago

疏忽了，readme中写错了，抱歉了

wentaozhu commented 3 years ago

还是达不到0.78，只能0.775+

LewPeng97 commented 3 years ago

还是达不到0.78，只能0.775+

batch_size设的多少呀，亲