New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
你好,checklist跑这套代码F1只有51左右,您是怎么做到近70的,大佬能传授下经验吗 #34
Comments
验证集51吗? |
训练51,测试46左右 |
噢 可以试试 |
我加载的luhua/chinese_pretrain_mrc_macbert_large和luhua/chinese_pretrain_mrc_roberta_wwm_ext_large模型权重,微调dureader2021比赛数据集,超参数和train_bert.sh脚本基本一致,就batch不一样,lr=2e-5、batch=4、epoch=2、max_answer_length=384、--version_2_with_negative,源代码也没修改,在验证集130条数据上,f1的变化[52.3077, 48.3666, 43.7441, 47.1442, 48.6656, 49.2478, 46.6051, 47.4777, 48.425, 47.5991],随着迭代,验证集跑了10次f1,但基本在40+,但是没有找到是什么原因,请问您有没有什么解决的思路提供我参考一下?感觉batch=4也不应该和您的结果有这么大的差距才对啊 |
验证集有两部分:checklist和indomain。checklist 130条,indomain1000条,所以你仅仅测试130条的checklist 得分低很正常。将checklist和indomain合并成一个(1130条)就行。 |
感谢,问题解决了 |
您好,这一部分代码在哪里,没有找到呀 |
你好!我也在尝试使用这个工程进行学习,但在过程中发现很多问题!您那边弄好了吗?是否可以提供比较完整的项目工程以供学习!谢谢! |
@kangyishuai 有试过在CMRC数据集上微调作者发布的模型吗?效果像作者说的能进一步提升吗? |
No description provided.
The text was updated successfully, but these errors were encountered: