-
Notifications
You must be signed in to change notification settings - Fork 21
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
Bert-LSR 结果异常 #31
Comments
| epoch 0 | step 50 | ms/b 2511.34 | train loss 5.877 | NA acc: 0.94 | not NA acc: 0.00 | tot acc: 0.91 |
所有的参数都跟更新版的一致,batch size = 10 |
谢谢 @Zeyu-Liang 反馈,我们尽快检查一下然后回复你,谢谢! |
@Zeyu-Liang we have fixed the issue and now you can try to reproduce the results on BERT. Noted that Bz =20 (for BERT, 15+ batch size is suggested for finetuning), hidden_size = 216 (should be divisible by 12 due to the constraint of the hyperparameter of GCN layers). Totally you may need about 50GB GPU memory with the setting for BERT-base. Thanks a lot! |
麻烦问一下 浮现的时候又遇到RuntimeError: 'lengths' argument should be a 1D CPU int64 tensor, but got 1D cuda:0 Long tensor这问题么 |
@SeaYM Try downgrade your PyTorch version to |
@nanguoshun 更新版Bert-LSR 发现跑出的结果发现异常,请问是什么原因呢?感谢您的工作。
The text was updated successfully, but these errors were encountered: