Skip to content
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

tinybert 在mnli任务不能复现 #229

Open
iMountTai opened this issue Dec 10, 2022 · 4 comments
Open

tinybert 在mnli任务不能复现 #229

iMountTai opened this issue Dec 10, 2022 · 4 comments

Comments

@iMountTai
Copy link

代码可以在qnli、QQP、sst2任务上正确复现,但是在mnli任务上结果很差劲,第一步layers_distill的准确率为43.8,第二步一直是10%左右,不清楚别的同学有没有遇到该问题?

@iMountTai
Copy link
Author

@gowtham1997 期待您的回复,谢谢~

@programmingLearner
Copy link

programmingLearner commented Nov 22, 2023

我也遇到了同样的问题。请问您的问题后来解决了吗?我现在怀疑是三个label的顺序teacher和student对应不上

@programmingLearner
Copy link

programmingLearner commented Nov 22, 2023

我猜可能需要修改原代码第178行,三个 label 的顺序。["contradiction", "entailment", "neutral"] 改为 ['entailment', 'neutral', 'contradiction']。参考 huggingface/transformers#12822
因为如果你的 BERT base Teacher 是按照第二个训练的,但 Student evaluation 却用第一个,就会对齐不上。

@programmingLearner
Copy link

programmingLearner commented Nov 22, 2023

改了那一行,现在 dev set 准确率可以轻松超过83%甚至更高了。应该就是这个问题。不过还需要改一下这个,这个代码现在 mismatched 覆盖了 matched 的输出结果:#115

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants