link-li / clmlf Goto Github PK

Great work !
i would like to know what parameters did you used for training bert on mvsa-multiple ( optimizer, lr, batch size ...)

Hi there!

Great paper, and thank you for the code. I had one question: I notice you don't l2 normalize the embeddings on contrastive loss calculation before einstien sum, which is generally done in all kinds of contrastive loss calculations. Any specific reason for not doing that?

Any help would be highly appreciated! Thank You!

您好，我有个问题比较困惑。为什么在训练时显存使用在逐渐增多？您训练时使用的显存是多少？

作者您好，我是一名学生，目前正在学习您的项目，请问您可以提供在MVSA-multiple的best-model参数吗？

Very interesting paper!
For reimplementation, could you share more details about how you conduct the back-translation operation? Thanks!

Dear author, I would like to obtain the source code of this paper from you, so as to facilitate my follow-up research work. I would be very grateful if you could see it and reply to me. Thank you