Do CPM-live support lr change during training #10
Answered
by
jayzzhou-thu
Kunlun-Zhu
asked this question in
Q&A 问答
-
|
We could try restore the model at certain checkpoint during training and change its lr or other hyper-parameters manually to prevent overfitting or unconvergence. |
Beta Was this translation helpful? Give feedback.
Answered by
jayzzhou-thu
May 31, 2022
Replies: 1 comment
-
|
Thanks for your suggestion! Now the model is stored every 500 steps and the hyper-parameters can be tuned if the training is unstable. |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
zibuyu
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Thanks for your suggestion! Now the model is stored every 500 steps and the hyper-parameters can be tuned if the training is unstable.