adds early stopping #238
Merged
adds early stopping #238
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
The documentation is not available anymore as the PR was closed or merged. |
lvwerra
reviewed
Mar 22, 2023
|
Good point, I ran a benchmark with early stopping enabled and a threshold of 0.001 here and the logging seems to work ok. |
edbeeching
force-pushed
the
early-stopping
branch
from
f4a1e3c to
368044f
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Adds early stopping to the PPO loop. Fixes #232
I used a value of 0.1 as the threshold as I observed an initial spike of 0.2 before we had instabilities in gpt2-xl earlier this week:
Note that RL4LM use a value of 0.5
My only other concern is gradient accumulation. I think it would it be better to zero the gradients as soon as we see a splike in KL, or to leave them in the as is. I have zeroed them for now, but it would be great to have feedback on this.