fix minillm trainer #4743
Open
fix minillm trainer #4743
Conversation
This file contains hidden or 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
Contributor
Author
|
@qgallouedec I have reduced #4731 to only fixing the convergence of the MiniLLM trainer. |
qgallouedec
reviewed
Dec 23, 2025
trl/trainer/grpo_trainer.py
Outdated
| if self.args.gradient_accumulation_steps % generate_every != 0 or ( | ||
| self.use_vllm and self.vllm_importance_sampling_correction | ||
| ): | ||
| ) or self.args.always_track_old_logps: |
Member
There was a problem hiding this comment.
I'm not sure to understand why you need this? When self.args.gradient_accumulation_steps % generate_every == 0, then old_per_token_logps == per_token_logps, why not just using per_token_logps.detach()?
Contributor
Author
There was a problem hiding this comment.
Nice comments! I have removed always_track_old_logps.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
What does this PR do?
Fix the convergence issues of MiniLLMTrainer. It can now smoothly optimize the Reverse KL Divergence between the teacher and the student:
Before submitting
Pull Request section?
to it if that's the case.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.