-
Notifications
You must be signed in to change notification settings - Fork 2.1k
GSPO parameters update from v2 #3798
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
Conversation
docs/source/paper_index.md
Outdated
beta=0.0, # GSPO set kl regularization to zero: https://github.com/volcengine/verl/pull/2775#issuecomment-3131807306 | ||
epsilon=3e-4, # GSPO paper (v2), section 5.1 | ||
epsilon_high=4e-4, # GSPO paper (v2), section 5.1 | ||
gradient_accumulation_steps=4, # partition rollout batch into 4 mini-batches. GSPO paper (v2), section 5.1 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
gradient_accumulation_steps=4, # partition rollout batch into 4 mini-batches. GSPO paper (v2), section 5.1 | |
steps_per_generation=4, # partition rollout batch into 4 mini-batches. GSPO paper (v2), section 5.1 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
can you also add somewhere that this value steps_per_generation
should be 4x gradient_accumulation_steps
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
you are right! done. thanks.
Oops, I missed yours, closing #3802 thanks! |
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
What does this PR do?
This PR updates the
training_args
of GSPO indocs/source/paper_index.md
to align with the hyperparameters described in the GSPO v2 paper and documented in trl/trainer/grpo_config.py.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.
@qgallouedec