Skip to content

fix(grpo): preserve prompt groups across rollout paths - #4013

Closed
ananthsub wants to merge 2 commits into
NVIDIA-NeMo:mainfrom
ananthsub:ananthsub/fix-prompt-grouping
Closed

fix(grpo): preserve prompt groups across rollout paths#4013
ananthsub wants to merge 2 commits into
NVIDIA-NeMo:mainfrom
ananthsub:ananthsub/fix-prompt-grouping

Conversation

@ananthsub

@ananthsub ananthsub commented Sep 5, 2026

Copy link
Copy Markdown
Contributor

Summary

GRPO currently identifies a prompt group by tokenizing the prompt returned from the rollout harness. A harness can add rollout-specific content, such as a unique workspace path, so sibling generations become singleton groups and receive zero group-relative advantages.

  • Preserve source-prompt identities through legacy synchronous GRPO and DAPO dynamic sampling.
  • Reconstruct prompt groups from replay-buffer group boundaries in legacy asynchronous GRPO, including groups restored from older checkpoints.
  • Derive single-controller grouping from stable {group_id}_g{index} sample IDs instead of harness-visible prompt tokens.
  • Use the same stable group identity in the synchronous data-plane rollout path.

PPO remains unchanged because its supported GAE and raw-reward estimators do not compute group-relative baselines from prompt IDs.

fyi @cmunley1

Group harness rollouts by their source prompt rather than model-visible prompt tokens, and carry those identities through dynamic sampling, replay, the data plane, and checkpoint restore.

Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
@ananthsub
ananthsub requested review from a team as code owners September 5, 2026 01:22
@copy-pr-bot

copy-pr-bot Bot commented Sep 5, 2026

Copy link
Copy Markdown

Auto-sync is disabled for ready for review pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
@ananthsub

Copy link
Copy Markdown
Contributor Author

closing in favor of #3882

@ananthsub ananthsub closed this Sep 8, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant