Drop invalid multimodal rollout rows before PPO training - #588
Merged
Zhiyuan He (hzy46) merged 1 commit intoSep 14, 2026
Merged
Zhiyuan He (hzy46) merged 1 commit into
Zhiyuan He (hzy46) merged 1 commit into
Conversation
Signed-off-by: ZenAlexa <zimingwang945@gmail.com>
Ziming Wang (ZenAlexa)
marked this pull request as ready for review
September 8, 2026 09:44
Contributor
Author
|
@microsoft-github-policy-service agree |
Contributor
Author
|
Zhiyuan He (@hzy46) could you approve the Test workflow for this PR? The CLA check is now passing. |
Contributor
|
Merged. Thank you for the contribution. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Mark failed image rows with the existing
is_drop_maskso the trainer removes them before PPO minibatch construction. This covers a missing processor, image preprocessing failures, missing image-grid metadata, and mRoPE position errors. Successful image rows and text rows retain their current behavior.Validation
The regression exercises
RolloutAdapter, PIL decoding, VERLDataProto, and the trainer's row selection. It verifies that failed image rows are removed and valid text tokens and rewards survive.The complete
tests/verlsuite passes on CPU with Python 3.12.14, PyTorch 2.13.0, and VERL 0.8.0. Ruff and formatting pass for the changed files.