feat(length-reward): support global reasoning profile bands - #3802
Closed
ehosseiniasl wants to merge 1 commit into
Closed
feat(length-reward): support global reasoning profile bands#3802ehosseiniasl wants to merge 1 commit into
ehosseiniasl wants to merge 1 commit into
Conversation
- clamp profile-band multipliers at the configured floor after the upper bound - derive reasoning length from generated assistant token IDs without prompt history - support global profile bands with row and agent overrides - cover async configuration propagation and profile-band behavior Signed-off-by: Ehsan Hosseini Asl <ehsan.hosseiniasl@gmail.com>
Contributor
Author
|
Closing for now while runtime validation and upstream dependency work continue. I will reopen or create the review PR only when it is ready. |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
grpo.length_bonus.defaultwhen dataset rows do not provide profile metadata.Motivation
Video async GRPO uses one shared reasoning-length profile across the cached dataset. Requiring every row to carry identical profile metadata is unnecessary and makes Arushi-style video recipes harder to configure safely.
Validation
length_adjustment_configreaches Gym rollouts.Full 2-node synchronous and 16-node asynchronous video validation is pending; this PR remains draft until those runs complete.
Dependency
The Arushi-aligned video recipes are being validated on the separate integration branch
ehsan/adi-video-profile-integrationbecause the current #3604 branch predates the merged video recipes from #3648.